Report prepared by:
The Campaign for Fiscal Equity, Inc.
May 2011
Taking Attendance Seriously
How School Absences Undermine Student
and School Performance in New York City
The Campaign for Fiscal Equity, Inc. (CFE) is a leading non-profit organization working to
protect and promote the constitutional right to a sound basic education - defined as a
meaningful high school education - for every public school child in the State of New York.
CFE was founded in 1993 by a coalition of concerned parents and education advocates who
filed the landmark case CFE v. State of New York, which established this right. To make this
right a reality, CFE works to ensure that the neediest students in low performing schools make
academic progress, graduate high school and become active civic participants who can
compete in the global economy. CFE works to educate and engage the public and policy
makers to ensure that the historic school budget increases, accountability reform and
meaningful public participation that resulted from the landmark CFE court decision and law
reform are fully implemented.
Helaine Doran
Deputy Director
Board of Directors
Luis Miranda, Chairperson, President, MirRam Group
Edward Fergus, Director, Applied Research, Evaluation and Policy, Metropolitan Center for
Urban Education, Steinhardt School of Education, New York University
Geri D. Palast, Managing Director, JFNA/JCPA Israel Action Network
Dennis Parker, Director, American Civil Liberties Union Racial Justice Program
Steven Sanders, Former Chairman, New York State Assembly Committee on Education
Ocynthia Williams, Parent Leader, NYC Coalition for Educational Justice
ACKNOWLEDGEMENTS
Taking Attendance Seriously: How School Absences Undermine Student and
School Performance in New York City
Director: Helaine Doran
Author: Dr. Martha Philbeck Musser
Consulting Editor: Phyllis Jordan
The Campaign for Fiscal Equity would like to thank the following supporters whose
grants made this report possible: The New York City Council, the Booth Ferris Foundation,
the Donors Education Collaborative, the New York Community Trust and the
Robert Sterling Clark Foundation.
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
Table of Contents
EXECUTIVE SUMMARY .............................................................................................. 1
INTRODUCTION ......................................................................................................... 7
METHOD .................................................................................................................... 12
OVERVIEW OF FINDINGS ........................................................................................ 15
STUDENT PROFILES................................................................................................. 22
SCHOOL PROFILES ................................................................................................... 27
ATTENDANCE AND PERFORMANCE AT THE SCHOOL LEVEL ............................. 34
ATTENDANCE AND PERFORMANCE AT THE STUDENT LEVEL ........................... 37
DISCUSSION .............................................................................................................. 53
RECOMMENDATIONS............................................................................................... 58
REFERENCES ............................................................................................................ 60
APPENDIX A: TESTS OF DIFFERENCES IN PERFORMANCE AMONG
ATTENDANCE QUINTILES CONTROLLING FOR GRADE 3 PERFORMANCE ......... 62
APPENDIX B: ELA MULTILEVEL MODEL ................................................................ 63
APPENDIX C: MATHEMATICS MULTILEVEL MODEL ............................................. 67
EXECUTIVE SUMMARY
1
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
Executive Summary
TAKING ATTENDANCE SERIOUSLY:
HOW SCHOOL ABSENCES UNDERMINE STUDENT AND
SCHOOL PERFORMANCE IN NEW YORK CITY
N
early two decades ago, the Campaign for Fiscal Equity (CFE) set out to ensure
that New York’s poorest children received the “sound basic education”
guaranteed in the state constitution. A court ruling in 2006 and subsequent
state legislation brought new resources to bear on struggling public school systems, and
the state began to narrow the equity gap that divides our children. Yet we will never
close the achievement gap, even with improved curriculum and instruction, if students
are not showing up for school.
Research has documented that as many as 90,000 New York City elementary students
missed a month or more of school in 2007-08.
1
CFE’s own rigorous study of the city’s
fourth-graders found that this excessive absenteeism is commonplace on elementary
campuses: In nearly 300 schools, at least 20 percent of fourth-graders were chronically
absent last year. Further, the study shows that these absences are dragging down
student achievement, lowering scores on the state’s math and English language arts
tests. Even a child with good attendance suffers a small loss academically when the
school has a high absentee rate, suggesting that excessive absences across the board can
undermine the quality of instruction for all students by creating classroom churn and
leaving teachers mired in review and remediation.
Raising attendance rates can boost test scores, for individuals and schools, the analysis
shows. In fact, the annual predicted test score gain from simply improving a child’s
attendance equals or exceeds the annual gain expected when a child attends a charter
school. Improving attendance and, consequently, increasing instructional time for
children, is a cost-effective intervention that every school in the city can adopt right
now.
CFE’s findings raise several important points for educators:
UÊ Attendance and achievement are inextricably linked. This research, which
focused on the connection between students’ third and fourth grade attendance
and their performance on New York State Testing Program grade 4 assessments,
confirms that student attendance is a statistically significant predictor of
performance. As such, increasing attendance becomes an essential tool for
improving achievement.
1 Nauer, Kim, White, Andrew, and Yerneni, Rajeev. (2008). Strengthening)Schools)by )Strengthenin g)Families:)Community)Strategies)to)Reverse)
Chronic)Absente eism) in)the) Early) Grade s)a nd)Imp rove)S uppo rts) for)Childre n)an d)Families.)Center for New York City Affairs, Milano the New School
for Management and Urban Policy.
EXECUTIVE SUMMARY
2
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
UÊ Attendance data can be an indicator of students and schools at risk."Researchers
have repeatedly identified chronic absence—defined as missing 10 percent of the
school year as a result of unexcused and excused absences—as a signal that
students are headed off track academically. Our study confirms that poor
attendance puts low-performing students at greater risk of educational failure.
UÊ Improving attendance can reduce the achievement gap."The association between
attendance and performance is found across socioeconomic and ethnic
backgrounds. It is important to note, however, that Black and Hispanic students,
the groups with the highest poverty rates, are more likely than White and Asian
students to be chronically absent. More than one in five Black and Hispanic
students is chronically absent. Similarly, students from low-income families had
lower attendance than their more affluent peers. This suggests that improving
attendance can help reduce the achievement gaps among ethnic and
socioeconomic groups.
UÊ Reducing chronic absence is essential to turning around under-performing
schools. In 298 New York City schools, at least 20 percent of fourth-graders were
chronically absent. These high rates of absenteeism correlated with low
performance. We suspect poor overall attendance reflects the lack of a high
quality, engaging curriculum. Improvements in curriculum and instruction are
critical to school reform. But they aren’t going to help if students aren’t in the
classroom.
WHAT WE DID
National research has established that students who are chronically absent as early as
kindergarten have lower achievement in later grades. To demonstrate that connection in
New York City schools, the Campaign for Fiscal Equity reviewed the attendance records,
state assessment scores and various demographic factors for 64,062 fourth-graders
attending 705 New York City public schools in the 2007-08 school year. We chose to look
at fourth-graders, since the state’s assessments in that grade are longer and, therefore,
considered more reliable than those in third and fifth grades.
The study considered attendance in both third and fourth grades. And we analyzed
other student factors that can weigh heavily on academic performance, including
poverty, ethnicity, disability, English language proficiency, racial or ethnic background,
mobility and past performance. We also considered school characteristics such as
average attendance and test scores, percentage of minority students, and teacher
education and turnover rates. Holding other student and school variables constant to
Executive Summary
EXECUTIVE SUMMARY
3
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
isolate the role of attendance, we examined the relationship between attendance and
performance at the student level.
To give readers a better understanding of the relationship between student demographics,
attendance and performance in New York City elementary schools, we also present profiles
of student and school statistics on these measures.
WHAT WE FOUND
Overall the best predictor of fourth-grade performance is third-grade performance. But
attendance in third and fourth grade played an important role, as well. In addition to
documenting the relationship between attendance and performance, the study revealed:
UÊ The average fourth-grade student attended school for almost 94 percent of enrolled
school days in third and fourth grades. Half of fourth-graders—about 32,000
students—attended at least 95.4 percent of school days in those grades.
UÊ On the other hand, 18 percent—more than 11,000 students—were chronically absent.
That means they missed more than 10 percent of school days during that period—the
equivalent of at least 19 days in a 185-day school year. Black, Hispanic, and Native
American students were more likely to be chronically absent than White and Asian
students.
UÊ Attendance patterns varied among schools. School average attendance in 2006-07
and 2007-08 ranged from 88.1 percent to 98.8 percent. In the seven schools with the
lowest attendance, the average student attended only 88 percent of enrolled school
days. In the four schools with the highest average attendance, the typical student
attended 98 percent of the time. Thus, students in the highest-attendance schools
received 18 additional days of instruction a year, compared with those in the lowest-
attendance schools.
UÊ Rates of chronic absence varied among schools. In seven schools, no fourth-grader
was chronically absent; in the school with the highest rate, 51.8 percent were chronic
absentees. In more than three quarters (539) of the 705 study schools, at least 10
percent of the fourth-grade class was chronically absent.
WHAT THE TEST SCORES SHOW
The CFE analysis demonstrates that the school-wide attendance rate affects how much of a
boost a student receives from improving his or her own attendance. As such, the research
suggests that students will gain more if their school has a higher attendance rate. If a fourth-
grader at a school with a high attendance rate (96.3 percent) increased his own attendance
rate from 86.8 to 95.4 percent—coming to school 16 more days—we predict he would see a
5.1-point gain in his English language arts score. This may seem modest, but consider that
three years of reform, from 2006 to 2009, raised the city’s average fourth-grade English
EXECUTIVE SUMMARY
4
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
language arts score by a total of 6 points. And a study by the New York City Charter Schools
Evaluation Project predicted a 3.6-point annual gain for students in fourth through eighth
grade who attend charter schools.
2
In math, the student would see a 6.9-point gain with
better attendance, compared with a 5-point annual increase at a charter school.
3
For
students at schools with low attendance rates (91.4 percent), the predicted gains are not as
large: 3.5 points in English language arts and 3.7 points in math.
Good attendance can not only
bring better scores for students,
but for schools as well. If the seven
schools with the worst third- and
fourth-grade absentee rates
brought their attendance up from
88 percent to the city average of
93.8 percent, the predicted average
scaled-score gain would be 4.8
points in English language arts
and 6.0 points in math. For some
schools, these gains could mean
the difference between meeting
the standards for yearly progress set under No Child Left Behind and failing.
WHY IT MATTERS
The results of the CFE analysis underscore the vital importance of attending school. A
growing body of research demonstrates the same. In Philadelphia, researcher Michael
Gottfried found similar associations between attendance and standardized test performance
in a study of public school students in third through eighth grade. His research
demonstrated that this association exists independent of other family characteristics, such
as parent education and involvement in school activities.
4
His research strongly suggests that
there is a direct link between attendance and performance.
In 2008, Hedy Chang and Mariajosé Romero at the National Center for Children in Poverty
analyzed U.S. Department of Education data for 21,260 children nationally from
kindergarten through fifth grade. They found that one in 10 kindergarten and first-graders
were chronically absent. By the end of first grade, these children were already slipping
behind in reading, math and general knowledge. Chronic absence in kindergarten was also
strongly associated with lower reading and math performance in fifth grade for poor
children.
5
An analysis that considered New York City students’ attendance from kindergarten
through grade 4 would likely show a stronger correlation of attendance with test scores than
documented by the CFE study.
2 Hoxby, Caroline M., Murarka, Sonali, and Kang, Jenny. (2009). How)New) York) C ity’s)Ch a r te r)S ch o ol s)Af fec t)Ach ie v e me n t ,)Augus t)20 09)Re p or t .
Cambridge, MA: New York City Charter Schools Evaluation Project.
3 Ibid.
4 Gottfried, M. A. (2011). The Detrimental Effects of Missing School: Evidence from Urban Siblings. American Journal of Education, 117, 147–182
5 Chang, H. & Romero, M. (2008). Present,)Engaged,)and)Accounted)for)–)The)Cri tical)Importance)of)Addressing)Chronic)Absence)in)the)Early)Grades.))
National Center for Children in Poverty, Mailman School of Public Health, Columbia University.
Scaled-Score Gain
Gain
PREDICTED SCALED-SCORE GAIN
EXECUTIVE SUMMARY
5
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
Improving attendance and performance are particularly important now that the New
York Board of Regents has instituted tougher standards for proficiency in reading and
math. These new standards resulted in smaller percentages of fourth-graders scoring at
the proficient levels in 2010. The percentage of proficient students decreased 17 points in
English language arts and 22 points in math. Based on historical data, we expect that
fewer of these fourth-graders will be proficient when they reach eighth grade. This is
ominous because a previous CFE study found that high schools with the largest
percentage of entering ninth-graders who, as eighth-graders, had frequent absences and
failed to reach the State learning standards had the lowest Regents diploma rates.
Clearly, test scores are the coin of the realm when it comes to education reform,
measuring school progress and, increasingly, teacher effectiveness. But the value of good
attendance extends far beyond standardized testing gains. For students, attending
school regularly can be a sign that they are engaged in learning, while poor attendance
as early as sixth grade can signal that a student will eventually drop out of high school.
6 7
For teachers, good attendance means working with a full classroom, rather than having
to repeat material for absentee students the next day. For communities, good attendance
has been linked to lower crime rates and higher graduation rates, which in turn bring
better employment and stronger local economies. The consequences of dropping out on
later income, dependence on welfare, and incarceration are widely documented. Each of
these consequences has serious implications for the larger community. Anecdotally,
schools find that good attendance begets good attendance: Students, especially in
secondary school, want to go where their friends are.
Recognizing the value of attendance to school improvement, Mayor Michael Bloomberg’s
office last fall launched a pilot program aimed at reducing chronic absenteeism and
truancy in 25 schools across the city. The program educates parents about the value of
good attendance, offers incentives for children to come to school and provides mentors
for students who are missing 10 percent or more of the school year. The efforts have
already borne fruit: In the first half of the school year, fully 22 of the 25 schools reduced
their absentee rates. The 10 elementary schools saw the best results, with a collective 24
percent decline in the percentage of students who were chronically absent. The seven
high schools showed little change.
CFE believes that this sort of program, as well as the community schools approach used
in many New York City schools, can substantially reduce chronic absence. When
attendance rises, performance will follow. If good curriculum and instruction are also in
place, we can start to make real progress for all of the city’s children.
6 Balfanz, Robert, Herzog, Lisa & MacIver, Douglas J. (2007). Preventing Student Disengagement and Keeping Students on the Graduation
Path in Urban Middle-Grades Schools: Early Identification and Effective Interventions. Educational Psychologist, 42, 223–235.
7 Ou, Suh-Ruu & Reynolds, Arthur J. (2008). Predictors of Educational Attainment in the Chicago Longitudinal Study. School )Psychology)
Quarterly,)23,199-229.
EXECUTIVE SUMMARY
6
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
RECOMMENDATIONS
UÊ Identify and share best practices for improving attendance. We must ensure
that all schools have effective policies for identifying and monitoring chronically
absent students, reaching out to those students and their families and ensuring
that they are on track academically. Most importantly, schools must create a
climate in which all staff, students and families understand the importance of
attendance, set high attendance goals and work to minimize absences.
UÊ Ensure that schools and teachers are looking at the right data. Schools need to
go beyond schoolwide attendance averages to analyze how many absences,
excused and unexcused, each student has accrued and to look for patterns in
neighborhoods, ethnic groups, grades or classrooms. Especially in the early
grades, absentee students often are not willfully skipping school but rather
missing days because of health and safety concerns, frequent moves or
unreliable transportation. Schools should identify these barriers and work with
parents and community organizations to address them.
UÊ Hold schools accountable for attendance and chronic absence rates at the city,
state and federal levels. It is important that schools be held accountable for
improving attendance. To that end, attendance and chronic absence rates should
be publicly available and reported for all federal accountability groups, including
racial and ethnic groups. We recommend that federal, state and city
accountability systems be revised to increase the value of attendance in
assessing school progress. Currently New York is one of five states that does not
include attendance data in its longitudinal student database. The state should
work with school districts to standardize and collect student-level attendance
data, and to develop standard definitions of chronic absence and truancy, so that
comparable measures are used statewide.
UÊ Engage parents and the community in improving school attendance. It goes
without saying that parents play an essential role in getting their children out of
bed and off to school each morning. Often, especially in kindergarten and first
grade, parents simply don’t realize the value of attendance. Community
organizations can help educate parents, support families who need help with
child care or transportation, and provide services to children and families who
need extra assistance.
