Expungement of Criminal Convictions: An Empirical Study

University of Michigan Law School University of Michigan Law School

University of Michigan Law School Scholarship Repository University of Michigan Law School Scholarship Repository

Articles Faculty Scholarship

2020

Expungement of Criminal Convictions: An Empirical Study Expungement of Criminal Convictions: An Empirical Study

J.J. Prescott

University of Michigan Law School

, [email protected]

Sonja B. Starr

University of Michigan Law School

, [email protected]

Available at: https://repository.law.umich.edu/articles/2165

Follow this and additional works at: https://repository.law.umich.edu/articles

Part of the Criminal Law Commons, Labor and Employment Law Commons, and the Law and Society

Commons

Recommended Citation Recommended Citation

Prescott, J.J. "Expungement of Criminal Convictions: An Empirical Study." Sonja B. Starr, co-author.

Harv.

L. Rev

. 133, no. 8 (2020): 2460-555.

This Article is brought to you for free and open access by the Faculty Scholarship at University of Michigan Law

School Scholarship Repository. It has been accepted for inclusion in Articles by an authorized administrator of

University of Michigan Law School Scholarship Repository. For more information, please contact

mlaw[email protected].

2460

VOLUME 133 JUNE 2020 NUMBER 8

2020

by The Harvard Law Review Association

ARTICLE

EXPUNGEMENT OF CRIMINAL CONVICTIONS:

AN EMPIRICAL STUDY

J.J. Prescott & Sonja B. Starr

INTRODUCTION ............................................................................................................................. 2461

I. E

XPUNGEMENT

OLICIES

AND

ELATED

ESEARCH

................................................ 2468

A. The Economic and Legal Aftermath of a Criminal Conviction ................................ 2468

B. Sealing of Criminal Records: The Legal and Policy Landscape .............................. 2472

C. Research Questions and Existing Empirical Research on Expungement ............... 2476

D. Our Empirical Setting and Data ................................................................................. 2480

II. T

PTAKE

: W

EEKS

AND

ECEIVES

XPUNGEMENTS

?...................... 2486

A. Estimating Uptake Rates............................................................................................... 2488

B. Who Receives Expungements? ..................................................................................... 2493

C. What Explains the Uptake Gap? .................................................................................. 2501

1. Lack of Information .................................................................................................. 2502

2. Administrative Hassle and Time Constraints ........................................................ 2502

3. Fees and Costs ............................................................................................................ 2504

4. Distrust and Fear of the Criminal Justice System ............................................... 2504

5. Lack of Access to Counsel ......................................................................................... 2505

6. Insufficient Motivation to Pursue Expungement .................................................. 2506

III. R

ECIDIVISM

UTCOMES

................................................................................................... 2510

A. Recidivism Among Expungement Recipients ............................................................. 2511

B. Interpretation and Implications ................................................................................... 2518

IV. E

MPLOYMENT

UTCOMES

............................................................................................... 2523

A. Employment and Wage Trajectories for Expungement Recipients ........................... 2524

B. Interpretation: Expungement Effect, Motivation, or Mean Regression? ................ 2533

V. O

THER

BJECTIONS

XPUNGEMENT

AWS

.......................................................... 2543

A. General Deterrence......................................................................................................... 2544

B. Statistical Discrimination or Stereotyping................................................................. 2548

CONCLUSION ................................................................................................................................. 2550

2020] EXPUNGEMENT OF CRIMINAL CONVICTIONS 2461

EXPUNGEMENT OF CRIMINAL CONVICTIONS:

AN EMPIRICAL STUDY

J.J. Prescott and Sonja B. Starr

∗

Laws permitting the expungement of criminal convictions are a key component of modern

criminal justice reform efforts and have been the subject of a recent upsurge in legislative

activity. This debate has been almost entirely devoid of evidence about the laws’ effects, in part

because the necessary data (such as sealed records themselves) have been unavailable. We were

able to obtain access to de-identified data that overcome that problem, and we use it to carry

out a comprehensive statewide study of expungement recipients and comparable nonrecipients

in Michigan. We offer three key sets of empirical findings. First, among those legally eligible

for expungement, just

% obtain it within five years of eligibility. Drawing on patterns in

our data as well as interviews with expungement lawyers, we point to reasons for this serious

“uptake gap.” Second, those who do obtain expungement have extremely low subsequent crime

rates, comparing favorably to the general population — a finding that defuses a common public-

safety objection to expungement laws. Third, those who obtain expungement experience a

sharp upturn in their wage and employment trajectories; on average, within one year, wages go

up by over

% versus the pre-expungement trajectory, an effect mostly driven by unemployed

people finding jobs and minimally employed people finding steadier or higher-paying work.

INTRODUCTION

oday, somewhere between 19 and 24 million Americans have felony

conviction records,

and an unknown — but presumably much

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

∗

Henry King Ransom Professor of Law and Henry M. Butzel Professor of Law, respectively,

and Co-Directors of the Empirical Legal Studies Center, University of Michigan Law School. The

authors gratefully acknowledge the generous financial support of the National Science Foundation

(Award No. SES 1023737) and the assistance of several Michigan state agencies in obtaining data:

the State Police; the Workforce Development Agency; the Unemployment Insurance Agency; the

Department of Technology, Management and Budget; and the State Court Administrative Office.

We are also grateful to Miriam Aukerman, Jeff Morenoff, and David Harding for early advising, to

conference and workshop participants at the NBER Summer Institute, Harvard Law School, the

University of Michigan, the Center for American Progress, and the Michigan District Judges

Association for helpful feedback, to colleagues at Michigan and elsewhere for fruitful discussions,

and to all the experts whose interviews and emails are cited herein for sharing their insights. We

are indebted to Margaret Love and David Schlussel for their advice and support on this project,

and we thank the many research assistants who contributed during the project’s ten-year history,

including Patrick Balke, Grady Bridges, Gabriella D’Agostini, David Do, Haley Dutch, Jonathan

Edelman, Nathaniel Givens, Seth Kingery, Rami Krispin, Elena Malik, German Marquez Alcala,

Charlotte McEwen, Chris Pryby, Chelsea Rinnig, Zehra Siddiqui, and especially Simmon Kim for

his extensive efforts.

See The Economic Impacts of the

2020

Census and Business Uses of Federal Data: Hear-

ing Before the J. Econ. Comm., 116th Cong. 12 (2019) (statement of Nicholas Eberstadt, Henry

Wendt Chair in Political Economy, American Enterprise Institute); Sarah K.S. Shannon et al.,

The Growth, Scope, and Spatial Distribution of People with Felony Records in the United

States,

1948

–

2010

, 54 D

EMOGRAPHY

1795, 1806 (2017). When arrests are added, 75 million

Americans — a third of adults — have criminal records. See FBI, N

OVEMBER

2018 N

EXT

2462 HARVARD LAW REVIEW [Vol. 133:2460

larger — number have misdemeanor conviction records.

In recent

years, policymakers, civil rights advocates, and scholars have paid in-

creasing attention to the substantial barriers to employment,

housing,

and social integration

that these records can pose, not to mention the

hundreds of collateral legal consequences that typically flow from crim-

inal convictions, such as restrictions on public-benefits eligibility and

occupational licensing.

Taken together, these hurdles have been de-

scribed as amounting to a “new civil death,”

and on a collective scale,

this phenomenon magnifies racial disparities in employment and other

outcomes due to disparities in the distribution of criminal records. For

all these reasons, a core part of this century’s emergent criminal justice

reform movement has been a search for effective policy levers to miti-

gate the reentry barriers faced by people with criminal records. This

effort is picking up steam in virtually every corner of the country, with two-

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

ENERATION

DENTIFICATION

(NGI) S

YSTEM

ACT

HEET

1 (2018), https://www.fbi.gov/file-

repository/ngi-monthly-fact-sheet/view [https://perma.cc/5KQJ-UBL5].

No studies currently document the total number of Americans with misdemeanor convictions.

However, statistics collected between 2008 and 2016 indicate that misdemeanors routinely make up

over 70% of a state’s criminal caseload. Megan Stevenson & Sandra Mayson, The Scale of Misde-

meanor Justice, 98 B.U.

L. R

. 731, 746 n.81 (2018).

See, e.g., N

STONE

, M

ICHAEL

ATZ

& J

ULIA

ELATT

, U

RBAN

NST

., I

NNOVATIONS

NYC H

EALTH

& H

UMAN

ERVICES

OLICY

: Y

OUNG

’

NITIATIVE

6 (2014),

https://www.urban.org/sites/default/files/publication/32651/413057-Innovations-in-NYC-Health-

and-Human-Services-Policy-Young-s-Men-s-Initiative.PDF [https://perma.cc/V954-Y23W]; Fact

Sheet: President Obama Announces New Actions to Promote Rehabilitation and Reintegration for

the Formerly-Incarcerated, O

BAMA

HITE

OUSE

RCHIVES

(Nov. 2, 2015),

https://obamawhitehouse.archives.gov/the-press-office/2015/11/02/fact-sheet-president-obama-

announces-new-actions-promote-rehabilitation [https://perma.cc/F89L-69K4].

See, e.g., A

NNE

ORRISON

IEHL

, H

AMILTON

ROJECT

, P

UTTING

IME

IMITS

THE

UNITIVENESS

THE

RIMINAL

USTICE

YSTEM

9 (2016), https://www.

hamiltonproject.org/assets/files/reducing_punitiveness_piehl_policymemo.pdf [https://perma.cc/

XQ5P-DWKF];

ARIE

LAIRE

RAN

-L

EUNG

, S

ARGENT

HRIVER

’

OVERTY

, W

HEN

ISCRETION

EANS

ENIAL

: A N

ATIONAL

ERSPECTIVE

RIMINAL

ECORDS

ARRIERS

EDERALLY

UBSIDIZED

OUSING

1 (2015), https://www.

povertylaw.org/wp-content/uploads/2019/09/WDMD-final.pdf [https://perma.cc/GV7K-GSEV].

See, e.g., Amy L. Solomon, In Search of a Job: Criminal Records as Barriers to Employment, NIJ

J., June 2012, at 42, 44, https://www.ncjrs.gov/pdffiles1/nij/238488.pdf [https://perma.cc/2T6Y-TK9F];

Letter from Eric H. Holder, Jr., U.S. Attorney Gen., to State Attorneys Gen. (Apr. 18, 2011) (available

at https://web.archive.org/web/20120227180437/http://www.nationalreentryresourcecenter.org/

documents/0000/1088/Reentry_Council_AG_Letter.pdf).

See M

ARGARET

OLGATE

OVE

., C

OLLATERAL

ONSEQUENCES

RIMINAL

ONVICTIONS

: L

, P

OLICY

AND

RACTICE

§§ 1:12, 2:8, 2:75, 6:16 (2018); see also Gabriel J.

Chin, The New Civil Death: Rethinking Punishment in the Era of Mass Conviction, 160 U.

. L.

. 1789, 1811–14 (2012); Alexandra Natapoff, Misdemeanor Decriminalization, 68 V

AND

. L.

. 1055, 1089–94 (2015). See generally National Inventory of Collateral Consequences of Con-

viction, CSG

UST

. C

, https://niccc.csgjusticecenter.org [https://perma.cc/N79E-5PFN] [herein-

after CSG Inventory].

Chin, supra note 6, at 1790; see also J

AMES

B. J

ACOBS

, T

TERNAL

RIMINAL

ECORD

4 (2015) (observing that criminal records are “for life” and “there is no statute of limitations”).

2020] EXPUNGEMENT OF CRIMINAL CONVICTIONS 2463

thirds of U.S. states adopting one or more such policies in 2018,

and forty-

three states, the federal government, and the District of Columbia passing

an “extraordinary 153 laws” aimed at this problem in 2019.

Perhaps the policy levers with the greatest theoretical potential to

improve reentry outcomes are laws that allow criminal conviction rec-

ords to be wholly expunged or, at least, sealed from public view. We

will refer to such laws collectively as “expungement laws,” and this pro-

cess as “expungement,” although such shorthand elides some differ-

ences.

Expungement offers the possibility of sweeping aside a wide

range of legal and socioeconomic consequences at once; these laws typ-

ically authorize individuals to apply for jobs, housing, schools, and ben-

efits as though their convictions did not exist.

Today, a substantial majority of U.S. states provide some form of

expungement procedure for otherwise-valid adult convictions.

Many

states have recently adopted, or are presently considering, new expunge-

ment laws or expansions to existing ones.

For example, New Mexico,

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

A major report by the Collateral Consequences Resource Center documents the “extraordi-

nary number of laws passed [by thirty-two states, the District of Columbia, and the U.S. Virgin

Islands] in 2018 aimed at reducing barriers to successful reintegration for individuals with a crimi-

nal record,” calling it the “most productive legislative year since a wave of ‘fair chance’ reforms

began in 2013.” Press Release, Collateral Consequences Resource Center, New Report on 2018 Fair

Chance and Expungement Reforms (Updated)

(Jan. 10, 2019), https://ccresourcecenter.

org/2019/01/10/press-release-new-report-on-2018-fair-chance-and-expungement-reforms/#more-18004

[https://perma.cc/4AGU-6CH5]; see also

ARGARET

OVE

& D

AV ID

CHLUSSEL

, C

OLLATERAL

ONSEQUENCES

., R

EDUCING

ARRIERS

EINTEGRATION

: F

AIR

HANCE

AND

XPUNGEMENT

EFORMS

2018, at 2–3 (2019), https://ccresourcecenter.org/wp-content/

uploads/2019/01/Fair-chance-and-expungement-reforms-in-2018-CCRC-Jan-2019.pdf

[https://perma.cc/9QA5-9PMM].

ARGARET

OVE

& D

AVI D

CHLUSSEL

, C

OLLATERAL

ONSEQUENCES

ATHWAYS

EINTEGRATION

: C

RIMINAL

ECORD

EFORMS

2019, at 1–2 (2020),

http://ccresourcecenter.org/wp-content/uploads/2020/02/Pathways-to-Reintegration_Criminal-

Record-Reforms-in-2019.pdf [https://perma.cc/35MG-734Y].

The details vary by jurisdiction, and many advocates use various terms such as “expungement,”

“sealing,” and “set-aside” interchangeably. Typically, a true “expungement” legally eliminates the record

from the state’s perspective. “Sealing” maintains the record for some limited state purposes (for exam-

ple, law enforcement investigations of future crimes) but insulates it from public view.

Two useful online resources contain periodically updated collections of these laws; at our last

check, we found slightly different information at the two sites, but both include close to forty states

with some form of expungement law for valid, nonpardoned, and nonvacated adult criminal con-

victions. See

-State Comparison: Expungement, Sealing & Other Record Relief, C

OLLATERAL

ONSEQUENCES

ESOURCE

., https://ccresourcecenter.org/state-restoration-profiles/50-state-

comparisonjudicial-expungement-sealing-and-set-aside [https://perma.cc/JA4B-2HUD] [hereinafter

CCRC State Survey]; States,

LEAN

LATE

LEARINGHOUSE

, https://cleanslateclearinghouse.

org/compare-states [https://perma.cc/94VU-UPRD].

For example, in 2018 alone, twenty states passed twenty-nine bills expanding expungement

eligibility. L

OVE

& S

CHLUSSEL

, supra note 8, at 2. In 2019, there was still further activity, with “31

states and D.C. enact[ing] no fewer than 67 bills creating, expanding, or streamlining record relief,” and

twenty states “authoriz[ing] diversion programs that produce non-conviction dispositions newly eligible

for record-clearing under existing law.” L

OVE

& S

CHLUSSEL

, supra note 9, at 10. Fifteen of these

states took “incremental” steps, and five made more dramatic eligibility-enhancing changes. Id. at

2464 HARVARD LAW REVIEW [Vol. 133:2460

for the first time, passed a law in 2019 to make a petition-based expunge-

ment process available,

and it is not alone.

In 2018, Pennsylvania be-

came the first state to adopt a sweeping program of automatic expunge-

ment of adult criminal convictions — specifically, nonviolent

misdemeanors after ten crime-free years.

In 2019, Utah, New Jersey,

and California also enacted automatic expungement laws,

which are

more ambitious in some ways. For example, Utah has only a five-year

waiting period in some instances,

and California’s recent legislation

has even shorter waiting periods (none in some cases) and encompasses

minor felonies as well as misdemeanors, although the law only operates

prospectively.

The New Jersey law extends automatic expungement

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

10–17. We note that many of the new or amended laws have features that roughly parallel the

Michigan law we study, and therefore, our work can be useful in predicting the likely consequences

of these reforms.

Danny W. Jarrett et al., New Mexico Adopts Ban-the-Box, Expungement Laws,

ACKSON

EWIS

(Apr. 22, 2019), https://www.jacksonlewis.com/publication/new-mexico-adopts-

ban-box-expungement-laws [https://perma.cc/96SE-ABY9]. The expungement law went into effect

on January 1, 2020. Id.

See L

OVE

& S

CHLUSSEL

, supra note 9, at 14 (“New Mexico, North Dakota, Delaware, West

Virginia, and Colorado made particularly dramatic changes to their petition-based systems to ex-

tend eligibility for relief to a range of non-conviction and conviction records. None of the first four

states had previously authorized relief for felony-level offenses, and Colorado had authorized seal-

ing only for drug convictions. The comprehensive schemes enacted by North Dakota and New

Mexico are noteworthy as the first laws in those states to authorize sealing of adult criminal records.

Both states extend relief to most felonies, but they also require the applicant to pay a filing fee and

make the case for relief at a court hearing. (North Dakota courts may dispense with the hearing if

the prosecutor agrees.)”).

Faith Karimi, Pennsylvania Is Sealing

Million Criminal Records as Part of Clean Slate

Law, CNN (June 28, 2019, 4:46 AM), https://www.cnn.com/2019/06/28/us/pennsylvania-clean-slate-

law-trnd/index.html [https://perma.cc/5XVN-RFFE]. Felonies, as well as violent and certain other

serious misdemeanors, are ineligible for automatic expungement. 18 P

. C

ONS

. S

TAT

. § 9122.3(a),

(b) (2019). But such offenses are still potentially eligible for expungement by petition. Id. § 9122.3(c).

Jessica Miller, Utah Lawmakers Pass the “Clean Slate” Bill to Automatically Clear the Criminal

Records of People Who Earn an Expungement, S

ALT

AKE

RIB

., (Mar. 16, 2019)

https://www.sltrib.com/news/2019/03/14/utah-lawmakers-pass-clean [https://perma.cc/HVV7-VFE6];

Press Release, State of N.J., Governor Murphy Signs Major Criminal Justice Reform Legislation (Dec.

18, 2019), https://www.nj.gov/governor/news/news/562019/approved/20191218a.shtml [https://

perma.cc/V5YE-RK8F]; CCRC Staff, California Becomes Third State to Adopt “Clean Slate” Record

Relief, C

OLLATERAL

ONSEQUENCES

ESOURCE

. (Oct. 10, 2019), http://ccresourcecenter.

org/2019/10/10/california-becomes-third-state-to-adopt-clean-slate-record-relief

[https://perma.cc/8U5C-XSRJ].

TAH

ODE

. § 77-40-102(5)(a)(iii)(A) (LexisNexis 2020). The waiting periods are six

years for more serious misdemeanors and seven years for certain drug crimes. Id. §§ 77-40-

102(5)(a)(iii)(B)–(C).

. P

ENAL

ODE

§ 1203.425(a)(2)(E) (West 2020) (stating that automatic relief is available

only for convictions “on or after January 1, 2021,” and only for misdemeanors, infractions, and

offenses resulting in a sentence of probation). For all crimes resulting in probation sentences (even

felonies), relief is automatic after completing probation, with no further waiting period. Id.

§ 1203.425(a)(2)(E)(i). Automatic expungement is also available for all misdemeanors, even those

resulting in jail time; in that case the waiting period is one year. Id. § 1203.425(a)(2)(E)(ii).

2020] EXPUNGEMENT OF CRIMINAL CONVICTIONS 2465

to some felonies as well as misdemeanors, without limits as to the num-

ber of convictions expunged, after a ten-year period with no subsequent

convictions.

Other states may soon follow suit.

More typically, ex-

pungement laws require individuals to go through a judicial process to

apply for relief, usually giving judges the discretion to deny the petition.

Many states have stringent eligibility requirements as to crime type or

severity or the number of convictions on the individual’s record, and

many have waiting periods.

Despite the considerable legislative ferment and the excitement that

surrounds these “Clean Slate” initiatives in the civil rights and criminal

justice reform worlds, what has been missing from the debate is hard

evidence about the effects and true potential of conviction expungement

laws. Empirical studies in this area have been difficult to carry out.

Expunged criminal records are, obviously, not typically available to

study — and other relevant outcome data, such as wage information or

employment status, are also protected by privacy laws. While there are

many persuasive theoretical reasons to believe that expungement laws

will have large and important effects across many domains,

the dearth

of empirical evidence is a significant hindrance to reform and experi-

mentation. It leaves policymakers almost entirely in the dark.

In this Article, we present the results of an unprecedented statewide

study that overcomes existing limitations on research on expungement and

seeks to fill various policy-relevant gaps in our empirical knowledge. Pur-

suant to a data-sharing agreement with the State of Michigan, we were

able to obtain complete, de-identified criminal records from the

Michigan State Police (MSP) on all individuals who had received convic-

tion expungements (known as “set-asides” in Michigan) as of March 2014,

as well as full criminal history records for much larger comparison groups

of individuals with convictions that were not expunged. In addition, the

state matched these criminal histories with detailed wage and employment

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

Criminal History Record Information — Expungement, 2019 N.J. Sess. Laws Serv. 11–12

(West) (to be codified at N.J.

TAT

. A

. § 2C:52–5.3). In the immediate term, this relief is petition

based, but a task force has been directed to develop an automated process. Id. at 12–14 (to be

codified at N.J.

TAT

. A

. § 2C:52–5.4).

See, e.g., Expungement Bill Package Passes House with Overwhelming Bi-partisan Support, M

ICH

OUSE

EMOCRATS

(Nov. 22, 2019), https://housedems.com/article/expungement-bill-

package-passes-house-overwhelming-bi-partisan-support [https://perma.cc/2PWU-2ZVL]; Kelan Lyons,

Winfield to Swap Out Lamont’s Clean Slate Bill with a Broader Measure, CT

IRROR

(Feb. 26, 2020),

https://ctmirror.org/2020/02/26/winfield-to-swap-out-lamonts-clean-slate-bill-for-a-broader-measure

[https://perma.cc/2AX2-BMWZ].

See sources cited supra note 11; see also infra section I.B, pp. 2472–76 (describing this legal

landscape in more detail).

See, e.g., Brian M. Murray, Unstitching Scarlet Letters?: Prosecutorial Discretion and Ex-

pungement, 86 F

ORDHAM

L. R

. 2821, 2824–26 (2018); Amy Myrick, Facing Your Criminal Rec-

ord: Expungement and the Collateral Problem of Wrongfully Represented Self, 47

& S

’

. 73, 74 (2013).

2466 HARVARD LAW REVIEW [Vol. 133:2460

data for the same individuals from the state’s unemployment insurance

program, sharing these data with us as well.

Michigan is an ideal setting in which to study expungement laws: it

is a large, diverse state with criminal justice challenges typical of the

United States today. Moreover, the version of Michigan’s expungement

law we study has many of the common features of a petition-based

record-clearing law, and it applies to a diverse set of records (e.g., vio-

lent/nonviolent, misdemeanor/felony), carries a fairly short waiting pe-

riod, has relatively straightforward eligibility rules, and went untouched

for more than two decades, allowing us to study results over time.

We use our unique data to investigate a number of interrelated em-

pirical questions, which can be grouped into three main areas of inquiry.

First, we examine the critical question of the “uptake rate”: the rate at

which those who are legally eligible for expungements actually receive

them.

We find that Michigan’s expungement uptake rate is discour-

agingly low; our best estimate is that only 6.5% of all eligible individuals

receive an expungement within five years of the date at which they first

qualify for one.

To better understand the uptake process, we examine

the characteristics of expungement recipients and their offenses and as-

sess whether some characteristics are predictive of uptake by eligible

individuals. We then use these data, plus interviews with Michigan ex-

pungement lawyers and advocates for people with criminal records, to

inform a discussion of why people might not apply for expungements

despite their potential benefits.

Second, we investigate expungement recipients’ subsequent criminal

offending.

We find very low rates of recidivism: just 7.1% of all expunge-

ment recipients are rearrested within five years of receiving their expunge-

ment (and only 2.6% are rearrested for violent offenses), while reconvic-

tion rates are even lower: 4.2% for any crime and only 0.6% for a violent

crime. Indeed, expungement recipients’ recidivism rates compare favor-

ably with those of the Michigan population as a whole.

We do not

claim that these low rates are necessarily because of expungements, al-

though there are several channels by which expungement receipt could

potentially contribute to lower recidivism risk. Another likely explana-

tion is that people who have limited criminal records and have gone at

least five years since their last conviction are simply very low risk to

begin with. This finding is consistent with the broader empirical liter-

ature on patterns of desistance from crime. Whatever the cause, the low

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

See infra Part II, pp. 2486–2510.

Although our data do not identify unsuccessful applicants, it is clear from follow-up inquiries with

the Michigan State Police that the low uptake rate can be primarily attributed to individuals’ failure to

apply, rather than to denials of applications by judges. See infra section II.C.6, pp. 2506–210.

See infra Part III, pp. 2510–23.

See infra p. 2514.

2020] EXPUNGEMENT OF CRIMINAL CONVICTIONS 2467

recidivism rates we observe help to defuse an otherwise potentially con-

vincing policy argument against expungement laws: that the public (in-

cluding employers and landlords) has a safety interest in knowing about

the prior records of those with whom they interact.

Third, we examine the employment consequences of criminal record

expungement.

After accounting for an individual’s prior employment

and wage history, as well as broader changes in the economy, we find

that expungement recipients experience considerable gains shortly after

receipt. Within one year, on average, an individual’s odds of being em-

ployed (earning any wages at all) increase by a factor of 1.13; their odds

of earning at least $100/week (a slightly more demanding employment

measure) increase by a factor of 1.23; and their reported quarterly wages

increase by a factor of 1.23 (and are sustained in subsequent years).

These results suggest that those with expunged records gain access to

more and better-paying jobs. To be sure, one has to be cautious about

drawing causal inferences here; it is very possible that some of the gains

come about because people choose to seek expungement at a time that

they are especially motivated to seek improvements in their economic

situation. Nonetheless, our data and other supporting evidence give us

some confidence that at least a large fraction of the improvement that

we observe stems from the clean record itself.

We provide background on expungement laws, existing relevant em-

pirical research, and our research setting (including our data and their

limitations) in Part I. We then present the three major components of

our work — our analyses of expungement uptake, the recidivism rates

among expungement recipients, and the relationship between expunge-

ment and subsequent employment outcomes — in Parts II, III, and IV,

respectively. In Part V, we respond to two potential objections to ex-

pungement, which might, if accurate, influence the policy takeaways of

our empirical results. In the Conclusion, we address policy implications

and consider future research possibilities.

Our findings tell a good news/bad news story: when expungement is

not automatic (and takes time, effort, and even money to apply), only a

very small share of the people eligible for relief actually apply for and

receive an expungement — but those who do experience clear improve-

ments in economic outcomes and pose little public-safety risk. Taken

together, these conclusions have a clear policy upshot: they support the

expansion of expungement availability, an easing of the procedural hur-

dles associated with seeking expungement, and, in particular, the emerg-

ing movement to make expungement occur automatically after a period

of time, rather than after an application process.

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

See infra Part IV, pp. 2523–43.

2468 HARVARD LAW REVIEW [Vol. 133:2460

XPUNGEMENT

OLICIES

AND

ELATED

ESEARCH

Several large bodies of scholarly research, as well as active policy

debates and commentary, inform and motivate our study. In section I.A,

we describe the many hurdles people with public criminal records face;

paring back these hurdles is the core policy motivation for expungement

laws and efforts to expand them. In section I.B, we add some further

detail to the Introduction’s description of the current legal and policy

landscape surrounding expungement. In section I.C, we discuss the very

limited empirical research that exists on expungement and identify the

key unanswered empirical questions that we seek to address. In section

I.D, we turn to our specific empirical setting, describing Michigan’s ex-

pungement law and the data that we use in our study.

A. The Economic and Legal Aftermath of a Criminal Conviction

The consequences of criminal convictions do not end when people

convicted of crimes complete their formal sentences. For many individ-

uals, punishments such as probation, fines, and even incarceration may

be dwarfed in importance by what comes next: exclusion from employ-

ment, obstacles to social integration, and a vast array of collateral legal

consequences that often last a lifetime.

A growing body of academic

research documents the scope, ubiquity, and size of these hurdles.

First, people with criminal records face serious employment barri-

ers — indeed, these barriers may exceed those facing any other disad-

vantaged group.

While many aspects of these individuals’ back-

grounds, as well as the interruptions to work and education experienced

by those who are incarcerated, may put them at greater risk of unem-

ployment than the general population,

the criminal record itself also

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

See Sarah B. Berson, Beyond the Sentence — Understanding Collateral Consequences, NIJ

J., Sept. 2013, at 25, 25 https://www.ncjrs.gov/pdffiles1/nij/241927.pdf [https://perma.cc/C2FM-

8GF3]; CSG Inventory, supra note 6; see also L

OVE

., supra note 6, §§ 1:11–1:12, 2:44.

See, e.g., P

HARITABLE

., C

OLLATERAL

OSTS

: I

NCARCERATION

’

FFECT

CONOMIC

OBILITY

(2010), https://www.pewtrusts.org/~/media/legacy/uploadedfiles/pcs_

assets/2010/collateralcosts1pdf.pdf [https://perma.cc/B7XB-MCEN]; Anastasia Christman &

Michelle Natividad Rodriguez, Research Supports Fair Chance Policies, N

’

. L. P

ROJECT

(Aug. 1, 2016), https://www.nelp.org/publication/research-supports-fair-chance-policies [https://

perma.cc/V7GG-YQNT].

See Harry J. Holzer et al., Perceived Criminality, Criminal Background Checks, and the Ra-

cial Hiring Practices of Employers, 49 J.L.

& E

CON

. 451, 471 (2006); Devah Pager, The Mark of a

Criminal Record, 108 A

. J. S

. 937, 960 (2003).

EBBIE

UKAMAL

, U.S. D

’

ABOR

, F

ROM

ARD

IME

ULL

IME

TRATEGIES

ELP

OVE

-O

FFENDERS

FROM

ELFARE

ORK

Part

III.B (2001),

https://hirenetwork.org/sites/default/files/From%20Hard%20Time%20to%20Full%20Time.pdf

[https://perma.cc/2XM8-H8KT]; A

L. S

OLOMON

., U

RBAN

NST

., F

ROM

RISON

ORK

: T

MPLOYMENT

IMENSIONS

RISONER

EENTRY

8–11 (2004),

https://www.urban.org/sites/default/files/publication/58126/411097-From-Prison-to-Work.PDF

[https://perma.cc/6W2R-CFQS]; J

EREMY

RAVIS

., U

RBAN

NST

., F

ROM

RISON

OME

2020] EXPUNGEMENT OF CRIMINAL CONVICTIONS 2469

seems to directly harm employment prospects.

Many employers report

that they take steps to avoid hiring individuals with records.

Their

motivations for doing so vary. A report prepared for the Department of

Labor has found that many employers are driven by “bias and stigma.”

Some believe that individuals with records cannot be trusted.

Others

fear a negligent-hiring lawsuit if an employee hired with a criminal rec-

ord commits a crime while on the job.

Experimental results confirm employers’ reluctance to hire individ-

uals with records. Devah Pager had matched pairs of testers, differing

only in criminal history, apply for a range of employment positions.

She found that applicants without records received more than twice as

many callbacks.

More recently, Amanda Agan and Sonja Starr sent

around 15,000 fictitious job applications paired by race to entry-level

jobs mainly in the restaurant and retail industries.

They found that

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

IMENSIONS

AND

ONSEQUENCES

RISONER

EENTRY

31–33 (2001), http://re-

search.urban.org/UploadedPDF/from_prison_to_home.pdf [https://perma.cc/K2Z6-RYGR].

HERRIE

UCKNOR

& A

LAN

ARBER

, C

FOR

CON

. & P

OLICY

ESEARCH

, T

RICE

: E

CONOMIC

OSTS

ARRIERS

MPLOYMENT

FOR

ORMER

RISONERS

AND

EOPLE

ONVICTED

ELONIES

10–13 (2016), http://cepr.net/images/

stories/reports/employment-prisoners-felonies-2016-06.pdf?v=5 [https://perma.cc/M58K-E5FE];

COTT

H. D

ECKER

., C

RIMINAL

TIGMA

, R

ACE

, G

ENDER

AND

MPLOYMENT

: A

XPANDED

SSESSMENT

THE

ONSEQUENCES

MPRISONMENT

FOR

MPLOYMENT

56 (2014), https://www.ncjrs.gov/pdffiles1/nij/grants/244756.pdf [https://perma.cc/M82H-3DHQ].

In a small study of expungement seekers in Chicago, “participants reported attempts to use face-to-

face contact with potential employers and landlords to convince them that, despite their criminal

records, they were trustworthy and dependable. However, participants with both extensive and

minor criminal record histories indicated that these attempts were ineffective.” Simone Ispa-Landa

& Charles E. Loeffler, Indefinite Punishment and the Criminal Record: Stigma Reports Among

Expungement-Seekers in Illinois, 54 C

RIMINOLOGY

387, 401 (2016) (internal citations omitted).

ICHELLE

ATIVIDAD

ODRIGUEZ

& M

AURICE

MSELLEM

, N

’

ROJECT

, 65 M

ILLION

“N

EED

PPLY

”: T

ASE

FOR

EFORMING

RIMINAL

ACKGROUND

HECKS

FOR

MPLOYMENT

1 (2011), https://www.nelp.org/wp-content/

uploads/2015/03/65_Million_Need_Not_Apply.pdf [https://perma.cc/8YS6-F2SR]; Holzer et al.,

supra note 30, at 455; Sarah Esther Lageson et al., Legal Ambiguity in Managerial Assessments of

Criminal Records, 40 L

& S

. I

NQUIRY

175, 191 (2015).

UKAMAL

, supra note 31, at III.B.

See Harry J. Holzer, Collateral Costs: Effects of Incarceration on Employment and Earnings

Among Young Workers, in D

RISONS

AKE

AFER

? T

ENEFITS

AND

OSTS

THE

RISON

OOM

239, 248 (Steven Raphael & Michael A. Stoll eds., 2009).

OVE

., supra note 6, §§ 6:18–6:29; see also M

UKAMAL

, supra note 31, at IV.A.2; S

’

FOR

UMAN

. M

GMT

. & C

HARLES

OCH

NST

., W

ORKERS

WITH

RIMINAL

ECORDS

(2018), https://www.shrm.org/hr-today/trends-and-forecasting/research-and-surveys/pages/second-

chances.aspx [https://perma.cc/JDQ8-T9ME]; Holzer, supra note 35, at 243.

Pager, supra note 30, at 946–48.

See id. at 958 fig.6.

Amanda Agan & Sonja Starr, Ban the Box, Criminal Records, and Racial Discrimination: A

Field Experiment, 133 Q.J.

CON

. 191, 192, 198 (2018) [hereinafter Agan & Starr, Ban the Box];

Amanda Agan & Sonja Starr, The Effect of Criminal Records on Access to Employment, 107 A

CON

. R

. 560, 560–61 (2017).

2470 HARVARD LAW REVIEW [Vol. 133:2460

when employers asked about criminal history, the applicants without rec-

ords received 63% more callbacks, even though the records in question

were relatively minor.

Many employers, influenced by the national “Ban

the Box” movement and resulting law changes, have removed questions

about criminal history from initial job applications — but these employers

typically still conduct background checks before finalizing a hire.

A sur-

vey in 2012 reported that 87% of randomly sampled employers performed

criminal background checks on at least some employees; 69% performed

background checks on all employees.

This represented an expansion

from an earlier era, driven by the availability of easier and less costly

internet-based searches.

Almost all states place court records on the

internet,

and private companies, such as Westlaw and LexisNexis, also

market criminal history databases.

In addition to these employment consequences, criminal convictions

bring with them a wide range of other “collateral” legal consequences —

that is, consequences that are not part of the sentence but a function of

a wide array of civil laws. Licensing restrictions categorically exclude

previously convicted individuals from hundreds of professions.

These

individuals are often prohibited from receiving various social services,

including welfare and health benefits, public housing, and food

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

Agan & Starr, Ban the Box, supra note 39, at 195, 200.

See id. at 193 n.3. The slogan refers to the fact that such questions are often set apart in a

box on the application form.

See S

’

FOR

UMAN

. M

GMT

., SHRM S

URVEY

INDINGS

: B

ACKGROUND

HECKING

— T

RIMINAL

ACKGROUND

HECKS

IRING

ECISIONS

(2012),

https://www.shrm.org/hr-today/trends-and-forecasting/research-and-surveys/Pages/

criminalbackgroundcheck.aspx [https://perma.cc/3EQU-KBQL].

