Dollars and Sense in Cook County
The fourth dependent variable, New Criminal Activity,
captured whether a released defendant had a new criminal
case filed in Cook County while on pretrial release (1=New
Criminal Activity; 0=No New Criminal Activity).
New
Criminal Activity was calculated only for defendants
released pretrial. If a defendant had any new case filed
during the follow-up period, they were considered to have
new criminal activity.
Finally, the fifth dependent variable, New Violent Criminal
Activity, captured whether a released defendant had a new
criminal case filed for a Person offense while on pretrial
release (1=New Violent Criminal Activity; 0=No New Violent
Criminal Activity). New Violent Criminal Activity was
calculated only for defendants released pretrial. If a
defendant had any new felony or misdemeanor case filed
during the follow-up period in which the most serious filed
charge was a Person offense, they were considered to have
new violent criminal activity.
Person offenses included
murder, kidnapping, criminal sexual abuse or assault,
assault, battery, domestic battery, reckless conduct,
robbery, vehicular hijacking, and violation of an order of
protection.
Case-level variables
Case-level variables in the present study included several
defendant, offense, and case processing characteristics.
Defendant characteristics were drawn from the court data
and included information on demographic attributes of
defendants. The analyses included a trichotomous variable
measuring defendant race/ethnicity (0=Defendant White
(reference), 1=Defendant Black, 3=Defendant
Hispanic/Other). The court data included one variable
capturing defendant race and ethnicity in the following
categories: American Indian and Alaskan Native,
Asian/Pacific Islander, Black Latino/Hispanic, Black non-
Latino/Hispanic, Other, Unknown/Missing, White
Latino/Hispanic, White non-Latino/Hispanic. These
categories were initially recoded into two variables
measuring race (White, Black, American Indian and Alaskan
Native, Asian/Pacific Islander, Other/Missing) and ethnicity
(Hispanic, Non-Hispanic). However, it was evident that
Hispanic ethnicity was not captured consistently in the
court data; just 1.7% of defendants in the pre-GO18.8A
cohort were Hispanic compared to 16.3% of defendants in
the post-GO18.8A cohort. Thus, in order to ensure the
inclusion of these defendants and the ability to examine the
effects of race and ethnicity, a single race/ethnicity variable
was created with Hispanic, American Indian, Alaskan Native,
and Asian/Pacific Islander defendants included in a third
race/ethnicity category. The analyses also included
defendant sex (0=Female (reference), 1=Male) and age (in
years). A variable capturing whether the defendant was in
violation of probation or bail bond was also included (0=No
Violation (reference); 1=Violation). A defendant was
determined to have a violation of probation if they had a
sentence of probation imposed within one year prior to the
initial felony bond court hearing date; a defendant was
determined to have a violation of bail bond if they had a
charge of violation of bail bond included in a prior case.
PSA scores measuring risk of FTA or New Criminal Activity
were coded as ordinal variables with three categories (1-2
(low risk) (reference), 3-4, and 5-6 (high risk)). A third
dichotomous PSA score measuring risk of New Violent
Criminal Activity was also included (0=No Risk, 1=Risk). PSA
scores were missing for 26.4% of the sample (35.5% of the
pre-GO18.8A sample and 16.0% of the post-GO18.8A
sample). In order to include PSA scores in the analyses,
missing values were recoded (Missing=99) to ensure the
cases were not dropped. Cook County Adult Probation and
the Office of the Chief Judge have found that over 90% of
felony defendants receive an assessment. The high missing
rate here is likely not due to Adult Probation - Pretrial
Services failing to meet assessment obligations; rather, it is
likely due to data quality issues and the study’s difficulty in
matching defendants across court and PSA datasets.
Offense characteristics included the number of filed charges
(continuous), number of filed felony charges (continuous),
and offense severity, which was coded as an ordinal
variable with seven categories (0=Unspecified Felony (least
serious) through 6=Murder (most serious)). The type of
offense was measured with a categorical variable
measuring the most serious offense in the defendant/case
combination (0=Property (reference), 1=Person, 2=Weapons,
3=Drugs, 4=Public Order/Other).
Lastly, one case processing variable was included: weeks at
risk (continuous), which measured the number of weeks a
defendant was released from jail during the follow-up
period. This was calculated only for defendants released
pretrial and included only the time that they were not
physically in the jail.
Analytical strategy
The impact of defendant, offense, and case characteristics
on case outcomes was analyzed using standard statistical
procedures to examine categorical data in multivariate
settings. Specifically, the baseline estimations relied on a
series of multivariate logistic regression models to estimate
the effect of these factors on the dependent variables listed
above. All models are estimated using Stata 14.
For each dependent variable, the analyses considered thee
models. The first model included the pre- and post-GO18.8A
cohorts with a control for the pre- and post-periods. This
allowed an estimation of the impact of GO18.8A on the odds
of the outcome (e.g., odds of I-Bond, Release, etc.),
controlling for defendant, offense, and case characteristics.
Two separate models were then run separately for each of
the pre- and post-GO18.8A cohorts; this was to examine
potential variation in the influence of defendant, offense,
and case characteristics during the two time periods.
Following each model, expected rates of each outcome
were estimated using predicted probabilities generated
using the Stata
margins
command. Predicted probabilities