UÊ Strengthen comprehensive school reform efforts so that when children come to
school they find an excellent curriculum, engaging teachers, a strong principal
and periodic assessments that ensure all students are on track toward meeting
graduation standards.
INTRODUCTION
7
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
Introduction
N
early two decades ago, the Campaign for Fiscal Equity (CFE) set out to ensure
that New York’s poorest children receive the “sound basic education”
guaranteed by the State constitution. A court ruling in 2006 and subsequent
State legislation brought new resources to bear on struggling public school systems, and
the State began to narrow the equity gap that divides our children. Yet we will never
close the achievement gap, even with improved curriculum and instruction, if students
are not showing up for school.
Regular school attendance is critical to academic success. National research establishes
that absence as early as kindergarten has an ongoing negative effect on achievement. In
many New York City schools, students are absent far too often and research links these
absences to diminished elementary school performance and lower high school
graduation rates. We also know that chronic absenteeism in elementary schools is
disproportionately a problem in poor and minority communities and contributes to the
achievement gaps among ethnic groups.
This rigorous study of fourth-grade students in New York City public schools documents
that, for individual students, higher attendance predicts higher performance on State
assessments of English language arts (ELA) and mathematics. In addition, higher mean
school attendance also predicts higher student performance. Therefore, the predicted
performance gain from higher individual attendance is greatest for students who attend
a school with high mean attendance. These associations are independent of the
relationships of other student and school variables with performance. That is, when all
other variables are held constant, there is a significant association between attendance
and performance at the student and school level. This report describes the method and
findings of this study and discusses their implications for improving performance in
New York City Public Schools.
ATTENDANCE PROBLEM IN NEW YORK CITY
Too many students in New York City are absent too often. In New York City, one in five
children misses at least a month of school each year—and in many neighborhoods the
number is much higher. According to the Center for New York City Affairs (the Center),
the City’s elementary schools have far more serious rates of absenteeism than had been
previously reported (Nauer, White, & Yerneni, 2008). While City school officials have
long been concerned by poor attendance rates in middle and high schools, an analysis
by the Center found that attendance problems begin with much younger students.
The Center’s analysis of Department of Education (DOE) data found that more than
90,000 children in grades K through 5—or 20 percent of total enrollment—missed at
least a month of school during the 2007–08 school year. They reported that 15.7 percent
INTRODUCTION
8
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
of students in grades kindergarten through 5—and 16.2 percent of students in grades 6
through 8—missed 30 or more days of school in the 2007-08 school year. In 165 schools
serving students in grades kindergarten through 8, at least 30 percent of students were
chronically absent. These schools were concentrated in areas of New York City with the
highest poverty rates.
PREVIOUS RESEARCH FINDINGS
Historically, most published research has examined attendance at the school rather than
the student level. These studies have generally found that higher average school
attendance is associated with higher performance (Roby, 2004). Several recent studies
have documented that for individual students being present at school more days is
associated with higher performance. Michael A. Gottfried’s research, by employing very
rigorous statistical methods, substantially advances our understanding of the link
between individual student attendance and performance. These rigorous methods
allowed him to eliminate alternative explanations of the link between attendance and
performance. One such alternative is that the apparent link between attendance and
performance results from the strong association of each with family characteristics,
such as socioeconomic status, mother’s education, and family involvement in education.
In this view, there is no direct link between attendance and performance. Gottfried’s
research discredits this possibility.
Gottfried has published three journal articles based on his analyses of a comprehensive
data set for elementary and middle school students in the Philadelphia School District
covering the 1994-95 through 2000-01 school years. In one study (Gottfried, 2010),
controlling for student, classroom, school, and neighborhood characteristics, as well as
past student performance, he found that the number of days the student was present
was positively related to both grade point average and standardized test results in the
elementary and middle grades. The relationship was somewhat stronger for middle than
elementary students and for mathematics than reading. The quasi-experimental
approach used in his research supports the premise that there is a causal relationship
between attendance and performance.
Using records for students in second through fourth grade from the same data set,
Gottfried examined the relationship of kind of absence—excused or unexcused—to
performance on reading and mathematics standardized tests (Gottfried, 2009). He found
that higher proportions of unexcused to total absences were associated with lower
performance, particularly in mathematics.
In a third study, Gottfried (2011) sought to isolate the effect of missing school on
elementary reading and mathematics performance. He controlled the effects of family
Introduction
INTRODUCTION
9
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
variables on both the number of days a student is absent and on student standardized-
test performance. Using the Philadelphia School District data set, he tracked five cohorts
of siblings for six school years beginning in the 1994-95 school year. As in previous
studies, he controlled for student, classroom, school, and neighborhood characteristics.
He eliminated family effects by coding each student’s absences and standardized test
scores as deviations from the family average. To illustrate this method in its simplest
form, consider two families. The siblings in one family were absent for four and eight
days and siblings in the second family were absent for 16 and 20 days. In each family,
one sibling was absent for two days more—and one sibling was absent for two days
fewer—than the family average of 6 or 18. The sibling in each family with fewer absences
would be coded -2 and the sibling with more absences would be coded +2. Using this
method, he eliminated family differences that caused the second set of siblings to be
absent for a greater number of days than the first set. By applying the same procedure to
standardized test scores, he eliminated family differences that caused some families to
achieve higher scores than other families. Using these deviation measures, he then
examined whether higher numbers of absences were related to lower test scores. He
found that eliminating the family effects resulted in a stronger relationship between
attendance and performance than he had found in previous studies without this control.
Hedy Chang and Mariajosé Romero (2008) at the National Center for Children in Poverty
(NCCP) reported on the importance of addressing chronic absence in the early grades.
They analyzed U.S. Department of Education national data for 21,260 children from
kindergarten entry in 1998 to grade 5. Chronic absence was defined as missing 10
percent or more of a school year: at least 18 days out of a 180-day school year. They
found that chronic absenteeism is disproportionately a problem in elementary schools
that serve mostly poor Black and Hispanic children. It contributes to the achievement
gap between these children and their White, Asian, and middle-class peers. Students
who have many absences in kindergarten are likely to have similar attendance problems
in first grade. By the end of first grade, these children are already slipping behind in
reading, math and general knowledge. Chronic absence in kindergarten was also
strongly associated with lower reading and math performance in fifth grade for poor
children.
Robert Balfanz and Vaughan Byrnes (2006) analyzed records for four cohorts of middle-
school students (grades 5-8) attending three high-poverty schools implementing whole-
school reform models in the Philadelphia School District. They divided students into two
groups according to whether they were closing the gap between achievement and grade-
level expectations during the middle school years or continuing to fall behind. Students
were counted as closing the gap, if according to standardized test scores, their gain in
grade equivalents was greater than their number of years in middle school. Better
INTRODUCTION
10
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
attendance was among the factors found to distinguish students who were closing the
gap from those who were not. Other factors that achieved similar results were more
years in home rooms where larger percentages of children made high gains, better
behavior marks, and higher self-rankings of effort in mathematics.
The Campaign for Fiscal Equity (CFE) has completed two analyses of the relationship of
school attendance and school performance. In the first study, we found average school
attendance to be a significant predictor of school grade 4 ELA performance. The study
documented the relationship between attendance and performance on the New York
State grade 4 English language arts (ELA) assessment at the school level. The
explanatory power of attendance was independent of other school variables: percentages
of economically disadvantaged students, English language learners, Black and Hispanic
students, and students with disabilities. These variables accounted for almost 78 percent
of the variation among schools in the performance of fourth-graders, with attendance
and the percentage of economically disadvantaged students being the best predictors of
performance.
CFE’s recent study, Diploma Dilemma: Rising Standards, the Regents Diploma, and
Schools that Beat the Odds
(Campaign for Fiscal Equity, 2010), documented the
importance of attendance in high schools. The study grouped New York City high
schools according to the grade 8 performance of entering ninth-graders. Within each
group of high schools serving similar students, schools with the lowest and highest
Regents diploma rates were distinguished by average daily attendance in the students’
expected graduation year. Schools with the highest Regents diploma rates had, on
average, the highest attendance rates.
RESEARCH OBJECTIVE
Improving performance is a critical goal and attendance is a key element in achieving
that goal. Before 2000, the United States had the largest percentage of college graduates
in the world. By 2010, according to the College Board, we had slipped to 12
th
place out of
36 countries in the percentage of 25 to 34 year olds with at least an associate’s degree. A
recent report released by the State Education Department (2011) showed that only 23
percent of New York City students who entered grade 9 in 2005-06 had graduated by June
2009 with performance that indicated adequate preparation for college; that is, scores of
75 or higher on the Regents English examination and 80 or higher on a Regents
mathematics examination. Our future prosperity depends on better preparing students
to meet future challenges, which will surely depend on higher levels of literacy,
enhanced technical and scientific skills, and greater problem-solving ability. We cannot
afford for students to leave high school without adequate preparation for college and the
workplace, as too many New York City students do today.
INTRODUCTION
11
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
All students now entering ninth grade in New York State except those with disabilities
must meet the more rigorous Regents diploma requirements to graduate. CFE’s research
suggests that entering ninth grade not having achieved the eighth-grade learning
standards substantially reduces a student’s chance of earning a Regents diploma. State
assessment results show that by third grade many students have fallen far behind grade-
level expectations and will fall farther behind without effective interventions, year by
year decreasing the probability of earning a Regents diploma.
The research objective is to document the relationship between individual student
attendance and performance in New York City elementary schools. To achieve this
objective, CFE performed multilevel regression analyses of performance on the grade 4
ELA and mathematics assessments. In examining this relationship, we accounted for the
association between previous performance (third grade) and fourth-grade performance
and for other student and school factors that are simultaneously related to attendance
and performance. These student factors include poverty, ethnicity, gender, disability,
English proficiency, and mobility. The school factors include average school attendance,
average grade 3 performance, percentage minority enrollment, and teacher
qualifications. These analyses define how these explanatory variables are associated
with the performance of individual students and how they vary among schools. We will
use the results to focus the attention of elementary-school staff and parents on the
importance of attendance in enabling students to meet learning standards and to
support the continuation and enhancement of comprehensive school reform in New
York City schools.
12
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
METHOD
DATA
W
e analyzed student data for 705 New York City Public
Schools that served fourth-graders in 2007-08. These
schools enrolled 64,062 fourth-grade students whose
records met the following criteria for study inclusion:
UÊ included scores for the 2006-07 third-grade and 2007-08 fourth-
grade assessments in ELA and/or mathematics.
UÊ indicated enrollment in fourth-grade in a New York City public
school—other than a school in District 75—at the end of the
2007-08 school year.
UÊ included attendance data for 2006-07 and 2007-08.
The New York City Department of Education (DOE) provided the
following data for third-graders in 2006-07 and fourth-graders in 2007-
08: A non-personally identifiable student tracking number; school
identification number where student was registered on October 31 of
each school year; school identification number where student was
registered on June 30 of each school year; gender; ethnicity; grade
level; identifiers of eligibility for free- or reduced-lunch, limited English
proficiency, and disability; days absent, present and released; and
scaled scores and performance levels for the State grade 3 and 4
assessments in ELA and mathematics.
We also obtained data on the teacher characteristics of study schools
in 2007-08 from the State Education Department’s Report Card Data
Base. These characteristics included lack of appropriate certification,
years of experience, graduate education beyond the master’s degree,
and teacher turnover rate.
The achievement data consisted of scaled scores and performance
levels on the 2007 grade 3 and the 2008 grade 4 ELA and mathematics
assessments of the New York State Testing Program. These assessments
measured the performance of students in grades 3 through 8 relative to
the State Learning Standards. In 2008 each of these assessments was
graded on a scale consisting of 280 to 360 points, with a scaled score of
650 indicating proficiency at each grade level. The scale is divided into
four performance levels: Level 1 identifies students with serious
academic deficiencies; Level 2, students partially meeting the
Method
13
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
METHOD
standards; Level 3, students meeting the standards and thus considered proficient; and
Level 4, students meeting the standards with distinction. A great majority of students in
the City and the State perform at Level 2 or 3 on each assessment.
GRADE 4 STATE ASSESSMENTS
Performance Levels, Corresponding Scaled-Score Ranges, and
Percentages of New York City Students Scoring at Each Level in 2007-08
PERFORMANCE
LEVEL
ELA MATHEMATICS
SCALED-SCORE
RANGE
PERCENT AT
LEVEL
SCALED-SCORE
RANGE
PERCENT AT
LEVEL
1 430-611 10.5 485-621 6.3
2 612-649 28.2 622-649 14.1
3 650-715 55.5 650-701 53.4
4 716-775 5.8 702-800 26.2
Using DOE data elements, we calculated the following attendance variables:
UÊ Student Days Enrolled is the sum of Days Absent, Days Present, and Days
Released. (Some students were not enrolled for the entire school year.)
UÊ Student School Year Attendance
was calculated by dividing the number of Days
Present
by Days Enrolled in each school year and multiplying by 100.
UÊ Student Cumulative Attendance was calculated by dividing the sum of Days
Present in 2006-07 and 2007-08 by Days Enrolled in 2006-07 and 2007-08 and
multiplying by 100.
UÊ School Mean Attendance is the arithmetic mean of Student Cumulative
Attendance
for fourth-graders enrolled in the school at the end of the 2007-08
school year.
UÊ A Chronic Absence Identifier was assigned to students who attended school for
fewer than 90 percent of enrolled school days in 2006-07 and 2007-08.
Students were considered to be continuously enrolled if the school identification
numbers on their records indicated that they were enrolled in the same school from
October 31, 2006 until the end of the 2007-08 school year. Continuous enrollment is the
measure of mobility.
ANALYSES
The focus of this research is multilevel regression analyses of the relationships between
attendance and performance on the grade 4 ELA and mathematics assessments. These
analyses are multilevel in that they consider both student and school characteristics in
predicting student performance. This statistical technique allows us to quantify the
association of each explanatory variable with performance, independent of the others.
The most powerful predictor of assessment performance is previous performance.
14
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
METHOD
Therefore, our analyses controlled for grade 3 ELA and mathematics performance.
Including grade 3 performance controlled for the time-invariant characteristics of
students that are simultaneously related to attendance and performance. These time-
invariant characteristics include those components of motivation, ability, and family
support that do not change over time. We also included factors that previous analyses
have shown to be associated with performance, that is, eligibility for free- or reduced-
price lunches (poverty), ethnicity, gender, disability, English proficiency, and continuous
enrollment. Because schools provide the context in which students are educated, school
characteristics can be expected to modify the relationship between student attendance
and performance. Therefore, these analyses also examined the mediating effects of
school contextual factors—mean grade 3 performance, school mean attendance, ethnic
composition, and a teacher quality variable. We selected turnover rate for ELA and
percentage of teachers with 30 credit hours beyond the master’s degree for mathematics
as the best predictors of performance. These analyses produced equations that predict
the grade 4 performance of individual students.
PRESENTATION OF FINDINGS
The next section provides an overview of research findings. The following three sections
provide background information to assist the reader in understanding the multilevel
analyses. “Student Profiles” presents demographic, attendance, and performance
profiles of the 64,062 fourth-graders included in the study. We also present similar
profiles for each ethnic group to show the associations between ethnicity and eligibility
for free-and-reduced-priced lunches, disability, English proficiency, attendance, and
performance. “School Profiles” presents enrollments, demographics, attendance,
performance, teacher qualifications—lack of appropriate certification, years of
experience, graduate education—and teacher turnover rates for the 705 schools included
in the study. This section also examines the relationships of student demographics and
teacher characteristics with attendance. “Attendance and Performance at the School
Level,” examines the relationship of school mean attendance with school mean ELA and
mathematics performance. The final section of findings reports on the multilevel
analyses of the relationships between individual attendance and performance on the
grade 4 ELA and mathematics assessments. The relationship of each explanatory
variable to grade 4 performance is described.