See J

ACOBS

, supra note 7, at 5; L

OVE

., supra note 6, § 5:2; Jeffrey Selbin et al.,

Unmarked? Criminal Record Clearing and Employment Outcomes, 108 J.

RIM

. L. &

RIMINOLOGY

1, 9 (2018).

See Privacy/Public Access to Court Records: State Links, N

’

FOR

. C

https://www.ncsc.org/topics/access-and-fairness/privacy-public-access-to-court-records/state-links

[https://perma.cc/V2HP-3J6M].

See Ben Geiger, Comment, The Case for Treating Ex-offenders as a Suspect Class, 94 C

ALIF

L. R

. 1191, 1198–99 (2006).

See C

HIDI

MEZ

& R

EBECCA

IRIUS

, N

’

ONFERENCE

TATE

EGISLATURES

ARRIERS

ORK

: I

MPROVING

MPLOYMENT

ICENSED

CCUPATIONS

FOR

NDIVIDUALS

WITH

RIMINAL

ECORDS

3 (2018), http://www.ncsl.org/Portals/1/Documents/

Labor/Licensing/criminalRecords_v06_web.pdf [https://perma.cc/6GKS-GJUF]; Alec C. Ewald,

Collateral Consequences and the Perils of Categorical Ambiguity, in L

UNISHMENT

EGULATION

77, 87–88 (Austin Sarat et al. eds., 2011); Policy Changes Needed to Unlock

Employment and Entrepreneurial Opportunity for

100

Million Americans with Criminal Records,

Kauffman Research Shows, B

USINESS

IRE

(Nov. 29, 2016, 8:00 AM), https://www.businesswire.com/

news/home/20161129005528/en/Policy-Needed-Unlock-Employment-Entrepreneurial-Opportunity-

100 [https://perma.cc/UA6K-VEEG]; Shoshana Weissmann & Nila Bala, Opinion, Criminal Justice,

Occupational Licensing Reforms Can Go Hand in Hand, T

ILL

(Apr. 15, 2018, 4:30 PM),

https://thehill.com/opinion/civil-rights/383262-criminal-justice-occupational-licensing-reforms-can-

go-hand-in-hand [https://perma.cc/R7SU-ZMWZ].

2020] EXPUNGEMENT OF CRIMINAL CONVICTIONS 2471

stamps.

Whole families can be evicted from public housing on the

basis of one member’s convictions.

Exclusion from housing renders

individuals with records homeless at high rates — for instance, in 1997,

the California Department of Corrections estimated that 10% of its pa-

rolees were homeless.

The specifics of these exclusions vary from state

to state, although some are encouraged or required by federal law.

They also vary based on crime type.

Because so many Americans have conviction records, these conse-

quences have a large aggregate impact. Repercussions include spillover

effects on family members never convicted of any crime; the Center for

American Progress estimates that almost half of U.S. children have a

parent with some form of criminal record (including arrests).

In addi-

tion, because criminal records are not equally distributed across the popu-

lation,

the effects of these collateral consequences are disproportionately

concentrated by race, gender, and poverty status, especially affecting black

men.

This concentration of criminal records may therefore be a signifi-

cant contributor to racial disparities in employment and other socioeco-

nomic outcomes.

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

See S

TANDARDS

FOR

RIMINAL

USTICE

ch. 19, introductory cmt. (A

. B

’

d ed.

2004

)

; P

AUL

AMUELS

& D

EBBIE

UKAMAL

, L

EGAL

CTION

., A

FTER

RISON

OADBLOCKS

EENTRY

12–13, 16 (2004), http://www.november.org/resources/LACReportCard.

pdf [https://perma.cc/FGV5-39ZZ]; Jeremy Travis, Invisible Punishment: An Instrument of Social

Exclusion, in I

NVISIBLE

UNISHMENT

: T

OLLATERAL

ONSEQUENCES

ASS

MPRISONMENT

15, 23–25 (Marc Mauer & Meda Chesney-Lind eds., 2002); Geiger, supra note 45,

at 1204–06.

Jenny Roberts, Why Misdemeanors Matter: Defining Effective Advocacy in the Lower Crim-

inal Courts, 45 U.C.

AVI S

L. R

. 277, 299 (2011).

. S

TATE

’

ORR

., P

REVENTING

AROLEE

AILURE

ROGRAM

: A

VA LUAT IO N

2 (1997), https://www.ncjrs.gov/pdffiles1/Digitization/180542NCJRS.pdf

[https://perma.cc/JX2E-V3BZ].

See M

UKAMAL

, supra note 31, at III.A.1; Gwen Rubinstein & Debbie Mukamal, Welfare and

Housing — Denial of Benefits to Drug Offenders, in I

NVISIBLE

UNISHMENT

, supra note 47, at

37, 41–42; Geiger, supra note 45, at 1204.

See SAMUELS & MUKAMAL, supra note 47, at 16.

Rebecca Vallas et al., Removing Barriers to Opportunity for Parents with Criminal Records

and Their Children, C

FOR

. P

ROGRESS

(Dec. 10, 2015, 12:01 AM), https://www.

americanprogress.org/issues/poverty/reports/2015/12/10/126902/removing-barriers-to-opportunity-

for-parents-with-criminal-records-and-their-children [https://perma.cc/KU9B-YUH7].

See E. A

ARSON

, U.S. D

’

USTICE

, P

RISONERS

2016, at 10 (2018),

https://www.bjs.gov/content/pub/pdf/p16.pdf [https://perma.cc/68J8-QFQT] (showing imprison-

ment rates in 2016 at 1609, 857, and 274 per 100,000 U.S. residents for black, Hispanic, and white

adults, respectively); Robert Brame et al., Demographic Patterns of Cumulative Arrest Prevalence

by Ages

and

, 60 C

RIME

& D

ELINQ

. 471, 476 (2014) (finding about half of black men have

been arrested by age twenty-three compared to 38% of white men).

See P

HARITABLE

., supra note 29, at 27; Andi Mullin, Banning the Box in

Minnesota — and Across the United States!, C

OMMUNITY

ATALYST

(Dec. 2, 2013),

https://www.communitycatalyst.org/blog/banning-the-box-in-minnesota-and-across-the-united-

states [https://perma.cc/69C6-TBU2].

2472 HARVARD LAW REVIEW [Vol. 133:2460

B. Sealing of Criminal Records: The Legal and Policy Landscape

More than two-thirds of states have adopted statutes that permit

adult criminal convictions to be sealed, set aside, or expunged.

These

laws generally require state agencies to ignore the affected individual’s

criminal record for most purposes other than law enforcement, thereby

lifting any statutory barriers to public employment, licensing, and ben-

efits. The laws also give an individual expungement recipient the legal

right to respond “no” when an employer or landlord asks if the applicant

has a criminal record.

In most states (including Michigan), sealed con-

victions remain available for law enforcement purposes, and for sen-

tencing in the event of a subsequent crime.

The specific eligibility requirements for criminal record expunge-

ment vary widely across jurisdictions. For example, some states’ laws

exclude certain classes of crimes, such as violent felonies.

Waiting pe-

riods also vary: In some states, at least for some categories of crime,

individuals can apply for expungement immediately after completing

their sentence (although evidence of rehabilitation must generally be

shown, which can be harder if little time has passed).

Other states, by

contrast, have minimum waiting periods ranging from one to twenty

years.

In some states (including Michigan), it is illegal for an employer

to discriminate on the basis of an expunged conviction if the employer

becomes aware of it.

The Collateral Consequences Resource Center

and the Clean Slate Clearinghouse provide comprehensive state-by-state

information on expungement laws.

Beyond adult conviction records, most states have other expunge-

ment policies covering at least some other types of criminal records, such

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

See supra note 11 and accompanying text; see also Ispa-Landa & Loeffler, supra note 32, at 392.

Adult Criminal Record FAQs, P

APILLON

OUND

., http://www.papillonfoundation.org/

information/expungement-faqs#Adult_Criminal_Record [https://perma.cc/SA3T-GWF6].

See, e.g., M

ICH

. C

OURTS

, N

ONPUBLIC

AND

IMITED

-A

CCESS

OURT

ECORDS

(2020),

https://courts.michigan.gov/administration/scao/resources/documents/standards/cf_chart.pdf [https://

perma.cc/9SVE-MJR5] (cataloging different kinds of court records and the access restrictions imposed

on those records “by statute, court order, or court rule,” id. at i, as of January 2020).

E.g., N.C. G

. S

TAT

. § 15A-145.5 (2019); O

KLA

. S

TAT

. tit. 22, § 18(A)(11) (2020).

See, e.g., A

RIZ

. R

. S

TAT

. A

. § 13-907 (2020); C

. P

ENAL

ODE

§§ 1203.4, 1203.4a

(West 2020).

See CCRC State Survey, supra note 11.

See, e.g., C

. L

. C

ODE

§ 432.7(a)(1) (West 2020) (prohibiting employment discrimination);

see also C

. G

’

ODE

§ 12952 (West 2020) (prohibiting disclosure in its new enactment). In

Michigan, to use or divulge information about an expunged conviction is a misdemeanor. M

ICH

OMP

. L

AWS

. § 780.623(5) (West 2020) (“Except as provided in subsection (2), a person, other

than the applicant . . . , who knows or should have known that a conviction was set aside under

this section and who divulges, uses, or publishes information concerning a conviction set aside un-

der this section is guilty of a misdemeanor punishable by imprisonment for not more than 90 days

or a fine of not more than $500.00, or both.”).

See sources cited supra note 11.

2020] EXPUNGEMENT OF CRIMINAL CONVICTIONS 2473

as juvenile records, nonconviction records, or convictions that have been

overturned or are subject to successful collateral attack.

Many also

have deferred-adjudication programs available for certain defendants

(for example, first-time drug crime defendants or youthful defendants),

in which no conviction is ever entered if the defendant completes certain

requirements.

These laws raise related empirical and policy questions,

and discussions about them might well be able to draw on our findings.

But we focus here on state laws that expunge otherwise-valid adult con-

victions that have become final; these policies have been expanding in

scope in recent years and are the subject of active political debates in

many states. Although supported by the American Bar Association,

and by advocacy groups such as the National Employment Law Project,

the Center for American Progress, and Community Legal Services,

they have also met with considerable political opposition, principally

from employers, who want access to information that they consider rel-

evant to hiring decisions.

The most recent wave of efforts to expand and improve expungement

laws — often referred to by advocates as the “Clean Slate” movement —

has focused to a large degree on the potential for automatic expungement

of certain criminal records after a certain period of time.

The ground-

breaking success in this area was the adoption of Pennsylvania’s Clean

Slate Act, which extends court-ordered criminal record sealing to encom-

pass a broader set of offenses and creates an automatic computerized pro-

cess for sealing certain eligible convictions: minor nonviolent misdemean-

ors after ten years without a subsequent conviction.

The Clean Slate

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

See id.

See id. (describing deferred-adjudication programs).

See S

TANDARDS

FOR

RIMINAL

USTICE

§ 19-2.5 (A

. B

’

d ed.

2004

)

; A

’

, R

EPORT

THE

OUSE

ELEGATES

, R

ESOLUTION

109B, at 1 (2019),

https://www.americanbar.org/content/dam/aba/administrative/house_of_delegates/resolutions/

2019-midyear/2019-midyear-109b.pdf [https://perma.cc/SJ5R-2ZV8].

See Press Release, Ctr. for Am. Progress, Removing Barriers to Economic Opportunity for

Americans with Criminal Records Is Focus of New Multistate Initiative by CAP, NELP, and CLS

(Sept. 12, 2017), https://www.americanprogress.org/press/release/2017/09/12/437592/release-removing-

barriers-economic-opportunity-americans-criminal-records-focus-new-multistate-initiative-cap-nelp-

cls [https://perma.cc/E73W-W8BK].

See, e.g., Alison Knezevich, New State Laws to Help Marylanders Clear Arrest Records, B

ALT

(Sept. 26, 2015, 11:02 PM), https://www.baltimoresun.com/maryland/bs-md-expungement-

changes-20150926-story.html [https://perma.cc/R3X3-4YFD]; Nancy Reardon, Activists Want Law

to Seal Criminal Records Sooner, P

ATRIOT

EDGER

(July 28, 2009, 9:11 PM), https://www.

patriotledger.com/x836550449/Activists-want-law-to-seal-criminal-records-sooner

[https://perma.cc/B5R6-W3MN].

See C

LEAN

LATE

, https://cleanslateinitiative.org [https://perma.cc/S3FF-JPGA].

See Act of June 28, 2018, No. 402, 2018 Pa. Laws No. 56; Press Release, Governor Tom Wolf

of Pa., Governor Wolf: “My Clean Slate” Program Introduced to Help Navigate New Law (Jan. 2,

2019), https://www.governor.pa.gov/governor-wolf-my-clean-slate-program-introduced-to-help-

navigate-new-law [https://perma.cc/ENF7-UDRE]; Frequently Asked Questions About Clean Slate,

2474 HARVARD LAW REVIEW [Vol. 133:2460

Bill received nearly unanimous legislative endorsement and was sup-

ported by an overwhelming majority of Pennsylvania residents.

Pennsylvania’s club became larger in 2019, when Utah, New Jersey,

and California enacted automatic expungement laws,

and similar reform

efforts are now underway in several other states,

including Michigan.

These newer laws have all been broader than Pennsylvania’s pioneering

statute along some dimensions. Of these laws, California’s is the most

ambitious, although (unlike the others) it does not apply retrospec-

tively.

The California law applies to all misdemeanors and to some

felonies resulting in probation,

and it imposes no lengthy waiting pe-

riod — indeed, none beyond sentence completion in some cases.

Utah’s

law, like Pennsylvania’s, is limited to nonviolent misdemeanors, but its

waiting periods for most crimes are only five to seven years depending

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

OMMUNITY

EGAL

ERVS

. P

HILA

. (June 26, 2018), https://clsphila.org/learn-about-issues/

frequently-asked-questions-about-clean-slate [https://perma.cc/3PY5-U3CM].

See J.D. Prose, Pennsylvania Becomes First State with “Clean Slate” Law for Nonviolent

Criminal Records, T

IMES

(June 28, 2018, 5:00 PM), https://www.timesonline.com/

news/20180628/pennsylvania-becomes-first-state-with-clean-slate-law-for-nonviolent-criminal-rec-

ords [https://perma.cc/2EVM-B5RE].

See C

. P

ENAL

ODE

§ 1203.425(a)(2)(E) (West 2020); N.J. S

TAT

. A

. § 2C:52-5.4 (West

2020); U

TAH

ODE

. § 77-40-102 (LexisNexis 2020).

See, e.g., David Brand, A Criminal Justice Reform Would Give Thousands a Clean Slate — If

Only They Would Apply, B

ROOKLYN

AILY

AGLE

(Nov. 25, 2019), https://brooklyneagle.com/

articles/2019/11/25/a-criminal-justice-reform-would-give-thousands-a-clean-slate-if-only-they-

would-apply [https://perma.cc/34H3-NDKR]; Gemma Gaudette, Bipartisan Proposal Would Give

Some Former Idaho Inmates a “Clean Slate,” B

OISE

. P

. R

ADIO

(Jan. 29, 2020),

https://www.boisestatepublicradio.org/post/bipartisan-proposal-would-give-some-former-idaho-

inmates-clean-slate#stream/0 [https://perma.cc/L6E3-8WXZ]; Asti Jackson & Phil Kent, A Criminal

Record Should Not Be a Lifetime Sentence; That’s Why We Need the Clean Slate Law, H

ARTFORD

OURANT

(Mar. 15, 2020, 6:00 AM), https://www.courant.com/opinion/op-ed/hc-op-jackson-kent-

clean-slate-0315-20200315-7bohb5by5nfmpkrbh6fngml4fy-story.html [https://perma.cc/9WZY-

Z3GH]. There is also activity at the federal level. Nila Bala & Rebecca Vallas, Opinion, State

Momentum in Criminal Record Sealing Fuels Federal Clean Slate Bill, T

ILL

(Mar. 2, 2020,

2:00 PM), https://thehill.com/opinion/criminal-justice/485477-state-momentum-in-criminal-record-

sealing-fuels-federal-clean-slate-bill [https://perma.cc/W5MN-T2GB]; Press Release, Ctr. for Am.

Progress, supra note 66.

See, e.g., Riley Beggin, Michigan Eyes Reform to Costly, Confusing System of Expunging Crimi-

nal Records, B

RIDGE

. (Nov. 4, 2019), https://www.bridgemi.com/michigan-government/

michigan-eyes-reform-costly-confusing-system-expunging-criminal-records [https://perma.cc/4D83-

CZXU]; Miriam Francisco, New Bill Would Automate Process of Criminal Record Expungement

for Certain Convictions, D

ETROIT

ETRO

IMES

(Sept. 20, 2019, 3:24 PM), https://www.me-

trotimes.com/news-hits/archives/2019/09/20/new-bill-would-automate-process-of-criminal-record-

expungement-for-certain-convictions [https://perma.cc/AC4M-DTGF].

. P

ENAL

ODE

§ 1203.425(a)(2)(E) (noting that the automatic conviction relief only ap-

plies to “conviction[s] [that] occurred on or after January 1, 2021”).

Id.

Id. (indicating that otherwise-qualifying individuals who receive probation become eligible

for such relief immediately after successfully completing probation while those with misdemeanor

or infraction convictions become eligible immediately after completing their sentence, so long as a

year has elapsed since the conviction’s date of judgment).

2020] EXPUNGEMENT OF CRIMINAL CONVICTIONS 2475

on the crime class.

New Jersey’s automatic expungement law has a

ten-year waiting period, but it allows some more serious crimes to be

automatically expunged; it delegates to a task force the development of

the process.

In addition, proposed federal legislation would require

automatic expungement one year after completing a sentence for mari-

juana offenses and certain other minor drug offenses.

Many states and local jurisdictions have also adopted other laws de-

signed to reduce barriers to employment of people with criminal records.

The most important category (along with expungement laws) are Ban-

the-Box laws and policies, which typically bar employers from asking

about records on initial job application forms and in initial interviews.

Thirty-five states (and the District of Columbia) and over 150 cities and

counties have passed Ban-the-Box laws governing public employers,

and a further thirteen states and eighteen cities and counties now extend

them to private employers.

In terms of the number of people affected,

Ban-the-Box laws are far more sweeping than typical expungement laws

because they apply to all criminal records and contain no eligibility re-

quirements. However, expungement, for those who do obtain it, offers

potentially far more significant relief from the consequences of criminal

convictions, cutting across different domains of life and, in effect, legally

erasing the conviction in question. In contrast, Ban-the-Box laws pri-

marily affect employer practices, and only change the timing of employ-

ers’ receipt of record-related information; they can still refuse to hire an

applicant after completing a background check.

Still, Clean Slate ad-

vocates typically see the two types of reforms as complementary, and

many advocacy organizations have pushed for both.

Some states have also adopted laws that substantially restrict employ-

ers’ use of criminal record information — for example, requiring that they

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

TAH

ODE

. § 77-40-105(2–3) (LexisNexis 2020).

N.J. S

TAT

. A

. § 2C:52-5.4 (West 2020).

Clean Slate Act of 2019, H.R. 2348, 116th Cong.

See L

INDA

VA NS

, A

ONE

, B

THE

MPLOYMENT

: A

RASSROOTS

ISTORY

8 (2016), http://www.prisonerswithchildren.org/wp-content/uploads/

2016/10/BTB-Employment-History-Report-2016.pdf [https://perma.cc/6P9Z-AP38]. The Ban-the-

Box campaign was initiated in 2004 by All of Us or None, a national civil rights movement, to

eliminate the box on employment forms (as well as other applications) that asks whether an appli-

cant has ever been convicted of a felony. See id. at 10.

ETH

VERY

, N

’

ROJECT

, B

THE

1 (2019),

https://s27147.pcdn.co/wp-content/uploads/Ban-the-Box-Fair-Chance-State-and-Local-Guide-

July-2019.pdf [https://perma.cc/33QT-FEZK].

See Agan & Starr, Ban the Box, supra note 39, at 193.

See, e.g., Angela Hanks, Ban the Box and Beyond, C

FOR

. P

ROGRESS

(July 27, 2017,

9:03 AM), https://www.americanprogress.org/issues/economy/reports/2017/07/27/436756/ban-box-

beyond [https://perma.cc/5T7V-58NU].

2476 HARVARD LAW REVIEW [Vol. 133:2460

only rely on information that is job relevant.

These laws essentially rep-

licate guidance long given by the federal Equal Employment Opportunity

Commission (EEOC), which holds that overly sweeping bans on employ-

ees with records can amount to disparate impact racial or national origin

discrimination.

Unfortunately, courts have been reluctant to enforce

these restrictions.

Many employers throughout the country continue

to implement blanket exclusions of individuals with records, notwith-

standing EEOC’s guidance

— which, in any event, the Fifth Circuit

recently ruled is not enforceable.

These practices, and the difficulty of

eliminating them through other legal mechanisms, are among the moti-

vations for expungement laws.

C. Research Questions and Existing Empirical

Research on Expungement

There has been very little empirical research on any of the many

questions surrounding expungement laws, despite the clear importance

of these inquiries to policymakers and the lives of millions of Americans.

In truth, most of these questions really cannot be answered effectively

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

See A

VERY

, supra note 81, at 1–2.

See U.S. E

QUAL

’T O

PPORTUNITY

OMM

’

, N

. 915.002, E

NFORCEMENT

UIDANCE

: C

ONSIDERATION

RREST

AND

ONVICTION

ECORDS

LOYMEN

ECISIONS

NDER

ITLE

VII

THE

IVIL

IGHTS

1964 (2012),

https://www.eeoc.gov/laws/guidance/arrest_conviction.cfm [https://perma.cc/UK3G-DAAH].

See Dallan F. Flake, When Any Sentence Is a Life Sentence: Employment Discrimination

Against Ex-offenders, 93 W

ASH

. U. L. R

. 45, 80 (2015) (“[T]he EEOC has yet to prevail in

court . . . .”); see also Benjamin Levin, Criminal Employment Law, 39 C

ARDOZO

L. R

. 2265,

2303 (2018); Geiger, supra note 45, at 1203 (indicating that at least early versions of state laws were

“severely underutilized”); id. at 1199 (“[R]egulations [may be] especially difficult to enforce since, in

order for an applicant to know that a rejection was based on the use of prohibited criminal history

information, either the applicant would have to sue and obtain subpoena power to see the job

application information or the employer would have to freely admit the illegal basis for rejecting

the applicant.”).

See R

ODRIGUEZ

& E

MSELLEM

, supra note 33, at 3; U.S. C

OMM

’

IVIL

IGHTS

OLLATERAL

ONSEQUENCES

: T

ROSSROADS

UNISHMENT

, R

EDEMPTION

AND

THE

FFECTS

OMMUNITIES

134 (2019), https://www.usccr.gov/pubs/2019/06-13-Collateral-

Consequences.pdf [https://perma.cc/WX86-2BMM] (“Employment is difficult to access for those

individuals with a criminal conviction as many employers choose to use a blanket ban on hiring

any person with a prior criminal conviction regardless of the offense committed by the person.”);

Miriam J. Aukerman, The Somewhat Suspect Class: Towards a Constitutional Framework for

Evaluating Occupational Restrictions Affecting People with Criminal Records, 7 J.L.

& S

’

18,

26 (2005); Mark Jones et al., Challenges Facing Released Prisoners and People with Criminal Rec-

ords: A Focus Group Approach, 2 C

ORR

ECTIONS: P

’

RAC

. & R

91, 97–98 (2017); see also

UKAMAL

, supra note 31, at III.B.

See Texas v. EEOC, 933 F.3d 433, 451 (5th Cir. 2019); CCRC Staff, Appeals Court Invalidates

EEOC Criminal Record Guidance, C

OLLATERAL

ONSEQUENCES

ESOURCE

. (Aug. 7, 2019),

http://ccresourcecenter.org/2019/08/07/appeals-court-invalidates-eeoc-criminal-record-guidance

[https://perma.cc/9PT6-245J]; Alonzo Martinez, Fifth Circuit to EEOC — Don’t Mess with Texas,

ORBES

(Aug. 13, 2019, 4:57 PM), https://www.forbes.com/sites/alonzomartinez/2019//13/fifth-circuit-

to-eeoc-dont-mess-with-texas/#4e17c8f6c897 [https://perma.cc/5GMY-Z9Y7].

2020] EXPUNGEMENT OF CRIMINAL CONVICTIONS 2477

absent comprehensive access to individual-level data on people whose

records have been expunged, and because those records are generally

unavailable, research has been stymied. Fortunately, by taking ad-

vantage of our unusual data access, we are able to examine several key

questions that have long remained unanswered.

First, there is essentially no research on the question of “uptake” in

the context of the expungement of convictions. The basic question re-

lates to the uptake rate: When people are legally eligible for expunge-

ment, how often do they actually apply for and receive it? Beyond this

question, we also seek to provide a richer picture of who receives re-

lief — what kinds of convictions they have, what types of sentences they

served, their demographics — and how their profiles compare to those

of the broader pool of eligible persons, that is, what factors predict ac-

tual receipt. Answering these questions can help us better understand

the problem of “uptake gaps” and can also inform the discussion of ex-

pungement’s effects. Given that the group that actually receives relief is

a highly selected one, in order to speculate as to the potential results of

extending expungement laws to a broader and possibly dissimilar popula-

tion, we need to understand that selection process.

Again, no published study addresses these questions with respect to

expungement laws.

However, in certain other post–conviction relief

contexts, there have been a few attempts to explore issues related to

uptake. The most comprehensive such effort is Colleen Chien’s research

estimating the uptake rates for several programs: President Obama’s

clemency initiative allowing certain federal inmates to apply for sen-

tence commutation; California’s Proposition 47, which allows some fel-

ony convictions to be reduced to misdemeanors; and California’s

Proposition 64, which legalized marijuana and provides various forms

of relief for those previously convicted on marijuana charges.

She

finds uptake rates of about 3%, 9%, and 3%, respectively (within a rel-

atively short time period, especially for the newer Proposition 64).

Chien also analyzes a sample of background-check data concerning

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

Very recently, Colleen Chien et al. released a short summary of preliminary findings from

their study of expungement uptake in Washington State. Colleen V. Chien et al., The Washington

State Second Chance Expungement Gap 1 (Feb. 28, 2020) (unpublished manuscript), https://

papers.ssrn.com/sol3/papers.cfm?abstract_id=3529777 [https://perma.cc/7AX6-PBE7]. The study

uses a random sample of cases to estimate how many cases are eligible for expungement in

Washington; comparison to reported statistics on the total number of expungements actually

granted implies that only about 3% of eligible individuals obtain expungement. See id. at 2, 5.

Colleen V. Chien, The Second Chance Gap, 119 M

ICH

. L. R

. 12 (forthcoming 2020) (manuscript

at 12), https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3265335 [https://perma.cc/R8SH-HY5T].

Id. (manuscript at 17–23). The study does not use individual-level data to identify eligible

cases and track their outcomes. Instead, it compares publicly reported figures on the estimated

numbers of people eligible for relief and on the number of applications. See id.

2478 HARVARD LAW REVIEW [Vol. 133:2460

“gig” jobseekers and finds evidence that many reports contain noncon-

viction records (for example, arrests not leading to convictions), at least

some of which should be clearable under state law.

Research by Chris-

topher Uggen and coauthors also uncovers large uptake gaps in voting

rights–restoration procedures for people disenfranchised due to felony

convictions.

Although none of these estimates focuses on the expunge-

ment of criminal convictions, they all suggest a general problem: when

criminal justice relief mechanisms require individuals to go through ap-

plication procedures, many people who might benefit from them will

not do so.

Second, we investigate subsequent recidivism outcomes for expunge-

ment recipients. We explore this particular outcome in order to address

the concern, often raised by opponents of record-sealing, that the public

has a safety interest in being able to identify people with records.

We are

unaware of any other empirical research on this question — a question we

are able to investigate because we have recipients’ full criminal records,

including post-expungement convictions.

Evidence on the recidivism

question is critical to addressing objections from those who fear endorsing

expungement policies because of the possibility that these policies could

hide or increase recidivism risk.

Third, we examine the employment consequences of expungement.

Although potential employment benefits are core to the policy case for

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

See id. (manuscript at 25–35). The rates at which such nonconviction records appeared varied

substantially by state. See id. at 32 n.130. Note that because cleared records do not appear in back-

ground check data, Chien’s approach does not identify the number or share of eligible people who

do clear nonconviction records; rather, it simply shows that there are many who have not done so.

To estimate uptake rates for expungement among those eligible requires access to cleared records,

which researchers typically have not had. For background on nonconviction criminal records, their

use, and their regulation, see generally C

OLLATERAL

ONSEQUENCES

, M

ODEL

-C

ONVICTION

ECORDS

at v–vii (2019), http://ccresourcecenter.org/model-law-on-non-con-

viction-records-3 [https://perma.cc/PY5W-BYSY].

HRISTOPHER

GGEN

., S

ENTENCING

ROJECT

, 6 M

ILLION

OST

OTERS

TATE

-L

EVEL

STIMATES

ELONY

ISENFRANCHISEMENT

2016

, at

13 (2016),

https://www.sentencingproject.org/wp-content/uploads/2016/10/6-Million-Lost-Voters.pdf

[https://perma.cc/E9GB-Y2NG].

See T. Markus Funk, The Dangers of Hiding Criminal Pasts, 66 T

ENN

. L. R

. 287, 298–99

(1998); Margaret Colgate Love, Starting Over with a Clean Slate: In Praise of a Forgotten Section

of the Model Penal Code, 30 F

ORDHAM

. L.J. 1705, 1726 (2003); Michael D. Mayfield,

Comment, Revisiting Expungement: Concealing Information in the Information Age, 1997 U

TAH

L. R

. 1057, 1065–66, 1069–70.

Some theoretical and qualitative work has explored a few potential mechanisms by which

expungement policies could reduce recidivism. See, e.g., Ericka B. Adams et al., Erasing the Mark

of a Criminal Past: Ex-offenders’ Expectations and Experiences with Record Clearance, 19

UNISHMENT

& S

’

23, 45–47 (2017) (finding, based on interviews with forty people, that ex-

pungement encourages attitudinal shifts that could potentially be associated with reduced recidi-

vism); Murat C. Mungan, Reducing Crime Through Expungements, 137 J.

CON

. B

EHAV

. & O

398, 399 (2017) (arguing that offering expungements but making them costly could allow govern-

ments to “separate generally-law-abiding-citizens” from career criminals, thereby reducing crime).

2020] EXPUNGEMENT OF CRIMINAL CONVICTIONS 2479

expungement laws, they have proven especially difficult to study be-

cause both expunged criminal records and wage and employment rec-

ords are typically confidential. The above-cited research showing that

criminal records impair employment opportunities provides a strong in-

tuitive reason to believe that expunging those records should have the

opposite effect. However, for several reasons, this intuition could prove

to be incorrect, and the magnitude of any effect is unknown. First,

many commentators have expressed concern that expungement may fail

to hide criminal history information from inquisitive employers — for

example, if news stories, mug shots, or other information about past of-

fenses can be easily found with a Google search or other digital means,

or if private criminal records databases do not effectively delete ex-

punged records.

Second, the waiting periods built into expungement

laws may undermine any potential benefits because these restrictions

mean that expungement cannot help during the critical period immedi-

ately after a conviction or during reentry.

Third, people with records

often also tend to face many other employment disadvantages besides

the record itself, potentially limiting the benefits of expungement.

To date, the most on-point research is a recent study by Jeffrey

Selbin, Justin McCrary, and Joshua Epstein, which tracks labor market

outcomes for 235 clients of a law clinic who pursued either expungement

or another form of relief (which allowed the reduction of felonies to mis-

demeanors).

100

The authors find suggestive evidence of employment

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

See Victoria Cumbow, Everything Posted Online Is There Forever, Even After It’s Been

Deleted, AL.

COM

(last updated Jan. 14, 2019), http://blog.al.com/breaking/2011/03/everything

_posted_online_is_th.html [https://perma.cc/6MHW-XTL5]; see also Sarah Esther Lageson, There’s No

Such Thing as Expunging a Criminal Record Anymore, S

LATE

(Jan. 7, 2019, 2:44 PM),

https://slate.com/technology/2019/01/criminal-record-expungement-internet-due-process.html [https://

perma.cc/H3EW-J396]; Experts: Deleted Online Information Never Actually Goes Away, C

. T

RIB

(Aug. 21, 2015, 11:26 AM), https://www.chicagotribune.com/business/blue-sky/chi-deleted-online-

information-never-goes-away-20150821-story.html [https://perma.cc/5A8N-GCBH].

See Clay Calvert & Jerry Bruno, When Cleansing Criminal History Clashes with the First

Amendment and Online Journalism: Are Expungement Statutes Irrelevant in the Digital Age?, 19

OMM

ONSPECTUS

123, 123–24 (2010); Alessandro Corda & Sarah E. Lageson, Disordered

Punishment: Workaround Technologies of Criminal Records Disclosure and the Rise of a New Penal

Entrepreneurialism, 60 B

RIT

. J. C

RIMINOLOGY

245, 247, 260–61 (2020); Marc A. Franklin & Diane

Johnsen, Expunging Criminal Records: Concealment and Dishonesty in an Open Society, 9

OFSTRA

L. R

. 733, 745–48 (1981); Eldar Haber, Digital Expungement, 77 M

. L. R

. 337,

355–57 (2018); Love, supra note 94, at 1719, 1725–26; Jenny Roberts, Expunging America’s Rap

Sheet in the Information Age, 2015 W

. L. R

. 321, 341–46; Mayfield, supra note 94, at 1068–69;

Sharon M. Dietrich, Ants Under the Refrigerator?, C

RIM

. J

UST

., Winter 2016, at 26, 26,

http://ccresourcecenter.org/wp-content/uploads/2017/03/Ants-under-the-Refrigerator-published.pdf

[https://perma.cc/2ZQH-BVBN].

See J

ACOBS

, supra note 7, at 131; Franklin & Johnsen, supra note 97, at 739; Selbin et al.,

supra note 43, at 52.

See sources cited supra note 31.

100

See Selbin et al., supra note 43, at 33, 38, 40.

2480 HARVARD LAW REVIEW [Vol. 133:2460

gains.

101

As they acknowledge, however, their small sample size does

not “allow for precise estimation,”

102

and the study does not present sta-

tistical estimates with standard errors (which permit readers to evaluate

the degree of imprecision). In addition, their sample is not necessarily

representative of those who obtain expungement,

103

and their data do

not allow them to answer the additional questions we discuss above re-

garding uptake and recidivism outcomes. So while their study provides

a welcome start, its objectives were relatively modest. In contrast, we

analyze a large, statewide sample that includes all expungement recipi-

ents as well as a large comparison group of individuals eligible for ex-

pungement, and we are able to use these data to address a variety of

interrelated questions with statistical power.

D. Our Empirical Setting and Data

We conduct our research in Michigan. Michigan is a large, demo-

graphically diverse state; studying its population, therefore, is likely to

yield empirical findings that are broadly generalizable to the rest of the

United States. For example, Michigan’s 14.1% poverty rate is roughly

comparable with the 13.1% national poverty rate;

104

its $56,697 median

household income is also close to the U.S. median household income of

$61,937.

105

Michigan has roughly the same percentage of people who

identify as African American (13.8%) and white (78.3%)

106

as the coun-

try overall, although it has significantly smaller Hispanic and Asian

populations.

107

Michigan’s criminal justice system is also reflective of

national trends. Michigan’s property crime rates are now slightly lower

than the national average, but, for most of the study period, they hewed

closely to national crime rates; Michigan’s violent crime rates are a bit

higher than average.

108

Similarly, Michigan’s state prison population

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

101

Id. at 50–51.

102

Id. at 49.

103

See id. at 45–46, 59.

104

Compare Michigan, U.S. C

ENSUS

UREAU

, https://data.census.gov/cedsci/profile?q=

Michigan [hereinafter Michigan Census Data] [https://perma.cc/XRH2-2853], with United States,

U.S.

ENSUS

UREAU

, https://data.census.gov/cedsci/profile?q=United%20States [hereinafter

National Census Data] [https://perma.cc/89M4-TSZ5].

105

Compare Michigan Census Data, supra note 104, with National Census Data, supra note 104.

106

ACS Demographic and Housing Estimates, U.S. C

ENSUS

UREAU

, https://data.cen-

sus.gov/cedsci/table?q=percent%20african%20american%20michi-

gan&g=0400000US26&hidePreview=false&tid=ACSDP1Y2018.DP05&t=Black%20or%20African

%20American&cid=DP05_0001E&vintage=2018 [https://perma.cc/5487-HNPS].

107

Compare id. (only 5.2% of the population identifying as Hispanic or Latino in Michigan), and

id. (only 3.3% identifying as Asian), with National Census Data, supra note 104 (18.3% of the na-

tional population identifying as Hispanic or Latino and 5.6% of the national population identifying

as Asian).

108

See Crime Data Explorer: Michigan, F

. B

UREAU

NVESTIGATION

, https://crime-data-

explorer.fr.cloud.gov/explorer/state/michigan/crime [https://perma.cc/M8MA-WJUV].

2020] EXPUNGEMENT OF CRIMINAL CONVICTIONS 2481

increased at roughly the same rate as the overall state prison population

in the nation through the mid-2000s.

109

Today, Michigan’s incarceration

rate of 641 per 100,000 people is the 29th highest in the country, only

slightly lower than the national rate of 698 per 100,000.

110

In Michigan, expungement of criminal convictions is provided for by

a longstanding statute, M.C.L. 780.621. The law actually labels this

process a criminal record “set aside,” rather than using the more recog-

nizable term “expungement.” And indeed, as in most states, the

Michigan process does not entail a complete expungement of the record

(which is retained by the police for limited purposes mainly related to

criminal justice), but rather a sealing and an elimination of many of the

record’s legal consequences. However, to maintain consistency with the

terminology we adopt elsewhere in this Article, we will continue to use

the term “expungement” as a convenient and familiar shorthand.

M.C.L. 780.621 has been on the books since the 1960s,

111

although

expungements were very rare before a significant statutory change that

took effect in 1983.

112

It has undergone some recent amendments,

mostly expansions in its eligibility provisions.

113

Because of the time

range of our data, the most relevant version of the law is the one that

was in effect (with no material changes) from 1983 until mid-2011. Our

empirical analyses focus on people eligible under that law’s require-

ments, and on expungements granted before the law changed in 2011.

The fact that Michigan has had an expungement law for decades distin-

guishes it from the majority of other states,

114

and it makes the state a

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

109

Compare Michigan Profile, P

RISON

’

NITIATIVE

, https://www.prisonpolicy.org/

profiles/MI.html [https://perma.cc/TE7G-2AMD], with State Prison Incarceration Rate,

1978

2012

RISON

’

NITIATIVE

, https://www.prisonpolicy.org/graphs/US_state_incrates_1978-

2012.html [https://perma.cc/2JL3-5SUM] (calculating the rate of increase as the slope of a line fitted

to the annual data).

110

Peter Wagner & Wendy Sawyer, States of Incarceration: The Global Context

2018

, P

RISON

’

NITIATIVE

(June 2018), https://www.prisonpolicy.org/global/2018.html [https://perma.cc/

U8DU-DABN].

111

ICH

. C

OMP

. L

AWS

§ 780.621 (2019); see also A Brief History of Expungement in Michigan,

AFE

& J

UST

ICH

(Aug. 22, 2018), https://www.safeandjustmi.org/2018/08/22/a-brief-history-of-

expungement-in-michigan [https://perma.cc/NP4V-96KR] [hereinafter A Brief History].

112

Only a handful of earlier expungements appear in our comprehensive dataset. For a related

discussion, see A Brief History, supra note 111. Advocating for amendments, one judge wrote in

1981 that the available procedures at the time were “segmented and confusing.” William S. Easton,

Commentary, Expunging Criminal Records: A Judge’s Perspective, 27 W

AY NE

L. R

. 1391, 1392

(1981). Prior to the statute’s amendment, the limited availability of expungement in Michigan was

driven by the difficulties of accessing arrest records, see id. at 1394, and the narrowness of the

procedures available to seek expungement, see People v. Upshaw, 283 N.W.2d 778, 779 (Mich. Ct.

App. 1979). The statute was amended in 1982. See 1982 Mich. Pub. Acts 1887.

113

See A Brief History, supra note 111.

114

See generally Love, supra note 94, at 1717–26 (surveying the expansion and contraction of

expungement laws and policies in the United States); CCRC State Survey, supra note 11.

2482 HARVARD LAW REVIEW [Vol. 133:2460

particularly useful subject for empirical study; Michigan’s long experi-

ence with record clearing can inform other states that have recently

adopted or are considering adopting similar policies. Cumulatively, tens

of thousands of people have received expungements under Michigan’s

law, and many received them long enough ago to allow us to observe

their post-expungement outcomes for a substantial period of time.

Michigan’s expungement law imposes a five-year waiting period,

115

which is in the middle of the range of the waiting periods that other

states’ statutes apply.

116

In the pre-2011 version, the five-year clock

began running at sentencing, unless the defendant served a term of in-

carceration after sentencing, in which case it ran from release.

117

Any

reconviction during this five-year period disqualified the individual

from an expungement.

118

Essentially, then, the law required five “clean”

years, excluding time behind bars. The statute covered (and still covers)

almost all types of crimes, including most violent felonies.

119

The prin-

cipal exceptions are traffic offenses, sex offenses that are subject to reg-

istration requirements, and the most serious class of felonies: those car-

rying potential life-imprisonment terms.

120

The wide range of crimes

otherwise covered is another advantage of Michigan’s law for research

purposes; it allows us to compare uptake rates and outcomes across

many important offense types and grades.

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

115

ICH

. C

OMP

. L

AWS

§ 780.621(5) (2019) (“An application [for expungement] shall only be filed

5 or more years after whichever of the following events occurs last: [listing imposition of sentence,

completion of probation, discharge from parole, or completion of a term of imprisonment].”); see

also M

ICH

. C

OMP

. L

AWS

§ 780.621(3) (2011) (“An application shall not be filed until at least 5 years

following imposition of the sentence for the conviction that the applicant seeks to set aside or 5

years following completion of any term of imprisonment for that conviction, whichever occurs

later.”).

116

See, e.g., I

. C

ODE

§ 35-38-9-2(c) (2019) (five-year waiting period); K

. S

TAT

. A

. § 21-

6614 (2019) (three to five years); L

. C

ODE

RIM

. P

ROC

. A

. arts. 977(A)(2), 978(A)(2) (2019)

(five-year waiting period for most misdemeanors and ten-year waiting period for felonies); M

ODE

., C

RIM

. P

ROC

. § 10-110(c) (West 2019) (ten to fifteen years); M

INN

. S

TAT

§ 609A.02(3)(a)(2)–(5) (2019) (one to five years, based on offense severity); M

. R

. S

TAT

§ 610.140(5)(1) (2019) (three years for misdemeanors; seven for felonies); N

. R

. S

TAT

§ 179.245(1) (2019) (effective July 1, 2020) (one to ten years, based on offense severity); N.Y.

RIM

ROC

. L

§ 160.59(5) (McKinney 2019) (ten years); see also M

ARGARET

OVE

, J

OSH

AINES

ENNY

SBORNE

, F

ORGIVING

AND

ORGETTING

MERICAN

USTICE

: A 50-S

TATE

UIDE

XPUNGEMENT

AND

ESTORATION

IGHTS

84–112 (2018), http://

ccresourcecenter.org/wp-content/uploads/2017/10/Forgiving-Forgetting-CCRC-Aug-2018.pdf

[https://perma.cc/SQ57-YQKN]; CCRC State Survey, supra note 11.

117

ICH

. C

OMP

. L

AWS

§ 780.621(3) (2011).

118

Id. § 780.621(4)(c).

119

Id. § 780.621(1) (“[A] person who is convicted of not more than 1 offense may file an

application . . . .”).

120

Id. § 780.621(2) (“A person shall not apply to have set aside, and a judge shall not set aside, a

conviction for a felony for which the maximum punishment is life imprisonment or an attempt to

commit a felony for which the maximum punishment is life imprisonment, a conviction for a vio-

lation or attempted violation of section 520c, 520d, or 520g of the Michigan penal code, . . . or a

conviction for a traffic offense.”).

2020] EXPUNGEMENT OF CRIMINAL CONVICTIONS 2483

However, Michigan does impose stringent limitations on expunge-

ment eligibility based on the length of the individual’s record — that is,

the number of convictions (nonconviction records like arrests are not

relevant to eligibility).

121

While this has now been loosened slightly, in

the pre-2011 version, expungements were strictly limited to people with

exactly one conviction on a single charge.

122

Prior convictions, subse-

quent convictions, and simultaneous convictions — even multiple

charges stemming from the same incident — were all disqualifying. On

a policy level, this has incongruous results; for example, a person with a

single very serious felony may be eligible for an expungement, while a

person with two simultaneous misdemeanor counts arising from one in-

cident is not. However, the simple, bright-line nature of these eligibility

requirements offers advantages from a research perspective;

123

for ex-

ample, it makes it easier to code expungement eligibility and other var-

iables related to the nature of the criminal record.

For those meeting the eligibility requirements in Michigan, expunge-

ments are not automatic. Rather, the law requires the expungement

seeker to go through an elaborate application process, which we describe

in detail in section II.C. The court is not required to grant the requested

expungement — it “may” grant the request if doing so is warranted by

the applicant’s subsequent behavior and “consistent with the public wel-

fare.”

124

Still, grant rates are fairly high. Although our data do not

identify unsuccessful applications, the Michigan State Police (MSP),

which processes applications and also implements the expungements

once they are granted, shared figures for 2016 and 2017 that indicate

that approximately 75% of applications are successful.

125

To carry out our study, we entered into a data-sharing agreement with

multiple Michigan state agencies, which worked with us to develop a

merged and de-identified dataset. MSP provided comprehensive

Michigan criminal histories (RAP sheets) on almost every expunged crim-

inal conviction in Michigan history through March 2014,

126

amounting to

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

121

Id. § 780.621(1) (“Except as provided in subsection (2), a person who is convicted of not more

than 1 offense may file an application with the convicting court for the entry of an order setting

aside the conviction.”); see also People v. Blachura, 440 N.W.2d 1, 1–2 (Mich. Ct. App. 1989) (person

convicted of five counts of perjury ineligible since each count deemed a separate conviction).

122

See M

ICH

. C

OMP

. L

AWS

§ 780.621(4)(c).

123

See Charles Tremper et al., Measuring Law for Evaluation Research, 34 E

VA LUAT ION

242, 252–54 (2010).

124

ICH

. C

OMP

. L

AWS

. § 780.621(9). Note that the text quoted has not changed. See 1982

Mich. Pub. Acts 1889.

125

Email from Ted Kilvington, Court Reporting Coordinator, Mich. State Police, to Simmon

Kim, Research Assoc., Univ. of Mich. Empirical Legal Studies Ctr. (Feb. 22, 2019, 10:47 AM) (on

file with authors). Although the figures MSP provided cover more recent years than do the data

we use in our study, discussions with MSP and with expungement lawyers do not give us any reason

to believe there have been substantial changes in grant rates in the intervening years.

126

A small share of the most recent records was inadvertently dropped, as we discuss below.

2484 HARVARD LAW REVIEW [Vol. 133:2460

nearly 30,000 cases, plus a larger group of other criminal records meet-

ing requirements that we discuss below. These were linked to wage and

employment information for the same individuals from the state Unem-

ployment Insurance Agency and Workforce Development Agency.

Matching was based on Social Security numbers, so those without such

numbers are excluded from our analysis of wage data.

127

In addition to

the records for individuals who actually receive expungements, we ob-

tained records for a large group of people who were, at the time of their

convictions, potentially legally eligible for expungements in the future,

if they subsequently succeeded in meeting the five-clean-years require-

ment. This group consists of individuals with a first criminal offense

(occurring between 1999 and 2008) that was on a list of legally eligible

offense codes, which we developed using the expungement statute and

provided to the MSP for a data query.

128

Our data have other limitations. For our criminal record and em-

ployment outcome data, we lack information on out-of-state convictions

(affecting our coding of expungement eligibility and recidivism out-

comes) and out-of-state income (affecting our wage and employment

analysis). Our wage and employment information comes from the state

unemployment insurance system, which covers wage and salary employ-

ees quite comprehensively but misses the self-employed and individuals

employed “under the table.”

129

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

127

Michigan’s Department of Technology, Management and Budget carried out the matching

process, and then the department de-identified all records before it turned them over to us. In the

initial stages of our work on this project, in order to shape our records requests for MSP, we also

received and analyzed court records from Michigan’s Judicial Data Warehouse. However, because

of the incompleteness of these records, we did not include them in the ultimate data analysis.

128

Because MSP asked us to keep our request reasonable in size, we excluded from this list

certain common petty misdemeanors — such as dog-leash law and hunting and fishing violations —

which are almost never expunged, even though they are legally eligible. We identified these as

among Michigan’s most common first offenses based on a preliminary analysis of court data from

the Judicial Data Warehouse, but these offenses are essentially not found at all in the complete

dataset on all expungements, perhaps suggesting that people do not bother to expunge records so

minor that they may impose no serious burdens. Some cases were also inadvertently excluded

because MSP formatted offense codes in inconsistent ways that did not match our list; these exclu-

sions seem substantively random. We also excluded minor-in-possession cases, which in Michigan were

(until the offense was decriminalized in 2016) often subject to a separate diversion process for youthful

individuals. We discuss the implication of these exclusions below, infra pp. 2492–93.

129

This means that our measured employment and wage levels will be biased downward relative

to total employment and wages. Although this is true to some degree both before and after ex-

pungement, it is possible that it is more true before expungement, if people with records are more

likely to seek under-the-table work. If so, we could overstate true wage gains. On the other hand,

there are reasons to focus on employment in the formal sector (as employment studies typically do).

There are distinct advantages to having formal labor market opportunities (for example, termina-

tion notice and other mandatory benefits). See Minhaj Mahmud et al., What Aspects of Formality

Do Workers Value? 3 (World Bank Grp., Knowledge and Strategy Team, Dev. Econ., Working Paper

No. 9108, 2020), http://documents.worldbank.org/curated/en/428331578944574540/

pdf/What-Aspects-of-Formality-Do-Workers-Value-Evidence-from-a-Choice-Experiment-in-

2020] EXPUNGEMENT OF CRIMINAL CONVICTIONS 2485

Our most serious data challenge involves missing records. The

state’s process for matching and de-identifying our criminal history and

employment outcomes data inadvertently dropped certain observa-

tions.

130

Fortunately, we were able to restore virtually all of the dropped

records that concern expungement recipients because we had already

received a complete version of that part of the dataset slightly earlier.

131

These drops only minimally affect our uptake analysis and should not

affect our employment analysis at all. However, the problem did lead us

to focus our primary recidivism outcome analysis on expungement recip-

ients alone, because the drops affect the accuracy of our recidivism

measures for nonrecipients.

132

Another data challenge relates to certain

missing dates in the criminal records data, a problem that requires us to

impute eligibility dates.

133

We believe this adds fairly minimal measure-

ment error. The imputation should not be far off, and should be correct

on average, so our resulting uptake estimates should not be biased, even

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

Bangladesh.pdf [https://perma.cc/RM65-FDGY]. Indeed, to the extent that people do move from

under-the-table work to formal employment after receiving an expungement, it is presumably be-

cause the formal-sector option is more appealing. We discuss these issues as they pertain to our

analyses below.

130

Data patterns make clear that their procedure successfully identified duplicate records pro-

duced by overlapping data pulls by MSP but then mistakenly deleted both copies.

131

We were unable to restore some records for individuals who had not received expungements,

however. By the time we diagnosed the problem, our data-sharing agreement and grant period had

expired, and the state authorities informed us that they could not reconstruct the original dataset

for us or repeat the process. Through information from MSP, we were able to identify the number

of lost records (about 9% of the nonrecipients in the original data pull).

132

The dropped nonrecipient cases were individuals with at least two convictions, so omitting

them would have distorted our analysis. We requested data on people with certain kinds of first

convictions or second convictions within a ten-year period; people with first and second convictions

meeting these criteria were included as duplicates and then doubly dropped. The other set of du-

plicates dropped was the set of expungement recipients who also met the eligibility criteria for the

comparison group, but this was the set that we were able to restore.

133

When MSP effectuates an expungement, they change the disposition field in the record to

denote an expungement, but their data entry system does not have a separate field for the expunge-

ment date. Instead, MSP’s practice in most years was to change a field called “judgment date” to

the expungement date. This practice caused the information originally stored in the judgment date

field to be lost — specifically, the conviction date, which is what we use in most cases to calculate

the start date for the five-year expungement eligibility clock. Expungement eligibility in nonincar-

ceration cases runs from sentencing, which is usually a few weeks after the case disposition, Email

from Jonathan Sacks, Appellate Def., State Appellate Defs.’ Office, to Sonja Starr, Professor of

Law, Univ. of Mich. Law Sch. (Feb. 22, 2019, 11:02 AM) (on file with authors); Email from Kim

Thomas, Clinical Professor of Law, Univ. of Mich. Law Sch., to J.J. Prescott, Professor of Law,

Univ. of Mich. Law Sch. (Feb. 17, 2019, 9:19 PM) (on file with authors), and has no recorded date

in the data. MSP informed us, however, that when they make their official determinations of ex-

pungement eligibility, they rely on the conviction date as a proxy for sentencing, so we do the same

here. See Email from Nick Romanov, IT Specialist, Mich. State Police, to J.J. Prescott, Professor

of Law, Univ. of Mich. Law Sch. (Nov. 7, 2018, 11:38 AM) (on file with authors). Because some of

our analyses (especially of uptake rates) require us to know both the expungement eligibility date

and the expungement grant date, we had to impute the lost conviction date based on the incident,

arrest, and charge dates.

2486 HARVARD LAW REVIEW [Vol. 133:2460

if we might get the date of eligibility wrong by a month or two for some

individual observations.

In a small subset of expungement cases, MSP changed the disposi-

tion code but did not change the judgment date field to the expungement

date, leaving no record of when the expungement occurred. These cases

are identifiable because the judgment date is too close to other key dates

in the case (for example, the arrest date) to be a plausible expungement

date given the required five-year waiting period. We eliminate these

cases from the recidivism and employment analyses, which require ex-

pungement dates; the lag time from other events in the case to expunge-

ment receipt is far too unpredictable to be imputed without introducing

significant measurement error. However, in our analyses of expunge-

ment uptake rates within certain periods of time, we cannot ignore these

cases or we would understate uptake, so instead we make assumptions

about their temporal distribution to estimate bounds on uptake rates.

Fortunately, the practical importance of this problem is small; there is

good reason to believe that most of these unknown-date expungements

occurred after June 2011, when we see a sudden drop in the number of

known-date expungements to virtually zero, reflecting a change in

MSP’s data-recording practice.

134

This date happens to coincide with

the substantive changes in the expungement law that (as we note above)

took place in 2011,

135

and the focus of our work is on expungements

taking place before that date anyway.

II.

PTAKE

: W

EEKS

AND

ECEIVES

XPUNGEMENTS

Challenges surrounding uptake may be the most underappreciated

problem concerning expungement policies — and indeed, similar prob-

lems appear in many access-to-justice contexts.

136

Uptake challenges

potentially arise whenever the government imposes significant bur-

dens — such as fees or onerous administrative requirements — on the

exercise of a right or opportunity, or whenever it is difficult for potential

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

134

MSP confirmed that expungement dates are not recorded in their current data and that the

number of expungements granted has only increased since 2011. See Email from Ted Kilvington,

supra note 125 (stating that 2594 expungements were processed in 2017). Assuming expungement

frequency at least did not drop, we believe at least half the unknown-date expungements in the

earliest version of our complete expungement dataset (which ended in December 2012) must have

occurred between June 2011 and December 2012, which leaves at most 3000 occurring before then

(about 13% of the number of known-date expungements).

135

See supra notes 111–114 and accompanying text.

136

See, e.g., J.J. Prescott, Assessing Access-to-Justice Outreach Strategies, 174 J.

NSTITUTIONAL

& T

HEORETICAL

CON

. 34, 36 (2018) (“Individuals often fail to take full ad-

vantage of beneficial government programs and policies. This take-up problem has been studied

in numerous contexts . . . .”).

2020] EXPUNGEMENT OF CRIMINAL CONVICTIONS 2487

beneficiaries to learn about those rights or opportunities.

137

And a right

or opportunity that is too difficult for most people to exercise is effec-

tively empty. Accordingly, in this Part, we address several questions

surrounding expungement uptake (here defined as the successful receipt

of an expungement by someone who is eligible to receive one), which

have never previously been examined empirically.

138

In section A, we estimate the overall five-year expungement uptake

rate among those who are eligible for such relief. In section B, we turn

to the question of who successfully obtains expungements, presenting

descriptive statistics on recipients and regression analyses to assess

which individual and case characteristics predict expungement receipt.

We use these analyses in section C, along with qualitative insights pro-

vided by our interviews with expungement experts, to inform a discus-

sion of uptake hurdles and their implications.

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

137

For a useful discussion of the “uptake” problem for social-benefit programs — referred to as the

“take-up” problem in many disciplines — see generally Janet Currie, The Take-Up of Social Benefits, in

UBLIC

OLICY

AND

THE

NCOME

ISTRIBUTION

80 (Alan J. Auerbach et al. eds., 2006).

138

Research on uptake is common in many contexts outside of the criminal justice domain, where

policy designers are very conscious of the issue. See generally, e.g., Saurabh Bhargava & Dayanand

Manoli, Psychological Frictions and the Incomplete Take-Up of Social Benefits: Evidence from an

IRS Field Experiment, 105 A

. E

CON

. R

. 3489 (2015) (tax benefits); Adam S. Booij et al., The

Role of Information in the Take-Up of Student Loans, 31 E

CON

. E

DUC

. R

. 33 (2012) (education);

Jeffrey R. Kling et al., Comparison Friction: Experimental Evidence from Medicare Drug Plans,

127 Q.J.

CON

. 199 (2012) (health); Brigitte C. Madrian & Dennis F. Shea, The Power of Suggestion:

Inertia in

401

(k) Participation and Savings Behavior, 116 Q.J. E

CON

. 1149 (2001) (saving incen-

tives); Giovanni Mastrobuoni, The Role of Information for Retirement Behavior: Evidence Based

on the Stepwise Introduction of the Social Security Statement, 95 J.

. E

CON

. 913 (2011) (Social

Security); Xiaopeng Pang et al., Does Women’s Knowledge of Voting Rights Affect Their Voting

Behaviour in Village Elections? Evidence from a Randomized Controlled Trial in China, 213

HINA

Q. 39 (2013) (voting rights); Dan Archer et al., Reducing Vulnerability to Human Trafficking:

An Experimental Intervention Using Anti-trafficking Campaigns to Change Knowledge, Attitudes,

Beliefs, and Practices in Nepal (U.S. Agency for Int’l Dev., Research and Innovation Grants

Working Papers Series, 2016) (legal access and compliance). This literature provides insight into

the general barriers to uptake. For example, Janet Currie identifies the costs of learning about and

making use of government programs, noting that these costs may be relatively high for individuals

who need programs the most. Currie, supra note 137, at 83. Saurabh Bhargava and Dayanand

Manoli emphasize the costs studied by behavioral economics, including psychological frictions due

to cognitive, motivational, and emotional constraints. Bhargava & Manoli, supra, at 3489–90. Peo-

ple may be confused about a program or policy, see, e.g., Jeffrey B. Liebman & Richard J.

Zeckhauser, Schmeduling 6–9 (2004) (unpublished working paper), https://scholar.harvard.

edu/files/jeffreyliebman/files/Schmeduling_WorkingPaper.pdf [https://perma.cc/2EFP-VRCB], may

be unaware that a particular program exists, see Currie, supra note 137, at 110–11, or may suffer

from other psychological biases or tendencies (such as procrastination, inattention, or distaste for

time-consuming processes) that inhibit their pursuit of a program’s benefits, see

Marianne Bertrand et al., Behavioral Economics and Marketing in Aid of Decision Making Among

the Poor, 25

J. P

. P

’

& M

ARKETING

8, 10 (2006); Madrian & Shea, supra, at 1150. Other

explanations include hyperbolic time discounting and stigma costs.

2488 HARVARD LAW REVIEW [Vol. 133:2460

A. Estimating Uptake Rates

The first step in estimating expungement uptake rates is identifying

which cases are legally eligible for relief. In our context, the relevant

eligible pool is defined first by the parameters of the records query that

we asked MSP to implement, as detailed in section I.D, and second by

some further sample refinements that we carry out thereafter. These

sample parameters are: (1) The individual has a first criminal conviction

on a single criminal count.

139

(2) The crime of conviction matches an

offense code on a list of offenses eligible for expungement (known as a

“set-aside” in Michigan).

140

(3) Sentencing for the eligible count took

place between January 1999 and May 2001.

141

(4) The individual is not

subsequently convicted for any crime within five years of sentencing.

142

(5) The individual was not sentenced to incarceration on the eligible

offense.

143

And (6) the individual has no out-of-state driver’s license

listed anywhere in their MSP arrest record.

144

The five-year uptake rate is the percentage of this eligible group that

receives an expungement within five years of the eligibility date (that is,

within ten years of sentencing). We make two further assumptions in

order to calculate this uptake rate, which relate to the data problems we

identify above in section I.D. First, we assume the distribution of

unknown-date expungements in our sample, in terms of the time elapsed

since the case’s disposition, is roughly similar to the distribution of

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

139

This was a pre-2011 eligibility requirement. See supra notes 119, 122 and accompanying text.

As we describe below, we are able to screen only for prior offenses that took place in Michigan.

140

We are confident that our offense list only contains offenses that are in fact legally eligible for

expungement. However, as we explain in section I.D and discuss below, our list is not inclusive of

all eligible offenses.

141

Thus, expungement eligibility kicks in between January 2004 and May 2006 in our uptake

sample. Our sample contains the most recent set of cases for which we can still track outcomes for

five years after eligibility accrues. The main estimates we give are five-year uptake rates, and we

treat May 2011 as the end of our tracking period because two important changes occurred in June

2011: the eligibility rules changed (making more people eligible but also rendering some formerly

eligible persons ineligible), and MSP ceased to record expungement dates. For cases receiving ex-

pungements, because of MSP’s data recording practices, we impute the sentencing date.

142

This is also a legal requirement, and again we implement this filter based on Michigan data

alone. See supra p. 2482. Our eligible pool consists of people who became eligible after five years,

and we estimate expungement receipt rates within the next five years. Some members of this sample

could have lost eligibility at some point during the second five-year period due to a subsequent

conviction, although as we will see in the recidivism analysis in Part III, this happens rarely.

143

We focus on nonincarceration cases (which, as we will see below, constitute the large majority

of all expungement cases) because we have a more accurate measure of the start date of the eligi-

bility clock, which runs from sentencing. In incarceration cases, the eligibility start date is based

on release from incarceration, and because our data tell us only the sentence and not the actual

release date, we cannot account for possible early release or credit for time served pretrial.

144

In our data, driver’s license numbers have been removed, but for most observations we know

whether the individual had a driver’s license and, if so, the state that issued that license. We exclude

individuals with out-of-state driver’s licenses to reduce the likelihood of miscoding eligibility on

account of unobservable out-of-state convictions, but we retain those observations of individuals

who had no driver’s license identified in our data. This restriction eliminates about 4% of our

sample.

2020] EXPUNGEMENT OF CRIMINAL CONVICTIONS 2489

known-date expungements. We show below that our empirical conclu-

sions are robust to — that is, unaffected by — alternative assumptions

about the unknown-date distribution. Second, we assume that the miss-

ing nonexpunged records would not have met the criteria for inclusion

in this sample, and thus we do not adjust our estimate to account for

them. We believe this latter assumption is largely correct; dropped cases

all had at least two convictions within the years covered by the pool of

eligible cases given to us by MSP (1999–2008), few of which would meet

the five-year nonrecidivism criterion. To the extent this assumption is

mistaken, we understate the number of eligible cases and thus overesti-

mate the uptake rate (but likely not by much).

In Table 1, we present our main uptake estimate, which is discour-

agingly low. Of eligible individuals, only 6.5% receive expungements

within five years of becoming eligible. The remaining 93.5% includes

people who do not apply for expungements and those who have their

applications denied by a judge. However, as we note above, we learned

from MSP that in 2016 and 2017 combined, 74% of the expungement

applications MSP received were ultimately granted by courts. If we

assume that all of these applicants were legally eligible and that the

same ratio applies during the relevant time period for our estimate

(2004–11), 6.5% of those eligible for expungement receive one within

five years, another 2.3% have their applications denied, and 91.2% do

not apply during the first five years of eligibility. This extrapolation may

actually overstate the share of eligible people who apply for expungement,

because some denials are based on ineligibility.

145

In section C, we discuss reasons for this poor uptake. But, first, how

confident can we be in our principal estimate? In particular, given that

our sample is both overinclusive and underinclusive in some ways, is

our main uptake estimate for this sample a good proxy for expungement

uptake for the full eligible population during these years?

146

To inform

this question, the remaining rows of Table 1 present estimates based on

alternative assumptions and sample definitions.

Given that we find very

low uptake, we use the term “conservative” below to refer to assump-

tions that likely lead to an overestimation of uptake.

First, in Rows 2 and 3 of Table 1, we alter our timing assumption

for the unknown-date expungements. Ninety such cases appear in this

sample. Our initial assumption that their timing distribution mirrors

that of known-date expungements is conservative; there is a strong rea-

son to believe that these expungements occur disproportionately after

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

145

Telephone Interview with Tracey Brame, Associate Dean, Cooley Law Sch. (Mar. 11, 2019) (on

file with authors).

146

In a similar study on uptake rates in the context of food stamp programs, the authors note

that “program takeup rates estimated from social science data sets with limited asset measures can

easily be mismeasured, and are likely to be too low.” Beth Osborne Daponte et al., Why Do Low-

Income Households Not Use Food Stamps? Evidence from an Experiment, 34 J.

ESOURCES

612, 622 (1999); see also id. at 622–24.

2490 HARVARD LAW REVIEW [Vol. 133:2460

May 2011 (and therefore after the five-year period), when MSP stopped

changing the disposition dates in their records. We estimate bounds on

the possible scope of any error from this assumption. If we assume that

none of the unknown-date expungements occurred within five years of

eligibility, or, instead, we apply the most conservative assumption (that

they all occurred within five years), the resulting uptake estimates range

from only 5.7% to 6.7%. Because our original approach is conservative,

we assume that the correct uptake rate (if all other assumptions are

valid, that is) lies somewhere between 5.7% and 6.5%.

In Row 4, we add to the sample all cases involving incarceration of

up to one year, testing the effect of our exclusion of these observations.

We assume that these individuals served their full term of incarceration,

commencing on the date of sentencing. This assumption is also con-

servative because it ignores credit for time served and possible early

release; it thus errs on the late side in estimating the release date and

may include slightly more than five years of eligible time in the “five-

year uptake” estimate. After adding these cases to the sample, the up-

take rate drops to 5.4%, indicating that people who have been sentenced

to incarceration receive expungements at lower rates than other eligible

people. Thus, focusing on nonincarceration cases has the effect of in-

creasing our main uptake estimate. We do not include a five-year esti-

mate for those serving longer incarceration terms because we do not

have a sufficient follow-up period for most of these cases. However,

given the evidence that those who were incarcerated for relatively short

terms have a lower uptake rate, it seems likely that including individuals

with longer incarceration terms would only lower the rate further.

There are a few uncertainties about our uptake estimates that we

cannot address directly with our data. In particular, we have no records

Specification

Five-Year

Uptake

Sample Size

(n)

1. Main 6.55% 9103

2. Expungement Date Unknown: Lower Bound 5.73% 9103

3. Expungement Date Unknown: Upper Bound 6.72% 9103

4. Including Incarcerated (

≤

1 year)

5.44% 14,223

ABLE

1. U

PTAKE

ATES

Receipt of Expungement Within Five Years of Eligibility

Notes: The uptake sample (n=9103) consists of all individuals who have a first conviction on a single

criminal count in Michigan for an expungement-eligible crime with a sentencing date between

January 1999 and May 2001, and who were not incarcerated, were not reconvicted for any crime

within five years of sentencing, and did not have an out-of-state driver’s license. The sample for Row

4 (n=14,223) adds individuals who were incarcerated for their expungement-eligible offense for up to

one year and were released by May 2001.

2020] EXPUNGEMENT OF CRIMINAL CONVICTIONS 2491

of out-of-state or federal convictions, and so our “eligible” pool probably

includes some people who are not in fact eligible due to such convictions.

Our exclusion of people with out-of-state licenses mitigates this problem

but does not eliminate it, as people can move or commit crimes across

borders. Including ineligible cases in the eligible pool will produce an

underestimate of the uptake rate.

However, back-of-the-envelope calculations suggest that this prob-

lem is minor. The Census Bureau estimates that about 1.8% of lower-

income Midwestern households (the best available proxy for our sample)

move across state lines each year.

147

When we apply this rate of attrition

to our sample and extrapolate from the within-Michigan reconviction

patterns in our data, we deduce that we might fail to observe disquali-

fying reconvictions within five years for about 1% of the purportedly

eligible population.

148

We have a similar problem with potentially dis-

qualifying prior adult convictions for people who might have previously

lived elsewhere; this problem might be twice as large or so as that of

subsequent convictions, given the age distribution of our sample. It is

also possible for Michigan residents to commit crimes in other states,

although the great majority of crimes are committed very close to home;

we estimate that we might miss prior or subsequent cross-border con-

victions for perhaps another 7% or 8% of the sample.

149

All in all, miss-

ing out-of-state convictions likely cause us to overstate the size of the

eligible group by no more than 10% to 12%, and therefore to understate

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

147

See Geographical Mobility:

2017

2018

tbl.9, U.S. C

ENSUS

UREAU

(Nov. 2018),

https://www2.census.gov/programs-surveys/demo/tables/geographic-mobility/2018/cps-2018/tab09.

xls [https://perma.cc/69NN-JP39]. 1.8% is based on households with incomes at $34,999 per year

or below. Id. If we use individual-level Census data instead, the interstate mobility estimate is

even lower — closer to 1.5%. See id.

148

To make this calculation, we assume that, per the state issuing their driver’s licenses, our

sample members were Michigan residents at the time of their convictions. The mobility estimate

implies that 7.3% of our sample would have moved away from Michigan after five years; for that

subgroup, we would miss on average the last 2.5 years of their criminal history. In our sample,

those who are not reconvicted within 2.5 years have a reconviction rate of 13% over the next 2.5

years; 13% times 7.3% is about 1%.

149

A large body of research finds that the average distance from an individual’s home to the

location of their crime is around one or two miles. See Michael Townsley & Aiden Sidebottom, All

Offenders Are Equal, but Some Are More Equal than Others: Variation in Journeys to Crime

Between Offenders, 48 C

RIMINOLOGY

897, 899–900 (2010) (reviewing literature). One piece of

evidence that most crimes are committed in-state comes from our own data: among arrestees with

driver’s licenses, 96% are from Michigan. If we make the conservative assumption that all of the

balance are out-of-state residents (not just people with old licenses from a state of previous resi-

dence) and we assume conversely that about 96% of the recidivism committed by Michigan resi-

dents is in-state (not a necessary inference but a plausible approximation), it would imply that we

overlook reconvictions before five years in about 2.7% of our sample. We could also miss prior

convictions over a somewhat longer period (eleven years on average in our sample), some of which

would presumably have involved the same individuals.

2492 HARVARD LAW REVIEW [Vol. 133:2460

the uptake rate by less than one percentage point. Meanwhile, the fed-

eral conviction problem is likely small enough to be ignored because

federal convictions are extremely rare relative to state convictions.

150

In addition, our eligible pool of people with records is defined by the

parameters of the data query that MSP conducted on our behalf, which,

as we discuss in section I.D, is itself limited by the list of statutory of-

fense codes of the eligible crimes that we provided.

151

This source of

error likely cuts in the opposite direction, causing us to overestimate

uptake, because the largest category of legally eligible offenses that we

exclude from the list are petty offenses that we know are very common

first offenses in Michigan but are almost never expunged.

152

If we were

to include them, the uptake rate we estimate would be lower. Although

we inadvertently omit some additional observations due to irregular for-

matting of statutory codes, these exclusions are very likely effectively

random — that is, there is no reason to expect that the observations

with irregular formatting have either higher or lower uptake rates com-

pared to the rates of those we include in the sample.

Overall, we are confident that our main five-year uptake estimate of

6.5% is quite accurate for our sample, and it is in the ballpark of the

correct figure for the true population of eligible individuals in Michigan

during the years in question (2004–11); our best guess is that the true

uptake rate is probably lower. While the lack of data from out-of-state

and federal courts may bias our estimate downward, these biases cannot

be very large. Meanwhile, the assumptions we make about the missing

cases and the unknown expungement dates are designed to bias our up-

take estimate upward, and our constraints on the statutory code and the

exclusion of cases with incarceration terms almost surely do so as well. In

any event, the substantive story on uptake from our data is clear: very

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

150

The federal district courts in Michigan sentenced 1233 people for felonies or nonpetty misde-

meanors in 2017. U.S.