15
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
OVERVIEW OF FINDINGS
T
his study is important because unlike previous studies of New York City students
it examines the relationship of attendance and performance at the student level,
while considering other student and school characteristics that are related to
performance. Our multilevel analyses document that both individual attendance and
school mean attendance in grades 3 and 4 are associated with performance on the grade
4 assessments in ELA and mathematics. Higher individual student attendance predicts
higher performance and attending a school with higher mean attendance increases the
predicted performance gain. These associations are independent of the relationship of
other student and school variables with performance. That is, when all other variables
are held constant, there is a significant association, for individual students, between
attendance and performance. These predicted results are based on data for 64,062
students in 705 schools and are highly reliable. The odds that there is no relationship
between attendance in grades 3 and 4 and grade 4 ELA and mathematics performance
are less than one in a thousand.
MULTILEVEL ANALYSIS FINDINGS
Individual Student Attendance
Our analyses produced equations that allow us to predict student performance under
various conditions. The relationship of individual student attendance with grade 4
performance is not uniform across schools. The relationship is stronger in schools with
higher average attendance in grades 3 and 4. To illustrate how performance varies
among schools and students with different attendance rates, we compared predicted
fourth-grade scaled scores in hypothetical schools with low-attendance—91.4 percent—
and high-attendance—96.3 percent. Only 10 percent of schools had lower attendance
than 91.4 percent, while 90 percent had lower attendance than 96.3 percent. The average
student in the low-attendance school missed more than twice as many days—16 in a 185
day school year—than the average student in the high attendance school—7 days.
Within each school, we compared the predicted grade 4 scores of three students with
different attendance rates: a chronically absent student (86.8-percent attendance), a
typical student (95.4-percent attendance), and a high-attendance student (99.0-percent
attendance). Students with different grade 3 performance and different demographic
characteristics will have different predicted scores. The predicted gains of students with
the same attendance improvement, however, depend only on their schools’ mean
attendance. For example, students with low and high grade 3 scores attending the school
with high mean attendance can expect the same gain by improving their attendance
from 86.8 to 95.4. The student with the higher grade 3 score will however have a higher
predicted grade 4 score because higher grade 3 performance predicts higher grade 4
performance. Table 1 provides the predicted gains achieved by improving attendance in
schools with the specified mean attendance rates, regardless of the student’s
demographic characteristics and grade 3 performance.
Overview of Findings
16
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
OVERVIEW OF FINDINGS
TABLE 1
The Predicted Scaled-Score Gains on the
Grade 4 ELA Assessment as Student Attendance Improves
STUDENT ATTENDANCE
SCHOOL MEAN ATTENDANCE
91.4 PERCENT (169 DAYS
IN 185 DAY SCHOOL
YEAR)
96.3 PERCENT (178
DAYS IN 185 DAY
SCHOOL YEAR)
Gain Predicted by Increasing Attendance
from 86.8 to 95.4 Percent
3.5 5.1
Gain Predicted by Increasing Attendance
from 95.4 to 99.0 Percent
1.5 2.1
Total Gain Predi cted by Increasi ng Attend ance
from 86.8 to 99.0 Percent
5.0 7.2
As shown in Table 1, in the low-attendance school, the predicted ELA scaled-score
difference between the chronically absent student and the student with typical
attendance was 3.5 scaled-score points. A student with very high attendance is predicted
to score an additional 1.5 points compared with the student with typical attendance.
Attendance had a stronger relationship with performance in high- than low-attendance
schools. In the high-attendance school, the predicted grade 4 ELA score of the student
with typical attendance is 5.1 points higher than that of the chronically absent student.
The predicted score of the high-attending student is another 2.1 points higher.
1
These predicted gains can be put in perspective by comparison to other benchmarks.
The predicted ELA scaled-score gain from improving a student’s attendance from 86.8 to
95.4 percent in a high-attendance school is 5.1 points. The mean grade 4 ELA score for
New York City students increased from 657 to 663 between 2006 and 2009; this gain of
six points was seen as indicating improvements in the school system. In 2007-08, more
than one quarter of fourth-graders scored at ELA Level 2, which includes scores from 612
to 649. The expected gain is 14 percent of that range. To cite another benchmark, the
average difference between minority students and White and Asian students on the
Grade 4 ELA assessment was about 30 points. The expected gain is 17 percent of that
difference. Finally, an evaluation of New York City charter schools (Hoxby, Murarka, &
Kang, 2009) estimated that the extra annual gain in ELA scaled score achieved by the
average student in grades 4 through 8 attending a charter school—rather than the
average City public school—would be 3.6 points.
1 Note that school average attendance increases when multiple enrolled students improve their attendance. As the average attendance of
the school increases, so will the predicted performance gain of individual students with improved attendance. If the low-attendance
school succeeds in increasing its average attendance to 92.4 percent, the predicted gain in ELA score increases to 4.5 points and the
predicted gain in mathematics score increases to 5.0 points.
Overview of Findings
17
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
OVERVIEW OF FINDINGS
TABLE 2
The Predicted Scaled-Score Gains on the
Grade 4 Mathematics Assessment as Student Attendance Improves
STUDENT ATTENDANCE
SCHOOL MEAN ATTENDANCE
91.4 PERCENT (169
DAYS IN 185 DAY
SCHOOL YEAR)
96.3 PERCENT (178 DAYS
IN 185 DAY SCHOOL
YEAR)
Gain Predicted by Increasing Attendance from
86.8 to 95.4 Percent
3.7 6.9
Gain Predicted by Increasing Attendance from
95.4 to 99.0 Percent
1.6 2.9
Total Gain Predicted by Increasing Attendance from
86.8 to 99.0 Percent
5.3 8.5
In Table 2, we see a similar pattern when we predict grade 4 mathematics scaled scores.
In the low-attendance school, the scaled-score difference between our chronically
absent student and our typical student is 3.7 points. The high-attendance student would
score an additional 1.6 points. Again the relationship between student attendance and
performance is stronger in high- than low-attendance schools. In the high-attendance
school, the predicted score of the student with typical attendance is 6.9 points higher
than that of the chronically absent student. The high-attendance student would score an
additional 2.9 points.
Again we place the gains in perspective. The predicted mathematics scaled-score gain
from improving a student’s attendance from 86.8 to 95.4 percent in a high-attendance
school is 6.9 points. Between 2006 and 2009, the mean grade 4 mathematics scaled score
for New York City students improved from 669 to 688.
2
In 2007-08, almost one in seven
fourth-graders scored at math Level 2, which includes scores from 622 to 649. The
expected gain is 25 percent of that range. Black, Hispanic, and Native American students
scored about 27 points lower than White and Asian students on the Grade 4 mathematics
assessment. The expected gain is about 26 percent of that difference. The predicted
average annual mathematics gain for students attending a charter school is 5.0 points.
School Attendance
The multilevel analyses results reported above document that higher student attendance
predicts better performance on ELA and mathematics assessments. The analyses also
document that increased school mean attendance predicts higher school mean scaled
scores on these assessments. Each one percentage point increase in school mean
attendance corresponds, on average, to a 0.803-point increase in a school’s grade 4 ELA
mean scaled score. This finding suggests that the seven schools with mean attendance
of only 88 percent could improve their mean ELA scaled score by about 4.8 points by
increasing their attendance to the mean of the schools’ means, 93.8 percent. The
improvement to be gained in mathematics by this increase is even larger: 0.991 points
for each one percentage point increase in school mean attendance. For the seven schools
2 The acknowledged increases in the predictability of mathematics test questions during those years suggest that some portion of that
increase does not reflect real increases in achievement.
18
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
OVERVIEW OF FINDINGS
with the lowest attendance, this amounts to a predicted increase of six scaled-score
points.
The multilevel analyses also documented the relationships of student grade 3
performance, student demographic variables, and school variables with grade 4
performance. These relationships are summarized below:
UÊ Student variables. The analysis of ELA data confirmed that eight student
variables are significantly associated with grade 4 performance. Higher
attendance, higher grade 3 performance, being female, and being continuously
enrolled in the same school in grades 3 and 4 predict better grade 4 performance.
Being Black, Hispanic, or Native American; being eligible for free- or reduced-
price lunches; being disabled; and having limited English proficiency predict
lower grade 4 performance. These associations are independent of the
relationships among these explanatory variables.
UÊ School variables. Three school variables, in addition to school mean attendance,
are associated with performance: school grade 3 ELA mean score, percentage of
minority students, and teacher turnover rate. A higher mean grade 3 ELA score
predicts higher grade 4 performance, while larger percentages of minority
students and higher teacher turnover predict lower grade 4 performance.
UÊ Cross-level interactions. Five cross-level interactions are associated with grade 4
ELA performance. For example, the relationship between individual attendance
and grade 4 performance is mediated by school mean attendance. The
interaction increases or decreases students’ predicted scores depending on the
mean attendance of the school they attend. Consider first students whose
attendance is above the school mean. If they attend a school with above average
school attendance, the interaction will result in a small addition to their
predicted scores. For such students who attend a school with below average
school attendance, the interaction will result in a small deduction from their
predicted scores. The opposite is true for students whose attendance is below
their schools’ mean.
UÊ Mathematics. The multilevel analysis of mathematics performance produced
similar findings but showed a stronger association between individual
attendance and performance. Seven of the eight student-level variables have
significant relationships with grade 4 mathematics performance, independent of
all other variables. The association of continuous enrollment with grade 4
mathematics performance is not statistically significant. Being female predicts
lower rather than higher performance. Further, no teacher quality variable is
significantly related to performance, once we account for all other explanatory
variables. Only three cross-level interaction variables meet the statistical criteria
for inclusion in the model.
19
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
OVERVIEW OF FINDINGS
OTHER STUDY FINDINGS
The findings listed below provide a context for understanding the results of the
multilevel analyses, summarized above.
The 64,062 fourth-graders in New York City Public Schools in 2007-08 were ethnically
and economically diverse. The prevalence of disability, English proficiency, and mobility
and patterns of attendance and performance varied among ethnic groups. (See “Student
Profiles” for details.)
UÊ The majority of fourth-graders (71 percent) were Hispanic or Black and 84
percent came from low-income families.
UÊ 18 percent of fourth-graders were disabled and 15.2 percent were limited English
proficient.
UÊ About 87 percent were continuously enrolled in the same school from October 31,
2006 to the end of the 2007-08 school year.
UÊ The average student attended school for almost 94 percent of enrolled school
days in grades 3 and 4. Half of fourth-graders—about 32,000 students—attended
at least 95.4 percent of school days in grades 3 and 4.
UÊ Eighteen percent—over 11,000 students—however were chronically absent; that
is, they missed more than 10 percent of school days during that period—the
equivalent of at least 19 days in a 185-day school year.
UÊ Each ethnic group had distinctive demographic characteristics. White students
were much less likely than other students to come from low-income families and
be eligible for free- or reduced-price lunches. Asian students were less likely to be
classified as disabled. Asian and Hispanic students were most likely to be English
language learners. Native American, Black, and Hispanic students were at least
five times as likely as Asian students to be chronically absent. White students
were most likely to be continuously enrolled in the same school in third and
fourth grade.
UÊ The average student scored at the proficient level (650 or higher) on the grades 3
and 4 State assessments in ELA and math in both third and fourth grade.
Underlying these average scores lies a continuum moving from scores signifying
no mastery of grade-level learning objectives to perfect test performance.
Most of the 705 schools had student enrollments that did not reflect the City’s ethnic and
economic diversity; rather their enrollments were drawn primarily from one or two
ethnic groups and families with similar income levels. Schools varied widely on
demographic indicators, on teacher qualifications, and in attendance rate and
performance. (See “School Profiles” for details.)
20
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
OVERVIEW OF FINDINGS
UÊ In half of study schools, 90 percent or more of fourth-graders were Native
American, Black, or Hispanic. In contrast, in 19 schools at least 90 percent were
White or Asian.
UÊ Schools diverged demographically: 5 enrolled no students with disabilities; 44 no
English language learners. In seven schools, all fourth-graders were continuously
enrolled; in six, none was continuously enrolled. In 219 schools, all fourth-
graders were eligible for free- and reduced-price lunches; in one school at the
other extreme fewer than four percent were eligible.
UÊ Attendance patterns varied among schools. School mean attendance in 2006-07
and 2007-08 ranged from 88.1 to 98.8 percent. In the seven schools with the
lowest attendance, students, on average, attended only 88 percent of enrolled
school days in 2006-07 and 2007-08. These students missed about 22 days in the
185-day school year. In the four schools with the highest average attendance,
students, on average, attended 98 percent of enrolled schools days, missing no
more than four days. Students in the highest-attendance schools received 18
additional days of instruction compared with those in the lowest-attendance
schools.
UÊ Similarly, rates of chronic absence varied among schools. In seven schools, no
fourth-grader was chronically absent; in the school with the highest rate of
chronic absence, 51.8 percent of fourth-graders attended school fewer than 90
percent of enrolled days in third and fourth grades; that is, they missed more
than 18 days in a 185-day school year. In more than three quarters (539) of the
705 study schools, at least 10 percent of fourth-graders were chronically absent.
UÊ Schools showed a range of performance on the State ELA and math assessments.
In schools with the lowest performance on the ELA assessment, the average
student barely scored at Level 2, indicating very limited achievement of the
learning standards; in the highest-performing schools, the average student
scored at Level 4. Students, on average, performed better on the mathematics
than the ELA assessment, but the range of performance among schools was
almost as great in mathematics as in ELA.
These differences among schools predict differences in attendance and performance.
(See “Attendance and Performance at the School Level” for details).
UÊ Demographic variables are significantly correlated with attendance at the school
level: larger percentages of fourth-graders who were eligible for free- and
reduced-priced lunches; who were Native American, Black, or Hispanic; and who
were classified as disabled are associated with lower attendance rates. Larger
percentages of students who were continuously enrolled are associated with
higher attendance.
21
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
OVERVIEW OF FINDINGS
UÊ The percentages of teachers without appropriate certification, the percentage
without three years’ experience, and the teacher turnover rate each have small
but highly significant negative associations with school attendance. As each
measure increased, school attendance decreased. The percentage of teachers
with a master’s degree plus at least 30 credit hours is significantly associated
with higher attendance.
UÊ Both school mean attendance in 2006-07 and 2007-08 and the percentage of
chronically absent students have moderate correlations with mean scaled scores
on the grades 3 and 4 assessments in ELA and mathematics. As average school
attendance increased, performance improved. As the percentage of chronically
absent students increased, school performance declined.
UÊ When schools are ranked according to attendance and divided on attendance
into five groups of equal size, the groups differ significantly on fourth-grade
performance even when third-grade performance is accounted for. The group
with the highest attendance, achieved an unadjusted mean grade 4 ELA score 32
points higher—and a mean grade 4 math score 35 points higher—than the group
with the lowest attendance.
22
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
STUDENT PROFILES
T
hese profiles reveal the diversity of fourth-graders in New York City
Public Schools. Table 3 shows the demographic profiles of 64,062
fourth-graders in 2007-08. The largest ethnic group was Hispanic
(40.4 percent), followed by Black (30.6 percent). The smallest group—with
only 0.4 percent of study group enrollment—was Alaskan Native or Native
American. In later analyses these three ethnic groups were combined and
referred to as the minority group. White and Asian students together made
up 28.5 percent of the study group,
3
which included slightly fewer females
than males.
A great majority of fourth-graders came from low-income families; 85
percent were eligible for free- or reduced-price lunches. In fourth grade, 18
percent were classified as disabled and 15.2 percent were limited English
proficient. Over 87 percent were continuously enrolled in the same school
from October 31, 2006 through the end of the 2007-08 school year. Eighteen
(18) percent—over 11,000 students—were chronically absent; that is, they
attended school for fewer than 90 percent of enrolled school days in 2006-07
and 2007-08—they missed the equivalent of at least 19 days in a 185-day
school year.
TABLE 3
Demographic Profile of 64,062 Fourth-Grade Students in 2007-08
STUDENT CHARACTERISTIC PERCENT
Native American 0.4%
Black 30.6%
Hispanic 40.4%
Asian 14.0%
White 14.5%
Female 49.5%
FRPL Eligible 85.0%
Students with Disabilities 18.0%
Limited English Proficient 15.2%
Continuously Enrolled 87.3%
Chronically Absent 18.0%
3 All but seven students were identified by DOE as belonging to a single ethnic category as, in compliance with federal
regulations, the State Education Department did not require the multi-racial classification in the 2007-08 school year.