ENTENCING

OMM

’

, S

TATISTICAL

NFORMATION

ACKET

ISCAL

EAR

2017: S

TATE

ICHIGAN

4 tbl.2, https://www.ussc.gov/sites/default/files/pdf/

research-and-publications/federal-sentencingstatistics/state-district-circuit/2017/mi17.pdf [https://

perma.cc/PB74-PYQW]. By comparison, in 2017, caseload figures suggest that the Michigan state

courts entered roughly 340,000 criminal convictions of any sort. See M

ICH

. C

OURTS

, 2017 C

OURT

ASELOAD

EPORT

11, https://courts.michigan.gov/education/stats/Caseload/2017/Statewide.pdf

[https://perma.cc/3EXJ-9YLF]. This figure is based on adding felony and misdemeanor (including

traffic) guilty pleas and trial verdicts in circuit and district courts, under the rough estimate that

80% of trial verdicts are convictions. The Michigan Department of Corrections reported approxi-

mately 47,000 people convicted of felonies entering corrections in 2016. See Kahryn Riley, Coping

with the Growing Number of Felons in Michigan, M

ACKINAC

FOR

. P

’

(Aug. 27,

2018), https://www.mackinac.org/coping-with-the-growing-number-of-felons-in-michigan

[https://perma.cc/C5UR-T7UV]. Although these numbers are not precisely comparable, the differ-

ence in scale is obvious and suggests that only a very small percentage of the single-Michigan-

conviction individuals in our sample are likely to have a federal record, making a negligible differ-

ence in our uptake estimates.

151

See supra note 128.

152

See id.

2020] EXPUNGEMENT OF CRIMINAL CONVICTIONS 2493

few of those who are legally eligible for expungements receive them

within five years of becoming eligible. Indeed, even if the true five-year

rate were 10% — substantially higher than any version of our esti-

mate — it would still be strikingly low.

What about after five years? Unfortunately, we cannot directly esti-

mate a longer-term uptake rate because of the time range of our data on

the eligible-group sample. However, we do have data on the full uni-

verse of expungement recipients in Michigan up through 2011. Among

nonincarceration cases in this more inclusive sample, at the time of ex-

pungement receipt, 44% have an elapsed time since eligibility of more

than five years.

153

Assuming the same pattern holds for the cases in our

eligible sample, extrapolating from our main uptake estimate would pre-

dict a lifetime uptake rate of 11.6% for those not sentenced to incarcer-

ation.

154

Even this rate means that 88% of those eligible will never

secure relief — and those who receive expungements after many years

still forfeit many years of potential benefits.

B. Who Receives Expungements?

Only a small minority of people with records in Michigan are eligible

for expungements,

155

and only a small minority of this eligible popula-

tion in fact receives them. What distinguishes eligible individuals who

receive an expungement from those who do not? To develop insight into

possible reasons for the low uptake rate we identify and to develop hy-

potheses about policies that might increase it, we assess the characteris-

tics of the individuals who do receive expungements and their cases, and

we investigate which of those characteristics are usefully predictive of

an eligible person receiving an expungement.

In Table 2, we provide descriptive statistics of the relevant popula-

tions for this exercise. The statistics in Column 1 refer to the complete

population of all Michigan expungement recipients with known dates

prior to June 2011 (when the law changed). Columns 2 and 3 report

summary statistics for the main sample we use in our uptake-rate anal-

ysis on display in Table 1. Column 2 covers individuals within that

sample who did receive expungements within five years of becoming

eligible, and Column 3 covers the entire uptake sample (most of whom

did not receive expungements). Columns 2 and 3 can be usefully com-

pared to one another to shed light on how expungement recipients differ

from the broader eligible group. We note that some of the figures for

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

153

See infra fig. 1 and accompanying text.

154

That is, we simply assume that the 6.5% share of the uptake sample who receive expunge-

ments within five years constitute 56% of those who will eventually receive expungements in their

lifetimes: 0.065/0.56=0.116.

155

Establishing the number of people with criminal records is notoriously difficult. Expungement

advocates consistently emphasized to us in interviews that very few people are actually eligible, which

is intuitive, given the strict limitations in terms of numbers of convictions (especially during the pre-

2011 period). See infra notes 204–212 and accompanying text for further analysis.

2494 HARVARD LAW REVIEW [Vol. 133:2460

(1) (2) (3)

Expungement

Recipients

(complete

population)

Expungement

Recipients

(uptake sample

only)

Eligible for

Expungement

(uptake sample

only)

Individual Characteristics

Race

Black 30.4% 39.8% 30.9%

White 65.6% 54.4% 67.0%

Other 4.0% 5.7% 2.1%

Mean Age at Conviction (years) 26.50 29.64 30.43

Male 60.9% 54.2% 64.9%

Employed (at eligibility) 66.0% 67.0% 64.6%

Employed (before expungement) 60.6% 53.8%

Quarterly Wages (at eligibility) $4,968 $5,013 $6,387

Quarterly Wages (before expungement) $5,160 $3,463

Case Characteristics

Felony 44.2% 48.9% 30.1%

Crime Type

Violent 14.9% 28.0% 26.2%

Drug 13.5% 14.0% 27.0%

Property/Economic 53.2% 39.8% 31.6%

Other 18.4% 18.2% 15.2%

Incarcerated 28.6%

Incarcerated ( > 1 year) 1.9%

Median Elapsed Time (years) 9.6

Number of Observations 22,004 522 9103

ABLE

2. D

EMOGRAPHIC

HARACTERISTICS

XPUNGEMENT

ECIPIENTS

Notes: Column (1) includes the complete population of Michigan known-date expungement recipients through

May 2011. Column (2) comprises all individuals in the uptake sample who received an expungement. The uptake

sample (n=9103) in Column (3) consists of all individuals who have a first conviction on a single criminal count in

Michigan for an expungement-eligible crime with a sentencing date between January 1999 and May 2001, and

who were not incarcerated, were not reconvicted for any crime within five years of sentencing, and did not have

an out-of-state driver’s license. The phrases “at eligibility” and “before expungement” refer to the quarter in

which an individual first becomes eligible for expungement and the quarter before expungement, respectively.

“Median Elapsed Time” is the number of years between sentencing and expungement receipt.

2020] EXPUNGEMENT OF CRIMINAL CONVICTIONS 2495

expungement recipients look a little bit different in Columns 1 and 2;

this may arise from the various constraints on the Column 2 sample.

While expungement recipients have diverse personal and case char-

acteristics, some patterns stand out in the data. Relative to the criminal

convictions of the entire eligible pool (as shown in Column 3), convic-

tions that are actually expunged (as shown in Column 2) are much more

likely to be felonies; they are more likely to be property offenses, less

likely to be drug crimes, and roughly equally likely to be violent offenses.

Although expungements are not limited to minor crimes under the law

we study, most expungement recipients were not sentenced to incarcer-

ation for their expunged offense. In fact, among all expungement recipi-

ents historically, only 28.6% were incarcerated for any time at all, and only

1.9% were incarcerated for more than one year.

156

Finally, compared to all

eligible individuals, expungement recipients are less likely to be male and

more likely to be black.

IGURE

1. T

IME

LAPSED

ETWEEN

ENTENCING

AND

XPUNGEMENT

Notes: The bars show the percentage of recipients who received their expunge-

ment before the end of the year(s) indicated and who did not receive it in an

earlier period. The sample consists of the complete population of Michigan

known-date expungement recipients through May 2011 (n=22,004).

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

156

Because our main uptake sample is defined to exclude those who were incarcerated, we do

not provide comparison figures in Columns 2 and 3.

10%

15%

20%

25%

Year 6 Years 7–8 Years 9–10 Years 11–15 Year s 16+

Year 6

Year s 7–8

Year s 9–10 Year s 11–15

Year s 16+

10%

15%

20%

25%

2496 HARVARD LAW REVIEW [Vol. 133:2460

We report employment measures for both the quarter that the recip-

ient became eligible for the expungement and the quarter before actually

receiving it, which may offer suggestive evidence about whether em-

ployment changes influence the timing of expungement applications.

157

Here, we discern notable differences between the uptake sample and the

full sample of expungement recipients: in the former, employment rates

and wages are far lower just before expungement than they were at the

time of eligibility, whereas in the full sample employment is modestly

lower and average wages are actually higher. This difference may re-

flect the fact that the uptake sample is narrowly defined temporally, and

the five-year observation period for the uptake outcome (ending between

2009 and 2011 for all members of the sample) includes a major economic

crash;

158

for this reason, these numbers may not actually suggest any

broader relationship between employment setbacks and expungement

timing. We explore that potential relationship in Part IV.

As Column 1 in Table 2 shows, at the time of expungement receipt,

the median time elapsed since sentencing is 9.6 years.

159

In Figure 1,

we show more detail on the distribution of elapsed time, focusing on

those not sentenced to incarceration, all of whom would have become

eligible after five years had transpired. Figure 1 makes clear that there

is wide variation in the time lag between eligibility and expungement

receipt. On the one hand, the single year with by far the most expunge-

ments granted is the first year after the five-year waiting period expires

(that is, the sixth year since sentencing — labeled “Year 6”), roughly a

quarter of the total. This flurry of expungement activity suggests a pent-

up demand effect: some expungement recipients likely anticipate becom-

ing eligible ahead of time and apply more or less as soon as they can.

160

The monthly rate, not shown in the graph, peaks at four months after

the date of eligibility; this too is consistent with the pent-up demand

theory, since expungements typically take a few months to process.

161

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

157

We discuss our wage data in more detail infra Part IV, pp. 2523–43.

158

See generally Chart Book: The Legacy of the Great Recession, C

UDGET

& P

’

RIORITIES

(June 6, 2019), https://www.cbpp.org/research/economy/chart-book-the-legacy-of-the-

great-recession [https://perma.cc/9K74-N79U] (reviewing the economic consequences of the reces-

sion of 2007–09).

159

We do not show this figure in Column 2 because that sample is already constrained to include

only those receiving expungements within five years of becoming eligible.

160

Cf. Jennifer Gerarda Brown, Competitive Federalism and the Legislative Incentives to

Recognize Same-Sex Marriage, 68 S.

. L. R

. 745, 754–59 (1995) (analyzing the potential ben-

efits of same-sex marriage legalization using a model assuming pent-up demand).

161

See Telephone Interview with Michael Kiehne, Attorney, Mich. Legal Help (Feb. 14, 2019)

(estimating that four to six months is typical for one large Michigan county); Telephone Interview

with Chioke Mose-Telesford, Deputy Dir. of Workforce Dev. for the City of Detroit (Feb. 27, 2019)

(explaining that Detroit’s legal assistance program has recently reduced its wait time to two months

through concerted efforts to accelerate the process but that previously 160 to 190 days would have

been typical). MSP advised us that typically about six to ten weeks elapse after MSP runs the

criminal background check and returns a report deeming the individual eligible before the expunge-

ment is granted by the judge. Email from Ted Kilvington, supra note 125.

2020] EXPUNGEMENT OF CRIMINAL CONVICTIONS 2497

The annual rate declines continuously thereafter (the last two bars in

the graph are taller only because they represent periods of five or more

years). Still, more than 44% of expungements take place after more than

five years of eligibility (year 11 or beyond), and almost 25% occur after

more than ten years of eligibility (year 16 or beyond).

IGURE

2. U

PTAKE

ATE

OUNTY

ICHIGAN

)

Notes: The bars indicate the percentage of expungement recipients by county. The sample

consists of all individuals who have a first conviction on a single criminal count in

Michigan for an expungement-eligible crime with a sentencing date between January 1999

and May 2001, and who were not incarcerated, were not reconvicted for any crime within

five years of sentencing, and did not have an out-of-state driver’s license (n=9103).

In Figure 2, we present uptake rates by county for the six most pop-

ulous counties in Michigan (in descending order of population from the

left), and for Michigan’s other seventy-seven counties combined. The

figure reveals considerable local variation. Michigan’s largest counties

are primarily urban and suburban; five of the six (all but Kent) are in

Southeastern Michigan, within an hour of Detroit. Nevertheless,

Oakland County’s rate is more than triple that of Genesee County,

which it borders. All of the larger counties have higher uptake rates

than the rest of the state combined. However, by 2014, courts in every

county in Michigan had granted at least one expungement, and at least

ten had been granted in seventy-seven out of eighty-three counties (with

the remaining six all being among Michigan’s least populous).

Many of the individual and case characteristics listed in Table 2 are

correlated with one another, and with the county as well, which makes

10%

12%

14%

Wayne Oakland Macomb Kent Genesee Washtenaw Other

10%

12%

14%

Wayne

Oakland

Macomb

Kent

Genesee Washtenaw

Other

2498 HARVARD LAW REVIEW [Vol. 133:2460

(1) (2) (3) (4) (5)

Male 0.663*** 0.654*** 0.671*** 0.693*** 0.692***

(0.064) (0.067) (0.058) (0.051) (0.052)

Black 0.989 0.825 0.846⁺ 0.936 0.948

(0.109) (0.098) (0.086) (0.079) (0.083)

Age at Conviction (years) 0.988** 0.986** 0.986*** 0.988*** 0.990**

(0.004) (0.004) (0.004) (0.003) (0.003)

Felony 2.552*** 2.129*** 2.211*** 2.262*** 2.446***

(0.284) (0.262) (0.227) (0.194) (0.219)

Offense Type

Violent Offense 0.464*** 0.545*** 0.525*** 0.554*** 0.571***

(0.068) (0.085) (0.070) (0.064) (0.069)

Drug Offense 1.243 1.487** 1.306* 1.400** 1.468***

(0.170) (0.221) (0.171) (0.150) (0.164)

Public Order/Other Offense 1.272* 1.297⁺ 1.331* 1.387*** 1.449***

(0.153) (0.175) (0.150) (0.128) (0.140)

Incarcerated (

≤

1 year) 0.553*** 0.676*** 0.672***

(0.057) (0.057) (0.058)

Wages (last quarter) 1.004

(0.007)

Employed (last quarter) 1.768***

(0.158)

Lost Wages (last quarter) 2.268***

(0.196)

Baseline Odds 0.077*** 0.187*** 0.120* 0.006*** 0.011***

(0.013) (0.048) (0.127) (0.006) (0.011)

Controls

Conviction Year Effects

XXXXX

County Effects

XXXX

Elapsed Time (years)

Number of Observations

9080 8406 13,465 515,964 474,670

ABLE

3. D

ETERMINANTS

XPUNGEMENT

ECEIPT

Notes: The results of these logistic regressions are reported as odds ratios. The outcome variable is an indicator

for expungement receipt. Columns (1) to (3) analyze expungement receipt within the first five years of eligibility.

Columns (1) and (2) consider individuals in our uptake sample; Column (3) adds individuals incarcerated for up

to one year. Columns (4) and (5) analyze expungement receipt in a given quarter using a panel-regression

framework; observations are person-quarters and standard errors (reported in parentheses) are clustered at the

person level.

⁺

, *, **, and *** represent significance at the 10%, 5%, 1%, and 0.1% level, respectively. The base

category for the offense type is property/economic crime.

Receipt:

Within Five Years of Eligibility

Receipt:

Current Quarter

2020] EXPUNGEMENT OF CRIMINAL CONVICTIONS 2499

inference from this table fraught. It could be, for instance, that the

higher uptake rate for black individuals can be entirely explained by

county-to-county variation; indeed, uptake rates are generally lower in

more rural, predominantly white Michigan counties. For this reason, in

Tabl e 3, we turn to regression analysis to test which individual and case

characteristics remain predictive of expungement uptake when we hold

other characteristics constant. We use logistic regression, a common ap-

proach for analyzing binary outcomes like expungement receipt.

162

present our results as odds ratios, which represent a multiplier of the

odds of the expungement occurring. An odds ratio greater than one

means that the variable is associated with increased odds of expunge-

ment receipt when other variables are fixed.

163

In Columns 1 through 3

of Table 3, we study the determinants of whether an individual receives

an expungement within five years of becoming eligible — the same out-

come we report in Table 1.

164

We estimate the following equation:

logit



𝐹𝑖𝑣𝑒𝑌𝑒𝑎𝑟𝑈𝑝𝑡𝑎𝑘𝑒





= 𝛼 + 𝛽



𝑀𝑎𝑙𝑒



+ 𝛽



𝐵𝑙𝑎𝑐𝑘



+ 𝛽



𝐶𝑜𝑛𝑣𝐴𝑔𝑒



+ 𝛽



𝐹𝑒𝑙𝑜𝑛𝑦



+ 𝛾



𝐶𝑟𝑖𝑚𝑒𝑇𝑦𝑝𝑒



+ 𝛿



𝐶𝑜𝑛𝑣𝑌𝑒𝑎𝑟



+ 𝜖



The outcome variable 𝐹𝑖𝑣𝑒𝑌𝑒𝑎𝑟𝑈𝑝𝑡𝑎𝑘𝑒 is an indicator for whether the

individual receives an expungement within five years of eligibility. 𝑀𝑎𝑙𝑒

and 𝐵𝑙𝑎𝑐𝑘 depict the individual’s reported gender and race; all nonblack

races are combined into one base category, which is 94% white.

𝐶𝑜𝑛𝑣𝐴𝑔𝑒 is the individual’s age in years at the time of the expungement-

eligible conviction. 𝐶𝑜𝑛𝑣𝑌𝑒𝑎𝑟 represents the year of that conviction.

𝐹𝑒𝑙𝑜𝑛𝑦 is an indicator for whether the crime of conviction is a felony,

and 𝐶𝑟𝑖𝑚𝑒𝑇𝑦𝑝𝑒 is a set of indicators for the type of crime: drug-related,

violent, and other, with property/economic crimes as the omitted refer-

ence category. Column 1 presents our baseline results. In Column 2

(and subsequent columns), we also control for county effects. In Column

3, we include individuals incarcerated for their crime for up to one year

and add an indicator for whether the individual was incarcerated.

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

162

However, linear (ordinary least squares) regression analysis produces substantively indistin-

guishable results when compared to the marginal effects from the logistic regressions.

163

Odds have a technical definition, and while they are not the same as probability, higher odds

do mean a higher probability. When probabilities are quite low (as they are here), the odds ratio is

a good rough approximation of the probability ratio. The odds that an event X (for example, an

expungement) occurs are equal to the probability that X occurs divided by the probability that X

does not occur. An odds ratio refers to the multiplier of these odds that applies with a one-unit

change in some explanatory variable.

164

For Columns 1 and 2 of Table 3, the sample is the same one we use for the main uptake

estimate in Table 1 (with constraints described in section II.A), and for Column 3, we add cases

with incarceration up to one year (the same sample we use in Table 1, Row 4). Our results are

robust to estimating Columns 1 through 3 on the same sample (in two different ways). The findings

in Column 4 are also unaffected by limiting the analysis to Column 5’s smaller sample. These

results are available upon request.

2500 HARVARD LAW REVIEW [Vol. 133:2460

The regression results in Column 1 show that, other things equal,

uptake rates are much higher among women and among people who

were relatively young at the time of their conviction. The impression

given in Table 2 of a higher uptake rate for black individuals completely

disappears in these regressions, implying that this racial difference is ac-

tually explained by other correlated characteristics. As for crime attri-

butes, the regressions all confirm that people with felony convictions

have, other things equal, more than twice the odds of receiving expunge-

ments. By contrast, people who have been incarcerated (even for short

periods) and people convicted of violent crimes have only about half the

odds of expungement receipt as otherwise similar individuals do; this pat-

tern is not apparent in the summary statistics in Table 2. A possible

theory for these results is that felony convictions make individuals much

more likely to apply (because felonies can trigger a wider range of col-

lateral consequences), but judges are less likely to grant expungements

in cases involving more serious or violent convictions.

165

Taken to-

gether, the pattern indicates that the people most likely to receive ex-

pungements are those with relatively minor felony convictions (espe-

cially women) — namely, felonies not resulting in jail time.

The regression results we present in Columns 4 and 5 also analyze

expungement receipt, but they replace the five-year uptake indicator

with a different outcome variable: whether an expungement is received

in any particular quarter. These analyses help us to understand the in-

fluences on expungement probability that vary over time — and in par-

ticular, to ask whether an individual’s immediate employment history

(such as a recent job loss) drives the decision to apply for an expunge-

ment. The dataset we use in these regressions has a panel structure,

meaning it contains separate observations for each quarter for each in-

dividual. The sample covers most of the people we include in the Col-

umn 3 analysis, depending on whether we have linked wage data for

them. The regression we estimate takes the following form:

logit



𝑆𝑒𝑡𝐴𝑠𝑖𝑑𝑒𝑅𝑒𝑐𝑒𝑖𝑝𝑡





= 𝛼 + 𝛽



𝑀𝑎𝑙𝑒



+ 𝛽



𝐵𝑙𝑎𝑐𝑘



+ 𝛽



𝐶𝑜𝑛𝑣𝐴𝑔𝑒



+ 𝛽



𝐹𝑒𝑙𝑜𝑛𝑦



+ 𝛽



𝐼𝑛𝑐𝑎𝑟𝑐𝑒𝑟𝑎𝑡𝑒𝑑



+ 𝛾



𝐶𝑟𝑖𝑚𝑒𝑇𝑦𝑝𝑒



+ 𝛿



𝐶𝑜𝑛𝑣𝑌𝑒𝑎𝑟



+ 𝜇



𝐶𝑜𝑢𝑛𝑡𝑦



+ 𝜃𝐸𝑚𝑝𝑙𝑃𝑟𝑒𝑣𝑄𝑡𝑟



+ 𝜋𝑊𝑎𝑔𝑒𝑃𝑟𝑒𝑣𝑄𝑡𝑟



+ 𝜌𝑇𝑖𝑚𝑒𝐸𝑙𝑎𝑝𝑠𝑒𝑑



+ 𝜖



The specifications underlying Columns 4 and 5 are similar to the

ones in the previous columns but also include variables that change over

time for a given individual. Column 4 includes whether an individual

was employed in the previous quarter (that is, whether any wages are

reported, which we discuss in greater detail in Part IV) and their re-

ported wages in the previous quarter, as well as the number of years

that have elapsed since the expungement-eligible conviction. We focus

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

165

See infra p. 2507.

2020] EXPUNGEMENT OF CRIMINAL CONVICTIONS 2501

on the role of the previous quarter in part because there is likely to be

at least a one-quarter lag between expungement application and receipt,

and we are interested in understanding the motivation for the applica-

tion. In addition, lagging this variable avoids the problem of reverse

causation: employment status and wages in the quarter of expungement

receipt could be affected by the expungement itself.

In Column 5, we focus on whether the individual recently experi-

enced an employment setback, which we proxy for with an indicator for

whether the individual’s wages had, in the quarter preceding the receipt

of the expungement, dropped by at least 20% compared to a year earlier

(denominated “Lost Wages” in Table 3). Our framework allows us to

assess the effects of wage loss on expungement receipt in a way that is

not confounded by the recession’s timing. The whole sample is affected

by the same economic trends, and the controls for year of conviction and

years since conviction (which together produce the calendar year) also

absorb those effects. The regression thus allows us to investigate whether,

among individuals subjected to the same broader economic conditions,

personal employment setbacks affect expungement probability.

We find, on the one hand, that being employed is a very strong pos-

itive predictor of expungement receipt in a particular quarter, increasing

the odds by a factor of 1.77 (Column 4). On the other hand, a recent

wage loss is an even stronger predictor, increasing the odds of expunge-

ment receipt by a factor of 2.27 (Column 5). Although these results are

in tension with each other, they each have plausible explanations: em-

ployed people may generally be more likely to have the resources, infor-

mation, and money to pursue an expungement; but people who have

experienced a recent employment setback may have more motivation to

pursue a new job. These results also help to inform our interpretation

of the wage and employment effects of expungement that we estimate

below; we refer back to these uptake findings in Part IV.

C. What Explains the Uptake Gap?

Given the many life disadvantages that come with a criminal record,

one might expect that most people who have records would jump at the

chance to leave their record behind. Yet strikingly few do so. Why?

Our quantitative data shed some light on this question but cannot really

answer it. To complement our data analysis, we sought out the insights

of experienced Michigan expungement lawyers and other advocates for

people with records, many of whom are actively involved with outreach

efforts (such as “expungement fairs”) that aim to encourage and facilitate

expungement applications.

166

We asked them for their perceptions of

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

166

See Telephone Interview with Anonymous Legal Aid Lawyer (Feb. 22, 2019); Telephone In-

terview with Miriam Aukerman, Senior Staff Attorney, ACLU of Mich. (Feb. 16, 2019); Telephone

2502 HARVARD LAW REVIEW [Vol. 133:2460

the underlying mechanisms of limited uptake. Our discussions consist-

ently pointed to a set of likely explanations.

. Lack of Information. — Every advocate we spoke to mentioned

this concern, and many thought it was the single most important uptake

barrier. Most people with records, even if they are eligible, lack the

information they need to pursue an expungement. As Tracey Brame,

who runs a law clinic that handles expungement cases, explained: “A lot

of people have absolutely no idea that they can do this.”

167

Many do

not know that the expungement law exists at all. Others may have a

vague idea that expungement is possible, but they do not know that they

are eligible or they are unfamiliar with what they need to do to pursue

one (or how to find out). Expungement law is complicated and not easy

for a layperson to grasp.

168

Many people do not understand their own

criminal records — for example, they may not know that their decision

to plead guilty to a traffic offense produced a criminal conviction. Sev-

eral advocates recounted running expungement fairs in which, even

though promotional materials identify the key eligibility requirements,

a substantial majority of those who turn up learn sometime after they

arrive that they are ineligible and walk out exasperated. It is certainly

plausible that there are just as many people who are eligible who assume

they are not. After all, the great majority of people with records are

ineligible for expungement, and in communities where many people

have records, to the extent that people know about expungement at all,

frustration with its stringent restrictions is common.

169

The shared im-

pression that “nobody is eligible” may be entrenched, including among

those who are actually eligible.

. Administrative Hassle and Time Constraints. — Obtaining an

expungement requires a nontrivial amount of organization, effort, and

time. The process is drawn out and requires patience and ongoing re-

solve. A would-be applicant, after overcoming the basic informational

hurdles we describe above, must track down the official application

form, which is available on the website of the State Court

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

Interview with Tracey Brame, supra note 145; Email from Josh Hoe, Co-Chair, Policy and Educ.

Comm., Nation Outside, to Sonja Starr, Professor of Law, Univ. of Mich. Law Sch. (Feb. 13, 2019,

12:25 PM) (on file with authors); Telephone Interview with Michael Kiehne, supra note 161; Tele-

phone Interview with Chioke Mose-Telesford, supra note 161; Email from John Shea, Private Prac-

titioner, to Sonja Starr, Professor of Law, Univ. of Mich. Law Sch. (Feb. 15, 2019, 1:33 PM) (on file

with authors); Email from Kim Thomas, supra note 133.

167

Telephone Interview with Tracey Brame, supra note 145.

168

This is particularly so given that the target population includes many people with significant

socioeconomic challenges, including limited literacy. Telephone Interview with Michael Kiehne,

supra note 161.

169

See Email from Josh Hoe, supra note 166; Telephone Interview with Chioke Mose-Telesford,

supra note 161.

2020] EXPUNGEMENT OF CRIMINAL CONVICTIONS 2503

Administrative Office.

170

Occupying two-thirds of the second page of

that form, in fairly small font, is the following set of instructions:

1. Determine whether you are eligible to apply to have your conviction set

aside according to MCL 780.621. You must complete a separate application

for each conviction if you are applying to have more than one conviction

set aside [as is allowed under the current statute].

2. Find out the exact date of conviction and the charge from the court. Get

a certified copy of the adjudication and attach it to your application.

3. Swear to the truth of the statements in this application and then sign it

in the presence of the court clerk or a notary public.

4. Make four copies of all attachments and this application. Take all copies

to the court clerk.

5. Depending on local practice, the clerk of the court may set a hearing date

at the time of filing. If a hearing date is set at the time of filing, the clerk

of the court will complete the Notice of Hearing.

6. Go to the local law enforcement agency for a fingerprint card and get

fingerprinted on the applicant card (RI-8) . . . .

7. Make out a money order or check to the State of Michigan for the appli-

cation. . . .

8. Mail a copy of the application packet, application fee, and the fingerprint

card to the Michigan State Police . . . .

9. Mail a copy of the application packet to the Attorney General of the State

of Michigan . . . .

10. Mail a copy of the application packet to the correct prosecuting official

where the conviction occurred (county, city, or township) . . . .

11. On both copies of the application, fill in the Proof of Service on the back

of the form. After you fill out and sign the Proof of Service, mail or take

one of the remaining application packets with the completed Proof of

Service to the court. Keep the other copy for your records . . . .

171

It is easy to see how this list would be daunting and confusing to

potential applicants (and these steps do not end the process; a court

hearing follows, usually many weeks later).

172

Advocates confirm that

this is so; everyone we interviewed independently mentioned the admin-

istrative burdens facing applicants. As many elaborated, people with

records are usually struggling with a variety of life challenges. Taking

time away from work and childcare responsibilities to go to a police

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

170

See M

ICH

. COUR

MC 227: A

PPLICATION

SIDE

ONVICTION

https://courts.michigan.gov/Administration/SCAO/Forms/courtforms/mc227.pdf

[https://perma.cc/EYV6-YJVF].

171

Id.

172

See id.; Common Questions About Setting Aside Adult Convictions (Expungements), M

ICH

EGAL

ELP

, https://michiganlegalhelp.org/self-help-tools/crime-traffic-and-id/common-

questions-about-setting-aside-adult-convictions-expungements [https://perma.cc/HYR8-9SB6]

(stating that it can take about six months or more for a conviction to be expunged).

2504 HARVARD LAW REVIEW [Vol. 133:2460

station to be fingerprinted, to make several trips to a courthouse, to find

a notary, and to mail all these materials to the right addresses may be

simply impossible, or at least difficult enough to be strongly discourag-

ing.

173

This is especially so when prospective applicants do not live near

the court that convicted them, where their expungements must be pro-

cessed and the hearing must take place. And, finally, one experienced

expungement lawyer at a public interest organization pointed out that

many of her clients are already overwhelmed with paperwork, such as

the forms associated with receiving public benefits.

174

. Fees and Costs. — Every expert we interviewed emphasized the

barriers erected by fees and other associated costs. The $50 expunge-

ment application fee cannot be waived, and it is not the only expense.

175

Michael Kiehne of Michigan Legal Help and Michigan Works estimated

that the total cost amounts to close to $100, including fingerprinting

($10–$20 at most local police stations), notary fees (up to $10), obtaining

a certified record of a conviction ($10–$12 if it is only one page), and

photocopies.

176

This does not include the costs of transportation and

possible loss of wages for time taken off work; Chioke Mose-Telesford

of Detroit’s Project Clean Slate emphasized that especially in a city with

poor public transit and low car ownership, the absence of transportation

options is a serious hurdle.

177

For applicants living in poverty, these accu-

mulated expenses are a serious financial impediment — consistent with

our finding that the unemployed are less likely to apply. In principle, one

could see the cost as an investment in future earnings; in Part IV, we

document average expected gains that would very quickly cover the

cost. But those without cash on hand may not have the liquidity or

ability to make such an investment or may be reluctant to do so when

the long-term benefits are speculative.

. Distrust and Fear of the Criminal Justice System. — For many

potential expungement applicants, their prior experience with the crim-

inal justice system may well have been among the worst experiences of

their lives. According to several advocates, this often amplifies the

daunting nature of the expungement process; individuals with records,

at least five years removed from an earlier criminal justice ordeal, may

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

173

This is consistent with the larger access-to-justice literature. The use of procedures that re-

quire travel to courthouses, for instance, can dramatically reduce participation. See Maximilian A.

Bulinski & J.J. Prescott, Online Case Resolution Systems: Enhancing Access, Fairness, Accuracy,

and Efficiency, 21 M

ICH

. J. R

ACE

& L. 205, 217–35 (2016); Prescott, supra note 136, at 38.

174

Telephone Interview with Anonymous Legal Aid Lawyer, supra note 166.

175

See M

ICH

COUR

, supra note 170 (calling the $50 fee a “required fee”); I Need a Fee Waiver

for Court, M

ICH

. L

EGAL

ELP

, https://michiganlegalhelp.org/self-help-tools/going-court/i-need-

fee-waiver-court [https://perma.cc/PCC3-FBK5] (not listing an expungement application fee as a

court fee that may be waived).

176

Telephone Interview with Michael Kiehne, supra note 161.

177

Telephone Interview with Chioke Mose-Telesford, supra note 161.

2020] EXPUNGEMENT OF CRIMINAL CONVICTIONS 2505

be strongly averse to returning to court or to a police station for any

reason.

178

This may be especially so if they expect the prosecutor or a

crime victim to come to court to contest an expungement petition, which

the law allows (and which sometimes happens).

179

In addition to fear

of the process itself, potential applicants may be pessimistic about the

likely outcome, even though most expungement applications are in fact

granted. As Kiehne explained:

Expungements are discretionary, and when you let people know that, they

tend to be pessimistic. We always try to tell people that many judges are

excited to grant these . . . . [But some] fear that it won’t work [for] them —

an issue based on their past experience with the courts. . . . [S]o it is hard to

get this information through.

180

A potential applicant who fears that any request for an expungement

will be denied may be particularly reluctant to undertake the effort, cost,

and stress of pursuing record-clearing opportunities.

181

. Lack of Access to Counsel. — Some of the obstacles above could be

overcome or rendered less discouraging with legal assistance. Although

expungement applications can be filed pro se, the process is far less diffi-

cult to navigate for an experienced attorney.

182

Yet, too often, none are

available. Criminal defense lawyers are typically long since out of touch

with their former clients by the time they become eligible for expunge-

ment, and most of the time defense lawyers do not advise their clients

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

178

See Email from Josh Hoe, supra note 166 (“Many formerly incarcerated and impacted people

have very little faith in judges, so there is also a suspicion of wasting their time going through the

process.”); Telephone Interview with Michael Kiehne, supra note 161. Alternatively, this fear and

apprehension may take the form of a stigma “friction.” Bhargava & Manoli, supra note 138, at

3490, 3505, 3512; see also D

AV ID

IBAR

, IZA W

ORLD

ABOR

, H

MPROVE

ARTICIPATION

OCIAL

SSISTANCE

ROGRAMS

6 (2014), https://wol.iza.org/uploads/

articles/104/pdfs/how-to-improve-participation-in-social-assistance-programs.pdf [https://perma.cc/

8APE-76MQ]; Jana Friedrichsen & Renke Schmacker, Fear of Stigmatization Prevents Individuals

from Claiming Benefits, 9 DIW

EEKLY

EPORT

215, 219 (2019); Robert Moffitt, An Economic

Model of Welfare Stigma, 73 A

. E

CON

. R

. 1023, 1024–27 (1983); David C. Ribar & Lauren A.

Haldeman, Changes in Meal Participation, Attendance, and Test Scores Associated with the Avail-

ability of Universal Free School Breakfasts, 87 S

. S

ERV

. R

. 354, 381 (2013).

179

See M

ICH

. C

OMP

. L

AWS

§ 780.621(11) (2019) (“[A]n opportunity shall be given to the attorney

general and to the prosecuting attorney to contest the [expungement] application . . . [and] [t]he

victim has the right to appear at any proceeding under this act concerning that conviction and to

make a written or oral statement.”).

180

Telephone Interview with Michael Kiehne, supra note 161.

181

A small survey of recipients of record clearance in California described respondents’ past

encounters with the court system as “consistently fraught with fear and shame.” Adams et al., supra

note 95, at 42. Notably, when respondents did go through with clearing their record, their experi-

ences were ultimately far more positive. See id. (noting that, given their past bad experiences,

respondents were often surprised by the respect they received from judges and district attorneys

during the clearance process).

182

See, e.g., Margaret (Peggy) Stevenson, Expungement: A Gateway to Work, C

LEARINGHOUSE

OMMUNITY

, Mar. 2015, at 1, 1–2 (“Clearing a record without an attorney is frequently difficult.

The absence of sufficient free legal help and the inability to pay for private attorneys mean that

many low-income people unnecessarily carry the burden of an expungeable conviction for years.”).

2506 HARVARD LAW REVIEW [Vol. 133:2460

about the possible prospect of an expungement five years in advance,

either. Paid attorneys are out of reach for most people with records.

183

And legal aid or pro bono attorneys with time on their hands are too

rare. Although there have been some recent improvements on this front

(which we discuss below), one legal aid lawyer with extensive expunge-

ment experience explained to us that many legal aid organizations are

simply overwhelmed with the many other needs they are asked to serve,

which makes it “hard to ask them to take on a whole new area.”

184

. Insufficient Motivation to Pursue Expungement. — Notwith-

standing the various hardships that tend to be associated with having

a criminal record, not everybody with a record is affected by it in the

same way. Virtually everyone with a criminal record would presumably

prefer not to have it, but not everybody necessarily experiences the re-

cord as a burden sufficient to motivate them to take on the investment

of time, money, and energy that an expungement demands. This may

be especially true for people with minor convictions — a theory sup-

ported by our finding that those with misdemeanors are less than half

as likely to obtain expungements as those with felonies are. It seems

implausible that judges would on average be less willing to grant ex-

pungements for misdemeanors, so the only credible explanation is that

people with misdemeanors are less likely to apply. This is intuitively

sensible. While misdemeanor records can certainly have nontrivial conse-

quences,

185

many collateral legal consequences apply only to felonies,

186

and some employment applications ask about only felony records.

187

It is

not surprising that those with felony records appear to be particularly

motivated to pursue expungement relief. In addition, motivation may

affect the timing of expungements; as Tracey Brame explained: “A lot of

people don’t prioritize it until [the conviction] is a problem,” such as

when they are trying to pursue an occupational license.

188

* * *

Our exchanges with practitioners and policymakers, who were con-

templating record clearing specifically when sharing their views with us,

produce a set of concrete reasons for low uptake. But record-clearing

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

183

See, e.g., Margaret Colgate Love, The Debt that Can Never Be Repaid: A Report Card on the

Collateral Consequences of Conviction, C

RIM

. J

UST

., Fall 2006, at 16, 22 (“[I]t is generally more

expensive for a criminal offender to hire a lawyer to go to court to seek expungement, than it is to

file an application for pardon . . . .”).

184

Telephone Interview with Anonymous Legal Aid Lawyer, supra note 166.

185

See, e.g., Charlie Gerstein & J.J. Prescott, Process Costs and Police Discretion, 128 H

ARV

. L.

. F. 268, 276, 278 (2015); Issa Kohler-Hausmann, Misdemeanor Justice: Control Without

Conviction, 119 A

. J. S

. 351, 369–73 (2013).

186

See U

MEZ

& P

IRIUS

, supra note 46, at 3.

187

See Agan & Starr, Ban the Box, supra note 39, at 193 n.2 (“Job application questions about

records are overwhelmingly limited to convictions (not arrests), and usually to felonies.”).

188

Telephone Interview with Tracey Brame, supra note 145.

2020] EXPUNGEMENT OF CRIMINAL CONVICTIONS 2507

policies — including the common requirement that eligible individuals

file a petition — are embedded in a larger criminal justice context, one

rife with access-to-justice and punishment disparities along economic

and racial dimensions.

189

To those familiar with that context, the uptake

gap, while stark, may actually be unsurprising. The gap aligns with

Sarah Brayne’s finding that people involved in the criminal justice sys-

tem are more likely to engage in “system avoidance” by self-selecting

away from institutions that keep formal records or carry the mantle of

government because they fear the prospect of adverse state surveil-

lance.

190

The burdens of seemingly unrelated policies, therefore, may

help us understand the failure of uptake with respect to expungement.

Nevertheless, addressing the immediate barriers to uptake we de-

scribe above is critical, and various organizations have taken important

steps to improve matters. Over the past fifteen years, as the challenges

facing people with records have drawn more attention nationally and in

Michigan,

191

significant efforts in Michigan to increase support for peo-

ple seeking expungement have sprouted. The City of Detroit, for exam-

ple, recently adopted an initiative called Project Clean Slate to support

expungement activity,

192

and Michigan Works, a statewide workforce

development program, has become increasingly involved in the issue.

193

Michigan Legal Help, a nonprofit that provides do-it-yourself tools to

pro se litigants, has developed an online tool that allows applicants to

assess their eligibility and to fill out the application form, although they

must still go through the various steps to file it.

194

Many of these organizations have sponsored expungement fairs,

which have become perhaps the central outreach tool in this space.

195

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

189

See D

AVI D

J. H

ARDING

., O

THE

UTSIDE

: P

RISONER

EENTRY

AND

EINTEGRATION

4–5 (2019) (acknowledging “a broader cultural, political, and institutional shift

toward order, control, and the use of punishment to solve social problems, particularly those asso-

ciated with poverty or ethnoracial minorities”).

190

Sarah Brayne, Surveillance and System Avoidance: Criminal Justice Contact and

Institutional Attachment, 79 A

. S

. R

. 367, 385 (2014).

191

We discuss recent expansions in expungement law above, supra notes 12–21 and accompany-

ing text. While the past couple of years have been particularly active, the trend of expansion actu-

ally goes back further. See Brian M. Murray, A New Era for Expungement Law Reform? Recent

Developments at the State and Federal Levels, 10 H

ARV

. L. & P

’

. 361, 369–73 (2016) (dis-

cussing reforms between 2009 and 2015).

192

See Project Clean Slate, C

ITY

ETROIT

, https://detroitmi.gov/departments/law-

department/project-clean-slate [https://perma.cc/ZJQ2-E4LY ] .

193

See Have You Been Convicted of a Crime?, M

ICH

. W

ORKS

! S

., https://www.mwse.org/

expungement [https://perma.cc/U4B7-SJCW]; Telephone Interview with Chioke Mose-Telesford,

supra note 161.

194

See Do-It-Yourself Expungement (Adult Conviction), M

ICH

. L

EGAL

ELP

, https://

michiganlegalhelp.org/self-help-tools/crime-traffic-and-id/do-it-yourself-expungement-adult-conviction

[https://perma.cc/5X47-88VN]; Telephone Interview with Michael Kiehne, supra note 161.

195

See Michigan Works! Southeast Expungement Fair, M

ICH

. L

EGAL

ELP

, https://

michiganlegalhelp.org/node/11324 [https://perma.cc/CUR6-MEQ5]; Project Clean Slate Fair This

2508 HARVARD LAW REVIEW [Vol. 133:2460

Expungement fairs are designed at a minimum to provide information,

and many also provide concrete assistance: applicants have been encour-

aged to fill out applications on site, and some fairs have had sheriffs

present to do the fingerprinting, as well as notaries.

196

Depending on

the sponsoring organization’s funding, these fairs have sometimes been

able to defray the costs for applicants — for example, by offering fin-

gerprinting for free or paying for the certified record of conviction.

197

Expungement fairs in Detroit have reportedly been quite large, some

perhaps close to a thousand participants, although most attendees have

proven to be ineligible.

198

Fairs elsewhere in the state have generally

not had anywhere close to this kind of attendance.

199

Miriam Aukerman, who worked on expungement cases at Western

Michigan Legal Aid until 2010, told us:

When I started doing this work back in 2003 or 2004, nobody was doing it.

People didn’t think it was important. When I left in 2010, it was a thing.

Legal services offices had started doing it; there were self-help packets avail-

able . . . . Expungement fairs were starting to happen. Advocacy organi-

zations started spreading the word.

200

Aukerman’s recollection is borne out by data. In 2010, the last full year

for which we have complete data, the number of known-date expunge-

ments granted in Michigan was 2044, which is 67% higher than the

number in 2004 (1224). MSP reported to us that 2594 expungements

were granted in 2017; this number (although down slightly from 2016)

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

Saturday to Provide Free Criminal Record Expungement for Detroiters, C

ITY

ETROIT

, https://

detroitmi.gov/news/project-clean-slate-fair-saturday-provide-free-criminal-record-expungement-

detroiters [https://perma.cc/P97U-43YU]; Telephone Interview with Michael Kiehne, supra note 161;

Telephone Interview with Chioke Mose-Telesford, supra note 161.

196

See, e.g., Hunter McLaren, Expungement Fair Gives Jackson Residents Second Chance at

Finding Work, Housing, ML

IVE

(Aug. 10, 2019), https://www.mlive.com/news/jackson/2019/08/

expungement-fair-gives-jackson-residents-second-chance-at-finding-work-housing.html [https://

perma.cc/LR6G-SFR5] (describing the provision of reduced fees and a local police officer to facili-

tate the expungement application process).

197

See id.

198

See Telephone Interview with Michael Kiehne, supra note 161 (indicating that some Detroit

fairs have had approximately 1000 attendees); Attorneys Participate in “Know Your Rights”

Program, Expungement Fair, L

EGAL

EWS

COM

(Feb. 10, 2020), http://legalnews.com/

detroit/1484605 [https://perma.cc/725R-DJQB]; Corey Williams, Detroit Jobseekers Look to Have

Nonviolent Records Expunged, A

SSOCIATED

RESS

, June 4, 2016; see also Project Clean Slate

Fair This Saturday to Provide Free Criminal Record Expungement for Detroiters, supra note 195

(noting that only 200 Detroit residents had records expunged that year).

199

See, e.g., Gabriella Galloway, Some with Criminal Records Get a Possible Second Chance, 9 & 10

EWS

(July 29, 2019), https://www.9and10news.com/2019/07/29/some-with-criminal-records-get-a-

possible-second-chance [https://perma.cc/UC8M-DYA3]; McLaren, supra note 196. But see, e.g.,

Jacqueline Francis, Criminal Expungement Fair Draws Hundreds, W

OODTV

COM

(Oct. 2, 2019, 11:12

PM), https://www.woodtv.com/news/grand-rapids/criminal-expungement-fair-draws-hundreds [https:

//perma.cc/VP6M-XRBJ] (describing a fairly large Grand Rapids fair with 435 attendees).

200

Telephone Interview with Miriam Aukerman, supra note 166.

2020] EXPUNGEMENT OF CRIMINAL CONVICTIONS 2509

represents another 27% growth since 2010.

201

These changes suggest

that efforts to help people surmount the various barriers to access ex-

pungement opportunities can successfully encourage applications.

That said, one should not be too sanguine; even if uptake doubled

from what we measure (for five-year periods ending between 2009 and

2011), it would still be only 13%. Expungement lawyers articulated to

us the frustration that many of their clients feel with the process, even

with legal help.

202

Rather than expect a massive influx of aid to help

people navigate the process, we should recognize that substantially clos-

ing the uptake gap will require legal changes to simplify that process.

We return to this point in the Conclusion.

Finally, it bears noting that, although our interviews focused mainly

on the problem of low uptake among eligible individuals, every advocate

we spoke to also stressed the stringency of the eligibility requirements,

which in each of their views exclude a great many worthy candidates.

203

Reforms in 2011 and 2015 slightly softened the requirements regarding

the allowable number of convictions,

204

but they are still strict and in

many ways quite arbitrary. Today, in Michigan, anyone with two felony

convictions is still excluded, even if they are low-level felonies and even

if they are two counts arising from the same incident.

205

Inexplicably,

people with three misdemeanors cannot seek expungements at all, even

though people with one felony plus two misdemeanors can seek to ex-

punge the felony.

206

Traffic misdemeanors cannot be expunged them-

selves but nevertheless count toward the limit and thus may disqualify an

individual from expunging another conviction.

207

These offenses are very

common in Michigan, accounting for more than half of all misdemeanor

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

201

Email from Ted Kilvington, supra note 125.

202

See Telephone Interview with Miriam Aukerman, supra note 166; Email from Josh Hoe, supra

note 166.

203

See, e.g., Telephone Interview with Anonymous Legal Aid Lawyer, supra note 166; Telephone

Interview with Miriam Aukerman, supra note 166 (“90% of the people would walk out [of an ex-

pungement fair] and be angry after you explained that you could only have one conviction [to be

eligible]. . . . It could be that the people who are most impacted by criminal records are those who

are not eligible.”); Email from Josh Hoe, supra note 166 (“A large number of folks are made ineligible

for expungement by either carve-outs in the legislation or by the fact that it only applies to people

with 1 felony and/or 2 misdemeanors.”).

204

See M

ICH

. C

OMP

. L

AWS

§ 780.621(4) (2020) (effective Jan. 12, 2015) (expanding eligibility for

expungement to victims of human trafficking); id. § 780.621(1)(c) (2020) (effective June 23, 2011)

(expanding eligibility for expungement of an offense to individuals who have two other “minor

offense[s]”).

205

See id. § 780.621(1)(a) (2020) (effective Jan. 12, 2015) (stating that individuals are only eligible

for an expungement if they “[have been] convicted of not more than [one] felony offense”).

206

See id. § 780.621(1)(a)–(b). This seems like a clear drafting error, but it is apparently enforced.

Telephone Interview with Michael Kiehne, supra note 161.

207

See M

ICH

. C

OMP

. L

AWS

§ 780.621(3)(d) (2020) (stating that expungements shall not be

granted for traffic offenses).

2510 HARVARD LAW REVIEW [Vol. 133:2460

convictions.

208

Petty misdemeanors, including violations of hunting and

fishing laws and dog-leash laws, likewise count toward the limit.

209

addition, the waiting period now runs five years from the completion of

all components of the sentence, which means that people with lengthy

probation or parole terms might actually have to wait out eight or ten

recidivism-free years in the community, not just five years as was re-

quired under Michigan’s old law.

210

All of these restrictions mean that the low uptake rate we estimate

is even starker when viewed in context: it is a very small fraction of a

very small fraction. For the past decade about 2000 expungements per

year have been granted in Michigan. Meanwhile, each year the Michigan

state courts add about 300,000 new criminal convictions.

211

On balance,

the population of people living with criminal records is continuing to

grow quickly; the existing expungement regime is like a bucket remov-

ing water from an ever-rising ocean.

212

III.

ECIDIVISM

UTCOMES

Petition-based expungement is not very common — but what happens

when it does occur? Are there significant public-safety costs to granting

record-clearing relief? For example, does expungement increase the risk

of recidivism or conceal other substantial dangers from the public? Pub-

lic-safety concerns underlie the most common objections to expunge-

ment laws, the idea being that the public — especially people with par-

ticular interests at stake, such as employers and landlords — have a

right to know when people have criminal records because of their

heightened risk of committing future crimes.

213

Here, our data paint a

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

208

See M

ICH

. C

OURTS

, supra note 150. An experienced Michigan expungement attorney told

us that it is common for her clients to believe they are eligible for an expungement because they do

not understand that a prior traffic offense happened to be a misdemeanor and because it does not

come up in ICHAT, the free background check tool that some use to check their eligibility (though

it does ultimately appear in the official background check to determine actual eligibility). Telephone

Interview with Anonymous Legal Aid Lawyer, supra note 166.

209

See M

ICH

. C

OMP

. L

AWS

§ 780.621(16)(f) (2020) (stating the definition of a misdemeanor for

purposes of the statute); cf., e.g., id.

§ 324.43560 (providing that a person who violates the hunting

and fishing licensing rules “is guilty of a misdemeanor”).

210

See id. § 780.621(5)(a)–(d) (effective Jan. 12, 2015).

211

See supra note 150.

212

See Riley, supra note 150 (estimating that tens of thousands of Michigan residents receive

first-time felony convictions each year).

213

See Murray, supra note 191, at 375 (describing the counterargument to expungement that

“federal crimes are serious, and actors within the community should be able to obtain information

about ex-offenders that might be relevant to an important decision, especially related to hiring”);

see also, e.g., Doe v. United States, 110 F. Supp. 3d 448, 449, 455 (E.D.N.Y. 2015) (memorializing

the judge’s belief that “employers are generally entitled to know about the past convictions of job

applicants,” id. at 449, and indicating that, in part for that reason, expungements may be granted

only when “extraordinary circumstances are present,” id. at 455); U.S.

’

USTICE

TTORNEY

ENERAL

’

EPORT

RIMINAL

ISTORY

ACKGROUND

HECKS

1 (2006),

2020] EXPUNGEMENT OF CRIMINAL CONVICTIONS 2511

sunnier picture. In section A, we estimate recidivism rates for expunge-

ment recipients, finding that reoffending is strikingly rare. In section B,

we consider the interpretation and policy implications of our findings.

In this discussion, we rely mainly on other criminological research for

guidance; as we explain below, the nature of the law and our data make

it difficult to directly assess the causal effect of receiving an expunge-

ment on criminal behavior.

A. Recidivism Among Expungement Recipients

In Table 4, we report two-year and five-year rearrest and reconvic-

tion rates for expungement recipients, starting from the date of receipt

of the expungement.

214

We analyze the criminal histories of the full

universe of individuals receiving expungements up through the begin-

ning of December 2010.

215

Most of the data concerns we raise with

regard to the “eligible” sample of individuals with records that we ana-

lyze in Part II do not arise here.

216

Our only significant data limitation

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

https://www.bjs.gov/content/pub/pdf/ag_bgchecks_report.pdf [https://perma.cc/4BY8-WT8M]

(stating that employers want access to criminal records so they can “protect employees, customers,

vulnerable persons, and business assets” and avoid negligent hiring suits).

214

Recidivism rates are a complicated concept, see generally M

ICHAEL

D. M

ALTZ

, R

ECIDIVISM

(1984), and one of the important questions in constructing estimates is the length of time over which

to consider an individual “at risk” of reoffending. At least within the domain of government statis-

tics, there is no real consensus on the appropriate at-risk period. See Bill Montgomery, We Have

No Standard for Measuring Recidivism, T

ILL

(Nov. 15, 2017, 3:00 PM),

https://thehill.com/opinion/criminal-justice/360520-we-have-no-standard-for-measuring-recidivism

[https://perma.cc/C56V-8RN7]. The two-year rates that we report are more typical of the recidivism

literature, but the time period varies. See, e.g., Seena Fazel & Achim Wolf, A Systematic Review of

Criminal Recidivism Rates Worldwide: Current Difficulties and Recommendations for Best Prac-

tice, 10 PL

S O

3–4, 5 tbl.3 (2015), https://journals.plos.org/plosone/article/

file?id=10.1371/journal.pone.0130390&type=printable [https://perma.cc/NJ77-2LQM]. Three-year

follow-up periods are also common, but selection of the length of time for evaluating recidivism is

often driven by data availability. See M

ARIEL

LPER

., U.S. D

’

USTICE

, 2018

PDATE

RISONER

ECIDIVISM

: A 9-Y

EAR

OLLOW

-U

ERIOD

(2005–2014) 14 (2018),

https://www.bjs.gov/content/pub/pdf/18upr9yfup0514.pdf [https://perma.cc/YT9K-MAQ9]. Longer

time spans are generally more beneficial for studying patterns of recidivism and their attributes.

See id. We are readily able to offer a longer-term measure, too, because our data cover a more

extensive span than many other researchers have been able to observe. We opt to include both two-

and five-year measures to provide both points of comparison.

215

We rely for this analysis on an MSP criminal history dataset that ends on December 6, 2012.

To allow a sufficient follow-up period, our analysis sample accordingly includes nearly all individ-

uals receiving expungements in Michigan with known dates up through December 6, 2010 (for the

two-year rates), and December 6, 2007 (for the five-year rates). The only individuals we drop are

232 people identified in the data as having out-of-state driver’s licenses, to reduce any bias that

might result from unobserved out-of-state recidivism. The resulting numbers of observations for

our analysis are 20,955 for the two-year sample and 15,256 for the five-year sample.

216

The rates we report are not really “estimates” for these populations; there is no sampling error

in the results that we report for pre-2011 Michigan expungement recipients. We need not worry

about the representativeness of the sample in thinking about how expungement recipients fared in

Michigan during our study period. However, our calculations are “estimates” if we think of past

expungement recipients as a “sample” that is representative of future expungement recipients (or,

2512 HARVARD LAW REVIEW [Vol. 133:2460

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

potentially, expungement recipients from other jurisdictions) about whom we wish to make predic-

tions. Accordingly, we include standard errors in Table 4, in parentheses; these show that our

estimates (if understood as such) are in any event quite precise.

(

)

(

)

(

)

(

)

Two-Year

Arrest

Rate

Two-Year

Conviction

Rate

Five-Year

Arrest

Rate

Five-Year

Conviction

Rate

A. Full Sample

1. Overall Rearrest/Reconviction Rates 3.4% 1.8% 7.1% 4.2%

(0.13%) (0.09%) (0.21%) (0.16%)

2. Violent Rearrest/Reconviction Rates 1.0% 0.2% 2.6% 0.6%

(0.07%) (0.03%) (0.13%) (0.06%)

3. Felony Rearrest/Reconviction Rates 1.2% 0.3% 2.7% 1.0%

(0.07%) (0.04%) (0.13%) (0.08%)

Number of Observations 20,955 20,955 15,256 15,256

B. Subsamples

1. Expungement After < 6 Years 3.8% 2.1% 8.1% 4.9%

(0.30%) (0.23%) (0.49%) (0.39%)

2. Expungement in Years 11 and 12 2.8% 1.6% 6.3% 3.6%

(0.50%) (0.37%) (0.86%) (0.66%)

3. Expunged Offense: Felony 4.0% 1.9% 8.1% 4.6%

(0.20%) (0.14%) (0.33%) (0.26%)

4. Expunged Offense: Misdemeanor 3.1% 1.8% 6.4% 3.8%

(0.17%) (0.13%) (0.28%) (0.22%)

5. Expunged Offense: Incarcerated 3.2% 1.7% 6.7% 3.9%

(0.23%) (0.17%) (0.38%) (0.29%)

6. Expunged Offense: Not Incarcerated 3.5% 1.8% 7.2% 4.3%

(0.15%) (0.11%) (0.25%) (0.19%)

7. Expunged Offense: Violent 4.4% 2.2% 8.4% 4.4%

(0.37%) (0.27%) (0.62%) (0.46%)

8. Expunged Offense: Violent 1.6% 0.3% 4.0% 0.8%

(violent rearrest/reconviction only)

(0.23%) (0.10%) (0.44%) (0.20%)

ABLE

4. R

EARREST

AND

ECONVICTION

ATES

FOR

XPUNGEMENT

ECIPIENTS

Notes: Rearrest and reconviction data come from Michigan State Police criminal history files. The full sample consists of

all individuals with no record of an out-of-state driver’s license who receive a known-date expungement in Michigan

through December 6, 2010, for two-year rates and through December 6, 2007, for five-year rates. In Panel B, all rearrest

and reconviction rates are “overall” rates — that is, they count any new criminal arrest or conviction — except for Row 8,

which shows only the rearrest and reconviction rates for violent offenses. Standard errors are reported in parentheses.

Subsample sizes are available from the authors upon request.

2020] EXPUNGEMENT OF CRIMINAL CONVICTIONS 2513

is our inability to observe out-of-state and federal arrests and convic-

tions,

217

which may mean that we slightly underestimate true recidivism

rates. We do not believe this omission has a large effect, however, for

reasons we already discuss in detail in Part II: federal arrests and con-

victions are rare, and interstate moves and interstate travel to commit

crimes are not common either.

218

In Row 1 of Table 4, Panel A, we report the overall two-year and

five-year rearrest and reconviction rates, while in the remaining rows

we give specific rates for certain types of crime that may be of particular

public and policy concern: violent offenses and felony offenses.

219

All of

these rates show a consistent pattern: recidivism among expungement

recipients is low.

220

Overall, 3.4% are rearrested and 1.8% are recon-

victed for crimes within two years; 7.1% are rearrested and 4.2% are

reconvicted within five years. The numbers are even lower when we

focus specifically on the types of crimes that worry people most. For

example, within five years, only 2.6% of expungement recipients are

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

217

See supra notes 147–150 and accompanying text.

218

Applying the assumptions we use in Part II that 1.8% of individuals move per year and 4%

of crime is committed across state borders, we expect we might be missing about 5–6% of the re-

arrests and reconvictions for our two-year outcome period and perhaps 7–8% in the five-year period

(when people have had more time to move). See supra notes 147–150 and accompanying text.

Given the very low recidivism rates we report in Table 4, the effect of this problem is negligible in

percentage-point terms. Another useful source is a 2005 study from the Bureau of Justice Statistics,

which finds nontrivial rates of out-of-state recidivism over the course of five years (whether due to

cross-border crime or individuals changing their state of residence). See M

ATTHEW

R. D

UROSE

., U.S. D

’

USTICE

, R

ECIDIVISM

RISONERS

ELEASED

30 S

TATES

2005:

ATTERNS

FROM

2005

2010, at 7 (2014), https://www.bjs.gov/content/pub/pdf/rprts05p0510.

pdf [https://perma.cc/J2BP-WTKN]. But still, even that out-of-state recidivism rate (10.9%) is only

about one-seventh of the overall recidivism rate (75.4%) observed over the same period. Id. at 7–

8. If we were to apply that same ratio here, our estimated recidivism rates would understate the

true rate by much less than a percentage point. Moreover, Michigan residents almost surely engage

in fewer cross-border crimes than residents in many states, simply because Michigan is surrounded

on most sides by water and has no interstate metropolitan areas.

219

Felonies are identified as such in the MSP records we use in our analysis; we classify an

offense as violent if it involves the use or threat of force according to its statutory definition. Stat-

utorily, a felony in Michigan is defined as an offense for which a person convicted of the charge

may be punished by death or imprisonment. See M

ICH

. C

OMP

. L

AWS

§ 750.7 (2020). Violent

offenses are a subset of both felonies and misdemeanors.

220

Recidivism risk varies by type of initial crime, age, poverty, and many other factors, but in

general, recidivism rates below 10% are relatively low. For example, in a recent multistate study

of prisoners released in 2005, over 60% are rearrested within two years, and 77% are rearrested

within five years. See A

LPER

., supra note 214, at 11 tbl.7. Limiting this to in-state arrests

(to more closely match our data), these numbers are 54.4% and 66.1%, respectively. See id. at 4

tbl.2. Recidivism rates for individuals on probation are also much higher. See P

ATRICK

ANGAN

& M

ARK

A. C

UNNIFF

, U.S. D

’

USTICE

, R

ECIDIVISM

ELONS

ROBATION

, 1986–89, at 5–6 (1992), https://www.bjs.gov/content/pub/pdf/rfp8689.pdf [https://

perma.cc/7BAU-4J4F]; see also Dana Goldstein, The Misleading Math of “Recidivism,”

ARSHALL

ROJECT

(Dec. 4, 2014, 11:15 AM), https://www.themarshallproject.org/2014/12/04/

the-misleading-math-of-recidivism [https://perma.cc/UAS7-BS2Q].

2514 HARVARD LAW REVIEW [Vol. 133:2460

rearrested and 0.6% are reconvicted for violent crimes; 2.7% are rearrested

and 1% are reconvicted for felonies.

We emphasize that these rates are much lower than those found in

most studies of criminal recidivism.

221

Indeed, our statistics suggest that

expungement recipients pose a lower crime risk than the general popu-

lation of Michigan as a whole. Although no general-population data are

available that directly parallel the figures in Table 4, the state does re-

port annual arrest figures. For comparison purposes, in the two-year

period from 2009–10, Michigan police made about 6.6 arrests per 100

adults in the population.

222

In contrast, using cases from approximately

the same time period, we calculate only 4.7 arrests per 100 expungement

recipients within two years of their having received an expungement.

223

This comparison is particularly striking given that expungement recipi-

ents (like people with criminal records generally) tend to have other

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

221

See, e.g., Donald P. Green & Daniel Winik, Using Random Judge Assignments to Estimate the

Effects of Incarceration and Probation on Recidivism Among Drug Offenders, 48 C

RIMINOLOGY

357, 370 (2010); Ilyana Kuziemko, How Should Inmates Be Released from Prison? An Assessment

of Parole Versus Fixed-Sentence Regimes, 128 Q.J. E

CON

. 371, 385–86 (2013); Cody Tuttle,

Snapping Back: Food Stamps Ban and Criminal Recidivism, 11 A

. E

CON

. J.: E

CON

. P

’

301,

315–16 (2019). For sources specific to Michigan, see, for example, David J. Harding et al., Short-

and Long-Term Effects of Imprisonment on Future Felony Convictions and Prison Admissions, 114

PNAS, 11,103, 11,106–07 (2017); C

hris

autz

, M

ichigan

ecidivism

ate

alls

owest

evel

ercent

ICHIGAN

GOV

(Feb. 13, 2018), https://www.michigan.gov/som/0,4669,7-

192-26847-459956--,00.html [https://perma.cc/M9GJ-WKHH].

222

Arrest figures are summed across “total arrests” for persons eighteen and up in M

ICH

. S

TATE

OLICE

, 2009 S

TATE

RREST

OTALS

(2010), https://www.michigan.gov/documents/

msp/2009Annual_StatewideArrests_332334_7.pdf [https://perma.cc/TAA9-D5XV], and M

ICH

TATE

OLICE

, 2010 S

TATEWIDE

RREST

OTALS

(2011), https://www.michigan.gov/documents/

msp/2010_Annual_StatewideArrests_358704_7.pdf [https://perma.cc/5HDB-CWKA]. Population fig-

ures come from the 2010 Census. QuickFacts: Michigan, U.S.

ENSUS

UREAU

, https://www.

census.gov/quickfacts/fact/table/MI/POP010210 [https://perma.cc/Q7KR-UZWB].

223

We calculate this number for the cohort of those receiving criminal-record expungements be-

tween December 2006 and December 2010, running for two years from each individual’s expunge-

ment date and excluding those individuals known to have out-of-state licenses; this is the most

recent complete expungement-receiving cohort for which we have two years of subsequent criminal

history data. Total arrests per capita over two years are higher than the two-year arrest rates for

the same population (like those we report in Table 4) because some people are arrested more than

once. See Alexi Jones & Wendy Sawyer, Arrest, Release, Repeat: How Police and Jails Are Misused

to Respond to Social Problems, P

RISON

’

NITIATIVE

(Aug. 2019), https://www.prisonpol-

icy.org/reports/repeatarrests.html [https://perma.cc/33J2-CV53]. There are no public data on the

number of unique individuals arrested in Michigan in any given period (for the closest we can find,

see Wanda Bertram & Alexi Jones, How Many People in Your State Go to Local Jails Every Year?,

RISON

’

NITIATIVE

(Sept. 18, 2019), https://www.prisonpolicy.

org/blog/2019/09/18/state-jail-bookings [https://perma.cc/2L9P-PTGM]), so the number of arrests is

the best metric to use for this comparison. We do not have a similarly authoritative source for the

number of convictions statewide. However, we can gain some insight into individuals who are

arrested multiple times from the Prison Policy Initiative: “Using nationally representative data from

the National Survey on Drug Use and Health (NSDUH), we find that at least 4.9 million individuals

were arrested and booked in 2017. Of those 4.9 million individuals, 3.5 million were arrested only

once in 2017; 930,000 were arrested twice; and 430,000 were arrested three or more times.” Jones

& Sawyer, supra.

2020] EXPUNGEMENT OF CRIMINAL CONVICTIONS 2515

characteristics that are generally associated with higher crime risk —

for example, they are more likely to be male and poor than the general

population is. In fact, if we compare expungement recipients to subsets

of the general population that look more like them in terms of observa-

ble characteristics such as age and gender, the comparison is even more

favorable to expungement recipients.

224

In Panel B of Table 4, we replicate the main results from Panel A,

Row 1 (overall arrest and reconviction rates) for subsets of the expunge-

ment population of particular interest. In Row 1, we examine individ-

uals who receive their expungements early, in their first year of eligibility

(that is, by the end of the sixth year since sentencing or release). The

waiting period for expungements in Michigan is five years, but the ac-

tual median elapsed time before receipt is nearly ten years after sentenc-

ing or release.

225

Because recidivism risk declines over time, this delay

should be expected to reduce expungement recipients’ recidivism

rates.

226

But policymakers might understandably wonder what would

happen if people did not wait those extra years — which would be the

consequence of a policy that made expungements automatic after five

years.

227

Row 1’s estimate in Table 4.B is designed to inform the ques-

tion whether five years is enough of a waiting period — and what we

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

224

These calculations are available upon request. Age and gender are two of the characteristics that

most strongly predict arrest rates, so it is helpful to begin by considering their role. One might expect

that expungement recipients’ low recidivism risk could be explained by the fact that, by the time they

qualify for an expungement, they have already “aged out” of crime. But while this may sometimes be

true (and indeed there are few very young adults who obtain expungements), the overall age distribution

of expungement recipients is similar to, and actually a bit younger than, the general adult population;

a majority are still in their twenties and thirties, ages when arrest rates are still higher than average.

See H

OWARD

N. S

NYDER

, U.S. D

’

USTICE

, A

RREST

THE

NITED

TATES

, 1990–2010,

at 3 (2012), https://www.bjs.gov/content/pub/pdf/aus9010.pdf [https://perma.cc/X6BZ-VXJ2]. Mean-

while, a majority of expungement recipients are male, see supra tbl.2, and males have higher arrest

rates, see S

NYDER

, supra, at 2. If one reweights the general population arrest rates so as to mirror the

expungement-recipient pool in terms of its age and sex distribution, the projected general-population

rate becomes even higher: about 7.6 arrests per 100 people in each two-year period (compared, again,

to 4.7 for expungement recipients). General-population arrest data are not available broken down by

many other characteristics, but we can be fairly certain that populations that are comparable to ex-

pungement recipients across other demographic and socioeconomic dimensions would have even higher

arrest rates, since expungement recipients tend to be disadvantaged in various ways, as we explain in

Part II. See supra section II.B, pp. 2493–2501.

225

See supra tbl.2.

226

See S

YNØVE

YGAARD

NDERSEN

& T

ORBJØRN

KARDHAMAR

, P

ICK

UMBER

APPING

ECIDIVISM

EASURES

AND

HEIR

ONSEQUENCES

15–17 (2014),

https://www.ssb.no/en/forskning/discussion-papers/_attachment/166596 [https://perma.cc/3UEH-

TGKQ]; N

ATHAN

AMES

, C

ONG

. R

ESEARCH

ERV

., RL34287, O

FFENDER

EENTRY

ORRECTIONAL

TATI STI CS

, R

EINTEGRATION

INTO

THE

OMMUNITY

AND

ECIDIVISM

& fig.3 (2014); Kuziemko, supra note 221, at 379; Fredrik Sivertsson, Catching Up in Crime? Long-

Term Processes of Recidivism Across Gender, 2 J.

EVELOPMENTAL

& L

IFE

-C

OURSE

RIMINOLOGY

371, 376 (2016).

227

See, e.g., Francisco, supra note 73; Kelan Lyons, Lamont to Introduce Clean Slate Legislation

in Next Session, CT

IRROR

(Jan. 8, 2020), https://ctmirror.org/2020/01/08/lamont-to-introduce-

2516 HARVARD LAW REVIEW [Vol. 133:2460

find in our data suggests that it is. Recidivism rates in all columns are

only very slightly higher than the main-sample numbers, indicating that

this subgroup remains low risk relative to the general public.

Meanwhile, Row 2 in Table 4.B shows rearrest rates for people who

receive expungements in the eleventh or twelfth year after sentencing or

release — a good proxy for the effect of providing for expungements

after ten clean years, as Pennsylvania’s new law does automatically for

minor offenses.

228

These recidivism rates are even lower than average

expungement-recipient rates (unsurprisingly, given the long clean rec-

ords of these individuals). All in all, though, the differences among these

various cohorts are quite small, suggesting that eligibility waiting-period

differences might not matter very much in terms of fostering lower post-

expungement recidivism rates.

In Rows 3 and 4 of Table 4.B, we show recidivism numbers for those

whose expungement conviction was a felony and a misdemeanor, respec-

tively. Recidivism rates are low for both groups, although slightly higher

for the felony group. In Rows 5 and 6 respectively, we compare the

behavior of expungement recipients who did and did not serve any time

behind bars, according to MSP records. One might expect recidivism

rates for the formerly incarcerated to be higher for two reasons: they

have more serious prior convictions,

229

and they may face additional

socioeconomic disadvantages arising from their period of incarcera-

tion.

230

These influences might be counterbalanced by the fact that

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

clean-slate-legislation-in-next-session [https://perma.cc/B75S-85JG]; cf. Margaret Love & David

Schlussel, Model Law Proposes Automatic Expungement of Non-conviction Records, C

RIME

(Dec. 18, 2019), https://thecrimereport.org/2019/12/18/model-law-proposes-automatic-

expungement-of-non-conviction-records [https://perma.cc/7L5G-FPRE].

228

See 18 P

. C

ONS

. S

TAT

. § 9122.2 (2018); see also R

ESTORATION

IGHTS

ROJECT

ENNSYLVANIA

ESTORATION

IGHTS

& R

ECORD

ELIEF

Part

III (2020), https://

ccresourcecenter.org/state-restoration-profiles/pennsylvania-restoration-of-rights-pardon-

expungement-sealing-2 [https://perma.cc/Q9CJ-BYV6]; supra note 69 and accompanying text.

229

See Blanton v. City of North Las Vegas, 489 U.S. 538, 542 (1989) (“[B]ecause incarceration is

an ‘intrinsically different’ form of punishment, it is the most powerful indication of whether an

offense is ‘serious.’” (quoting Muniz v. Hoffman, 422 U.S. 454, 477 (1975))); N

’

ESEARCH

OUNCIL

, T

ROWTH

NCARCERATION

THE

NITED

TATES

: E

XPLORING

AUSES

AND

ONSEQUENCES

19 (Jeremy Travis et al. eds., 2014) (“Prison terms usually are re-

served for those found guilty of more serious crimes, defined as felonies by state and federal legis-

latures.”). But see S. Union Co. v. United States, 567 U.S. 343, 351 (2012) (“But not all fines are

insubstantial, and not all offenses punishable by fine are petty.”); N

’

ESEARCH

OUNCIL

supra, at 86 (“Considerations of proportionality and parsimony have fallen into neglect in the United

States. Many laws enacted in the 1980s and 1990s required less serious crimes to be punished more

severely than more serious ones. Examples include mandatory minimum sentence laws requiring

longer terms for people convicted of small sales of drugs than terms typically imposed for many

violent offenses, and the sentencing of people to 25-year minimum terms for property misdemeanors

under California’s three strikes law.”).

230

See John Bronsteen et al., Happiness and Punishment, 76 U. C

. L. R

. 1037, 1038 (2009)

(“Even large fines are likely to have only minor effects on the well-being of those who receive them,

because people adjust quite easily to their new financial circumstances. . . . On the other hand,

2020] EXPUNGEMENT OF CRIMINAL CONVICTIONS 2517

during the five-year period in which we observe them, formerly incar-

cerated expungement recipients are further beyond their

conviction/release dates (because the clock runs from the later of the two

events) and are older on average.