In total, records for 52 students did not have a useable ethnic identity.
Student Profiles
23
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
STUDENT PROFILES
Table 4 shows the mean 2006-07 and 2007-08 attendance rates of students in the study
sample. It also shows their mean cumulative attendance for the two school years. Mean
attendance in 2007-08 was 0.4 percentage points higher than in 2006-07.
TABLE 4
Attendance Profile of 2007-08 Fourth-Grade Students
ATTENDANCE MEAN
STANDARD
DEVIATION
2006-07 93.7% 5.99%
2007-08 94.1% 5.80%
Cumulative 93.9% 5.51%
Figure 1 shows the percentage of students at each of five levels of attendance,
ranging from severe chronic absence to good attendance. Three percent of
fourth-graders (1,821 students) had severe attendance problems; they attended
fewer than 80 percent of enrolled days—fewer than 148 days in a 185-day
school year. One student attended only 44 percent of school days in grades 3
and 4. The remaining group of chronically absent students (9,714 students)
attended at least 80 percent—but fewer than 90 percent—of school days. The
third group attended fewer than 92.5 percent of school days; these 7,018
students were not chronic absentees but their low attendance may have
compromised their performance. The fourth group (10,759 students) attended
at least 92.5 but fewer than 95 percent of school days. The final and largest
group, 54.3 percent of fourth-graders (34,763 students) attended school regularly—at
least 95 percent of enrolled days.
FIGURE 1
The Percentage of Students at Each Attendance Level
11%
17%
54%
3%
15%
Less+than+80+percent
80.00+to+89.99+percent
90.00+to+92.49+percent
92.50+to+94.99+percent
95.00+percent+or+higher
THE STANDARD DEVIATION (SD)
measures the variation of values
around the mean value. A small
SD indicates that values tend to
be close to the mean, while a
large SD indicates that the values
are spread over a larger range.
24
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
STUDENT PROFILES
Table 5 reports the mean scaled score achieved by study students on the State
assessments in ELA and mathematics in third and fourth grades. On each test, the
average student scored at the proficient level, 650 or above. In both grades, mean
mathematics scores were substantially higher than mean ELA scores. Note that the
percentages of students meeting the proficiency standards increased substantially
statewide in 2008 and 2009, leading the Board of Regents to raise standards on the 2010
tests. The rationale for this decision is described in the Discussion.
4
TABLE 5
Performance Profile of 2007-08 Fourth-Grade Students
ASSESSMENT
MEAN SCALED
SCORE
STANDARD
DEVIATION
NUMBER
TESTED
Grade 3 ELA (2006-07) 658.8 39.7 62,651
Grade 3 Math (2006-07) 685.8 37.4 63,919
Grade 4 ELA (2007-08) 658.2 39.3 63,778
4
Grade 4 Math (2007-08) 681.0 38.5 64,944
STUDENT PROFILES BY ETHNICITY
Ethnic groups varied substantially in the prevalence of disability, limited English
proficiency, and eligibility for free- and reduced-price lunches. Table 6 shows that White
students were less likely than others to be eligible for free- or reduced-price lunches
(FRPL Eligible
), 52.6 percent compared with 85.0 percent of all students. Asian students
were less likely than other students to be classified as disabled; only 7.5 percent of
Asians were classified compared with 18.0 percent of all fourth-graders. Hispanic
students were more likely than others to be limited English proficient; more than a
quarter of Hispanic students were so identified. The Asian group had the second highest
prevalence of limited English proficient students; 16.9 percent.
TABLE 6
The Incidence of Free- and-Reduced-Price-Lunch Eligibility, Disability, and
Limited English Proficiency among 2007-08 Fourth-Graders by Ethnic Group
ETHNIC GROUP NUMBER
FRPL ELIGIBLE
STUDENTS WITH
DISABILITIES
LIMITED ENGLISH
PROFICIENT
NUMBER PERCENT NUMBER PERCENT NUMBER PERCENT
Native American 253 224 88.5% 63 24.9% 22 8.7%
Asian 8,993 7,613 84.7% 675 7.5% 1,517 16.9%
Black 19,575 17,497 89.4% 3,706 18.9% 470 2.4%
Hispanic 25,894 24,216 93.5% 5,480 21.2% 7,187 27.8%
White 9,295 4,886 52.6% 1,601 17.2% 544 5.9%
All Students 64,062 54,472 85.0% 11,528 18.0% 9,742 15.2%
4 The increase in the number of students tested in 2008 compared with 2007 can be accounted for by the revised No Child Left Behind
requirement to test limited English proficient students with fewer years of English instruction.
25
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
STUDENT PROFILES
Table 7 shows differences among ethnic groups in the percentage of fourth-graders who
were continuously enrolled in the same school from fall 2006 until the end of the 2007-
08 school year. White students were most likely—and Black students least likely—to be
continuously enrolled.
TABLE 7
The Number and Percentage of Continuously
Enrolled Fourth-Graders by Ethnic Group
ETHNIC GROUP
CONTINUOUSLY ENROLLED
NUMBER PERCENT
Native American 225 88.9%
Asian 8,067 89.1%
Black 16,470 84.1%
Hispanic 22,472 86.8%
White 8,675 93.3%
All Students 55,955 87.3%
Students of different ethnic backgrounds had different attendance patterns (Table 8).
Asian students had the highest mean and median cumulative attendance and were least
likely to be chronically absent. Half of Asian students attended school at least 98.1
percent of school days. Native American, Black, and Hispanic students were five times as
likely as Asian students and almost twice as likely as White students to be chronically
absent. The Black student group included the largest percentage of students who were
chronically absent. This being said, many students in each ethnic group attended school
regularly. While more than one-fifth of Black and Hispanic students were chronically
absent, almost half (48 percent) of each group attended school regularly; that is, they
were present on 95 percent of enrolled days. While almost one quarter of Native
American students were chronically absent, more than half attended at least 95 percent
of enrolled days.
TABLE 8
Student Attendance in 2006-07 and 2007-08 by Ethnic Group
ETHNIC
GROUP
COUNT
CUMULATIVE ATTENDANCE 2006-07 AND 2007-08
MEAN MEDIAN
NUMBER 95
PERCENT
OR ABOVE
PERCENT
95 PERCENT
OR ABOVE
NUMBER
CHRONICALLY
ABSENT
PERCENT
CHRONICALLY
ABSENT
Native
American
253 93.3% 95.1% 131 51.8% 63 24.9%
Asian 8,993 96.9% 98.1% 7,279 80.9% 381 4.2%
Black 19,575 93.0% 94.8% 9,398 48.0% 4,477 22.9%
Hispanic 25,894 93.3% 94.8% 12,450 48.1% 5,473 21.1%
White 9,295 94.7% 95.8% 5,486 59.0% 1,114 12.0%
All Students 64,062 93.9% 95.4% 34,763 54.3% 11,522 18.0%
26
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
STUDENT PROFILES
Figure 2 graphically displays the differences in average attendance between students who
are White and Asian and students who are Black, Hispanic, and Native American. White
and Asian students were more likely to attend school regularly—at least 95 percent of
school days—and less likely to be chronically absent than other ethnic groups.
FIGURE 2
Comparison of Attendance Patterns of Students Who Are White or
Asian with Students Who Are Black, Hispanic, or Native American
Asian and White Students Black, Hispanic and Native American Students
8%
22%
70%
22%
30%
48%
95% or higher 90–95% Less than 90%
On average, Hispanic, Black, and Native American students achieved similar scores on the
grades 3 and 4 ELA and mathematics assessments (Figure 3).
5
Their mean scores on each
assessment were 25 to 35 points lower than those of Asian and White students. With the
exception of Asian students, the average score of each ethnic group was lower in fourth
than third grade. The improving scores of Asian students may reflect the increasing
English proficiency of students who began kindergarten as English language learners.
FIGURE 3
Grade 3 (2006-07) and Grade 4 (2007-08) School Mean Scaled Scores
Grade&3&ELA
652.3
675.5
652.5
649.8
680.9
658.8
Native
American
Asian Black Hispanic White All
Students
Grade&4&ELA
648.4
676.9
651.5
649.5
678.7
658.2
Native
American
Asian Black Hispanic White All
Students
Grade&3&Math
678.5
707.2
677.8
678.6
702.1
685.8
Native
American
Asian Black Hispanic White All
Students
Grade&4&Math
673.2
708.1
670.4
673.1
699.4
681.0
Native
American
Asian Black Hispanic White All
Students
5 Performance levels and corresponding score ranges are shown on page 13.
27
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
SCHOOL PROFILES
Our examination of schools shows that the enrollment in many schools was not
representative of the demographic, performance, and attendance profiles of the fourth-
grade population. Many schools instead served only segments of this population—
students with similar demographic, attendance, and performance patterns. This
segregation of students reflects the neighborhoods from which schools draw students,
neighborhoods that are frequently segregated by ethnicity and income. We find that
school minority composition is significantly associated with attendance. We show that
schools varied in the qualifications and experience of their teachers and their ability to
retain teachers and that these teacher characteristics are significantly associated with
attendance.
The profile in Table 9 reveals substantial variations among schools in enrollment and
student demographics. These variations are described below.
TABLE 9
Demographic Profile of 705 Study Schools
SCHOOL CHARACTERISTIC MEAN MINIMUM MAXIMUM
STANDARD
DEVIATION
Fourth-Grade Enrollment 91 14 287 42.7
Percent Minority 72.9% 3.7% 100.0% 31.3%
Percent Native American 4.3% 0.0% 11.8% 1.0%
Percent Black 33.1% 0.0% 98.3% 30.8%
Percent Hispanic 39.3% 0.0% 100.0% 27.0%
Percent Asian or White 27.0% 0.0% 96.3% 31.3%
Percent Asian 12.2% 0.0% 89.8% 18.1%
Percent White 14.8% 0.0% 91.8% 22.6%
FRPL Eligible 2007-08 84.2% 3.6% 100.0% 22.5%
Percent SwD 18.9% 0.0% 58.3% 9.4%
Percent ELL 14.0% 0.0% 81.8% 12.5%
Percent Continuously Enrolled 86.8% 0.0% 100.0% 10.8%
UÊ The average school enrolled 91 fourth-graders meeting the study inclusion
criteria. Eighty-six (86) percent of study schools had enrollments between 31 and
150.
UÊ Ethnic composition varied from school to school, with many schools enrolling
students almost exclusively from one or two ethnic groups. In half of study
schools, 90 percent or more of fourth-graders were Native American, Black, or
Hispanic. In 19 schools at least 90 percent were White or Asian.
School Profiles
28
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
SCHOOL PROFILES
UÊ In the average school, 84.2 percent of fourth-graders were eligible for free- or
reduced-price lunches. Some schools, however, served relatively affluent
students. In one such school, only 3.6 percent were eligible.
UÊ Similarly, students with disabilities and limited English proficient students were
not equally distributed among schools. Five schools enrolled no students with
disabilities while in five other schools the majority of fourth-graders were
disabled. Forty-four (44) schools served no limited English proficient students,
while in one school almost 82 percent of students were limited English proficient.
UÊ Some schools had stable enrollments; others served more mobile populations.
We counted the number of fourth-graders who were continuously enrolled from
October 31, 2006 to the end of the 2007-08 school year. At one extreme, a school
enrolled no such students; at the other extreme, almost 87 percent were
continuously enrolled.
Attendance varied substantially among schools. We calculated the school mean
attendance in third and fourth grade of 2007-08 fourth-graders (Table 10). Figure 4
shows the number of schools by the school mean attendance in one-percentage-point
bands. School mean attendance ranged from a low of 88.1 percent to a high of 98.8
percent. In 21 schools, the average student was chronically absent. In more than three
quarters (539) of the 705 study schools, at least 10 percent of fourth-graders were
chronically absent.
In the seven schools with the lowest attendance, students attended only 88 percent of
enrolled school days in 2006-07 and 2007-08. These students missed about 22 days—or a
month—in the 185-day school year. In the four schools with the highest average
attendance, students attended at least 98 percent of enrolled schools days, missing no
more than four days. Students in the highest-attendance schools received 18 additional
days of instruction, on average, compared with those in the lowest-attendance schools.
TABLE 10
Attendance Profile of 705 Study Schools
SCHOOL CHARACTERISTIC MEAN MINIMUM MAXIMUM
STANDARD
DEVIATION
2006-07 Attendance 93.6% 87.9% 98.4% 2.0%
2007-08 Attendance 94.0% 88.0% 99.2% 1.9%
Cumulative Attendance 93.8% 88.1% 98.8% 1.9%
Chronic Absence Rate 18.8% 0.0% 51.8% 11.1%
School Profiles
29
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
SCHOOL PROFILES
7
14
34
83
102
118
140
126
60
17
4
88 89 90 91 92 93 94 95 96 97 98
School+Mean+Attendance+in+Percents
Number+of+Schools
FIGURE 4
Distribution of Schools by Mean Attendance
The range of school performance on the ELA and mathematics assessments was wide
(Table 11). On all tests except grade 3 mathematics, students in the highest-performing
school scored more than 100 point higher, on average, than students in the lowest-
performing. On the grade 3 mathematics test, the difference was 95 points. In the
lowest-performing schools, the average student scored at the bottom of the Level 2
range
6
on the ELA assessments, indicating that the majority of students had significant
deficiencies in reading, writing, and listening comprehension.
TABLE 11
Performance Profile of 705 Study Schools
STATE ASSESSMENT MEAN SCORE MINIMUM MAXIMUM
STANDARD
DEVIATION
Grade 3 ELA 658.5 619.4 738.9 17.7
Grade 3 Math 684.8 645.1 740.5 16.1
Grade 4 ELA 657.8 612.1 729.2 17.7
Grade 4 Math 680.2 636.7 745.4 18.3
TEACHER CHARACTERISTICS
Teachers were not distributed equally among schools by education and experience
(Table 12). In one school, 45 percent of teachers were not appropriately certified for the
class or classes they were teaching. In 89 schools, all teachers were appropriately
certified. In one school, 54 percent of teachers had fewer than three years’ experience. In
14 schools, all teachers had at least three years’ experience. In the school with the
greatest percentage of highly educated teachers, 87 percent had earned at least 30
credits beyond the master’s degree; three schools had no teachers with such credentials.
6 On the grade 3 ELA assessment, Level 2 included scores from 616 to 649. On the grade 4 ELA assessment, Level 2 included scores from 612
through 649.
30
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
SCHOOL PROFILES
Teacher turnover also varied from school to school. In the school with the highest rate, 55
percent of teachers employed in 2007-08 had not been employed in that school in the
previous year. Four schools experienced no turnover of teachers from 2006-07 to 2007-08.
TABLE 12
The Characteristics of Teachers in Study Schools
TEACHER
CHARACTERISTIC
MEAN MINIMUM MAXIMUM
STANDARD
DEVIATION
Percent without
Appropriate Certification
8..3% 0.0% 45.0% 6.8%
Percent Fewer than
3 Years’ Experience
14.1% 0.0% 54.0% 8.4%
Percent Master’s Plus
30 Credit Hours
38.2% 0.0% 87.0% 13.3%
Turnover Rate 14.5% 0.0% 55.0% 7.3%
ATTENDANCE, PERFORMANCE, AND STUDENT DEMOGRAPHICS
School mean attendance in 2006-07 and 2007-08 is associated with a number of the
demographic indicators that distinguish schools (Table 13). We examined the relationship
between attendance and the following variables: the percentages of students eligible for
free- or reduced-price lunches (FRPL eligible), of minority students, of students with
disabilities, of limited English proficient students, of females, and of students who were
continuously enrolled in 2006-07 and 2007-08.
School mean attendance decreased as the percentages of students who were minority or
who came from families in poverty increased. The scatter plot in Figure 5 illustrates the
relationship between minority composition and attendance. The regression line on this
figure shows the best estimate of attendance as minority enrollment increased from 0 to
100 percent. As the percentage of minority students increased, attendance fell. Note,
however, the greater variability in attendance among schools with high rather than low
percentages of minority students. Among schools with no more than 70 percent minority
students, the lowest attendance rate was 92 percent. Among schools where at least 95
percent of fourth-graders were minority, attendance ranged from 88 to 97 percent. Some
schools with large percentages of minority students had high attendance; 65 of the 308
schools in this category had attendance rates of 94 percent or greater; 20 had attendance
rates of 95 percent or higher.