231

On balance, we find very slightly

lower recidivism rates for those who have been incarcerated.

Finally, in Rows 7 and 8 of Table 4.B, we present recidivism numbers

for people who have had violent-offense convictions expunged. In Row

7, we demonstrate that the overall rearrest and reconviction numbers

for this subsample are slightly higher than the average for the entire

sample, although still quite low. In Row 8, we report the rearrest and

reconviction numbers for new violent crimes alone (as we did in Table

4.A, Row 2). This comparison addresses the specific fear that employers,

landlords, and policymakers might have about people with a past vio-

lent offense: that they will commit another violent crime.

232

Although

the rates of violent reoffense that we calculate are very slightly higher

in the subsample with violent crime convictions, they are still extremely

low in absolute terms.

233

Only 0.3% of those who have had a violent

offense expunged are reconvicted of another violent offense within two

years, and only 0.8% within five years.

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

virtually any period of incarceration, no matter how brief, has consequences that negatively affect

prisoners’ lives in ways that resist adaptation, even after they have been released. Prisoners are

often abandoned by their spouses and friends, face difficulty finding and keeping employment, and

may suffer from incurable diseases contracted during their incarceration.”); see also Manudeep

Bhuller et al., Incarceration, Recidivism, and Employment, 128 J.

. E

CON

. 1269, 1313 (2020)

(noting that incarceration cripples a previously employed person’s post-release employment pro-

spects); Flake, supra note 86, at 55–62; David J. Harding et al., Imprisonment and Labor Market

Outcomes: Evidence from a Natural Experiment, 124 A

. J. S

. 49, 75–76 & tbl.2 (2018); Steven

Raphael, The Effects of Conviction and Incarceration on Future Employment Outcomes, in

ABELING

HEORY

: E

MPIRICAL

ESTS

237, 259 (David P. Farrington & Joseph Murray eds.,

2014); supra section I.A., pp. 2468–71.

231

Cf. J.J. Prescott et al., Understanding Violent-Crime Recidivism, 95 N

OTRE

AME

L. R

1643, 1688–97 (2020) (finding that age is negatively related to recidivism risk among prison releasees).

232

See, e.g., id. at 1643–45; Megan Denver, Justin T. Pickett & Shawn D. Bushway, The Language

of Stigmatization and the Mark of Violence: Experimental Evidence on the Social Construction and

Use of Criminal Record Stigma, 55 C

RIMINOLOGY

664, 675–76 (2017) (reporting a “substantial”

increase in the perceived risk of recidivism for those convicted of violent crimes as compared to

those convicted of nonviolent crimes, id. at 676); see also supra p. 2513. But see Michael M. O’Hear

& Darren Wheelock, Violent Crime and Punitiveness: An Empirical Study of Public Opinion, 103

ARQ

. L. R

. (forthcoming 2020) (manuscript at 47–49) (on file with Marquette University Law

School) (finding no evidence that public support for the punitive sentencing of individuals who

commit violent crimes results from fear of violent-crime recidivism).

233

According to a national study of recidivism, 11% of prisoners released for a violent offense

are arrested during their first year following release for having committed a new violent offense,

compared to 9% of those released for a property offense and 7% of prisoners released for a drug

offense. A

LPER

., supra note 214, at 11.

2518 HARVARD LAW REVIEW [Vol. 133:2460

B. Interpretation and Implications

Our recidivism analysis should, at the very least, dispel any notion

that individuals who receive expungements under Michigan’s law pose

any particular risk of reoffending. And the same is likely to be true in

other states considering similar record-clearing laws. Employers and

landlords should rest assured that petition-based expungement laws

with comparable requirements and waiting periods are unlikely to deny

them access to information that they need to protect themselves or their

employees, customers, or tenants.

But what are the broader implications of this analysis for expunge-

ment policy, both in Michigan and beyond? In particular, do the data

allow us to say anything about the effects of broadening the availability

of expungement — either procedurally (for example, by rendering it au-

tomatic) or substantively (by loosening eligibility requirements)? Ex-

pungement recipients are, again, an unrepresentative sample of individ-

uals with criminal records. Primarily, they are self-selected — they had

the motivation to apply for expungement, the belief that their case could

be convincing to a judge, and the energy, money, and organizational

skills to complete the application process.

234

They are also filtered by

judges’ discretion, and judges presumably look for candidates who seem

to be at low risk of committing future crime. So we cannot assume that

subsequent crime rates for people who are not selected in the same way

would be as low. And if they were not as low, would the public-safety

objection then become legitimate?

To address these questions, we must consider two separate influences

on post-expungement recidivism rates. The first is baseline risk: the risk

of recidivism, absent expungement, of those same individuals.

235

The

second is the causal effect of expungement on these individuals’ subse-

quent criminal behavior.

236

Either or both could contribute to the low

recidivism rates that we see in our expungement sample — that is, re-

cipients could have been low risk regardless and/or the expungement of

their record could have lowered their risk. Unfortunately, we do not

have a good empirical technique for using our data to disentangle these

two theories. This is mainly because of the key legal requirement for

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

234

See supra p. 2490 & tbl.1; section II.C, pp. 2501–10 (showing that the vast majority of indi-

viduals who are legally eligible for an expungement do not succeed in obtaining one and arguing

that many potential recipients do not apply for record-clearing relief because of various institu-

tional, psychological, and practical barriers).

235

See, e.g., C

HRISTOPHER

LAPHAM

& J

AMES

ICHOLSON

, T

ONCISE

XFORD

ICTIONARY

ATHEMATICS

39 (5th ed. 2014) (defining baseline risk); cf. Natalie J. Jones et

al., Predicting Criminal Recidivism in Adult Male Offenders: Researcher Versus Parole Officer

Assessment of Dynamic Risk, 37 C

RIM

. J

UST

& B

EHAV

. 860, 861 (2010) (distinguishing baseline or

counterfactual risk from dynamic risk that incorporates changes in an individual’s situation over

time and is more tightly related to recidivism).

236

See, e.g., Adams et al., supra note 95, at 45 (arguing that expungement itself may cause a

reduction in recidivism); Selbin et al., supra note 43, at 57 (same).

2020] EXPUNGEMENT OF CRIMINAL CONVICTIONS 2519

expungement: nonrecidivism. Recidivism rates are by definition zero

before people become eligible for expungement and before they receive

one, which means that a study with a “pre-post” design cannot mean-

ingfully evaluate expungement’s effects on recidivism.

Accordingly, to inform a discussion of what seems most likely to ex-

plain the low post-expungement recidivism risk we estimate, we turn

for guidance to the criminological literature more broadly. Scholarship

in this area provides strong support for both explanations for our em-

pirical results. Expungement recipients probably have low baseline re-

cidivism risk at the time they receive an expungement, and the receipt

of an expungement probably lowers this risk further.

As to baseline risk, first, a key finding of the relevant research is that

most people who commit crimes do not continue to do so forever. Within

the large criminological literature on recidivism and patterns of de-

sistance from crime, a small but important subset of work focuses on

the extent to which older criminal records remain usefully predictive of

future offending.

237

In particular, it asks: When people have remained

crime-free for a given period of time, at what point do they become no

riskier, or at least not notably riskier, than the general population or

people without criminal records? Alfred Blumstein and Kiminori

Nakamura have labeled this concept “redemption” — suggesting that

the passage of some number of recidivism-free years should be under-

stood to “redeem” the individual in the eyes of society.

238

Estimated

times to redemption appear to vary across samples, crime types, age

cohorts, and methods, but they are usually in the range of four to ten

years, at least when juxtaposing individuals with records to the same

age cohort within the general population.

239

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

237

See, e.g., A

LFRED

LUMSTEIN

& K

IMINORI

AKAMURA

, E

XTENSION

URRENT

STIMATES

EDEMPTION

IMES

: R

OBUSTNESS

ESTING

, O

-S

TATE

RRESTS

AND

ACIAL

IFFERENCES

1–3 (2012), https://www.ncjrs.gov/pdffiles1/nij/grants/240100.pdf

[https://perma.cc/BD4R-4GMV]; Alfred Blumstein & Kiminori Nakamura, Redemption in the

Presence of Widespread Criminal Background Checks, 47 C

RIMINOLOGY

327, 331–39 (2009);

Shawn D. Bushway et al., The Predictive Value of Criminal Background Checks: Do Age and

Criminal History Affect Time to Redemption?, 49 C

RIMINOLOGY

27, 28–30 (2011); Megan C.

Kurlychek et al., Enduring Risk? Old Criminal Records and Predictions of Future Criminal In-

volvement, 53 C

RIME

& D

ELINQ

. 64, 71–78 (2007); Megan C. Kurlychek et al., Scarlet Letters and

Recidivism: Does an Old Criminal Record Predict Future Offending?, 5 C

RIMINOLOGY

& P

’

483, 492–98 (2006).

238

See B

LUMSTEIN

& N

AKAMURA

, supra note 237, at 8. This religiously derived language has

potential connotations (for example, the idea that those with records are “fallen”) that are probably

not intended by any of the scholars using it and that we do not endorse; still, we use the term here

in deference to the leading work on desistance from crime.

239

See generally Blumstein & Nakamura, supra note 237. If instead the contrast is drawn with

only people who have no prior criminal record (instead of the general population, which includes

people with records), then redemption studies tend to find that the recidivism risk of people with

records never converges completely to the risk levels posed by individuals without records, see id.

2520 HARVARD LAW REVIEW [Vol. 133:2460

One could easily interpret the low recidivism rates we find in section

III.A — and the favorable comparisons to general-population arrest

rates — as indications of redemption. Expungement recipients have all

gone at least five years without a subsequent conviction before receiving

their expungement; the median period between conviction and expunge-

ment is nearly ten years.

240

Moreover, by definition, expungement re-

cipients in our data were all individuals without any prior record when

they were convicted of their expunged offense — and recidivism studies

consistently find that the length of one’s existing criminal record is a

strong predictor of subsequent recidivism.

241

And, as we note above,

the group that receives expungements is self-selected and selected by

judges. Taken together, then, there are many reasons to believe ex-

pungement recipients may be at a low baseline risk for recidivism.

It is, however, also quite likely that receiving an expungement re-

duces a recipient’s recidivism risk below their previous baseline. This

possibility is supported by criminological research on the factors pre-

dicting recidivism.

242

Recidivism-related benefits of expungements are

likely to be mediated primarily by their effects on employment and wage

levels — which we assess below in Part IV. Unemployment is a moder-

ately strong predictor of recidivism.

243

Likewise, higher weekly wages

significantly reduce recidivism risk.

244

A study of women with records

indicates that “poverty status increases the odds of rearrest by a factor

of 4.6” and is a better predictor of rearrest than a recidivism-risk index

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

at 340–44; however, the difference eventually comes sufficiently close to zero that a reasonable ob-

server would conclude that the old record provides no useful or actionable information, see id. at

343 fig.4.

240

See supra tbl.2.

241

See, e.g., U.S. S

ENTENCING

OMM

’

, M

EASURING

ECIDIVISM

: T

RIMINAL

ISTORY

OMPUTATION

THE

EDERAL

ENTENCING

UIDELINES

10 (2004).

242

Indeed, although direct empirical evidence has long been lacking, this argument has often

been part of the policy case made for expungement laws. See, e.g., S

TANDARDS

FOR

RIMINAL

USTICE

ch. 19, introductory cmt., at 13, § 19-2.5, cmt., at 33 (A

. B

’

3d ed. 2004);

Deborah N. Archer & Kele S. Williams, Making America “The Land of Second Chances”: Restoring

Socioeconomic Rights for Ex-offenders, 30 N.Y.U.

. L. & S

. C

HANGE

527, 527–30 (2006);

Andrew von Hirsch & Martin Wasik, Civil Disqualifications Attending Conviction: A Suggested

Conceptual Framework, 56 C

AMBRIDGE

L.J. 599, 605 (1997); Jeremy Travis et al., Prisoner

Reentry: Issues for Practice and Policy, C

RIM

. J

UST

., Spring 2002, at 12, 17–18.

243

See, e.g., C

. M

AYORAL

OLICY

AUCUS

, R

EBUILDING

IVES

. R

ESTORING

OPE

TRENGTHENING

OMMUNITIES

15 (2006); P

AUL

ENDREAU

., C

FOR

RIMINAL

USTICE

TUDIES

, C

ASE

EEDS

EVIEW

: E

MPLOYMENT

OMAIN

5–7, 12 (2000),

https://www.csc-scc.gc.ca/research/092/r90_e.pdf [https://perma.cc/P89C-L8UM]; J

OAN

ETERSILIA

, W

HEN

RISONERS

OME

40–41 (2003); T

RAVIS

., supra note 31, at

31–33; Nicholas Freudenberg et al., Coming Home from Jail: The Social and Health Consequences

of Community Reentry for Women, Male Adolescents, and Their Families and Communities, 95 A

J. P

. H

EALTH

1725, 1729 (2005).

244

Samuel L. Myers, Jr., Estimating the Economic Model of Crime: Employment Versus

Punishment Effects, 98 Q.J.

CON

. 157, 163 (1983).

2020] EXPUNGEMENT OF CRIMINAL CONVICTIONS 2521

commonly used in corrections.

245

Homeless individuals with criminal

records are especially likely to recidivate,

246

while people who receive

educational programming while incarcerated are less likely to recidivate

upon release.

247

To the extent that criminal records limit access to qual-

ity housing, student loans, satisfying employment, and decent wages,

expungement should reduce recidivism by mitigating each of these soci-

oeconomic contributors to criminal behavior.

Another possibility is that expungement reduces recidivism by alle-

viating the social exclusion that is often associated with criminal records.

People with records face significant social stigma,

248

which is exacer-

bated by employers’ growing reliance on increasingly available online

criminal records databases.

249

Sociologists and criminologists have long

argued that social stigma and exclusion can contribute to criminal re-

cidivism.

250

Although little empirical support has been found for the

strongest claims of 1970s-era labeling theorists (for example, that label-

ing individuals renders criminal justice interventions actually counter-

productive on net),

251

stigma and exclusion may still be an important

part of the dynamic underlying recidivism.

252

All in all, it is highly plausible that expunging criminal records could

benefit public safety, given what we know from a large body of research

about recidivism patterns generally. In contrast, no similarly plausible

empirical support exists for the opposite claim, one that lies behind re-

sistance to expungement laws: that sealing records increases recidivism

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

245

Kristy Holtfreter et al., Poverty, State Capital, and Recidivism Among Women Offenders, 3

RIMINOLOGY

& P

. P

’

185, 198 (2004).

246

See Stephen Metraux & Dennis P. Culhane, Homeless Shelter Use and Reincarceration

Following Prison Release, 3 C

RIMINOLOGY

& P

. P

’

139, 151–53 (2004).

247

See S

TEPHEN

J. S

TEURER

& L

INDA

G. S

MITH

, C

ORR

. E

DUC

. A

’

, E

DUCATION

EDUCES

RIME

: T

HREE

-S

TATE

ECIDIVISM

TUDY

20–21 (2003), https://files.eric.ed.gov/

fulltext/ED478452.pdf [https://perma.cc/DA7J-KHDF]; John Nuttall et al., The Effect of Earning a

GED on Recidivism Rates, 54 J.

ORRECTIONAL

DUC

. 90, 92–94 (2003); Kristen M. Zgoba et

al., The Influence of GED Obtainment on Inmate Release Outcome, 35 C

RIM

. J

UST

. & B

EHAV

. 375,

376–77 (2008).

248

See Marc A. Franklin & Diane Johnsen, Expunging Criminal Records: Concealment and Dis-

honesty in an Open Society, 9 H

OFSTRA

L. R

. 733, 736–37 (1981).

249

See N

’

REEDOM

NFO

OAL

, P

ORTAL

OMPLIANCE

: A Q

UALITATIVE

NALYSIS

NLINE

UBLIC

ECORD

EQUEST

ERVICES

AJOR

U.S. C

ITIES

2–3

(2019), https://www.nfoic.org/sites/default/files/pages/2019-09/NFOIC%20Portal%20to%20Compliance.

pdf [https://perma.cc/PN2G-WC4G]; Kurlychek et al., Enduring Risk? Old Criminal Records and

Predictions of Future Criminal Involvement, supra note 237, at 65, 68.

250

See, e.g., William D. Payne, Negative Labels: Passageways and Prisons, 19 C

RIME

& D

ELINQ

33, 33–34 (1973); Charles R. Tittle, Deterrents or Labeling?, 53 S

. F

ORCES

399, 399–400 (1975)

(surveying the existing literature at the time on labeling “deviance” and whether labeling results in

recidivism or deterrence).

251

See Tittle, supra note 250, at 402–04 (refuting the labeling argument).

252

See David R. Karp, The New Debate About Shame in Criminal Justice: An Interactionist

Account, 21 J

UST

. S

. J. 301, 305–07 (2000).

2522 HARVARD LAW REVIEW [Vol. 133:2460

risk. To our knowledge, those raising this objection have never pre-

sented evidence supporting it, and its rationale is not obvious.

The intuition seems to be that those who know an individual’s crim-

inal background can take steps to protect themselves — for example, by

choosing not to hire the person. And as the empirical evidence we dis-

cuss in section I.A indicates, many U.S. employers do exactly that. But

while such a step might be in an individual employer’s or landlord’s

self-interest, it is hard to see how it would be in society’s interest overall,

given that lack of employment, unstable housing, and social stigma are

all risk factors for recidivism. Even if a person with a record is at an

elevated risk of committing crime, the employer who declines to hire

that person simply shifts risks to others, or even amplifies those risks.

253

No empirical research shows that the collective effect of decisions not to

hire or rent to people with records makes society safer. Notably, the

purported public-safety case against expungement follows the same

logic as the purported public-safety case in favor of sex offender regis-

tries with public notification requirements — yet empirical research has

shown that these laws may actually increase recidivism.

254

With that in mind, let’s return to the question we raise at the begin-

ning of this section. Suppose that a state were considering making ex-

pungement more broadly available to those with records: making it au-

tomatic for some subset of individuals with records, shortening the

waiting period, or loosening eligibility requirements. If these new clas-

ses of recipients have a higher baseline risk of recidivism than current

recipients do, their post-expungement recidivism rates will also proba-

bly be higher — maybe substantially so.

But from a policy perspective, this possibility is a red herring. The

higher recidivism risk would not be caused by the expungements or by

the law change that allowed them; it is simply a baseline risk, which

would exist with or without expungements. A higher recidivism rate is

not relevant to the cost-benefit analysis unless it is an effect of the law

change or the resulting expungements. But there is no credible evidence

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

253

There is at least one plausible counterargument. There may be situations in which someone

with a history that includes a particular crime applies for a position that would put them in an

especially favorable position to recommit the same or a similar offense. Further, that criminal op-

portunity would not be available at another location or with another employer. Many employers

and landlords might prefer to use a public criminal record to reduce the likelihood that this person

reoffends in that place or in that situation. However, while theoretically possible, situations of this

sort in which crime is not simply displaced seem likely to be very few, and, more importantly, the

negative effect that employment and housing difficulties can have on general recidivism factors is

likely to overshadow any benefit. Cf. J.J. Prescott, Portmanteau Ascendant: Post-release Regulations

and Sex Offender Recidivism, 48 C

ONN

. L. R

. 1035, 1040, 1057 (2016) (identifying this dynamic in

the context of sex offender post-release regulations).

254

J.J. Prescott & Jonah E. Rockoff, Do Sex Offender Registration and Notification Laws Affect

Criminal Behavior?, 54 J.L.

& E

CON

. 161, 181 (2011).

2020] EXPUNGEMENT OF CRIMINAL CONVICTIONS 2523

to support the idea that risk might grow because of the greater availa-

bility of expungement relief. Meanwhile, existing research provides firm

grounds to believe that, if anything, recidivism would fall. Put another

way, if expungement reduces recidivism, then granting it to a pool of

new recipients with higher underlying recidivism risk simply offers a

greater potential public-safety upside.

IV.

MPLOYMENT

UTCOMES

Although record clearing can simultaneously address a range of col-

lateral consequences of criminal convictions, probably the most im-

portant motivation for most expungement applicants

255

— and the most

important policy rationale for expungement laws — is to improve access

to employment. As we discuss in Part I, well-developed theoretical ac-

counts convincingly explain why criminal record expungement should

have positive employment consequences for recipients, but the empirical

evidence on the question remains extremely limited.

256

This is a huge

problem as advocates and policymakers negotiate the future of expunge-

ment policies. In this Part, we hope to at least begin to resolve it. Sec-

tion IV.A presents estimates and figures comparing pre- and post-

expungement wage and employment trajectories for our large sample of

recipients; these show large gains in both employment rates and wages

following an individual’s receipt of an expungement. Section IV.B ad-

dresses the question whether these gains can be interpreted as causal

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

255

A sample of individuals who received record-clearing relief in California “expressed frustra-

tion with their inability to gain employment [with their criminal record], noting that the limited

jobs open to them predominantly involved poor working conditions and low pay . . . . They genu-

inely hoped record clearance would increase their employment potential and facilitate their success-

ful reintegration into society.” Adams et al., supra note 95, at 30. In another survey of people

seeking expungement of criminal records in Chicago, participants reported suffering ongoing stigma

from their record: “Even participants with extensive histories who had remained ‘clean’ for lengthy

periods reported criminal record related stigma and material hardship.” Ispa-Landa & Loeffler,

supra note 32, at 398. For example, “a 44-year-old man who had 12 arrests on his rap sheet [and]

had been convicted of drug possession and unlawful use of a weapon 19 years earlier” said that “he

could easily find several references who would vouch for his work ethic and skills, but even these

personal testimonies did not compensate for his criminal record.” Id.

256

See Megan Denver et al., A New Look at the Employment and Recidivism Relationship

Through the Lens of a Criminal Background Check, 55 C

RIMINOLOGY

174, 176–77 (2017); see also

id. at 196; Pager, supra note 30, at 939–44; supra section I.C, pp. 2476–80. A handful of studies

have been conducted at the county level and have concluded that expungement leads to positive

employment outcomes overall. See Adams et al., supra note 95, at 45–47; Selbin et al., supra note

43, at 57–58; see also M

EYLI

HAPIN

., A C

OST

-B

ENEFIT

NALYSIS

RIMINAL

ECORD

XPUNGEMENT

ANTA

LARA

OUNTY

4 (2014), https://publicpolicy.

stanford.edu/publications/cost-benefit-analysis-criminal-record-expungement-santa-clara-county

[https://perma.cc/M2GZ-E5CB]; C

LEAN

LATE

ROGRAM

, O

FFICE

THE

. D

., C

ITY

S.F., 2007–2008 E

VA L UAT IO N

INDINGS

6 (2009), http://sfpublicdefender.org/wp-

content/uploads/sites/2/2009/05/clean-slate-evaluation-final-report.pdf [https://perma.cc/J6V2-9XWF].

2524 HARVARD LAW REVIEW [Vol. 133:2460

effects of receiving an expungement, offering additional results to help

us come to grips with this important question of interpretation.

A. Employment and Wage Trajectories for Expungement Recipients

The outcome data we use in this work are quarterly Unemployment

Insurance Agency (UIA) wage data collected for expungement recipi-

ents. The data run from the third quarter of 1997 through the second

quarter of 2013. The sample inclusion criteria we use to conduct our

analysis are as follows: (1) The individual received an expungement on

a known date between January 1998 and May 2011.

257

(2) The individ-

ual did not possess a non-Michigan driver’s license identified in our

criminal history data. (3) The individual matched to the UIA data in at

least one quarter during the period from 1997 to 2013.

258

And (4) The

individual is of working age (eighteen to sixty-four) for the entire time

over which we observe their wages for a particular regression.

We examine both employment status and wages as outcomes. We

structure our dataset for the research in this Part as a panel — that is,

we follow the same individuals before and after they receive their ex-

pungement, and each observation is a person-quarter.

259

For each anal-

ysis, we use a balanced panel, meaning that for every individual in our

analysis (regardless of the date of receipt), we include the same number

of quarters before and after the expungement.

260

Essentially, we use

each individual as their own control; we compare an individual’s post-

expungement employment-related trajectories with that same individ-

ual’s pre-expungement trajectories.

In Table 5, we report estimated average changes in employment

trends before and after an individual receives an expungement. The

different columns of Table 5 present these estimated changes in trends

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

257

Within each column in Table 5, the sample is further constrained to ensure that outcome data

exist for all person-quarters in the analysis. For example, if we have only five quarters of pre-

expungement data for particular individuals, we include them in the sample for Column 2, which

considers two quarters of data before and after expungement, but we exclude them from the sample

for Column 3, which includes eight quarters of data on either side of the expungement.

258

For those individuals who do match, we interpret an absence of reported wage data for a

particular quarter as implying the individual has no employment in that quarter. By contrast, we

take the total absence of someone from the wage data to imply a failure in the matching process

(for example, a missing or incorrect Social Security number, or a non-Michigan resident).

259

For a brief summary of the advantages of using panel data, see C

HENG

SIAO

, A

NALYSIS

ANEL

ATA

1–8 (2d ed. 2003).

260

See C

HRISTOPHER

F. B

AUM

, A

NTRODUCTION

ODERN

CONOMETRICS

SING

TATA

46–47 (2006); J

EFFREY

A. E

DWARDS

, B

UILDING

ETTER

CONOMETRIC

ODELS

SING

ROSS

ECTION

AND

ANEL

ATA

23 (2014); M

ARKUS

AY ER

, U

NBALANCED

ANEL

ATA

ODELS

(2010), https://homepage.univie.ac.at/robert.kunst/pan2010_pres_mayer.pdf

[https://perma.cc/K92A-FSXJ].

2020] EXPUNGEMENT OF CRIMINAL CONVICTIONS 2525

over different time windows of various widths; we place the expunge-

ment at the center of each window. We estimate the following equation

using ordinary least squares:

𝐸𝑚𝑝𝑙𝑜𝑦𝑒𝑑



= 𝛼 + 𝛾𝐸𝑙𝑎𝑝𝑠𝑒𝑑



+ 𝛿𝐸𝑙𝑎𝑝𝑠𝑒𝑑



× 𝑃𝑜𝑠𝑡



+ 𝜃𝑇𝑜𝑡𝑎𝑙𝐸𝑚𝑝𝑙𝑜𝑦𝑚𝑒𝑛𝑡



+ 𝜗𝑈𝑅𝑎𝑡𝑒



+ 𝜖



Our main outcome of interest, 𝐸𝑚𝑝𝑙𝑜𝑦𝑒𝑑, is an indicator for whether

an individual has any positive wages reported to UIA by any employer

in the quarter.

261

𝐸𝑙𝑎𝑝𝑠𝑒𝑑 is a linear measure of time, increasing by one

for each quarter; it is set to zero in the quarter before the individual

receives their expungement.

262

The coefficient we estimate on 𝐸𝑙𝑎𝑝𝑠𝑒𝑑

represents the underlying linear trend in employment in the quarters

prior to the expungement — that is, it is the average change per quarter

in the employment rate for the sample. A negative coefficient on

𝐸𝑙𝑎𝑝𝑠𝑒𝑑 would indicate that the group’s average employment rate was

declining prior to expungement receipt, and a positive coefficient would

indicate that it was increasing.

Our primary variable of interest, 𝐸𝑙𝑎𝑝𝑠𝑒𝑑 × 𝑃𝑜𝑠𝑡, captures the

change in the linear employment-rate trend after expungement receipt.

A positive coefficient on 𝐸𝑙𝑎𝑝𝑠𝑒𝑑 × 𝑃𝑜𝑠𝑡 implies that the employment-

rate trend improves post-expungement, and a negative coefficient con-

versely means it gets worse.

263

We estimate changes in the employment

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

261

“Any positive wages” is a very minimal threshold for deeming an individual “employed,” as

there is no minimum number of hours or total compensation requirement. Furthermore, the wages

could come only from one part of the quarter; the wage data are not further broken down tempo-

rally. On the other hand, absence of employment/wages as measured in these data does not neces-

sarily mean that the individual is truly “unemployed” in the sense that economists use the term —

that is, actively looking for a job, but so far unsuccessful. Being unemployed in our data simply

means that the individual does not have a wage- or salary-paying job. Virtually all employers in

Michigan are required to report wages, but the data do not include people who are self-employed,

nor do they differentiate between unemployed people who are looking for work and people who are

out of the workforce. See M

ICH

. C

OMP

. L

AWS

§ 421.13(2) (West 2020). If our real outcome

of interest is total lawful employment (including self-employment) among the working-age popula-

tion, then presence in the UIA data is a rough proxy for it. UIA-reported wages may also understate

total earnings. That said, these issues affect the data throughout the study period, and we do not

think they should substantially affect our estimation of trend changes from before an expungement

to after its receipt. If anything, classical measurement error like this tends generally to bias regres-

sion estimates toward zero — specifically, it may cause us to understate the true effect size of ex-

pungement. See J

EFFREY

M. W

OOLDRIDGE

, E

CONOMETRIC

NALYSIS

ROSS

ECTION

AND

ANEL

ATA

73–75 (1st ed. 2002); Andrew B. Abel, Classical Measurement Error with

Several Regressors 16–17 (May 9, 2017) (unpublished manuscript), http://finance.wharton.

upenn.edu/~abel/pdf_files_papers/attenuation%20bias%2016.pdf [https://perma.cc/VL72-2LG8].

262

Positive values of 𝐸𝑙𝑎𝑝𝑠𝑒𝑑 represent the number of quarters since the individual received

their expungement. We count the quarters prior to the expungement, running from negative values

up through quarter zero, which is the last quarter unaffected by the expungement.

263

The sum of the estimated coefficients of 𝐸𝑙𝑎𝑝𝑠𝑒𝑑 and 𝐸𝑙𝑎𝑝𝑠𝑒𝑑 × 𝑃𝑜𝑠𝑡 represent the post-

expungement trend in the employment rate for this group; a positive sum means employment levels

are increasing, and a negative sum means they are declining.

2526 HARVARD LAW REVIEW [Vol. 133:2460

trend, rather than a simple one-time change in the employment level, on

the assumption that the effects of an expungement may accumulate over

time rather than being instantaneous.

264

We estimate linear trends (as

opposed to more complicated curves) for simplicity of interpretation.

But estimating these trends over a variety of time windows can help us

to detect nonlinear patterns in the actual trajectory.

265

The remaining variables in our estimating equation are measures

for economic conditions during the calendar quarter of each observa-

tion. The time period of the analysis (1997–2013) includes considerable

economic fluctuations, most importantly the 2008 financial crisis and

subsequent deep recession; overall it was a period of economic decline

and rising unemployment in Michigan.

266

To account for these

changes in the labor market, we include Michigan’s quarterly unem-

ployment rate and the total number of people employed in the state as

reported by the Bureau of Labor Statistics.

267

We control for these

changing economic conditions because they could potentially affect the

pre- and post-expungement periods differently. In contrast, because of

the balanced-panel structure, fixed attributes of given individuals or

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

264

In addition, especially if there is an underlying trend over time before the expungement, esti-

mating changes in levels alone can be misleading. For example, if there is a continuous upward

trend throughout the sample period, the average level would be higher in the post-expungement

period even if the growth has nothing to do with expungement receipt; at the same time, if there is

a downward trend before the expungement that reverses after expungement receipt, the average

employment levels might look the same before and after, even though there may have been a dra-

matic change in trajectory.

265

The results are substantively similar if we use logistic regression for the binary outcome var-

iables, as we did in the Table 3 uptake analysis. Here, we prefer the linear probability model

because our key variable of interest is an interaction term, and interactions are notoriously difficult

to interpret in logistic regression. See Edward C. Norton et al., Computing Interaction Effects and

Standard Errors in Logit and Probit Models, 4 S

TATA

J. 154, 154–56 (2004). Linear probability

models also tend to perform well in the middle of the probability distribution, a condition that is

satisfied here (but not in the Table 3 analysis). See Paul von Hippel, Linear vs. Logistic Probability

Models: Which Is Better, and When?, S

TATISTICAL

ORIZONS

(July 5, 2015), https://

statisticalhorizons.com/linear-vs-logistic [https://perma.cc/7XZB-EAA3].

266

See Erin Duffin, Unemployment Rate in Michigan from

1992

2019

, S

TATISTA

(Mar. 5,

2020), https://www.statista.com/statistics/189438/unemployment-rate-in-michigan-since-1992 [https://

perma.cc/YA9U-4LWY]; Dustin Walsh, A Decade After Financial Crisis, Michigan Still Recovering,

RAIN

’

ETROIT

. (Sept. 16, 2018, 12:01 AM), https://www.crainsdetroit.com/economy/

decade-after-financial-crisis-michigan-still-recovering [https://perma.cc/7RA7-UNSF]. We observe

each individual in the sample only for much shorter — and differing — parts of this period, so

fluctuations will affect each of them differently, which should reduce the aggregate effect of this

variation, if any, on our estimates.

267

See generally Geographic Profile of Employment and Unemployment: Archive, U.S. B

UREAU

. S

TAT

., https://www.bls.gov/opub/geographic-profile/archive.htm [https://perma.cc/ZAS2-

LE52] (reporting, among other things, annual state averages for total employment and the unem-

ployment rate among the civilian noninstitutional population).

2020] EXPUNGEMENT OF CRIMINAL CONVICTIONS 2527

their prior criminal records can be safely ignored, since they are

equally present before and after the receipt of the expungement.

268

In each version of the regression results in Table 5,

269

we estimate a

substantial and statistically significant (p < 0.001) upward turn in the

employment-rate trajectory of expungement recipients after they receive

their expungements.

270

The coefficients, representing per-quarter gains,

decline in magnitude as the window gets larger (although they remain

significant), which intimates that the improvements are steepest in the

first two quarters and slowly become more gradual; this can also be seen

in the net gain and proportional net gain calculations we report in Table

271

By the end of one year, after controlling for the pre-receipt trend

and other potentially confounding labor-market variables, expungement

recipients gain nearly eight percentage points in their employment rate;

proportionally, they are 1.13 times as likely to be employed. The net

gain following expungement is about the same in the two-year observa-

tion period and slightly lower in the three-year period, suggesting not

only that the employment improvements following an expungement

come relatively quickly, but also that they are largely sustained in the

following years.

In Panel B of Table 5, we show the same type of analysis, but sub-

stitute a different, less minimalist definition of the employment outcome

variable. Instead of requiring only that any wages have been earned at

any time in the quarter,

mployed

now requires individuals to have

earned an average of at least $100 a week ($1300 total for the quarter)

before they count as employed. This is still a threshold that falls well

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

268

The standard errors that we estimate for these regressions are clustered at the person level;

clustering in this way means that the reported precision of our estimates accounts for the fact that

our quarterly observations for each person are not independent of one another.

269

The “window” range we indicate at the top of each column in Table 5 reflects the number of

quarters before and after expungement receipt and thus the sample over which we estimate our

regressions; for example, “+/- 4” means that we observe individuals for one year before (four quar-

ters) and one year after the expungement.

270

The p-value represents the probability that a randomly drawn sample of the size used by the

researcher would produce the observed coefficient (or a more extreme one) if the true coefficient of

the relationship in the underlying population were zero. It is a way of expressing statistical impre-

cision due to sampling error. See B.S.

VERITT

& A. S

KRONDAL

, T

AMBRIDGE

ICTIONARY

TATISTICS

307 (4th ed. 2010) (defining “null hypothesis”); Valen E. Johnson, Revised Standards

for Statistical Evidence, 110 PNAS 19313, 19313 (2013). A p-value of .05 or less is generally con-

sidered statistically significant, though even lower is better. See Johnson, supra, at 19313, 19315.

271

Near the bottom of each Panel in Table 5, “net gain” and “proportional net gain” show the

estimated percentage-point and proportional gains in employment experienced by the average ex-

pungement recipient by the window’s end, relative to what would have occurred had they stayed

on their pre-expungement trajectory. We calculate “net gain” by multiplying the estimated

𝐸𝑙𝑎𝑝𝑠𝑒𝑑 × 𝑃𝑜𝑠𝑡 coefficient by the number of quarters observed in the post-expungement period in

the column. We calculate “proportional net gain” by dividing the net gain by the average employ-

ment level in the quarter before expungement receipt (which is 61.7%).

2528 HARVARD LAW REVIEW [Vol. 133:2460

short of representing full-time employment.

272

Even so, by using this

definition, we at least exclude truly trivial earnings that may be causing

people who are for all practical purposes unemployed to be coded for

our analysis as employed. And, importantly, it may allow us to pick up

on some differences that the main

mployed

coding misses.

And indeed, the results in Panel B show substantially larger trend

changes than we display in Panel A. The general pattern is the same —

steeper trend changes in the narrower windows — but the magnitudes

are larger. By the end of one year, we now see a twelve-percentage-

point net gain in employment relative to the pre-expungement trajec-

tory. This implies that expungement recipients are 1.23 times as likely

to be making at least $100/week as they would have been (53% are em-

ployed at baseline using this measure of employment). As before, the net

gains are similar in the two-year window and slightly lower in the three-

year window, indicating that the employment gains largely endure but

do not continue to increase after the first year.

In Panel C of Table 5, we estimate changes in the average wage tra-

jectory of individuals following expungement receipt. We find a similar

pattern, with a steep gain in the first year followed by a subsequent

plateau in gains. By the end of the first year, relative to the pre-receipt

trend, recipients gain an average of $1111 in quarterly wages (that is,

$4444 per year), which is a 23% improvement over the pre-expungement

average; the net gain increases to $1234 in the two-year estimate, a 25%

improvement. The proportional increase in wages is much larger than

the increase in the employment rate in Panel A, but very close to the

gain we calculate using the alternative employment measure in Panel B.

This pattern implies that the increase in average wages may substan-

tially be explained by unemployed people obtaining employment or by

very minimally employed people securing more hours or higher-paying

work following their expungement. In Figures 3A, 3B, and 3C, we

demonstrate visually the employment and wage patterns that underlie

these regression estimates.

273

These graphs show only raw averages by

quarter before and after expungement receipt; they are not regression-

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

272

The minimum wage in Michigan was in the vicinity of $7 per hour throughout our sample

period. See Changes in Basic Minimum Wages in Non-farm Employment Under State Law:

Selected Years

1968

2019

, U.S. D

’

.: W

AGE

& H

OUR

IVISION

(Jan. 2020),

https://www.dol.gov/agencies/whd/state/minimum-wage/history [https://perma.cc/MVJ7-ZELL].

Therefore, this measure of employment represents about fourteen hours of work per week at min-

imum wage, or less at a higher wage; $100/week is less than one-fifth of the median wage among

persons with any wage in our sample.

273

In these figures, we do not model changes in employment outcomes as linear trends during

the pre-expungement and post-expungement periods. We simply show the employment rate and

average wage for each quarter beginning two years before and ending two years after the expunge-

ment (including quarter 0, the quarter immediately preceding the expungement quarter, which is

marked with a vertical dotted line), and then we connect the dots. The sample we use to create the

graphs is the same that we use for Column 3 in Table 5, Panels A–C. The graphs all have similar

patterns — a clear V-shape immediately surrounding the expungement.

2020] EXPUNGEMENT OF CRIMINAL CONVICTIONS 2529

(1) (2) (3) (4)

Window (Quarters) +/- 2 +/- 4 +/- 8 +/- 12

A. Employment Rate

(= any wages)

Elapsed × Post 0.0260*** 0.0198*** 0.00945*** 0.00521***

(0.003) (0.002) (0.001) (0.0006)

Elapsed -0.00710*** -0.00649*** -0.00346*** -0.00237***

(0.002) (0.001) (0.0006) (0.0004)

Net Gain

0.052 0.079 0.076 0.063

Proportional Net Gain

8.4% 12.8% 12.3% 10.2%

Number of Observations 71,899 125,451 226,525 301,089

B. Employment Rate

(> $100/week)

Elapsed × Post 0.0367*** 0.0308*** 0.0149*** 0.00838***

(0.003) (0.002) (0.001) (0.001)

Elapsed -0.0144*** -0.0124*** -0.00562*** -0.00320***

(0.002) (0.001) (0.0006) (0.0004)

Net Gain

0.073 0.123 0.119 0.101

Proportional Net Gain

13.8% 23.2% 22.4% 18.9%

Number of Observations 71,899 125,451 226,525 301,089

C. Quarterly Wages

(U.S. dollars)

Elapsed × Post 314.5*** 277.8*** 154.2*** 92.22***

(36.77) (18.12) (10.38) (7.80)

Elapsed -160.5*** -113.5*** -43.57*** -14.95*

(22.26) (11.78) (7.32) (6.10)

Net Gain

$628.98 $1,111.35 $1,233.37 $1,106.69

Proportional Net Gain

12.8% 22.6% 25.1% 22.5%

Number of Observations 71,899 125,451 226,525 301,089

Controls

BLS Total Employment

XXXX

BLS Unemployment Rate

XXXX

TABLE 5. CHANGES IN EMPLOYMENT AND WAGE TRENDS

AFTER EXPUNGEMENT

(Full Employment-Outcomes Sample)

Notes: Outcome data come from Michigan’s Unemployment Insurance Agency (UIA). The “window” refers to

the length of the evaluation period and indicates the number of quarters before and after the expungement —

for example, “+/- 2” means a four-quarter (one year) evaluation period, with two quarters before and two

quarters after the expungement. The sample includes quarterly data from individuals who received a known-

date expungement between January 1998 and May 2011, did not have an out-of-state driver’s license, matched

to the UIA data in at least one quarter of the sample period, and were of working age (eighteen to sixty-four) for

the entire observation period indicated in the column’s window. Standard errors clustered on the person are

reported in parentheses.

⁺

, *, **, and *** represent significance at the 10%, 5%, 1%, and 0.1% level, respectively.

2530 HARVARD LAW REVIEW [Vol. 133:2460

adjusted to control for economic trends in Michigan. And, notably, the

downward overall pattern in each of the graphs reflects the declining

state of Michigan’s economy over the course of our study period. But

what is important to understand for our purposes is not that underlying

decline but the change in the direction of the graph in the period imme-

diately after the expungement — the sudden, V-shaped upturn.

The results that we show here are robust; that is, similar patterns

consistently continue to appear even if we vary the details of the regres-

sion specification or the sample definition.

274

For example, there are

certain additional economic control variables that could potentially im-

prove our ability to account for the influence of economic fluctuations,

but which we do not add to the main specification because we do not

have the data for all years. These include the average quarterly earnings

for Michigan from the Quarterly Workforce Indicators series (available

from 2001) and average quarterly wages and employment rates for ex-

pungement nonrecipients in our own data (available from 2004). For

the years that they are available, adding these variables into our analysis

does not change the results. In addition, one might worry that, nonetheless,

IGURE

MPLOYMENT

ATE

(

AGES

)

EFORE

AND

FTER

XPUNGEMENT

Notes: Quarter 1 (immediately after the dotted line) is the quarter of expunge-

ment receipt. The sample consists of individuals who received a known-date

expungement between January 1998 and May 2011, did not have an out-of-

state driver’s license, matched to our UIA data in at least one quarter during

the period from 1997 to 2013, and were of working age (eighteen to sixty-four).

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

274

Cf. R

ICARDO

ARONNA

OBUST

TATISTICS

HEORY

AND

ETHODS

(

WITH

R) 1–2 (2d ed. 2019) (“Classical statistical inference quantities . . . can be adversely influ-

enced by the presence of even one outlier in the data. In contrast, appropriately constructed robust

versions of those inference quantities are little influenced by outliers.”).

50%

60%

70%

80%

Employment Rate

-8 -6 -4 -2 0 2 4 6 8

El apsed Quarters

Elapsed Quarters

Employment Rate

80%

70%

60%

50%

2020] EXPUNGEMENT OF CRIMINAL CONVICTIONS 2531

IGURE

MPLOYMENT

ATE

$100/

WEEK

)

EFORE

AND

FTER

XPUNGEMENT

IGURE

VERAGE

UARTERLY

AGES

EFORE

AND

FTER

XPUNGEMENT

Notes: Quarter 1 (immediately after the dotted line) is the quarter of expunge-

ment receipt. The sample consists of individuals who received a known-date

expungement between January 1998 and May 2011, did not have an out-of-

state driver’s license, matched to our UIA data in at least one quarter during

the period from 1997 to 2013, and were of working age (eighteen to sixty-four).

50%

60%

70%

80%

Empl oyment Rate (> $100/w eek )

-8 -6 -4 -2 0 2 4 6 8

El apsed Quarters

3000

4000

5000

6000

7000

Average Quar terl y Wages ($)

-8 -6 -4 -2 0 2 4 6 8

Elapsed Quarters

80%

70%

60%

50%

Employment Rate (> $100/week)

Elapsed Quarters

Average Quarterly Wages ($)

7000

6000

5000

4000

3000

2532 HARVARD LAW REVIEW [Vol. 133:2460

these controls might somehow not fully account for the major swings

introduced by the financial crisis and recession. But the same patterns

in our data persist if we simply drop that entire time period, such that

the windows for each analysis end before 2008 begins. The same sub-

stantive story also surfaces if we leave in the sample all of the individu-

als who do not match to the wage data at all; for our main sample def-

inition, we assume their absence was likely just a failure of the

matching process, but they could also be people who were not working

for the entire period. If we include them and count them as having

zeroes for all of the outcome variables, the employment rate and average

wages drop, and our point estimates decline proportionally, but the pro-

portional net gain estimates remain the same.

We also conduct these analyses separately for different subpopula-

tions to see whether expungement affects different groups of recipients

differently. We find substantial, and largely similar, trend changes in

every subgroup. The most conspicuous pattern we discover is that wage

and employment gains are much larger for women, at least in propor-

tional terms (women have lower baseline wages). For example, in the

one year after expungement, we estimate that men’s wages increase by

17%, while women’s wages appear to increase by 30%. We are unable

to explain the difference, as there are many possible explanations; for

example, perhaps women apply disproportionately to job types for

which a criminal record is a particularly serious barrier.

275

Studies of

other kinds of job-access interventions, like job training, also generally

find much larger effects for women, so the difference in our data may be

a byproduct of a more general phenomenon.

276

We also find larger ef-

fects for black expungement recipients than for white recipients, but

this difference is not as large (one-year wage gains of 25% versus 18%).

Among both men and women, people who expunge a felony see only

slightly larger gains than those who expunge a misdemeanor (one-year

wage gains of 33% versus 29% among women, and 18% versus 16% for

men). This difference seems surprisingly small, given that felonies carry

more collateral consequences and presumably more labor-market

stigma.

277

However, relatively few misdemeanors are expunged in the

first place, so they may represent a highly selected subset — for example,

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

275

A 2019 Bureau of Labor Statistics survey reports that women dominate fields like social work

(81.9% women), teaching young children (98.7% women), and home health care (88.3% women) —

they are more likely to work with sensitive information and vulnerable populations. Labor Force

Statistics from the Current Population Survey, U.S.

UREAU

TAT

, https://www.

bls.gov/cps/cpsaat11.htm [https://perma.cc/ENU7-EML5] (last updated Jan. 22, 2020). More em-

ployers in these fields probably conduct background checks — in fact, background checks are re-

quired by law for any childcare provider. Background Checks: What You Need to Know,

HILD

ARE

GOV

, https://www.childcare.gov/index.php/consumer-education/background-checks-

what-you-need-to-know [https://perma.cc/GZ59-BVNR].

276

See infra note 341 and accompanying text.

277

See Christopher Uggen et al., The Edge of Stigma: An Experimental Audit of the Effects of

Low-Level Criminal Records on Employment, 52 C

RIMINOLOGY

627, 647 (2014).

2020] EXPUNGEMENT OF CRIMINAL CONVICTIONS 2533

perhaps the misdemeanors people choose to expunge tend to be partic-

ularly stigmatizing for them. In any case, the estimates for all crime

categories and for all categories of individuals are large and significant.

B. Interpretation: Expungement Effect, Motivation,

or Mean Regression?

Our analysis demonstrates that expungement is associated with large

improvements in the employment rate and wages on average — and, in

particular, a reversal of the pre-expungement downward trend that we

observe for recipients as a group. Can we conclude from this evidence

that expungements cause these improvements? The answer is not obvi-

ous. A causal interpretation of the trend-change estimates above de-

pends on the assumption that, in the absence of these individuals receiv-

ing expungements, their collective pre-expungement employment and

wage trajectories would have continued on the same path. But is that

true? Absent expungement, is there some other force that might reverse

the downward slide at about the same time as receipt and give rise to

the same V-shaped pattern we observe? The control variables in the

regression already account for the role of changing labor-market trends,

which would otherwise be one plausible explanation. But there is still

the possibility of omitted variable bias — there could be something else

going on that we cannot measure or observe that accounts for these pat-

terns, other than the receipt of an expungement.

278

The most likely candidate is motivation. As we detail in Part II, it

takes a lot of effort, as well as a financial investment, to obtain an ex-

pungement.

279

Moreover, most expungement recipients choose not to go

through this effort as soon as they become eligible — three-quarters wait

more than a year, and 44% wait more than five years.

280

So, what mo-

tivates people who have waited several years to decide that now is the

time to apply? From an inference perspective, it would be ideal if their

decision were random, but it probably is not. For many, it may be that

they are trying to get a job — perhaps because they have been laid off,

or because they are seeking out better pay.

281

This latter theory finds

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

278

Omitted variable bias (OVB) (or selection bias, in this case) is almost always a challenge in

observational studies that seek to identify treatment effects. See Helene Starks et al., The Challenge

of Selection Bias and Confounding in Palliative Care Research, 12 J.

ALLIATIVE

. 181, 181

(2009), https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2982715/pdf/jpm.2009.9672.pdf [https://

perma.cc/YJH8-EKUZ]. However, in at least some employment-related contexts, OVB may not ulti-

mately result in significant bias. See generally Marco Caliendo et al., Unobservable, but

Unimportant? The Relevance of Usually Unobserved Variables for the Evaluation of Labor Market

Policies, 46 L

. E

CON

. 14 (2017).

279

See supra section II.A, pp. 2488–93; supra section II.C, pp. 2501–10.

280

See supra fig.1 and accompanying text.

281

Cf. Adams et al., supra note 95, at 30 (noting that prior to record clearance, “[q]uintessential

good jobs — those in specialized fields such as health care; jobs that provided good benefits such

2534 HARVARD LAW REVIEW [Vol. 133:2460

some support in our findings in Column 5 of Table 3: People are more

likely to receive an expungement within a year of a job loss or substan-

tial decline in wages. And a person who is motivated to improve their

employment situation to the point that they are willing to go through

the burdensome expungement process is probably taking other steps to

better their prospects, too — like applying for jobs, obviously. So, if we

see a turnaround in their employment trajectory just after the receipt of

their expungement, could the change actually be due to these other steps,

with the expungement merely incidental?

A related causal-inference concern is the possibility that our empiri-

cal findings are simply the product of statistical regression toward the

mean.

282

Consider a classic example often used to explain the concept:

both very tall parents and very short parents tend, on average, to have

children who are a little closer to the population’s average height than

they are.

283

Although height is substantially hereditary, people do not

turn out exactly like their parents; there is randomness involved in gene

mixing and in the operation of other influences on height. Parents who

are extremely tall or short are likely that way partly because of random

noise, and their children are unlikely to get the same roll of the dice.

284

Similar patterns are found in other real-world phenomena involving

random fluctuations — extreme outcomes tend to be followed by less

extreme ones

285

for the straightforward reason that most outcomes are,

by definition, not extreme.

286

A similar phenomenon could explain some of our results. Suppose

there is a certain amount of random variation in individual wages and

employment status (or as good as random for our purposes — uncorre-

lated with the other variables of interest in our analysis). Suppose fur-

ther that people tend to apply for expungement when they have recently

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

as health insurance, retirement plans, etc.; and jobs with the local, state, or federal government —

were particularly difficult to secure”); Ispa-Landa & Loeffler, supra note 32, at 389 (“Historically,

[criminal justice system (CJS)–marked individuals] could find employment in the manufacturing

sector and other portions of the economy that had less direct contact with members of the general

public. However, with deindustrialization, retail and other service sector jobs have largely replaced

manufacturing jobs. These pay lower initial wages and offer fewer opportunities for wage progres-

sion. Furthermore, because retail and service sector jobs involve extensive contact with the public,

employers may be especially reluctant to hire CJS-marked individuals.” (citations omitted)).

282

For an explanation and examples of the statistical phenomenon, see generally Adrian G.

Barnett et al., Regression to the Mean: What It Is and How to Deal with It, 34 I

’

PIDEMIOLOGY

215 (2005).

283

Charlotte M. Wright & Tim D. Cheetham, The Strengths and Limitations of Parental Heights

as a Predictor of Attained Height, 81 A

RCHIVES

ISEASE

HILDHOOD

257, 257 (1999) (“How-

ever, where parents were unusually tall or short, their children were relatively less tall or short,

respectively . . . .”).

284

See J. Martin Bland & Douglas G. Altman, Statistics Notes: Regression Towards the Mean,

308 B

RIT

. M

. J. 1499, 1499 (1994) (“This is a statistical, not a genetic phenomenon.”).

285

See Barnett et al., supra note 282, at 217.

286

See John R. Nesselroade et al., Regression Toward the Mean and the Study of Change, 88

SYCHOL

. B

ULL

. 622, 624–25 (1980).

2020] EXPUNGEMENT OF CRIMINAL CONVICTIONS 2535

experienced an employment setback, so they are more likely to seek re-

lief when they are on the downside of one of these random fluctuations.

Then, even leaving to one side any broader job-search motivation that

may be correlated with an individual interest in expungement, one

might expect that downward trend to reverse itself at around the time

an individual receives an expungement. Our analysis assumes that an

existing downward trend in employment outcomes observed across a

sample should be expected to continue without some clear reason for it

to change — like receiving an expungement. Such an expectation is

sensible when the underlying trend is driven by durable, real-world fac-

tors. But if the trend is not substantive but rather the product of ran-

dom noise, then such an expectation can confound causal inference and

lead one astray, especially when a trough in employment circum-

stances activates an understandable tendency among those eligible to

apply for an expungement, inducing a spurious correlation.

Unfortunately, there is no entirely clean way to unravel these causal

conundrums in our data. In truth, it is likely that all of these factors

contribute to the large apparent employment effects we find — that is,

expungements do bring about employment gains, but the gains caused

by expungement receipt are not quite as large as those we estimate

above. Indeed, the effects of motivation spillovers and record clearing

might be expected to be mutually reinforcing: perhaps the motivation

drives the job search, but the expungement makes the search more likely

to be successful (and that prospect in turn may increase motivation),

suggesting an even more reticulated causal story.

Fortunately, our data do provide a couple of reasons to believe that

causal effects of expungements probably explain at least a large part of

the trend changes we observe. The first reason concerns the exact tim-

ing of the upturn. The motivation and mean-regression theories both

hinge on the idea that people apply for expungements at particular times:

when they are motivated to seek work and/or when they have recently

experienced the loss of a job or wages. But there is a significant period

of time between application and receipt, and our analysis focuses on the

date of receipt of the expungement. It takes at least a few months to

receive an expungement,

287

and this time lag implies that, in our quar-

terly wage data, any motivation or mean-regression effects should begin

to be visible even two quarters before and definitely one quarter before

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

287

Lamentably, we do not know the date of the application for expungement in the great ma-

jority of observations in our data, and the time lag between application and receipt reportedly

varies by a fair amount, depending on geography and so forth. But our conversations with ex-

pungement lawyers in Michigan and with representatives of the Michigan State Police suggest

that it is very hard to complete the process in less than two months, and that (while some jurisdic-

tions have recently achieved efficiency improvements) three to six months was a more common lag

during the years covered by our study. See supra note 161 and accompanying text.

2536 HARVARD LAW REVIEW [Vol. 133:2460

expungement receipt. We might expect to detect motivation effects even

earlier because the impetus to seek a job might often be around for a

while before any expungement application.

But what we see in the data does not seem to match this alternative

story — or at least, it matches it at best only weakly. In Figures 3A, 3B,

and 3C, the first quarter in which there is a change in trajectory — an

upswing from the prior quarter — is quarter 1, the quarter in which the

individual actually receives the expungement.

288

In that quarter, we

spot a very slight improvement from the previous quarter — which does

represent a substantial change, stopping the previous downward trajec-

tory — and a much steeper upward pivot between quarters 1 and 2 for

the first employment curve (Figure 3A), and between quarters 2 and 3

for Figures 3B and 3C. This type of pattern is very much what we

would expect to observe if expungement receipt (and not other factors

correlated with an expungement application) drives the change in tra-

jectory.

289

If the shift in the trend were instead triggered by something

happening a quarter or two earlier (e.g., the motivation accompanying

the choice to apply for an expungement), we would likely see a similar

pattern, only shifted a quarter or two to the left. But we see no trend

change until quarter 1 in any of these graphs.

290

It is possible that what

occurs between quarter 0 and quarter 1 includes some component of the

mean-reversion or motivation effect — we cannot rule this out. But the

pattern is more consistent with changes being driven at least quite sub-

stantially by expungement receipt.

Second, we provide a further test of this causal theory in Table 6 by

analyzing a specific subset of our sample: those who receive expunge-

ments within one year of becoming eligible to apply. As we explain

above, we consider these cases (accounting for 25% of all expungements,

and 19% of the wage-linked sample) to be primarily “pent-up demand”

cases — people just waiting to become eligible.

291

These individuals are

obviously motivated to obtain an expungement, but there is no reason

to believe that this motivation would have materialized just at the time

of their application or, alternatively, as a consequence of some recent

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

288

See supra section IV.A, pp. 2524–33.

289

We only discern about half of the trend change in the partial post-expungement quarter

(quarter 1); the trend change appears in the figures in full bloom beginning only in the first quarters

that fall fully in the post-expungement period.

290

Indeed, in the wage graph (Figure 3C), the fall between quarters –1 and 0 is particularly

steep.

291

See supra notes 160–161 and accompanying text.

2020] EXPUNGEMENT OF CRIMINAL CONVICTIONS 2537

(1) (2) (3) (4)

Window (Quarters) +/- 2 +/- 4 +/- 8 +/- 12

A. Employment Rate

(= any wages)

Elapsed × Post 0.0231*** 0.0173*** 0.00884*** 0.00521***

(0.007) (0.004) (0.002) (0.0016)

Elapsed -0.00151 -0.00529* -0.00217+ -0.00213*

(0.004) (0.002) (0.0013) (0.0010)

Net Gain 0.046 0.069 0.071 0.063

Proportional Net Gain 7.6% 11.4% 11.6% 10.3%

Number of Observations 13,386 23,184 42,075 55,293

B. Employment Rate

(> $100/week)

Elapsed × Post 0.0230** 0.0215*** 0.0125*** 0.00742***

(0.007) (0.004) (0.002) (0.002)

Elapsed -0.00364 -0.00605** -0.00217 -0.00132

(0.004) (0.002) (0.0013) (0.0011)

Net Gain 0.046 0.086 0.100 0.089

Proportional Net Gain 8.8% 16.5% 19.2% 17.1%

Number of Observations 13,386 23,184 42,075 55,293

C. Quarterly Wages

(U.S. dollars)

Elapsed × Post 227.8** 211.1*** 128.6*** 59.70***

(70.00) (39.96) (23.11) (17.24)

Elapsed -17.76 -25.56 9.292 31.45*

(41.44) (24.46) (15.95) (13.08)

Net Gain $455.68 $844.25 $1,028.88 $716.45

Proportional Net Gain 10.6% 19.6% 23.9% 16.6%

Number of Observations 13,386 23,184 42,075 55,293

Controls

BLS Total Employment

XXXX

BLS Unemployment Rate

XXXX

TABLE 6. CHANGES IN EMPLOYMENT AND WAGE TRENDS

AFTER EXPUNGEMENT

(Expungement in First-Eligible-Year Subsample)

Notes: Outcome data come from Michigan’s Unemployment Insurance Agency (UIA). The “window” refers to

the length of the evaluation period and indicates the number of quarters before and after the expungement —

for example, “+/- 2” means a four-quarter (one year) evaluation period, with two quarters before and two

quarters after the expungement. The sample period runs from the third quarter of 1997 through the second

quarter of 2013. The sample includes quarterly data from individuals who received a known-date

expungement between January 1998 and May 2011, did not have an out-of-state driver’s license, matched to the

UIA data in at least one quarter of the sample period, were of working age (eighteen to sixty-four) for the entire

observation period indicated in the column, and received an expungement within their first year of eligibility.

Standard errors clustered on the person are reported in parentheses.

⁺

,*,**,and***representsignificance atthe

10%, 5%, 1%, and 0.1% level, respectively.

2538 HARVARD LAW REVIEW [Vol. 133:2460

employment setback.

292

Motivation that is longer term rather than spe-

cific to the particular time around the expungement is unlikely to con-

found our results because it would not explain a turnaround in the trend

at the time of expungement receipt. Likewise, we worry less about re-

gression to the mean when we study this sample because there is less

reason to believe that an individual’s choice to apply at the culmination

of the waiting period is motivated by finding themselves at the bottom

end of a random fluctuation in their employment outcomes; rather, they

are simply applying the first chance they get.

The regressions in Table 6, Panels A–C, directly parallel those in

Tabl e 5, Panels A–C; we carry out the same analyses on the smaller

“pent-up” sample. Table 6 demonstrates very strong gains in employ-

ment and wages after receiving an expungement — changes that remain

statistically significant (p < 0.001 for almost all estimates) despite the

loss of power accompanying the sample’s smaller size.

293

Early ex-

pungement recipients become, by two years after expungement receipt,

1.12 times as likely to be employed at all using our “any wages” measure

(a seven percentage point gain) and 1.2 times as likely to be employed

for at least $100 per month (ten percentage points), and they earn 1.24

times their prior wages on average.

The most important difference between these results and our Table 5

findings concerns the pre-expungement underlying trend, which is nega-

tive in all Table 5 specifications. By contrast, the early-expungement

group exhibits no significant prior trend for most windows and outcome

variables, consistent with our theory of pent-up demand.

294

In other

words, the arrival of the five-year threshold for these individuals is ef-

fectively random; it does not coincide with any particular pattern in pre-

expungement employment outcomes. Thus, mean regression cannot ex-

plain the subsequent trend changes because there is no prior downturn

from which to regress. It is also hard to explain by a job search — there

is no prior setback to explain its arrival, and moreover, the timing of the

expungement for early recipients seems to be driven mechanically by the

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

292

Cf. Peter M. Gollwitzer, Mindset Theory of Action Phases, reprinted in 1 H

ANDBOOK

HEORIES

OCIAL

SYCHOLOGY

526, 527 (Paul A.M. Van Lange et al. eds., 2012) (explaining

that under the Rubicon model for analyzing motivation, a person’s course of action may be sepa-

rated into phases that include a predecision phase — where a person begins setting goals — and a

preaction phase — where a person begins goal-directed behaviors).

293

The timing of the largest gains is slightly later (the curves are slightly less steep in the six-

month window), but by the end of two years, the net gain in proportional terms is essentially

identical to what we observe for the broader group in Table 5.

294

Table 6 offers evidence in support of our causal account of expungement receipt: the flat prior

trend, which is followed by a steeply positive post-expungement trend, is inconsistent with early-

expungement recipients applying for expungement in the wake of an employment downturn. Ad-

mittedly, a few of our specifications hint at a small negative pretrend in employment outcomes,

but others actually produce positive coefficients on 𝐸𝑙𝑎𝑝𝑠𝑒𝑑, our trend variable.

2020] EXPUNGEMENT OF CRIMINAL CONVICTIONS 2539

eligibility clock.

295

Perhaps the absence of mean regression and motiva-

tion effects slightly dampens the immediate effects we estimate in the

shortest time window, but by two years the gains are just as large. This

result provides quite good reason to believe that the employment and wage

gains are substantially caused by expungement.

In Figure 4, we present a variant of the wage graph from Figure 3C;

this version, perhaps, provides more compelling visual evidence of the

causal explanation of our results. This analysis is limited to the early-

expungement sample, and it is also residualized — that is, we use re-

gression analysis to remove the underlying downward trends attributa-

ble to secular changes in the economy. This graph shows little evidence

of any downward trajectory for individual wages before expungement.

Rather, it presents a noisy but essentially flat pattern, which abruptly

turns upward after expungement receipt.

296

The results from the early-expungement sample are encouraging in

another sense as well: they suggest that similar employment benefits

might accrue in states that make expungements automatic after a fixed

period of time post-conviction.

297

One might otherwise worry that the

employment gains depend on the fact that expungement recipients like

those in our data are a self-selected group who choose to apply just

when they are hunting for a job. Even if the expungement does help

(that is, it has a causal effect) in obtaining employment or better wages,

one would not necessarily expect a similar effect to emerge if a recipient

is not job hunting in the first place.

298

But in the early-expungement sample, the timing of the expunge-

ment is less likely to be determined by an individual’s search for better

employment circumstances; rather, for most of this group, expungement

timing is dictated by the arbitrary five-year waiting period. And that

perhaps explains why the gains for this group are slightly slower to ar-

rive — but still, they arrive fully within two years of the expungement.

If expungement kicked in automatically after five years (or some other

time threshold), presumably most recipients would not happen to be

looking for a job (or a better-paying job or one with more hours) at that

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

295

See supra notes 115–124 and accompanying text (describing in detail the eligibility require-

ments — and in particular the five-year waiting period — for expungement in Michigan).

296

For space reasons, we show only the wage graph here and not the other outcome variables.

The graphs of the employment measures, when we construct them using the early expungement

sample and residualized, do show a slight decline in the quarters before expungement receipt, but

it is much smaller than what we observe for the full sample.

297

For a list of states that have led this policy shift, see supra section I.B, pp. 2472–76.

298

In other words, we worry that receipt of an expungement might have heterogeneous effects

on employment outcomes, depending on whether someone is hunting for a new job, which might

be unobservable. Cf. Patrick Arni & Amelie Schiprowski, Job Search Requirements, Effort

Provision and Labor Market Outcomes, 169 J.

. E

CON

. 65, 75–76 (2019); James J. Heckman &

Edward Vytlacil, Policy-Relevant Treatment Effects, 91 A

. E

CON

. R

. 107, 107 (2001).

2540 HARVARD LAW REVIEW [Vol. 133:2460

IGURE

VERAGE

ESIDUALIZED

UARTERLY

AGES

EFORE

AND

FTER

XPUNGEMENT

(Expungement in First-Eligible-Year Subsample)

Notes: Quarter 1 (immediately after the dotted line) is the quarter of expunge-

ment receipt. The sample consists of individuals who received a known-date

expungement between January 1998 and May 2011, did not have an out-of-

state driver’s license, matched to our UIA data in at least one quarter during

the period from 1997 to 2013, and were of working age (eighteen to sixty-four).

very time or within moments of learning about their expungement. But

almost everyone looks to improve their employment situation eventu-

ally — and probably more frequently among people with records, who

tend to be in lower-paying jobs with higher turnover.

299

In sum, the causal interpretation of our results is on all fours with

many patterns in our data that are otherwise difficult to explain — and

with what we should expect based on extensive research showing that

criminal records pose substantial hurdles to employment success.

300

Our

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

299

See Adams et al., supra note 95, at 30 (“Respondents, at times, expressed frustration with their

inability to gain employment, noting that the limited jobs open to them predominantly involved

poor working conditions and low pay . . . .”).

300

See Franklin & Johnsen, supra note 248, at 736–37; supra section I.A, pp. 2468–71.

-2000

-1000

1000

2000

Average Quar ter l y Wages ($) (Residualized)

-8 -6 -4 -2 0 2 4 6 8

Elapsed Quarters

2000

1000

-1000

-2000

Average Quarterly Wages ($) (Residualized)

Elapsed Quarters

2020] EXPUNGEMENT OF CRIMINAL CONVICTIONS 2541

findings are consistent with the intuition that expungement of those rec-

ords should help to reduce those hurdles.

Notably, our analysis is based entirely on expungements received in

the age of the internet.

301

Thus, our results suggest that it is possible

for record clearing to generate substantial benefits for individuals with

records notwithstanding the search tools currently available to employ-

ers. And this, perhaps, should not be surprising. Many arrests and

convictions are not especially newsworthy and do not create a long trail

of adverse Google hits.

302

Employers, after all, overwhelmingly pay to

carry out background checks using criminal records databases; they do

so presumably because they believe that the database they patronize will

provide them with much more comprehensive and reliable information

than they could otherwise obtain for free.

303

And Michigan’s experience demonstrates that expungement can, in

fact, succeed in excising convictions from these databases.

304

This, too,

is probably to be expected under the circumstances. Companies that

manage these databases obtain their records from the state and update

them frequently;

305

their legal obligation not to share expunged records

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

301

Dates range from 1998 to 2011, with a median of 2006. Internet usage was less ubiquitous

early in this period, but the gains from expungements are similar (if anything, larger) if one looks

only at dates after the median of this span of years, which was well into the period when internet-

based background research had become pervasive.

302

See Sara Tiegreen & Elana Newman, Violence: Comparing Reporting and Reality, D

ART

FOR

OURNALISM

& T

RAUMA

(Feb. 18, 2009), https://dartcenter.org/content/violence-

comparing-reporting-and-reality [https://perma.cc/FX48-4Y56] (noting that “[r]eports of crime

don’t match actual rates of crime” and that less serious crimes receive much less attention from the

media). But see Lageson, supra note 96 (“Even if a record is officially wiped clean, it’s legal for

criminal justice agencies and other websites to keep criminal records online. Arrest records, mug

shots, and court records are classified as in the public record in most states.”).

303

One 2010 survey found that 92% of employers perform criminal background checks on some

potential hires and that 73% perform criminal background checks on all potential hires. Two years

later, this number declined to 86% of employers conducting criminal background checks for some

potential hires and 69% of employers for all potential hires. S

’

FOR

UMAN

. M

GMT

supra note 42.

304

Michigan criminal records are readily available for direct search by employers and other

members of the public via state databases from which expunged records are purged. See Criminal

History Records, M

ICH

. S

OLICE

https://www.michigan.gov/msp/0,4643,7-123-1878_

8311---,00.html [https://perma.cc/4QQB-SML8] (explaining the different kinds of searches available

and noting the unavailability of suppressed criminal records); M

ICH

. S

ECS

, https://

michigan.staterecords.org [https://perma.cc/R573-2XR3] (providing a searchable database).

305

See M

ARINA

UANE

., U

RBAN

NST

., C

RIMINAL

ACKGROUND

HECKS

: I

MPACT

MPLOYMENT

AND

ECIDIVISM

7–8 (2017), https://www.urban.org/research/

publication/criminal-background-checks-impact-employment-and-recidivism/view/full_report

[https://perma.cc/6GWF-SESW].

2542 HARVARD LAW REVIEW [Vol. 133:2460

is clear;

306

and thus failing to observe this requirement potentially ex-

poses them to criminal prosecution.

307

In all of our conversations with

expungement advocates in Michigan, none mentioned any concerns

about the practical effectiveness of expungements at limiting employer

access to records, even though similar concerns have been raised often

in the course of debates about expungement reform elsewhere.

308

To be sure, our data do not establish that expungement works per-

fectly in all cases to hide records from the public eye; it is likely that

some individuals really cannot escape the digital trail of their prior con-

victions, and mistakes could well persist for some individuals who have

had their records cleared in some criminal history databases.

309

More-

over, a clean record obviously does not automatically translate into a

new or better job; the search for employment involves many other fac-

tors, and not every expungement recipient will end up experiencing em-

ployment benefits (although some will presumably experience benefits

in other areas of life as well). The employment and wage gains that we

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

306

Third-party vendors are less important given the availability of public records, but in any

event, they too are constrained by law not to share expunged records — indeed, doing so is a mis-

demeanor. M

ICH

. C

OMP

. L

AWS

§ 780.623(5) (2020).

307

See id.

308

Many data collection agencies that offer background checks to companies and employers

qualify as consumer reporting agencies (CRAs) that are regulated by the Fair Credit Reporting Act

(FCRA). See L

OVE

., supra note 6, §§ 5:14–5:15, 5:19; Roberts, supra note 97, at 345. How-

ever, “not all CRAs use reliable sources for their reports, and they certainly do not always update

expunged records from their databases through removal. Thus the potential for, and indeed inci-

dents of, error can be high.” Id.; see also R

EBECCA

ALLAS

& S

HARON

IETRICH

, C

FOR

. P

ROGRESS

, O

TRIKE

AND

’

: H

LIMINATE

ARRIERS

CONOMIC

ECURITY

AND

OBILITY

FOR

EOPLE

WITH

RIMINAL

ECORDS

14 (2014),

https://cdn.americanprogress.org/wp-content/uploads/2014/12/VallasCriminalRecordsReport.pdf

[https://perma.cc/PXV4-ANWP] (“Common errors include reporting mismatches of cases belonging

to someone else, reporting expunged cases, and failure to report outcomes of old arrests.”). Jenny

Roberts mentions that some states do have their own credit reporting acts, and these are a way of

“limiting access to and use of sealed and expunged records.” Roberts, supra note 97, at 345. Mug-

shot websites do not qualify as CRAs and are therefore unregulated by the FCRA. See id. Eldar

Haber argues that digital access to records poses a constant threat to record clearing, Haber, supra

note 97, at 356, and expungements “must be revisited and revised to address the challenges of digital

technology to rehabilitation,” id. at 384. “‘Mugshot’ websites post photos from the day’s arrests,

often free of charge except to those who must pay to get expunged charges taken off the site. Web-

sites like ‘Instant Checkmate’ encourage a quick check of criminal records and social media before

a date.” Roberts, supra note 97, at 329 (footnotes omitted). Roberts argues that sealing and ex-

pungement laws can be made more effective by “ensuring the accuracy of records, restricting ways

in which decision makers with access can use records, and effective regulation of background

screening companies and private companies that sell mugshots and other arrest and conviction

information.” Id. at 344. Clay Calvert and Jerry Bruno argue that journalists are positioned to

balance the competing interests of expungement laws and the First Amendment guarantee of a free

press: “[I]f a newspaper publishes a story regarding the arrest of an individual, then it takes on an

ethical obligation to subsequently report on the expunction of that same individual’s record should

such a resolution occur.” Calvert & Bruno, supra note 97, at 143–44.