Similarly, there is a negative relationship between disability and attendance: As the
percentage of students with disabilities increased, attendance decreased. There is,
however, no significant correlation at the school level between the percentages of either
limited English proficient students or females and attendance. One explanatory variable
was positively correlated with attendance: Schools with the highest percentages of
students who were continuously enrolled in the 2006-07 and 2007-08 school years tended
to have the highest attendance.
31
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
SCHOOL PROFILES
FIGURE 5
Relationship between School Percent Black, Hispanic,
or Native American and School Mean Attendance
32
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
SCHOOL PROFILES
TABLE 13
Correlations of School Mean Attendance with
School Demographic Characteristics
PERCENT OF STUDENTS
WITH CHARACTERISTIC
PEARSON
CORRELATION
SIGNIFICANCE
(2-TAILED)
FRPL Eligible -.408 .0005
Minority Students -.644 .0005
Students with Disabilities -.365 .0005
Limited English Proficient -.036 .337
Female .019 .609
Continuously Enrolled .263 .0005
Teacher credentials, experience, and education have small, but highly
significant correlations with school attendance (Table 14). Schools with the
highest percentages of teachers who lacked appropriate certification or who
had fewer than three years’ experience tended to have the lowest attendance
rates. Schools with the highest percentage of teachers with 30 credit hours
beyond the master’s degree and schools with the lowest rates of teacher
turnover tended to have higher attendance. The positive relationship between
teacher education and attendance is illustrated in Figure 6. These variations
in teacher qualifications and turnover rate are likely to play a role in
perpetuating differences in attendance and performance.
TABLE 14
Correlations of Teacher Characteristics with School Mean Attendance
PERCENT OF TEACHERS
WITH CHARACTERISTIC
PEARSON
CORRELATION
SIGNIFICANCE
(2-TAILED)
Without Appropriate Certification -.292 .0005
Without Three Years’ Experience -.148 .0005
Master’s Degree + 30 Hours .393 .0005
School Turnover Rate -.244 .0005
THE PEARSON CORRELATION
describes the degree of
relationship between two
variables on a scale from -1 to +1.
It indicates how well a straight
line describes the relationship
between the variables. Positive
correlations, such as that
between the percentage of
continuously enrolled students
and school mean attendance,
indicate that two variables
increase and decrease together.
Negative correlations, such as
that between the percentage of
minority students and school
mean attendance, indicate that
as one variable increases the
other decreases. A correlation of
+1 indicates that for every one-
unit increase in Variable A,
Variable B increases by one unit.
A correlation of -1 indicates that
for every one-unit increase in
Variable A, Variable B decreases
by one unit. A correlation of 0
indicates no relationship between
two variables. The significance
level of a correlation coefficient
indicates the probability that the
correlation coefficient would be
obtained by chance if there were
no relationship between the two
variables.
33
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
SCHOOL PROFILES
FIGURE 6
Relationship between Teacher Education and School Mean Attendance
34
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
ATTENDANCE AND PERFORMANCE AT THE SCHOOL LEVEL
S
chool grade 4 performance is strongly correlated with school
attendance, when the relationships of other variables with both
attendance and performance are not considered. Over 43 percent of
the variability among schools in performance on each assessment can be
accounted for by school mean attendance (Table 15). (Percentage of
variability accounted for is calculated by squaring the correlation
coefficient; for example, the square of the correlation between grade 3 ELA
scores and cumulative attendance, .657, equals .43.) The correlations
between chronic absence rate and performance are equally strong. The
relationship between school mean attendance and performance on the 2007-
08 grade 4 mathematics assessment is illustrated in Figure 7. As school mean
attendance increased from 88 to 99 percent, the predicted grade 4 mean
score increased from 640 to 720. These relationships are not independent of
the simultaneous relationships of poverty, disability, ethnicity and other
variables with attendance and performance. Therefore, only a part of the
80-point difference can be attributed to attendance.
TABLE 15
Correlations of School Mean Attendance and School Percentage
Chronically Absent with Mean Scaled Score on State Assessments
STATE
ASSESSMENT
CUMULATIVE
ATTENDANCE
PERCENT
CHRONICALLY
ABSENT
SIG.
(2-TAILED)
Grade 3 ELA .657 -.647 .0005
Grade 3 Math .674 -.652 .0005
Grade 4 ELA .678 -.662 .0005
Grade 4 Math .691 -.666 .0005
Attendance and Performance at the School Level
35
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
ATTENDANCE AND PERFORMANCE AT THE SCHOOL LEVEL
FIGURE 7
Relationship between School Mean Attendance and Performance
on the Grade 4 Mathematics Assessment
To further examine the relationship between school attendance and performance on the
grades 3 and 4 ELA assessments, we divided schools into quintiles—five groups of equal
size—according to school mean attendance. Table 16 shows the ELA mean and standard
deviations in grades 3 and 4 by quintile. In both years, the quintile with the lowest
attendance (below 92.040 percent) achieved the lowest mean score and the quintile with
the highest attendance (above 95.490 percent) achieved the highest mean score.
Performance improved systematically with increases in attendance: As the attendance
of the quintile increased, so did the mean scaled score on each assessment. In both
grades, schools in the highest quintile achieved unadjusted mean scores about 32 points
higher than those in the lowest quintile.
36
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
ATTENDANCE AND PERFORMANCE AT THE SCHOOL LEVEL
TABLE 16
Grade 4 ELA Mean Scaled Score and Standard Deviation by School-Mean-Attendance Quintile
ATTENDANCE
QUINTILE
NUMBER OF
SCHOOLS
GRADE 3 ELA GRADE 4 ELA
MEAN STD. DEVIATION MEAN STD. DEVIATION
88.05 – 92.039 141 644.4 8.5 643.3 10.1
92.040 – 93.32 141 649.8 9.9 648.4 10.5
93.34 – 94.38 141 655.4 14.0 655.2 13.8
94.39 – 95.490 141 666.5 14.9 666.3 14.2
95.492 – 98.78 141 676.3 17.8 675.6 16.1
All Schools 705 658.5 17.7 657.8 17.7
Table 17 presents the same analysis for school performance on the grades 3 and 4
mathematics assessments. Again, performance at both grade levels increased with each
higher attendance quintile. The unadjusted mean grade 3 score of schools in the highest
attendance quintile was 30 points higher than that of schools in the lowest attendance
quintile. The difference between these quintiles increased in grade 4 to more than 35
points.
TABLE 17
Grade 4 Math Mean Scaled Score and Standard Deviation by School-Mean-Attendance Quintile
ATTENDANCE
QUINTILE
NUMBER OF
SCHOOLS
GRADE 3 MATH GRADE 4 MATH
MEAN STD. DEVIATION MEAN STD. DEVIATION
88.05 – 92.039 141 671.8 9.0 664.9 9.6
92.041 – 93.32 141 676.9 11.1 671.4 11.1
93.34 – 94.38 141 682.3 13.4 676.8 14.9
94.39 – 95.490 141 691.3 12.9 687.8 14.3
95.492 – 98.78 141 701.9 13.2 700.0 16.2
All Schools 705 684.8 16.1 680.2 18.3
To more accurately estimate the relationship between attendance and performance, we
need to control for previous performance, the most powerful predictor of grade 4
performance. We examined the differences among attendance quintiles in grade 4 ELA
scores controlling for school grade 3 performance; that is, we asked whether the
attendance quintiles would still differ if all quintiles had the same grade 3 performance.
This analysis shows that 76 percent of the variance in grade 4 ELA mean scores is
explained by the school’s grade 3 ELA mean score. An additional 4.7 percent is explained
by the attendance quintile to which the school belongs. The model explains more than
86 percent of the variance among schools.
We also analyzed the relationship between attendance quintiles and mean grade 4 math
score, controlling for grade 3 math scores. The analysis shows that 66 percent of the
variance in grade 4 math mean scores is explained by the school’s grade 3 math mean
score. An additional 6.4 percent is explained by the attendance quintile to which the
school belongs. The model explains more than 82 percent of the variation in grade 4
math performance among schools. (The details of these analyses are presented in
Appendix A.)
37
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
ATTENDANCE AND PERFORMANCE AT THE STUDENT LEVEL
W
e have established that attendance and grade 4 performance are related at the
school level, even when grade 3 performance is controlled. These multilevel
analyses will examine this relationship at the student level for ELA and for
mathematics. The goal is to create an equation that predicts student performance on
each grade 4 assessment from a set of student- and school-level variables.
These analyses differ from those reported in previous sections in that they tell us the
relationship of each variable to performance independent of all other specified variables.
For example, we documented that the percentage of minority students and poverty
(measured as eligibility for free- and reduced-price lunches) are each negatively related
to attendance. But minority percentage and poverty are also correlated (r = .572,
p <.0005, 2-tailed). The higher absence rate of minority students is at least partly
attributable to the greater incidence of poverty in minority populations. Controlling for
poverty—that is, statistically holding poverty constant in all schools—reduces the
correlation between minority percentage and attendance from -.647 to -.583. Another
portion of the relationship between minority percentage and attendance is explained by
the relationship between teacher characteristics and minority percentage. To provide an
example, higher percentages of minority students are associated with smaller percentage
of teachers with master’s degree plus 30 credit hours (r = -.548, p <.0005, 2-tailed).
Controlling for both poverty and teacher education further reduces the correlation
between percent minority and attendance to -.499. This exercise is intended to illustrate
the difficulty of disentangling the complex associations among these explanatory
variables. The higher absence rates of Black, Hispanic, and Native American students
can be attributed to a variety of school, environmental, and cultural factors only some of
which are directly measurable. If all of the relevant variables were controlled, the
correlation between percentage minority and attendance might be reduced to zero.
These are multilevel analyses in that they consider the relationships of both student-
and school-level variables to performance. They employ student data, but account for
the effect of the school on student performance. The school provides the context in
which students learn and that context influences both attendance and performance. We
hypothesize, for example, that average attendance in a school has an association with
student performance independent of individual student attendance. Multilevel modeling
accounts for school contextual variables and for the interaction between school and
student characteristics in predicting student performance. It acknowledges that students
with the same grade 3 score may achieve different grade 4 scores based on the quality of
curricula and instruction in the school they attend or on the composition of its student
body. Further, the predicted gain from improving individual attendance or grade 3
performance may vary among students attending the same school because of the
interaction between student and school variables. The quality of a school’s instructional
Attendance and Performance at the Student Level
38
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
ATTENDANCE AND PERFORMANCE AT THE STUDENT LEVEL
and support programs determine its potential ability to improve students’ performance
from year to year. In these analyses, we use school mean attendance, mean grade 3
performance, and teacher quality variables as measures of program quality.
THE ELA MODEL
Based on correlations found previously between demographic variables and
performance, we selected the following student-level variables as predictors of grade 4
ELA performance: grade 3 ELA performance, cumulative attendance, minority status
(Black, Hispanic or Native American), gender, free- or reduced-price lunch eligibility,
disability status, limited English proficient status, and being continuously enrolled. The
relationships of the first three variables with grade 4 ELA performance were assumed to
depend on the school context and, in the final model, were allowed to vary among
schools. Four school-level variables, school mean grade 3 ELA score, school mean
attendance, school percent minority, and teacher turnover rate were selected as
potentially affecting the relationship of these three variables with students’ grade 4
performance. We also assumed that these school-level variables might interact with the
student-level variables to predict performance. The multiplication of the three student-
level variables with the four school-level variables creates 12 potential interactions. In a
series of statistical tests, we eliminated all but five as not improving the explanatory
power of the model. The remaining interactions were 1) student grade 3 ELA score with
school mean grade 3 ELA score, 2) student attendance with school mean attendance, 3)
student minority status with school percent minority, and 4) student minority status
with teacher turnover rate, and 5) student minority status with mean school grade 3 ELA
score. (The theoretical model tested is provided in Appendix B along with details of the
analysis.)
We developed the model to predict grade 4 ELA scaled scores in four steps, each building
on the results of the previous (Table 18). In the first step, we entered no explanatory
variables, only the 705 schools. This procedure tells us that the mean predicted grade 4
ELA scaled score across schools is 658.38. It also tells us that some of the variation
among students results from differences among schools. As documented in the school
profiles, students in a given school tend to be more similar to each other than to
students in other schools. In Step 3, we entered school-level variables to account for
some of this intraschool similarity.
In the second step, we entered the eight student level-variables, but we did not allow the
coefficient of any variable to vary across schools. A coefficient tells you how much the
grade 4 ELA score is expected to change when that explanatory variable increases by one
unit, holding all the other variables constant. We confirmed that each student variable
Attendance and Performance at the Student Level
39
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
ATTENDANCE AND PERFORMANCE AT THE STUDENT LEVEL
has a highly significant association with the grade 4 ELA scaled score. As predicted, four
variables tend to decrease the predicted scaled score—being Black, Hispanic or Native
American; being eligible for free- or reduced-price lunches; being disabled; and being
limited English proficient. The other four variables—higher attendance, a higher grade 3
ELA score, being female, and being continuously enrolled—predict better grade 4 ELA
scores.
In the third step, we added the four school contextual variables. Each student- and
school-level variable is a highly significant predictor of grade 4 ELA scaled score.
In the final model, we allowed the relationship between three student-level variables—
grade 3 ELA score, attendance, and minority status—and performance to vary among
schools. We examined the way that the relationships of these variables with grade 4 ELA
performance were modified by the four school-level variables. For example, we asked
whether the relationship between a student’s grade 3 and grade 4 ELA performance
changed when we controlled for measures of school quality, that is, mean school score,
school attendance, and teacher turnover. This is tantamount to asking the following:
given students with the same grade 3 performance, will the students attending the
higher quality school reliably achieve higher scores in grade 4? The cross-level
interactions we entered allowed us to determine if the change in grade 4 performance
predicted by a student’s attendance, grade 3 performance, or minority status depended
on the school they attended. For example, did the increase in a student’s grade 4
performance predicted by a higher grade 3 score depend on the mean grade 3 score of
the student’s school?
Each of the five cross-level interactions has a significant relationship with grade 4 ELA
performance. The addition of these interactions modified the coefficients of the student-
level and school-level variables slightly, but they remained highly significant. The
estimated coefficient for each variable—describing its association with grade 4 ELA
performance—at each step is provided in Table 18.
40
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
ATTENDANCE AND PERFORMANCE AT THE STUDENT LEVEL
TABLE 18
Model Predicting A Student’s Grade 4 ELA Scaled Score
COEFFICIENT ESTIMATES
STEP 1 STEP 2 STEP 3 STEP 4
Intercept 658.38 658.2 658.240 658.236
STUDENT VARIABLES
Grade 3 ELA score 0.574 0.574 0.580
Cumulative Attendance 0.486 0.486 0.498
Minority Status -4.731 -4.714 -5.204
Female 3.177 3.174 3.159
Free- or Reduced-Price Lunch Eligible -2.798 -2.777 -3.099
Student with Disability -15.300 -15.304 -14.759
Limited English Proficient -7.718 -7.741 -7.320
Continuously Enrolled 1.436 1.435 1.560
SCHOOL VARIABLES
Grade 3 ELA Mean Score 0.786 0.794
Mean Attendance 0.871 0.803
Percentage Minority -0.045 -0.043
Teach er Turn over Rate -0.159 -0.148
CROSS-LEVEL INTERACTIONS
Grade 3 ELA Student Score
by School Mean Score
-0.003
Student Attendance by
School Mean Attendance
0.038
Student Minority Status
by School Percent Minority
-0.054
Student Minority Status
by Teacher Turnover Rate
-0.109
Student Minority Status
by School Grade 3 ELA Mean
-0.143
Note. The standard errors, degrees of freedom, t-values, and probability values for each model can be found in
Appendix B. Each step produced a better fit than the previous step as confirmed by deviance differences. All
coefficients for individual and school variables were significant at the .0005 level. The significance levels for
the cross-level interactions were as follows:
Grade 3 ELA Student Score by School Mean Score: .0005
Student Attendance by School Mean Attendance: .0010
Student Minority Status by School Percent Minority: .0010
Student Minority Status by Teacher Turnover Rate: .0380
Student Minority Status by School Grade 3 ELA Mean: .0005
INTERPRETATION OF ELA RESULTS
The intercept and coefficients in the final model (Table 18) allow us to build an equation
for predicting student performance on the grade 4 ELA assessment based on the
explanatory variables considered in the analysis. The intercept (658.236) is our best
estimate of the value of the student’s grade 4 ELA score when the value of all student
41
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
ATTENDANCE AND PERFORMANCE AT THE STUDENT LEVEL
and school variables is zero. The coefficient of each variable helps to explain why
students score above or below the school mean. It predicts the change in performance as
the value of the explanatory variable increases or decreases, independent of the
association of that variable with all other explanatory variables. The coefficient for
student attendance (0.498), for example, represents the predicted change in grade 4 ELA
score for each one percentage point increase in attendance. The coefficients of the
different variables are not directly comparable because the variables are measured on
different scales.