309

See supra notes 96–97 and accompanying text.

2020] EXPUNGEMENT OF CRIMINAL CONVICTIONS 2543

find are averages for expungement recipients as a group, and to the ex-

tent that they can be interpreted as causal effects, they represent achiev-

able gains even if expungement does not work flawlessly.

THER

BJECTIONS

XPUNGEMENT

AWS

This Article does not seek to address comprehensively all policy con-

cerns surrounding reform of expungement laws. Some arguments raised

by both proponents and opponents do not depend on empirical claims

and are not subject to evaluation by data-driven studies like ours. For

example, expungement advocates often argue that people who have com-

pleted their sentences following a conviction deserve a second chance and

should not be punished and stigmatized forever.

310

The case for a “clean

slate” is sometimes framed in terms of forgiveness and redemption, lan-

guage that, as we note above, alludes to various religious traditions.

311

The affirmative case for expungement reform can also be framed in terms

of equality, as seeking to help people with records escape the permanent

form of second-class citizenship that effectively results from the collateral

consequences of convictions.

312

At the same time, some expungement

opponents contend that sealing records (and allowing people who have

been convicted of crimes to explicitly state the opposite) is dishonest and

at odds with freedom of information, government transparency, and the

public’s right to know.

313

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

310

See, e.g., N

’

RIMINAL

. L

AWYERS

, C

OLLATERAL

AMAGE

MERICA

’

AILURE

ORGIVE

ORGET

THE

RIME

12–13, 15–18 (2014),

https://www.nacdl.org/getattachment/4a1f16cd-ec82-44f1-a093-798ee1cd7ba3/collateral-damage-

america-s-failure-to-forgive-or-forget-in-the-war-on-crime-a-roadmap-to-restore-rights-and-

status-after-arrest-or-conviction.pdf [https://perma.cc/EL3N-LHU4] [hereinafter C

OLLATERAL

AMAGE

]; V

ALLAS

& D

IETRICH

, supra note 308, at 34, 37, 49.

311

See, e.g., V

ALLAS

& D

IETRICH

, supra note 308, at 34 (“Enabling Americans with criminal rec-

ords to obtain a clean slate upon rehabilitation would permit them to redeem themselves and move on

with their lives after they pay their debt to society.”); supra note 238 and accompanying text.

312

E.g., C

OLLATERAL

AMAGE

, supra note 310, at 9 (“[C]ollateral consequences, whether based

on specific legal provisions or the general discrimination they encourage, have produced untold collat-

eral damage in the war on crime. Collateral consequences affect jobs and licenses, housing, public

benefits, voting rights, judicial rights, parental rights, the right to bear arms, immigration status, and

even volunteer opportunities. Each individual consequence may have seemed prudent and logical

when enacted, but their overall effect is to consign millions to second-class status.”); Beth Feldstein, The

Steep Price of a Clean Slate, E

QUAL

UST

. U

NDER

L. (June 20, 2019), https://equaljusticeunder-

law.org/thejusticereport/the-steep-price-of-a-clean-slate [https://perma.cc/9JFM-D68Y] (“[O]ur laws

perpetuate a cycle of poverty where financially-struggling people with criminal records are stuck as

second-class citizens.”); Eric Westervelt & Barbara Brosher, Scrubbing the Past to Give Those with a

Criminal Record a Second Chance, NPR (Feb. 19, 2019, 4:58 AM), https://www.

npr.org/2019/02/19/692322738/scrubbing-the-past-to-give-those-with-a-criminal-record-a-second-

chance [https://perma.cc/68EJ-56WH] (“With background checks ubiquitous for jobs, schools, mort-

gage applications and more, even one conviction — and sometimes even just one arrest — can dog

people for years, critics say, relegating them to permanent second-class status.”).

313

See Funk, supra note 94, at 302–04 (arguing that aggressive expungement policies hamper police

investigations and present a risk of liability or theft to employers); see also J

ACOBS

, supra note 7, at 302

2544 HARVARD LAW REVIEW [Vol. 133:2460

These arguments essentially involve questions of principle, values,

and preferences, not data or empirics, and fully engaging them is beyond

our Article’s scope. Instead, in this Part, we briefly address two other

possible objections to expungement laws that do connect more directly to

the substance of our evidence and argument: the crime and labor-market

consequences of expungement laws.

A. General Deterrence

When evaluating the effects of expungement on crime, one might ob-

ject that our focus on subsequent recidivism rates of expungement recip-

ients is incomplete. Record-clearing policies can influence crime rates

through other channels — in particular general deterrence, especially of

those who do not presently have records. Perhaps some people are de-

terred from committing crimes by the knowledge of the many ways in

which having a criminal record might harm them, such as embarrassing

them in their communities or interfering with their future employment

and housing prospects.

314

The concern, therefore, is that expanding ex-

pungement opportunities (especially through a much broader and/or au-

tomatic expungement policy) may conflict with general deterrence —

that is, deterrent effects on the population as a whole — and do so in a

way that produces a net increase in crime.

Such a consequence is theoretically possible, but once we inspect the

argument’s basic premises, we are skeptical of it. Our study does not

evaluate general deterrence effects, nor has any study of expungement

laws.

315

But other existing research provides many reasons to expect

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

(“First Amendment purists and open government advocates . . . argue that when it comes to combatting

crime, criminal justice agencies and courts do a better and fairer job of preventing and solving crime

when they have more information at their disposal.”); Alessandro Corda, More Justice and Less Harm:

Reinventing Access to Criminal History Records, 60 H

. L.J. 1, 36 (2016).

314

See Eric Rasmusen, Stigma and Self-Fulfilling Expectations of Criminality, 39 J.L. & E

CON

519, 532–36 (1996) (arguing conviction stigma may contribute to deterrence). But see Murat C.

Mungan, Stigma Dilution and Over-criminalization, 18

. L. & E

CON

. R

. 88, 88 (2016) (“Crim-

inalizing an act that provides weak signals about a person’s productivity and character can dilute

the stigma attached to having a criminal record.”). Indeed, this possibility has been raised in the

context of sex offender community notification laws, and there is at least some evidence that concern

over being publicly labelled as a “sex offender” might reduce the number of first-time offenses,

although the evidence, even in this context, is mixed. See Prescott & Rockoff, supra note 254, at

181 (finding some evidence that community notification may reduce the frequency of first-time sex

offenses but also that it may exacerbate recidivism among those subject to it).

315

Recently, there has been some experimental research on record expungement, however. See

Romain Espinosa et al., The General Versus Specific Deterrence Effects of Expungements:

Experimental Evidence 1–3 (George Mason Univ., Law & Econ. Research Paper No. 19-10, 2019),

https://ssrn.com/abstract=3379701 [https://perma.cc/LGF7-9C6J] (reporting results from a lab ex-

periment on expungement and its influence on criminal behavior and suggesting that the effect of

expungements on general deterrence varies based on cost but that “[t]here is no effect of expunge-

ments on specific deterrence,” id. at 1).

2020] EXPUNGEMENT OF CRIMINAL CONVICTIONS 2545

such effects to be minimal (at least when the record-clearing laws in ques-

tion are similar to those most states have in place or are presently consid-

ering). If anything, they are very likely to be outweighed by salutary spe-

cific deterrence effects on those who already have records.

First, general deterrence effects are relevant only if potential offend-

ers are not only aware of the collateral consequences of criminal convic-

tions,

316

but also cognizant of the availability of expungement. But even

people who already have criminal records and are legally eligible for ex-

pungement under existing record-clearing laws rarely know about or un-

derstand the opportunity. We doubt that — even if broader laws were

enacted — general public knowledge of it would be particularly wide-

spread, especially prior to a criminal conviction.

317

Similarly, studies of

the effects of other legal shocks on general criminal deterrence emphasize

that deterrence depends on perceived consequences, not actual ones,

318

and they stress that most people do not have a very good understanding

of the law that governs them, a phenomenon that tends to undermine

deterrence effects.

319

Even in a world without record clearing, individu-

als considering committing crimes may be overoptimistic when thinking

about potential labor-market and other collateral consequences. They

may be unaware of many of those consequences and also dismiss the like-

lihood of getting caught.

Second, the existence of waiting periods ought to almost completely

eliminate general deterrence concerns. This is especially so for fairly long

waiting periods, like Michigan’s current five years, and even more so for

the up-to-ten years the new Pennsylvania and Utah automatic expunge-

ment laws require.

320

It might be less true where waiting periods are

shorter, such as under California’s new law, but even in that case, the law

requires an individual to at least complete their sentence, delaying the

availability of expungement.

321

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

316

See Rasmusen, supra note 314, at 520; see also Murat C. Mungan, Gateway Crimes, 68 A

L. R

. 671, 681 (2017) (“Both stigma and formal sanctions have the desirable effect of increasing

the expected cost” — that is, a cost assumed to be accurately predicted in advance — “to committing

crime and thereby reducing people’s incentives to commit crimes.”).

317

See supra section II.C, pp. 2501–10.

318

See Daniel S. Nagin, Criminal Deterrence Research at the Outset of the Twenty-First Century,

23 C

RIME

& J

UST

. 1, 5 (1998) (“The perceptual deterrence literature was spawned by the recogni-

tion that deterrence is ultimately a perceptual phenomenon. While great effort has been committed

to analyzing the links between sanction risk perceptions and behavior, comparatively little attention

has been given to examining the origins of risk perceptions and their connection to actual sanction

policy.”); see also Aaron Chalfin & Justin McCrary, Criminal Deterrence: A Review of the Literature,

55 J.

CON

. L

ITERATURE

5, 10–12 (2017).

319

See Robert Apel, Sanctions, Perceptions, and Crime: Implications for Criminal Deterrence, 29 J.

UANTI TATI VE

RIMINOLOGY

67, 78–79 (2013); Chalfin & McCrary, supra note 318, at 10–11.

320

See M

ICH

. C

OMP

. L

AWS

§ 780.621(5) (2020); 18 P

. C

ONS

. S

TAT

. § 9122.2(a) (2019); U

TAH

ODE

. § 77-40-105(3)(c)(i) (West 2020).

321

See C

. P

ENAL

ODE

§ 1203.425(a)(2)(E) (West 2020).

2546 HARVARD LAW REVIEW [Vol. 133:2460

Intuitively, we have a hard time imagining the potential offenders

who would (absent expungement) avoid committing a crime out of fear

of social stigma, job loss, or other collateral consequences, but who would

decide to commit the crime after all if they became aware that five or ten

years later they might, if they keep their record clean and are otherwise

eligible, be relieved of some of those consequences.

322

This possibility

seems especially unlikely under a petition-based regime involving uncer-

tain judicial discretion. What is far more likely to be salient to these po-

tential offenders is the immediate consequence of losing the job they have

now, or embarrassing themselves or their families now (in addition to the

sentencing consequences). The possibility of expungement many years

down the road is comparatively remote and indefinite, and even at best

it does not undo many of those consequences; it simply stops them from

continuing to accrue over time.

Support for this intuition derives from the substantial literature ad-

dressing the effects of differences or changes in expected sentences on

crime rates.

323

In general, this research finds that criminal behavior is

not particularly sensitive to changes in the expected sentence, especially

not to additional years of incarceration beyond the first few.

324

Research

instead indicates that deterrence is best achieved through certainty and

celerity of punishment — for example, by visibly increasing policing.

325

Partly, this failure of deterrence is a function of people not understanding

the relevant laws, but it is also because most people have high, even hy-

perbolic, “discount rates” — we simply do not care nearly as much about

things that will happen to us years from now as we do about things that

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

322

In support of our conjecture, research indicates that individuals living in states that decrimi-

nalize marijuana often remain — at least for a time — not fully aware of the change in the law.

Rather, they continue to believe that they could be jailed for marijuana possession. Robert

MacCoun et al., Do Citizens Know Whether Their State Has Decriminalized Marijuana? Assessing

the Perceptual Component of Deterrence Theory, 5 R

. L. & E

CON

. 347, 366–67 (2009).

323

See Chalfin & McCrary, supra note 318, at 5–6, 24–28 (providing an excellent recent review

of the literature); J.J. Prescott, Criminal Sanctions and Deterrence, in 1 E

NCYCLOPEDIA

AND

CONOMICS

498, 508–09 (Alain Marciano & Giovanni Battista Ramello eds., 2019).

324

See, e.g., Marc Mauer, Long-Term Sentences: Time to Reconsider the Scale of Punishment, 87

UMKC

L. R

. 113, 123 (2018); Michael Tonry, An Honest Politician’s Guide to Deterrence:

Certainty, Severity, Celerity, and Parsimony, in 23 D

ETERRENCE

, C

HOICE

AND

RIME

365, 365

(Daniel S. Nagin et al. eds, 2018); Giovanni Mastrobuoni & David Rivers, Criminal Discount

Factors and Deterrence 1 (Inst. of Labor Econ., Discussion Paper No. 9769, 2016), https://

papers.ssrn.com/sol3/papers.cfm?abstract_id=2730969 [https://perma.cc/ZN8M-FHWZ] (“Our esti-

mates imply that the majority of deterrence is derived from the first few years in prison.”). But see,

e.g., Francesco Drago, Roberto Galbiati & Pietro Vertova, The Deterrent Effects of Prison: Evidence

from a Natural Experiment, 117 J.

. E

CON

. 257, 259–60, 267–73 (2009).

325

See Chalfin & McCrary, supra note 318, at 5 (“[T]here is considerable evidence that crime is

responsive to police . . . .”).

2020] EXPUNGEMENT OF CRIMINAL CONVICTIONS 2547

will happen today.

326

This basic feature of human psychology, which

may be even more present in lawbreakers,

327

has been documented in

wide-ranging contexts.

328

And its pervasiveness strongly implies that the

possibility of an expungement years later is unlikely to sway a person’s

decision to commit a crime today.

329

Furthermore, even beyond the ef-

fects of the waiting period’s implicit delay, scholarship on general deter-

rence suggests that people are simply less responsive to the consequences

of being caught for a crime in the first place than one might expect based

on traditional economic theory.

330

Third, if one broadens the analysis to encompass general deterrence

effects, it also makes sense to take into account yet another channel by

which expungement might influence crime: the specific deterrence of po-

tential expungement recipients. That is, one must consider how the pos-

sibility of expungement affects the incentives of the many people who

already have criminal records, not just those who are contemplating the

commission of first offenses.

331

After all, most crime is committed by re-

peat offenders. And these effects, intuitively, cut in the opposite direction

— and not just after expungement, which is the subject of our recidivism

discussion in Part III. Just about every state that makes expungement

available conditions it on some period of good behavior — which creates

incentives for good behavior before the waiting period expires.

332

To be

sure, such effects have not been studied either, and they too depend on

potential offenders knowing about the expungement law and caring

about the consequences of a clean record. But even though knowledge

of expungement laws is limited even among this subpopulation, people

who are already convicted and potentially expungement-eligible are, in-

tuitively, at least somewhat more likely than members of the general pub-

lic to be aware of such policies and to consider their benefits germane to

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

326

See Chae Mamayek et al., Temporal Discounting, Present Orientation, and Criminal

Deterrence, in T

XFORD

ANDBOOK

FFENDER

ECISION

AKING

209, 214–15

(Wim Bernasco et al. eds., 2017).

327

See David Åkerlund et al., Time Discounting and Criminal Behavior, 113 PNAS 6160, 6160 (2016).

328

Murat Yilmaz, An Extended Survey of Time-Inconsistency and Its Applications, 32

G

AZIÇI

J. R

. S

., E

CON

. & A

DMIN

. S

TUD

. 55, 56–58 (2018).

329

J.J. Prescott and Jonah Rockoff, supra note 254, do find some evidence that the possibility of

becoming a publicly registered sex offender many years in the future can deter first-time sex of-

fenses, see id. at 181. However, inclusion on a sex offender registry is an unusually well-known and

widely feared collateral consequence; Prescott, supra note 253, observes that other collateral conse-

quences (and even sentencing consequences such as those we discuss here) are likely less salient to

potential offenders, see id. at 1040.

330

See Chalfin & McCrary, supra note 318, at 37–40.

331

See Michael Mueller-Smith & Kevin T. Schnepel, Diversion in the Criminal Justice System:

Regression Discontinuity Evidence on Court Deferrals 3–6 (Aug. 3, 2017) (unpublished manuscript),

https://sites.lsa.umich.edu/mgms/wp-content/uploads/sites/283/2017/08/Diversion_in_the_Criminal_

Justice_System.pdf [https://perma.cc/F98G-SGCH].

332

It might be possible to strengthen incentives for ongoing good behavior by providing that a

subsequent conviction following an expungement could lead to the unsealing of a previously sealed

conviction. See D

EVAH

AGER

, M

ARKED

: R

ACE

, C

RIME

AND

INDING

ORK

ASS

NCARCERATION

156–57 (2007).

2548 HARVARD LAW REVIEW [Vol. 133:2460

their decision making. After all, these people are living at that moment

with the often-significant collateral consequences of their records; if they

are struggling to find a job or housing, the possibility that another clean

year or two could entitle them to an expungement might be much more

salient to them than expungement opportunities are to the general public.

Finally, it bears noting that the theoretical possibility that general de-

terrence of criminal behavior might decrease in a world with expunge-

ment is an argument that, at its logical extreme, can be deployed against

any reform meant to reduce the harshness of the criminal justice system’s

consequences. But it cannot be the case that anything that makes the

lives of people with records worse is better; this notion is at odds with all

reentry efforts and all policy-based limits on punitiveness. At some point,

the costs of such an approach to individuals and communities outweigh

the benefits. In this case, the purported general deterrence concern is

purely theoretical, tentative at best, and fragile on its own terms. Absent

strong evidence supporting its practical importance in the expungement

context, policymakers ought to give the general deterrence argument

very little weight.

B. Statistical Discrimination or Stereotyping

A different objection relates to the potential unintended labor-

market consequences of expungement laws. The precise worry is that if

employers really do not want to hire people with criminal records, wall-

ing off information about applicants’ records will simply cause them to

pursue that objective in other, potentially more pernicious ways. Specif-

ically, employers might instead rely more heavily on perceived proxies for

criminal records, such as race — which could amplify racial disparities

in employment.

333

This phenomenon is referred to as statistical discrim-

ination, or — if employers’ assumptions about who has criminal records

are inaccurate or exaggerated — stereotyping.

334

The strongest empirical support for this concern comes from recent

research by one of us on the effects of Ban-the-Box laws. In a large, ran-

domized field experiment that varied the race of fictitious job applicants,

Agan and Starr find that after Ban-the-Box went into effect in two

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

333

See Jennifer L. Doleac & Benjamin Hansen, The Unintended Consequences of “Ban the Box”:

Statistical Discrimination and Employment Outcomes when Criminal Histories Are Hidden, 38 J.

CON

. 321, 324–26 (2020).

334

See generally Kevin Lang & Jee-Yeon K. Lehmann, Racial Discrimination in the Labor

Market: Theory and Empirics, 50 J.

CON

. L

ITERATURE

959, 985–98 (2012). Landlords and other

decision makers could in theory have similar responses, although we focus our discussion here on

employment. See, e.g., David Bjerk, Racial Profiling, Statistical Discrimination, and the Effect of

a Colorblind Policy on the Crime Rate, 9 J.

. E

CON

. T

HEORY

521, 522 (2007).

2020] EXPUNGEMENT OF CRIMINAL CONVICTIONS 2549

jurisdictions, white applicants’ advantage in receiving callbacks from af-

fected employers skyrocketed by more than a factor of six.

335

These find-

ings suggest that large numbers of employers appear simply to assume

that most or at least many black applicants have criminal records. Un-

derstandably, we are often asked whether we ought to expect expunge-

ment laws to have a similar effect.

This question is legitimate and worthy of study in the future; we are

aware of no empirical evidence on this concern yet. But we believe that

at least under the vast majority of current and proposed expungement

policies, there is little reason to fear a substantial increase in racial dis-

crimination. An important difference between expungement and Ban-

the-Box restrictions is that Ban-the-Box explicitly takes away all crimi-

nal record information on all applicants, at least during the early stages

of the hiring process, when significant screening occurs.

336

Employers

reviewing applications are left completely in the dark as to who has a

criminal record — even as to very serious, very long, and very recent

criminal records. Expungement policies, in contrast, are much more tar-

geted in their approach; most conviction records will never be expunged,

and recent (and thus much more informative) records categorically can-

not be expunged under the law in most states. Moreover, in a jurisdiction

with an expungement law, employers can collect information on criminal

records, and anybody with a non-expunged record can be required to

provide that information.

So employers under an expungement policy still have access to a lot

of information about criminal history; this should reduce their incentive

to rely on factors like race as a proxy for it. In Michigan, for instance, if

an employer sees that an applicant reports having no record, it is vanish-

ingly unlikely at present that the applicant actually has an expunged rec-

ord; there are ten million people in Michigan, and only around thirty

thousand expungement recipients. Even if expungement were made au-

tomatic, expanding the size of the latter group substantially, it would still

be fairly unlikely for any particular applicant to have received an ex-

pungement.

337

Moreover, even if the applicant does have an expunged

record, at worst, it would be an old and likely less serious one, followed

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

335

See Agan & Starr, Ban the Box, supra note 39, at 191.

336

See Doleac & Hansen, supra note 333, at 323–24 (“Even though ex-offenders could be weeded

out after the interview process, interviewing candidates is costly. Employers would rather not spend

time interviewing candidates who they are sure to reject when their criminal history is revealed.”

(footnote omitted)).

337

See Amy Shlosberg et al., Expungement and Post-exoneration Offending, 104 J. C

RIM

. L. &

RIMINOLOGY

353, 355–62 (2014) (surveying state and federal expungement policies and noting

how few crimes are eligible for expungement).

2550 HARVARD LAW REVIEW [Vol. 133:2460

by years of nonrecidivism, given the waiting period and eligibility re-

quirements.

338

This possibility provides little incentive for employers to

engage in statistical discrimination.

What’s more, all of this assumes employers know expungement laws

exist at all. Ban-the-Box laws have been very salient to employers, who

were forced to change their applications and were mandated to refrain

from asking about something that had always been a traditional part of

their process.

339

Expungement law is not salient in the same way. Em-

ployers need do nothing, and they can continue to ask the questions (and

run the background checks) that they always have. And, if most people

with expungeable records are unaware of record-clearing opportunities,

it seems very likely that many employers will be in the dark about them

as well.

All of that said, there can be no doubt that the landscape of expunge-

ment reform is changing, and the more sweeping expungement policies

become, the more reason there will be to worry about the possibility of

statistical discrimination by decision makers. If an expungement policy

is so broad that it leaves employers effectively without much useful crim-

inal record information at all, they might indeed react much as employers

do to Ban-the-Box policies. Among current expungement laws, Califor-

nia’s may come closest to raising this concern, since it appears to apply

to a broad swath of criminal offenses, albeit relatively minor ones, and

has (for some crimes) no post-sentence waiting period.

340

Therefore, the

effect of this and similarly expansive laws on racial discrimination is

worth assessing via future research.

ONCLUSION

As states throughout the country continue to debate the adoption and

expansion of expungement laws, it is important for their decisionmaking

to be guided not by hype or supposition, but by empirical evidence. Prior

to the work we present in this Article, policymakers have had very little

at their disposal. Fortunately, Michigan’s experience with expungement

provides a great opportunity to evaluate how these laws work in practice

and what their consequences are likely to be. The challenges faced by

people with records in Michigan are fundamentally similar to those faced

by their counterparts in other states, and many features of Michigan’s

law — for example, its coverage of felonies and violent crimes and its

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

338

See Pager, supra note 30, at 963 (noting that employers may sometimes only ask about incar-

ceration within the past year); Kathy Gurchiek, Research: Employers Willing to Overlook a Criminal

Record to Hire the Right Person, SHRM (May 17, 2018), https://www.shrm.org/resourcesandtools/hr-

topics/behavioral-competencies/global-and-cultural-effectiveness/pages/research-employers-willing-

to-overlook-criminal-record-to-hire-right-person.aspx [https://perma.cc/U3Z3-7B92] (reporting survey

data showing that employers are less worried about — and therefore relatively more likely to hire —

individuals with the kinds of records that expungement laws typically address).

339

See Hanks, supra note 83.

340

See C

. P

ENAL

ODE

§ 1203.425(a)(2)(E)(i) (West 2020).

2020] EXPUNGEMENT OF CRIMINAL CONVICTIONS 2551

middle-of-the-road waiting period — make the state’s experience with

expungement particularly relevant to assessing the likely effects of re-

forms nationwide. We hope the findings we document in this study, the

first of its kind, will provide helpful guidance for those crucial policy

decisions.

Taken together, our findings strongly support increasing the availa-

bility of expungement — and particularly efforts to make expungements

automatic, or at least procedurally easy to obtain. Those whose records

are expunged experience large gains in employment rates and wages —

and while some of these gains may have their source in other factors such

as underlying motivation, expungements very likely cause a substantial

portion of these improvements. The expungement effects we uncover

imply that record clearing compares favorably as a strategy to other pol-

icy interventions that seek to improve employment outcomes. Take, for

example, job training — a common public investment. A meta-analysis

of thirty-one studies covering fifteen different publicly funded job-train-

ing programs finds average unweighted gains in annual wages of $832

for women and $471 for men.

341

Like expungement recipients in

Michigan, participants in these programs were self-selected; they pur-

sued the programs because they thought they would benefit from them.

And they did benefit — but not nearly as much as expungement recipi-

ents do (annualized, the wage gains for our sample come to $4594 for

women and $4295 for men).

342

Meanwhile, the average cost of these job-

training programs to the government was $6600 per person.

343

Expunge-

ment, in contrast, has comparatively minimal costs (running a back-

ground check, holding a court hearing, processing the paperwork),

344

and

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

341

David H. Greenberg et al., A Meta-Analysis of Government-Sponsored Training Programs, 57

NDUS

. & L

. R

. 31, 42 (2003) (reporting weighted means across studies).

342

The job-training study reports results in 1999 dollars, id. at 33, whereas our figures are nom-

inal gains over one-year periods falling between 1997 to 2011, with a median year of 2005. Adjust-

ing the job-training averages to 2005 dollars comes to $1611 for women and $373 for men.

343

Id. at 50.

344

For example, the cost of ink fingerprinting, one of the procedure’s requirements, is $15 at the

Washtenaw County Sheriff’s Office. Fee Schedule, S

HERIFF

’

. W

ASHTENAW

OUNTY

ICH

., (effective Jan. 4, 2019), https://www.washtenaw.org/1533/Fee-Schedule [https://perma.cc/

YMF9-927B]. Certified records in Washtenaw County are $10 plus $1 per page. Records Request,

RIAL

. W

ASHTENAW

OUNTY

ICH

., https://www.washtenaw.org/1420/Records-Request

[https://perma.cc/LBS4-GDPX]. In Jackson, Michigan, “United Way of Jackson County [wrote]

$50 checks to cover the cost of sending paperwork to the Michigan State Police and $10 checks to

cover the cost of retrieving official court documents for individuals trying to set aside their record.”

McLaren, supra note 196. One hidden cost is the cost of hiring an attorney to help convince a judge

to grant the expungement request: “[A]n expungement can be between $1,000 to $2,500, although it can

be done without an attorney.” Jameson Cook, State Makes It Easier to Expunge Prior Convictions,

AKLAND

RESS

(Mar. 17, 2015), https://www.theoaklandpress.com/news/nation-world-news/state-

makes-%20it-easier-to-expunge-prior-convictions/article_74afe689-de8a-554c-bc00-

2dec246d40c5.html [https://perma.cc/8TY3-FVJ4].

2552 HARVARD LAW REVIEW [Vol. 133:2460

these could very likely be reduced much further if the process were sim-

plified or automated. As an employment intervention, therefore, ex-

pungement appears superior to job training in terms of both its effective-

ness and its price tag.

The arguments of expungement opponents do not typically focus on

program expenditures, of course, but rather on public safety.

345

Fortu-

nately, our findings on the crime front are equally encouraging. Subse-

quent offending rates after expungement are extremely low. Ninety-nine

percent of those who receive expungements in Michigan are not con-

victed of a felony anytime in the next five years; 99.4% are not convicted

of any violent crime; and 95.8% are not convicted of any crime at all,

even a petty misdemeanor. In fact, expungement recipients appear to be

lower risk than the general public. To be sure, recidivism rates would

not be as low if states made expungement available immediately follow-

ing the completion of a sentence, removed judicial discretion, or extended

it to a generally riskier pool, such as people with more extensive criminal

records. But even in those scenarios, there is still no evidence to suggest

that access to expungement would increase the recidivism risk of those

groups; if anything, due to the benefits we find in terms of employment

outcomes (and possible benefits in other areas), one should probably ex-

pect their recidivism rates to decline.

The discouraging element of our findings is that, despite the apparent

benefits of expungement, very few people — even among those who are

eligible — actually obtain them. Our best estimate is that 6.5% of people

who meet the legal requirements for expungement in Michigan obtain

one within five years — a small fraction of what is already a small frac-

tion of all those living with records, given the tight eligibility require-

ments. This low uptake rate is troubling, but not shocking, given the

procedural hurdles and expenses involved, the lack of legal counsel, the

dearth of public information, and the fact that most people with records

have limited resources for overcoming these challenges. Unfortunately,

all but a handful of states with expungement laws require individuals to

apply for relief and give judges the discretion to deny them, so the situa-

tion is unlikely to be better in other states.

346

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

345

“The most potent argument for broad access to criminal history data is to preserve public

safety. The right to a safe and peaceful environment is a core value for a civilized society, and

employers cannot allow known dangers in the workplace.” Cynthia Diane Stephens, Keeping an

Arrest from Resulting in a Life Sentence in an Age of Full Disclosure of Criminal Records, M

ICH

B.J., Nov. 2008, at 29, 31. But see Doe v. United States, 110 F. Supp. 3d 448, 452 (E.D.N.Y. 2015)

(“Simply put, the public safety is better served when people with criminal convictions are able to

participate as productive members of society by working and paying taxes.”).

346

See CCRC State Survey, supra note 11. For examples of the pitfalls in the application process,

see Simone Ispa-Landa, Believing in a Positive Future as a Form of Stigma Resistance: Narratives

of Denied Expungement-Seekers, 40 D

EVIANT

EHAV

. 1428, 1435 (2019).

2020] EXPUNGEMENT OF CRIMINAL CONVICTIONS 2553

The policy upshot of our research is clear: the process of obtaining

expungement should be made as simple as possible, or ideally automatic

once an individual’s record meets the legal requirements. The path-

breaking Pennsylvania automatic-sealing law, while still at the forefront

of developing legislation, has illustrated a way forward; indeed,

California’s, Utah’s, and New Jersey’s newly adopted automatic ex-

pungement policies are further dramatic advances.

347

With luck, the

next wave of Clean Slate legislation will continue to move in this direc-

tion.

348

Pennsylvania’s law is a watershed in terms of expungement proce-

dure, but it is unfortunately quite limited in its substantive scope (that is,

in terms of its eligibility rules): automatic expungement applies only to

people with minor, nonviolent misdemeanors after ten crime-free

years.

349

Other earlier adopters are less limited on a few dimensions, but

not on many and not by much. Presumably, Pennsylvania clean-slate

advocates concluded that the bill’s substantive constraints were politi-

cally necessary to secure automatic process.

350

Likely, this limitation was

meant to defuse the concern that people with more substantial or more

recent convictions posed a greater public-safety risk. But our findings

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

347

See Sharon Dietrich & Rebecca Vallas, The Left, the Right, and the Football Players: How

Clean Slate Automated Sealing Was Passed in Pennsylvania, C

LEARINGHOUSE

OMMUNITY

(Nov. 2018), https://www.povertylaw.org/clearinghouse/stories/dietrich_vallas [https://perma.cc/

TE6D-AFJE] (“Pennsylvania catapulted to the forefront of record clearing, becoming the first state

in the nation to adopt a groundbreaking new innovation: automated sealing of minor records by

use of technology — a concept known as ‘Clean Slate.’”); supra note 16 and accompanying text.

348

One factor that makes automatic expungement laws technologically feasible, though, is an

orderly records system that transmits records easily from judicial districts to a central office. See

Maura Ewing, Simplifying How the Courts Seal Criminal Records, T

TLANTIC

(Apr. 6, 2017),

https://www.theatlantic.com/politics/archive/2017/04/simplifying-how-the-courts-seal-criminal-

records/521964 [https://perma.cc/N6ER-TKMR]. States like Michigan and Pennsylvania are

primed for ambitious reforms because of their centralized record-keeping systems, while most other

states have “siloed” systems that do not transmit records from one court to another. Id. Record-

keeping systems (and the infrastructure they require) are an additional hurdle that states without

centralized systems should consider in crafting their reforms.

349

18 P

. C

ONS

. S

TAT

. § 9122.2 (2018).

350

For example, those with past felony convictions, multiple first-degree misdemeanor convictions,

and four or more second-degree misdemeanors are not eligible for automatic sealing. Mark Scolforo,

Wolf Signs Bill Sealing Some Criminal Records After Decade, A

SSOCIATED

RESS

(June 28, 2018),

https://apnews.com/9fb7a96015944642bdf796c30cee35a5/Wolf-signs-bill-sealing-some-criminal-

records-after-decade [https://perma.cc/Z2QF-8EBJ]. See also Dietrich & Vallas, supra note 347, for a

description of the political compromises the bill made: “The bill . . . expanded sealing eligibility to reach

first-degree misdemeanors. Although a modest expansion, this was the most that appeared politically

feasible, given that misdemeanor sealing had just been introduced by Act 5 earlier in 2016.” Id. District

attorney support was vital to the passing of the law, leading to more exclusions in the bill: “Critical

stakeholders we needed on board — or at least to remain neutral — were the district attorneys, from

whom opposition on a record-clearing bill in Pennsylvania was likely enough to kill it. One major area

of focus for them and others was which criminal records would be eligible for sealing both through

Clean Slate and through expanded eligibility for filing by petition, and the list of exclusions became

longer as we brought them and other key parties on board.” Id.

2554 HARVARD LAW REVIEW [Vol. 133:2460

indicate that these concerns are unfounded; people who receive expunge-

ments after just five years pose a very low recidivism risk, even though

Michigan does not exclude people with felonies or violent offenses.

351

The somewhat broader approaches now being pursued in states like

California — and perhaps soon in Michigan as well — are more ambi-

tious, and studying their effects will help us to learn more about what

happens when expungement is granted broadly and automatically, and

not to a narrow, self-selected sample.

For now, however, it is likely that most states will continue to require

an application-based, discretionary procedure for at least some subsets of

expungement applicants. Where this is so, policymakers should consider

how to at least render that process as easy and accommodating as possi-

ble. Courts, public defenders’ offices, or prosecutors’ offices could auto-

matically notify individuals, after the requisite waiting period, that they

may be eligible for expungement, and provide them with links to online

tools for determining eligibility. Court hearings should not be necessary,

especially when prosecutors and victims do not oppose expungements.

Online applications should be permitted. Serving the application on mul-

tiple entities (the court, the prosecution, the police) should not be re-

quired, given that the information could be passed along automatically

via computerized processes. Fingerprinting may be essential for back-

ground checks (although this too perhaps could be done via touchscreen),

but there is no reason an applicant should have to make a fingerprinting

visit to a police station and multiple courthouse visits, not to mention a

possible visit to a notary. For example, the court clerk’s office should be

set up to provide whatever records are necessary, provide the form or a

computer terminal, take the fingerprints, and collect the forms all at once.

Fees should be eliminated; they are minor in their impact on the public

fisc (especially given the cost-effectiveness of expungements at improving

employment outcomes), but they remain substantial barriers for many

people with records.

Long after they have served their sentences, tens of millions of

Americans and their families face the serious challenges of life with a

criminal conviction record, and this number increases daily. Collectively,

these challenges contribute to many significant public policy concerns,

making it harder for these families to avoid poverty and contributing to

racial disparities in employment and other domains. Our empirical re-

sults suggest that expungement is a powerful policy lever for redressing

these negative consequences, without risk (and possibly with benefits) to

public safety. But expungement will only realize its full potential and

–––––––––––––––––––––––––––––––––––––––––––––––––––––––––––––

351

See supra Part III, pp. 2510–23 (describing the low recidivism rates for expungement recipi-

ents despite Michigan allowing expungements for those with violent felony convictions); see also

ICH

. C

OMP

. L

AWS

§ 780.621(3) (2020) (allowing expungements for those with one felony convic-

tion, with the exceptions of sex offenses that are subject to registration requirements and offenses

carrying potential life-imprisonment terms).

2020] EXPUNGEMENT OF CRIMINAL CONVICTIONS 2555

make a serious dent in these large-scale social problems if we make it avail-

able much more broadly and much more easily. Legislatures throughout

the country have been taking up the issue, which provides reason for opti-

mism. They should now consider the empirical evidence, which makes a

clear case for action.