Predicting Attendance
In the analysis, student variables were not entered as raw scores, but rather were
represented as deviations from the school mean. Each student’s grade 3 scaled score, for
example, was represented as the difference between the student’s score and the school
mean. Similarly, school variables were represented as the difference between the school
mean and the mean of all school means. Because deviation scores were used, the mean
of each student and school explanatory variable is zero. Therefore, the grade 4 score of a
student at the school mean on all variables except student attendance who attends a
school at the mean on all school-level variables can be predicted by the following
equation:
Y
GR4ELA
= 658.236+ 0.498X
ATTEND
Where: Y
GR4ELA
is the predicted Grade 4 ELA scaled score and X
ATTEND
is the difference
between the student’s cumulative attendance and school mean attendance. (If the
school mean is 93.8 percent and the student’s attendance is 95.8 percent, the value of
X
ATTEND
is +2.) We assume that the value of all other explanatory variables is 0.
Figure 8(a) (page 43) illustrates the predicted grade 4 ELA scores based on this equation.
The graph shows the predicted increase in scores as student attendance improves. The
predicted grade 4 ELA score of a student whose attendance is 12 percentage points below
the school mean attendance is 652.3. The predicted score of a student whose attendance
is equal to the school mean is the intercept, 658.236. The predicted score of the student
whose attendance is six points above the school mean is 661.3. We assume that these
students are identical on all explanatory variables except student attendance and attend
a school with mean attendance equal to the mean of school means.
Effect Size
Researchers calculate effect size to provide a standardized estimate of the strength of
the relationship between two variables. Standardization is important because it allows
comparison of effects among studies with different outcome measures and is achieved
by measuring the change in the dependent variable in standard deviations achieved by a
change of one standard deviation in the explanatory variable. The effect size of
attendance on ELA performance is calculated by dividing the product of the coefficient
42
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
ATTENDANCE AND PERFORMANCE AT THE STUDENT LEVEL
and standard deviation of attendance by the standard deviation of the grade 4 ELA scaled
scores. We see that one standard deviation change in individual attendance yields a change
in grade 4 ELA scaled score equal to .07 standard deviations.
Effect Size = (.498 * 5.510) / 39.28 = .070
Variations among Schools
Multilevel analysis estimates the range of intercepts and random coefficients among schools
that we would obtain if we created regression equations for each school. Because they are
normally distributed, we can construct an interval to capture school-to-school variability in
the intercept and coefficients. We find that 95 percent of all intercept values fall in the range
648.411 to 669.182. These intercepts represent the range of schools on mean grade 4
performance for students at the school mean on all explanatory variables. Among schools,
ninety-five percent of the coefficient values for the three random variables fall in the
following ranges:
Grade 3 ELA: 0.382 to 0.777
Cumulative attendance: 0.097 to 0.898
Minority status: -9.903 to -0.505
We see that the relationship of each variable to grade 4 ELA performance is greater in some
schools than others. It appears that some schools were more effective than others in
building on students’ past strengths or overcoming obstacles that reduced achievement in
grade 3.
Similarly, student attendance has a much stronger relationship with grade 4 performance in
some schools than in others. We suggest that attendance is more important in schools where
instruction is effective and fast-paced than in schools where it is less effective.
Finally, some schools were much more effective than others at overcoming the deficits
associated with minority status. The predicted decrease in grade 4 ELA scores for Black,
Hispanic, and Native American students is about 20 times as great in some schools than
others.
The differences in attendance coefficients are illustrated in Figures 8(b) and (c) for schools
with coefficients near the end points of the expected range, that is, coefficients of 0.1 and
0.9. For schools with a coefficient of 0.1, the predicted grade 4 ELA scaled score increases by
only 0.1 for each one point increase in attendance. For schools with a coefficient of 0.9, the
increase is 0.9 for each one point increase in attendance. These graphs are interpreted in the
same way as Figure 8(a) but illustrate the effects of smaller and larger coefficients on the
relationship between student attendance and performance.
43
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
ATTENDANCE AND PERFORMANCE AT THE STUDENT LEVEL
FIGURE 8A.
Predicting Grade 4 ELA Scores from Student Attendance with Coefficient =0.486
FIGURE 8B.
Predicting Grade 4 ELA Scores from Student Attendance with Coefficient =0.1
FIGURE 8C.
Predicting Grade 4 ELA Scores from Student Attendance with Coefficient =0.9
44
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
ATTENDANCE AND PERFORMANCE AT THE STUDENT LEVEL
The predicted effect of each explanatory variable on grade 4 ELA performance is
described below.
Student Level
1. Each one point increase in grade 3 ELA scaled score corresponds, on average,
to a 0.580 increase in grade 4 ELA scaled score.
2. Each one percentage point increase in a student’s grade 3 and 4 cumulative
attendance corresponds, on average, to a 0.498 increase in grade 4 ELA scaled
score.
3. Native American, Black, and Hispanic students, on average, score 5.204 points
lower than White and Asian students.
4. Female students, on average, score 3.159 points higher than male students.
5. Students who are eligible for free- or reduced-price lunches, on average, score
3.099 points lower than students who are not eligible.
6. Students with disabilities, on average, score 14.759 points lower than students
who are not disabled.
7. Limited English proficient students, on average, score 7.320 points lower than
English proficient students.
8. Students who are continuously enrolled score 1.560 points higher, on average,
than students who were not continuously enrolled.
Contextual Variables
9. Each one point increase in a school’s grade 3 ELA mean scaled score
corresponds, on average, to a 0.794-point increase in the school’s grade 4 ELA
mean scaled score.
10. Each one percentage point increase in school mean attendance rate, on
average, corresponds to a 0.803-point increase in the school’s grade 4 ELA
mean scaled score.
11. Each one percentage point increase in the percentage of Native American,
Black, and Hispanic students, on average, corresponds to a 0.043-point
decrease in the school’s grade 4 ELA mean scaled score.
12. Each one percentage point increase in school teacher turnover rate
corresponds, on average, to a 0.148-point decrease in the school’s grade 4 ELA
mean scaled score.
45
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
ATTENDANCE AND PERFORMANCE AT THE STUDENT LEVEL
Cross-Level Interactions
13. Grade 3 ELA Student Score by School Mean Score. For each one point increase
in school mean score, the grade 4 ELA scaled score is modified by the product
of the grade 3 ELA scaled score (school-mean centered) and -0.003. This table
illustrates the effect of the interaction on the predicted score of students at the
mean on all variables except grade 3 ELA student score and school mean score
in schools at the 10
th
and 90
th
percentiles with students at the 10
th
, 50
th
, and 90
th
percentiles on the grade 3 ELA assessment.
GRADE 3 ELA SCHOOL
MEAN SCORE
GRADE 3 ELA SCORE
614 657 708
680 641.7 663.6 689.6
640 626.9 654.4 687.0
Difference 14.8 9.2 2.6
For students below the grade 3 ELA school mean, the negative coefficient on
this interaction increases the advantage of attending a high-performing school
and the disadvantage of attending a low-performing school. The opposite is
true for students above the mean: their advantage of attending a school above
the mean and their disadvantage of attending a school below the mean are
diminished. The table shows the differential effect of attending a school with a
high mean scaled score (680) versus a school with a low mean scaled score
(640) on low-performing students with grade 3 score of 614. The predicted
score of students in the high-performing school is nearly 15 points higher than
that of students in the low-performing school. In contrast, for students who
achieved a score of 708 in grade 3, school performance makes little difference.
14. Student Attendance by School Mean Attendance. For each one point increase
in school mean attendance, the grade 4 ELA score is modified by the product
of the student attendance rate (school-mean centered) and 0.038. This table
illustrates the effect of the interaction on the predicted score of students at the
mean on all variables except student attendance and school mean attendance
in schools at the 10
th
and 90
th
percentiles and with students at the 10
th
, 50
th
,
and 90
th
percentiles of student attendance.
MEAN SCHOOL ATTENDANCE
STUDENT ATTENDANCE
86.8% 95.4% 99.0%
96.3% 654.63 659.7 661.9
91.4% 654.45 658.0 659.4
DIFFERENCE 0.2 1.8 2.4
46
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
ATTENDANCE AND PERFORMANCE AT THE STUDENT LEVEL
For students with very low attendance, school mean attendance matters little.
The predicted grade 4 score of students with very low attendance (86.8
percent) attending a high attendance school (96.3 percent) is almost identical
to that of similar students attending a low attendance school. For very high
attendance students, attending a high-attendance school (96.3 percent)
increases their predicted grade 4 score.
15. Student Minority Status by School Percent Minority. For each one percentage
point increase in Native American, Black, and Hispanic enrollment, the grade
4 ELA score is modified by the product of minority status (school-mean
centered) and -0.054. The difference between the predicted grade 4 ELA scores
of minority students and other students increases as the percentage of
minority students in a school increases. This table illustrates the effect of the
interaction on the predicted score of students at the mean on all variables
except student minority status and school percent minority in schools at the
10
th
and 50
th
percentiles of minority enrollment.
% MINORITY
OTHER
ETHNIC
MINORITY DIFFERENCE
20% 661.0 658.6 2.3
88% 662.9 656.9 6.0
In schools with low-percentages of minority students, the difference between
minority and other students is small. In schools with large percentages of
minority students, the difference is greater.
16. Student Minority Status by Teacher Turnover Rate. For each one point
increase in teacher turnover rate, the grade 4 ELA score is modified by the
product of minority status (school-mean centered) and -0.109. This table
illustrates the effect of the interaction on the predicted score of students at the
mean on all variables except student minority status and teacher turnover rate
in schools at the 10
th
, 50
th,
and 90
th
percentiles.
TURNOVER RATE
OTHER
ETHNIC
MINORITY DIFFERENCE
6% 662.6 658.4 4.2
14% 662.0 657.0 5.1
24% 661.3 655.2 6.1
As the school turnover rate increases, the difference between the predicted
grade 4 ELA score of Native American, Black, and Hispanic students compared
with that of other students increases. Attending a school with low teacher
turnover increases the predicted score of minority students (658.4 compared
47
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
ATTENDANCE AND PERFORMANCE AT THE STUDENT LEVEL
with 655.2). For other students, turnover rate makes a smaller difference in
predicted score. The difference between minority and other students increases
with turnover rate.
17. Student Minority Status by School Grade 3 ELA Mean. For each one point
increase in grade 3 ELA mean scaled score, the grade 4 ELA score is modified
by the product of minority status (school-mean centered) and -0.143. This table
illustrates the effect of the interaction on the predicted score of students at the
mean on all variables except student minority status and school grade 3 ELA
mean in schools at the 10
th
, 50
th,
and 90
th
percentiles.
GRADE 3 ELA SCHOOL
MEAN SCORE
OTHER
ETHNIC
MINORITY DIFFERENCE
640 645.1 642.6 2.5
657 660.6 655.6 5.0
680 681.8 673.5 8.4
For Native American, Black, and Hispanic students, the interaction decreases
the predicted effect of school grade 3 ELA mean; for other students the effect is
increased. The higher the overall school grade 3 ELA performance, the better
minority students perform. However, the gap between minority students and
others widens as school grade 3 ELA mean scaled score increases.
MATHEMATICS MODEL
We developed this model (Table 19) using the same procedures and variables used in the
development of the ELA model but found some differences from the ELA model. (The
details of this analysis are presented in Appendix C.) While seven of the student-level
variables were significant predictors of grade 4 math performance, the eighth,
continuous enrollment, fell short of reaching significance at the .05 level. We selected
the percentage of teachers with 30 credit hours beyond the master’s degree, rather than
teacher turnover rate, as the fourth school-level variable because it had the highest
correlation with grade 4 math performance. This variable was not a significant predictor
of grade 4 performance, when all other explanatory variables were accounted for. Only
three interactions significantly improved the explanatory power of the model: 1) student
grade 3 mathematics score with school mean grade 3 math score, 2) student attendance
with school mean attendance, and 3) student minority status with mean school grade 3
math score.
48
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
ATTENDANCE AND PERFORMANCE AT THE STUDENT LEVEL
TABLE 19
Model Predicting A Student’s Grade 4 Mathematics Scaled Score
COEFFICIENT ESTIMATES
STEP 1 STEP 2 STEP 3 STEP 4
Intercept 680.243 680.194 680.100 680.124
STUDENT VARIABLES
Grade 3 Math Score 0.639 0.639 0.640
Cumulative Attendance 0.569 0.569 0.613
Minority Status -7.561 -7.561 -6.972
Female -2.700 -2.698 -2.674
Free- or Reduced-Price Lunch Eligible -2.095 -2.094 -2.011
Student with Disability -11.448 -11.449 -11.146
Limited English Proficient -4.256 -4.256 -4.340
Continuously Enrolled 0.205 0.206 0.352
SCHOOL VARIABLES
Grade 3 Math Mean Score 0.779 0.785
Mean Attendance 1.008 0.991
Percentage Minority -0.123 -0.119
Teach ers wi th Mas ter’s + 30 C red its -0.043 -0.041
CROSS-LEVEL INTERACTIONS
Grade 3 Math Student Score by
School Mean Score
-0.002
Student Attendance by School Mean Attendance 0.076
Student Minority Status by
School Grade 3 Math Mean
-0.088
Note. The standard errors, degrees of freedom, t-values, and probability values for each model can be found
in Appendix C. Each step produced a better fit than the previous step as confirmed by deviance differences.
All coefficients for individual variables except Continuously Enrolled were significant at the .0005 level. The
Continuously Enrolled variable was not statistically significant in any step. All coefficients for school variables
except Teachers with Master’s Plus 30 credits were significant at the .0005 level. The significance level of the
latter variable was .069 in Step 3 and .083 in Step 4. The significance levels for the cross-level interactions were
as follows:
Grade 3 Math Student Score by School Mean Score: .0005
Student Attendance by School Mean Attendance: .0005
Student Minority Status by School Grade 3 Math Mean: .0010
49
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
ATTENDANCE AND PERFORMANCE AT THE STUDENT LEVEL
INTERPRETATION OF MATHEMATICS RESULTS
The coefficient for student attendance was somewhat higher in the mathematics model
than in the ELA model: all else being equal, a one percentage point change in student
attendance predicts a greater change in the grade 4 mathematics score than in the grade
4 ELA score. The decrement in predicted grade 4 score for students with disabilities is
smaller for mathematics than for ELA, 11.146 compared with 14.749. In the ELA model,
being female is positively associated with grade 4 performance; in the math model, the
association is negative, -2.674 compared with 3.159. The coefficient associated with
being in a school with a large minority enrollment is more than twice as large for math
as for ELA, -0.119 compared with -0.043. The coefficient for the interaction of student
minority status and school percentage minority, however, is not significant in the
mathematics model. This finding indicates that the association of student minority
status and mathematics performance is independent of the percentage of minority
students in the school.
Effect Size
The effect size of attendance on mathematics performance is calculated by dividing the
product of the coefficient and standard deviation of attendance by the standard
deviation of the grade 4 mathematics scaled scores. We see that one standard deviation
change in individual attendance yields a predicted change in grade 4 mathematics
scaled score equal to .088 standard deviations.
Effect Size = (.613 * 5.510) / 38.483 = .088
Variations Among Schools
The analysis estimated the range of intercepts and random coefficients among schools
that would be obtained by doing school-by-school analyses. We found that 95 percent of
all intercept values fall in the range 666.908 to 693.340. These intercepts represent the
range of predicted grade 4 math school mean scores for students at the school mean on
all explanatory variables. Among schools, ninety-five percent of the coefficient values
fall in the following ranges:
Grade 3 mathematics: 0.461 to 0.820
Cumulative attendance: 0.341 to 0.885
Minority status: -15.903 to 1.171
The range of coefficients for cumulative attendance is smaller in the math than the ELA
model. The range of coefficients for minority status is greater and includes some positive
values, indicating that the school-level explanatory variables that we employed were not
sufficient to account for the association between minority status and grade 4 math
50
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
ATTENDANCE AND PERFORMANCE AT THE STUDENT LEVEL
performance. Additional explanatory variables are necessary to describe this
relationship.
The associations of each explanatory variable and interaction with grade 4 mathematics
are described below.
Student Level
1. Each one point increase in grade 3 math scaled score corresponds, on average,
to a 0.640 increase in grade 4 math scaled score.
2. Each one percentage point increase in a student’s grade 3 and 4 cumulative
attendance corresponds, on average, to a 0.613 increase in grade 4 math scaled
score.
3. Native American, Black, and Hispanic students, on average, score 6.972 points
lower than White and Asian students.
4. Female students, on average, score 2.674 points lower than male students.
5. Students who are eligible for free- or reduced-price lunches, on average, score
2.011 points lower than students who are not eligible.
6. Students with disabilities, on average, score 11.146 points lower than students
who are not disabled.
7. Limited English proficient students, on average, score 4.340 points lower than
English proficient students.
8. Students who are continuously enrolled score 0.352 points higher, on average,
than students who were not continuously enrolled. This coefficient is not
statistically significant.
Contextual Variables
9. Each one point increase in a school’s grade 3 math mean scaled score
corresponds, on average, to a 0.785-point increase in the school’s grade 4 math
mean scaled score.
10. Each one percentage point increase in school mean attendance, on average,
corresponds to a 0.991-point increase in the school’s grade 4 math mean scaled
score.
11. Each one percentage point increase in the percentage of Native American,
Black, and Hispanic students, on average, corresponds to a 0.119-point
decrease in the school’s grade 4 math mean scaled score.
51
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
ATTENDANCE AND PERFORMANCE AT THE STUDENT LEVEL
12. Each one percentage point increase in the percentage of teachers with 30
credits beyond the master’s degree corresponds, on average, to a 0.041-point
decrease in the school’s grade 4 math mean scaled score. This coefficient is not
statistically significant.
Cross-Level Interactions
13. Grade 3 Math Student Score by School Mean Score. For each one point
increase in school mean score, the grade 4 math scaled score is modified by
the product of the grade 3 math scaled score (school-mean centered) and
-0.002. This table illustrates the effect of the interaction on the predicted score
of students at the mean on all variables except grade 3 math student score and
school mean score in schools at the 10th and 90th percentiles with students at
the 10th, 50th, and 90th percentiles on the grade 3 math assessment.
GRADE 3 MATH
SCHOOL MEAN SCORE
GRADE 3 MATH SCORE
642 685 739
707 663.3 688.4 719.9
667 653.1 681.9 718.0
Difference 10.2 6.6 1.9
For students below the grade 3 math school mean, the negative coefficient on
this interaction increases the advantage of attending a high-performing school
and the disadvantage of attending a low-performing school. The opposite is
true for students above the mean: their advantage of attending a school above
the mean is diminished and their disadvantage of attending a school below the
mean is diminished. (Remember that the product of two negative numbers is
positive.) The table shows the differential effect of attending a school with a
high mean scaled score (707) versus a school with a low mean scaled score
(667) on low-performing students with grade 3 scores of 642. The predicted
score of the student in the high-performing school is over 10 points higher
than that of the student in the low-performing school. In contrast, for students
who achieved a score of 739 in grade 3, school performance makes little
difference.
52
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
ATTENDANCE AND PERFORMANCE AT THE STUDENT LEVEL
14. Student Attendance by School Mean Attendance. For each one point increase
in school attendance rate, the grade 4 math score is modified by the product of
the student attendance rate (school-mean centered) and 0.076. This table
illustrates the effect of the interaction on the predicted score of students at the
mean on all variables except student attendance and school mean attendance
in schools at the 10th and 90th percentiles and with students at the 10th, 50th,
and 90th percentiles.
MEAN SCHOOL
ATTENDANCE
STUDENT ATTENDANCE
86.8% 95.4% 99.0%
96.3% 675.0 681.9 684.8
91.4% 675.8 679.5 681.1
Difference -0.8 2.4 3.7
For students with very low attendance, school mean attendance matters little.
The predicted grade 4 score of students with very low attendance (86.8 percent)
attending a high attendance school (96.3 percent) is almost identical to that of
similar students attending a low attendance school (91.4 percent). For very high
attendance students, attending a high-attendance school increases their
predicted grade 4 score.
15. Student Minority Status by School Grade 3 Math Mean. For each one point
increase in grade 3 math mean scaled score, the grade 4 math score is
modified by the product of minority status (school-mean centered) and -0.088.
This table illustrates the effect of the interaction on the predicted score of
students at the mean on all variables except student minority status and
school grade 3 math mean in schools at the 10th, 50th, and 90th percentiles.
GRADE 3 MATH
SCHOOL MEAN SCORE
OTHER
ETHNIC
MINORITY DIFFERENCE
667 669.2 663.8 5.3
685 689.2 681.8 7.4
707 703.9 695.0 8.9
The higher the overall school grade 3 math performance, the better minority
students perform. The gap between minority students and others, however,
widens as school grade 3 math mean scaled score increases.
53
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
DISCUSSION
T
his research confirms that student and school attendance are related to
performance on the grade 4 State assessments in ELA and mathematics; that is,
as individual student and school mean attendance increase so does performance.
The relationship between an individual student’s attendance and grade 4 performance
depends on the school the student attends. Comparing two students with the same
attendance, the student attending the school with higher mean attendance will have
better grade 4 performance, all else being equal. These associations are independent of
other explanatory variables measured in these analyses; that is, they do not result from
the simultaneous association of any measured variable with both attendance and
performance.
The finding that student attendance is related to performance is consistent with findings
obtained by Gottfried (2009, 2010, 2011), using data for students in the Philadelphia
School District. In fact, this study’s effect sizes—standardized measures of the
relationship of attendance with performance—are very similar to those in his sibling
study, described in the Introduction. The sibling study also strongly suggests that the
documented association between attendance and grade 4 performance is direct and not
the result of simultaneous relationships of unmeasured family characteristics—such as
parent education and family involvement in school activities—with attendance and
performance. The convergence of these results with Gottfried’s increases our confidence
that there is a direct causal link between attendance and performance.
Gottfried’s research strongly suggests that the relationship between attendance and
performance is not confined to grade 4 but exists in grades 3 through 8. He originally
studied five student cohorts for a six-year period and found that attendance was linked
to performance in the elementary and middle grades. His research also suggests that
attendance may become more important as students progress through the elementary
and middle grades.
Our analyses also suggest that the improvement in grade 4 performance gained through
increased attendance would be further accelerated in grade 5 by virtue of both higher
grade 4 performance and sustained or improved attendance in grade 5. Students who
master more of the learning standards in a grade have a better foundation for
achievement in the next grade. As such, we may expect that the achievement gap
between persistently low- and high-attending students will widen over time.
We acknowledge that a substantial increase in attendance is required to obtain
moderate increases in performance. We also know that rates of chronic absence are
alarmingly high: 18 percent of fourth-graders were chronically absent; 3 percent
attended fewer than 80 percent of school days. In three-quarters of schools, at least 10
percent of students were chronically absent. CFE believes that intensive and
Discussion
54
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
DISCUSSION
comprehensive school reform measures can substantially reduce chronic absence.
Performance improvements will then accrue directly from increased attendance and
from more effective curricula and instruction.
HOW ATTENDANCE AFFECTS PERFORMANCE
While CFE’s results are based on correlations and cannot prove a causal relationship
between attendance and performance, the potential causal mechanism is readily
evident: students who attend school regularly receive more instruction. A study of
charter schools by Caroline Hoxby,
!Sonali Murarka, and Jenny Kang, discussed in the
Introduction,
found that one characteristic of successful charter schools is increased
instructional time through a longer school day or year. Improving attendance is an
efficient way of increasing effective instructional time in an era of limited budgets.
While the preponderance of evidence strongly supports an association, probably causal,
between attendance and performance, it is likely that this association varies from
student to student. Attendance measures time in school. Time in school however does
not equate perfectly to hours of instruction in the skills and knowledge assessed on
standardized tests of ELA and mathematics. The amount, quality, and pace of such
instruction vary from day to day and school to school. On some days no such instruction
may be given. Further, a student’s ability to focus on and benefit from instruction varies
from day to day and student to student. Therefore, the value of an instructional day as
measured by assessments varies across days, students, and schools. Similarly, the
performance decrement caused by missing a day will vary depending on several factors:
what relevant instruction did the student miss? Were missed lessons made up? What did
the student do when out of school? A student who spends hours at home reading
challenging text will experience a smaller decrement, if any, than a student who spends
that time watching cartoons. Gottfried’s finding that higher proportions of unexcused
absences are associated with lower performance supports the hypothesis that the
decrement associated with missing school varies from student to student. We suggest
two mechanisms that contribute to the difference between excused and unexcused
absences: teachers are more likely to assist students with excused absences and students
are more likely to make up work when their absence is excused rather than unexcused.
We examined two indicators of school quality—school mean attendance and mean
grade 3 performance—that are significant predictors of both grade 4 ELA and
mathematics performance. We found that the gain in individual performance predicted
by increased attendance is greater in schools with higher mean attendance. Similarly,
the predicted scores of students with the same grade 3 scores depend on the average
performance of the school they attend. Those attending schools with higher mean grade
3 scores are predicted to have higher grade 4 scores. The lower the student’s grade 3
Discussion
55
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
DISCUSSION
score, the more grade 4 performance is influenced by the school they attend. For
students with very high performance in grade 3, the school they attend makes little
difference in their predicted grade 4 scores. We can predict from these findings that
students with low grade 3 performance and attendance who improve their grade 4
attendance will make greater performance gains if they attend a high performing, high
attendance school. This makes sense because the benefits of attendance can only be
fully realized on days when the student is engaged in high-quality instruction targeted
to the ELA and mathematics learning standards. As such, we suggest that students will
benefit fully from improved attendance only if schools provide high quality instruction
and if they and the community ensure that obstacles to student engagement in learning,
such as ill health, lack of family involvement, and family and personal difficulties, are
minimized.
WHY ATTENDANCE IMPROVEMENTS ARE CRITICAL
The strengthening of the State standards in 2010 substantially decreased the percentage
of fourth-graders scoring at the proficient level and suggests that the 2008 results used
in this study overestimate the percentage of fourth-graders who were proficient and on
track toward meeting the grade 8 learning standards. In 2008, citywide, 61.4 percent of
New York City fourth-graders scored at the proficient level or higher in ELA; 79.6 percent
did so in mathematics. Under the strengthened standards, in 2010, the percentage
scoring at the proficient level or higher in ELA decreased by 17 percentage points; in
math, by 22 points. Further, under the strengthened standards, the achievement gap
among ethnic groups increased.
Based on past trends, we can expect, even without more rigorous standards, that when
these fourth-graders reach eighth grade, fewer will meet the State standards in ELA and
mathematics. Comparison of grade 4 and 8 results on 2008 State assessments shows that
smaller percentages of students met the learning standards in grade 8; 61.3 percent met
the ELA standards in grade 4, compared with 43.1 percent in grade 8. In mathematics,
79.6 percent met the standards in grade 4, compared with 59.8 percent in grade 8.
The strengthening of learning standards by the Board of Regents makes improving grade
8 performance both more challenging and more important. CFE’s study of high school
graduation and Regents diploma rates found that schools with the largest percentages of
entering ninth-graders who, in grade 8, failed to reach the State learning standards and
attended irregularly had the lowest Regents diploma rates. The data indicate that
attendance drops during the high school years, particularly in schools with low-
graduation rates, and better school attendance is associated with higher Regents
diploma rates. Several studies provide evidence that poor attendance as early as grade 6
indicates a high risk of dropping out before graduation (Balfanz, Herzog & McIver, 2007;
Ou & Reynolds, 2008). Ruth Neild and colleagues (2007) documented that students in
Philadelphia with attendance below 80 percent in sixth grade had a three in four chance
of dropping out of high school. The consequences of dropping out on later income,
56
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
DISCUSSION
dependence on welfare, and incarceration are widely documented. Each of these
consequences has serious implications for the larger community. CFE believes that
making the necessary changes to improve and sustain attendance from the elementary
to the high school years is an important step in increasing graduation rates and
preparing students for the future.
MINORITY STUDENTS
In an era of rising standards, the performance of minority students is of particular
concern. We found that, all else being equal, the predicted scores of Native American,
Black, and Hispanic students were about five points lower in ELA—and seven points
lower in mathematics—than those of White and Asian students. In both subjects, these
predicted decrements are independent of additional decrements applied to some
minority students for poor grade 3 performance, inadequate attendance, low family
income, disability, and limited English proficiency. In ELA, they are also independent of
decrements for changing schools and attending schools with high teacher turnover
rates.
The unexplained decrements for minority students merit further research to identify
remedial steps to close the achievement gap. Other potential causes of the gap lie in the
school, the student, the home, and the community. They include school programs not
designed to meet the needs of minority students, ineffective teachers, poor
communication with parents, teacher attendance, inadequate student effort and
behavior problems, health or nutritional status, maternal education, unsafe
neighborhoods, and persistent poverty—which subsidized lunch eligibility cannot
distinguish from transient poverty.
While these performance decrements for minority students were independent of
decrements in predicted scores attributable to poor attendance, 22 percent of minority
students were chronically absent, further diminishing their ability to achieve at their
full potential. We believe that increasing the attendance rates of minority students and
eliminating chronic absence will reduce the achievement gaps documented among
ethnic groups. For minority students to fully benefit from improved attendance, schools
must ensure that they are engaged in learning through high quality instruction and
curricula.
CITY EFFORTS TO IMPROVE ATTENDANCE
DOE in cooperation with school staff are implementing many programs to improve
attendance, including one of the most sophisticated attendance tracking systems in the
country. Principals have access to a daily report that can instantly show them which
students are chronically absent or in danger of becoming so. Top DOE officials point out,
accurately, that overall attendance has been improving and that citywide, fourth-grade
attendance improved from 93 to 94 percent between 1999-00 and 2007-08.
57
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
DISCUSSION
Recognizing the value of attendance to school improvement, in June 2010 Mayor Michael
Bloomberg’s office created the Interagency Task Force on Truancy and Chronic
Absenteeism (Bloomberg, 2011). In September 2010, the task force launched a pilot
program aimed at reducing chronic absenteeism and truancy in 25 schools across the
City. The program educates parents about the value of good attendance, offers
incentives for children to come to school and provides mentors for students who are on
track to miss more than 10 percent of the school year. The efforts have already borne
fruit: In the first half of the school year, fully 22 of the 25 schools reduced their absentee
rates. The 10 elementary schools saw the best results, with a collective 24 percent
decline in the percentage of students who are chronically absent. The seven high schools
showed little change. In February 2011, the Mayor announced the expansion of this pilot
to include a multimedia campaign in which the media and celebrities stress the
importance of attendance to students and parents. Students with excessive absences
receive inspirational wake-up messages and congratulatory messages for improved
attendance. Parents and students receive warning messages if attendance does not
improve. Media participants air public service announcements stressing the importance
of attendance to educational success.
CONCLUSION
In recent years, New York City has made significant strides in hiring qualified staff,
providing professional development, placing effective curricula in schools, and
providing supplemental programs to increase time on task. These changes have
translated into improved performance. Still, too many New York City students are below
the State learning standards and too many leave school without earning a high school
diploma. To realize the full benefit of these ongoing improvement efforts, schools must
place greater focus on improving attendance.
Despite the City’s ongoing effort to improve attendance, 18 percent of students in our
study group were absent for more than 10 percent of school days in third and fourth
grade and in 539 schools at least 10 percent of fourth-graders were chronically absent.
These findings suggest that many schools must re-invigorate their efforts to improve
attendance. It is too soon to evaluate the effectiveness of Mayor Bloomberg’s latest
initiatives. We applaud his attention to this important issue.
We believe that substantially reducing the number of students who are chronically
absent and increasing the attendance of all students would raise achievement, reduce
the percentage of students retained in grade, and increase graduation rates. To benefit
fully from increased attendance, however, students must be fully engaged in high-
quality instruction.
58
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
RECOMMENDATIONS
DEVELOP STRONG ATTENDANCE PROGRAMS
T
he results of this study are intended to inform the discussion about
the role of attendance improvement in New York City’s
comprehensive school reform effort. New York City must begin by
setting clear standards and high expectations for attendance. We
recommend that DOE identify schools that have been most successful in
improving attendance and ensure that their best practices are shared
among schools.
At a minimum DOE must ensure that schools do the following:
UÊ Value high attendance in the school and classroom.
UÊ Use the attendance tracking system to promptly identify students
who are missing too many days of school.
UÊ Examine patterns of attendance in neighborhoods, ethnic groups,
grades or classrooms to identify and address systemic causes of
absence.
UÊ Develop and implement policies for reaching out to identified
students and their families to determine the reasons for absence
and to mitigate those reasons where possible. Especially in the early
grades, absentee students often are not willfully skipping school but
rather miss days because of health and safety concerns, frequent
moves or unreliable transportation. Where poverty or homelessness
is identified as a cause of excessive absence, we advocate efforts to
mitigate their consequences through health and nutrition programs,
parent education, preschool and prekindergarten, homework
centers, extended-day and other programs specifically designed to
meet the needs of such children.
UÊ Put policies in place to minimize the effect of missing school by
ensuring that students make up missed work and are kept on track
toward acquiring the skills and knowledge expected for their grades.
Most importantly, schools must create a climate in which all staff, students
and families understand the importance of attendance and work to
minimize absences.
Recommendations
59
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
RECOMMENDATIONS
ACCOUNTABILITY FOR ATTENDANCE
It is important that schools be held accountable for improving attendance. To that end,
attendance rates and chronic absence rates should be publicly available and, reported
for all federal accountability groups, including racial/ethnic groups. We recommend
that federal, State, and City accountability systems be revised to increase the value of
attendance in assessing school progress. New York is one of five states that does not
include attendance data in its longitudinal student database. The State should work with
school districts to standardize and collect student-level attendance data, and to develop
standard definitions of chronic absence and truancy, so that comparable measures are
used statewide.
COMPREHENSIVE SCHOOL REFORM
Attendance is only one of many factors linked to performance. Programs to improve
attendance in the elementary and middle school grades should be part of comprehensive
school-wide reforms designed to ensure that all students enter ninth grade prepared for
coursework leading to a Regents diploma. The importance of challenging curricula and
effective teaching cannot be overestimated. To be fully effective, attendance
improvement programs should be part of comprehensive programs that include
initiatives in the following areas.
This research shows that the most important predictor of student performance in grade
4 is grade 3 performance. Therefore, we believe that to improve student achievement,
schools must improve both attendance and the quality of instruction. Chang and
Romero (p. 17) suggest that “…chronic absence might be, at least partially, remedied by
high-quality educational programs. …[Data from one locality suggest] that when school
quality was high, children were less likely to be chronically absent in the early grades
despite living in a high risk neighborhood in which many of their peers are missing
extended periods of school.” To ensure academic success, children from such families
need high-quality schools with effective academic programs and experienced, highly
qualified teachers appropriately trained to meet their needs.
We also recommend that student achievement be monitored periodically—rather than
annually—using standardized tests to identify students who are not on track to master
the skills and knowledge expected for their grade. Students may fall behind because of
absence, failure to engage in learning and complete assignments, or from inadequate
instruction. Whatever the cause, early identification is most likely to lead to successful
intervention.
60
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
REFERENCES
Balfanz, R., Herzog, L. & MacIver, D. J. (2007). Preventing Student
Disengagement and Keeping Students on the Graduation Path in Urban
Middle-Grades Schools: Early Identification and Effective Interventions.
Educational Psychologist, 42, 223–235.
Balfanz, R. & Byrnes, V. (2006). Closing the Mathematics Achievement Gap in
High-Poverty Middle Schools: Enablers and Constraints. Journal of Education
for Students Placed at Risk, 11, 143-159.
Bloomberg, M. (2011). Mayor Bloomberg Launches Wake Up! NYC Campaign To
Reduce Chronic Absenteeism and Truancy in Schools. (Downloaded from
http://www.mikebloomberg.com/index.cfm?objectid=10A681CB-C29C-7CA2-
F5782EA5FC7DE65B on March 28, 2011.)
Chang, H. & Romero, M. (2008). Present, Engaged, and Accounted for – The
Critical Importance of Addressing Chronic Absence in the Early Grades. National
Center for Children in Poverty, Mailman School of Public Health, Columbia
University. (Accessed in November 2008 on the NCCP Website: www.nccp.
org.)
Campaign for Fiscal Equity. (2010). Diploma Dilemma: Rising Standards, the
Regents Diploma, and Schools that Beat the Odds. New York City, NY:
Campaign for Fiscal Equity.
Gottfried, M. A. (2009). Excused Versus Unexcused: How Student Absence in
Elementary Schools Affect Academic Achievement. Educational Evaluation
and Policy Analysis, 31, 392-415.
Gottfried, M. A. (2010). Evaluating the Relationship between Student
Attendance and Achievement in Urban Elementary and Middle Schools: An
Instrumental Variables Approach. American Educational Research Journal, 47,
343-465.
Gottfried, M. A. (2011). The Detrimental Effects of Missing School: Evidence
from Urban Siblings. American Journal of Education, 117, 147-182.
Hoxby, C. M., Murarka, S., & Kang, J. (2009). How New York City’s Charter
Schools Affect Achievement, August 2009 Report. Cambridge, MA: New York
City Charter Schools Evaluation Project.
Nauer, K., White, A., & Yerneni, R. (2008). Strengthening Schools by
Strengthening Families: Community Strategies to Reverse Chronic Absenteeism
in the Early Grades and Improve Supports for Children and Families. Center for
References
61
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
REFERENCES
New York City Affairs, Milano the New School for Management and Urban Policy.
(Accessed in November 2008 on the Center’s Website: www.centernyc.org.)
Neild, R. C., Balfanz, R. & Herzog, L. (2007). An Early Warning System. Educational
Leadership, 65, 28-33.
New York State Education Department. (2011). “College and Career Ready” Graduation
Rate Data. (Downloaded from http://graphics8.nytimes.com/packages/pdf/nyregion/201
10208regentsGradRates.PDF, February 2011.)
Ou, S. & Reynolds, A. J. (2008). Predictors of Educational Attainment in the Chicago
Longitudinal Study. School Psychology Quarterly, 23, 199-229.
Roby, D. E. (2004). Research on School Attendance and Student Achievement: A Study of
Ohio Schools. Educational Research Quarterly, 28, 3-16. (Downloaded from http://www.
eric.ed.gov/PDFS/EJ714746.pdf on February 15, 2011.)
62
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
APPENDIX A: TEST OF DIFFERENCES IN PERFORMANCE AMONG ATTENDANCE QUINTILES
GRADE 4 ELA SCHOOL MEAN SCORE WITH
CO-VARIATE SCHOOL MEAN GRADE 3 ELA SCORE
AND FACTOR SCHOOL ATTENDANCE QUINTILE
TESTS OF BETWEEN-SUBJECTS EFFECTS
Dependent Variable: Grade 4 ELA Mean Score
Source
Type III Sum of
Squares
df Mean Square F Sig.
Partial Eta
Squared
Corrected Model 189,922.565
a
5 37,984.513 903.378 .000 .866
Intercept 2796.574 1 2796.574 66.510 .000 .087
Grade 3 ELA
Mean Score
91,546.005 1 91,546.005 2,177.219 .000 .757
Attend Quintile 1,719.280 4 429.820 10.222 .000 .055
Error 29,391.000 699 42.047
Total 305,232,693.796 705
Corrected Total 219,313.565 704
a
R Squared = .866 (Adjusted R Squared = .865)
GRADE 4 MATHEMATICS SCHOOL MEAN SCORE WITH
CO-VARIATE SCHOOL MEAN GRADE 3 MATH SCORE
AND FACTOR SCHOOL ATTENDANCE QUINTILE
TESTS OF BETWEEN-SUBJECTS EFFECTS
Dependent Variable: Grade 4 Math Mean Score
Source
Type III Sum
of Squares
df Mean Square F Sig.
Partial Eta
Squared
Corrected Model 193,044.319
a
5 38,608.864 632.156 .000 .819
Intercept 711.549 1 711.549 11.650 .001 .016
Grade 3 Math
Mean Score
84,321.398 1 84,321.398 1380.624 .000 .664
Attend Quintile 2931.308 4 732.827 11.999 .000 .064
Error 42,691.328 699 61.075
Total 326,413,287.477 705
Corrected Total 235,735.647 704
a
R Squared = .819 (Adjusted R Squared = .818)
Appendix A
Tests of Differences in Performance Among Attendance
Quintiles Controlling for Grade 3 Performance
63
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
APPENDIX B: ELA MULTILEVEL MODEL
The model for predicting grade 4 ELA performance includes eight student variables.
Three student variables are allowed random coefficients; that is, they are allowed to vary
among schools: grade 3 ELA scaled score (GR3ELA1), attendance in grades 3 and 4
(ATTEND1), and identification as Black, Hispanic, or Native American (MINOR1).
Historically, Black, Hispanic, and Native American students have demonstrated very
similar performance in New York City. Therefore, for parsimony, students in these three
ethnic groups are coded 1. In addition, the model includes five student-level variables
with fixed coefficients. All have documented associations with ELA performance. These
variables are gender (X
GENDER1
), eligibility for free and reduced-price lunch (X
FRPL1
),
identification as a student with disability (X
SwD1
), identification as limited English
proficient (X
LEP1
), and being continuously enrolled in a school from October 31, 2006 until
the end of the 2007-08 school year (X
CONT_ENROLL1
). Four contextual (school-level) variables
are considered: school mean grade 3 ELA scale score for study cohort (X
3ELA2
); school
mean attendance in grades 3 and 4 for the study cohort (X
ATTEND2
); school percentage
Black, Hispanic or Native American (X
MINOR2
); and teacher turnover rate (X
TURNOVER2
).
Teacher turnover rate correlates significantly with the percentage of teachers having
fewer than three years experience (r = .559, p<.0005), having significant graduate credits
beyond the master’s degree (r = -.383, p<.0005), and without appropriate certification for
teaching assignment (r = .292, p<.0005). The study includes 705 schools with a total
cohort enrollment of 64,062. Because of missing data, not all students were included in
the multivariate analyses.
THE THEORETICAL MODEL
ELA Level 1 Model
Y
4ELA
= ࢠ
01
+
ࢠ
1j
GR3ELA1
+ ࢠ
2j
ATTEND1 + ࢠ
3j
MINOR1 + a
40
X
GENDER1
+ a
50
X
FRPL1
+ a
60
X
SwD1
+ a
70
X
LEP1
+ a
80
X
CONT_ENROLL1
+ (+
0j
+ +
1j
3ELA1 + +
2j
ATTEND1 + +
3j
MINOR1 + e
IJ
)
ELA Level 2 Model
ࢠ
0j =
a
00
+
a
01
X
3ELA2
+ a
02
X
ATTEND2
+ a
03
X
MINOR2
+ a
04
X
TURNOVER2
+ +
0j
ࢠ
1j =
a
10
+
a
11
X
3ELA2
+ a
12
X
ATTEND2
+ a
13
X
MINOR2
+ a
14
X
TURNOVER2
+ +
1j
ࢠ
2j =
a
20
+
a
21
X
3ELA2
+ a
22
X
ATTEND2
+ a
23
X
MINOR2
+ a
24
X
TURNOVER2
+ +
2j
ࢠ
3j =
a
30
+
a
31
X
3ELA2
+ a
32
X
ATTEND2
+ a
33
X
MINOR2
+ a
34
X
TURNOVER2
+ +
3j
Appendix B
ELA Multilevel Model
64
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
APPENDIX B: ELA MULTILEVEL MODEL
ELA Full Equation
Y
4ELA
= a
00
+ a
10
GR3ELA1
+ a
20
ATTEND1 + a
30
MINOR1
+ a
40
X
GENDER1
+ a
50
X
FRPL1
+ a
60
X
SwD1
+ a
70
X
LEP1
+ a
80
X
CONT_ENROLL1
+ a
01
X
3ELA2
+ a
02
X
ATTEND2
+ a
03
X
MINOR2
+ a
04
X
TURNOVER2
+a
11
X
3ELA2
* GR3ELA1 + a
12
X
ATTEND2
* GR3ELA1 + a
13
X
MINOR2
* GR3ELA1
+ a
14
X
TURNOVER2
* GR3ELA1 + a
21
X
3ELA2
*ATTEND1 + a
22
X
ATTEND2
*ATTEND1
+ a
23
X
MINOR2
* ATTEND1 + a
24
X
TURNOVER2
*ATTEND1 + a
31
X
3ELA2
*MINOR1
+ a
32
X
ATTEND2
*MINOR1 + a
33
X
MINOR2
* MINOR1 + a
34
X
TURNOVER2
*MINOR1 ++
0j
+ +
1j
3ELA1
+ +
2j
ATTEND1 + +
3j
MINOR1 + e
IJ
THE ELA EMPIRICAL MODEL (STEP 4)
Using the SPSS Mixed Model procedure, we first determined the intraclass correlation
for the null model with a random intercept (0.187). Second, we entered the eight level 1
variables with fixed coefficients. Third, we added the school contextual variables, X
GR3ELA
,
X
ATTEND2
, X
MINOR2
, and X
TURNOVER2
, with fixed coefficients. This step reduced the intraclass
correlation to 0.046. In the fourth iteration, three student variables, GR3ELA1, ATTEND1,
and MINOR1, were allowed random coefficients and five cross-level interaction terms
implied by the school contextual variables were entered. In each case, there are
statistically significant increases in -2 Log Likelihood using r
2
with degrees of freedom
equal to the change in number of parameters from the previous model. Seven potential
cross-level interactions were eliminated from the model because they did not result in
statistically significant increases in -2 Log Likelihood.
The methodology described above resulted in the following empirical model:
Y
4ELA1
= 658.236
+ 0.580GR3ELA1
+0.498ATTEND1 – 5.204MINOR1
+ 3.159X
GENDER1
(0.225) (0.005) (0.022) (0.409) (0.194)
– 3.099X
FRPL1
– 14.759X
SwD1
– 7.320X
LEP1
+1.560X
CONT_ENROLL1
+ 0.794X
gr3ELA2
+ 0.803X
ATTEND2
(0.345) (0.285) (0.316) (0.317) (0.020) (1.164)
– 0.043X
MINOR2
– 0.148X
TURNOVER2
– 0.003 X
GR3ELA2
*GR3ELA1 + 0.038X
ATTEND2
*ATTEND1
(0.011) (0.030) (0.0003) (0.011)
–.054X
MINOR2
* MINOR1 – 0.109X
TURNOVER2
*MINOR1 – 0.143X
GR3ELA2
*MINOR1
(0.016) (0.052) (0.028)
Note. The standard error of each coefficient is shown in parenthesis.
Appendix B
ELA Multilevel Model
65
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
APPENDIX B: ELA MULTILEVEL MODEL
ELA MULTILEVEL ANALYSIS RESULTS
Table 18 Expansion: Model Predicting Grade 4 ELA Scaled Score
Step I: Estimates of ELA Fixed Effects
Parameter Estimate Std. Error df t Sig.
Intercept 658.382 .656588 694.697 1002.732 .000
Step 2: Estimates of ELA Fixed Effects
Parameter Estimate Std. Error df t Sig.
Intercept 658.232 .658864 699.062 999.042 .000
Grade 3 ELA score .574 .003096 61739.866 185.259 .000
Cumulative Attendance .486 .019366 61740.783 25.081 .000
Minority Status -4.731 .300347 61740.372 -15.750 .000
Female 3.177 .195932 61740.208 16.213 .000
Free- or Reduced-Price Lunch
Eligible
-2.798 .343743 61740.487 -8.139 .000
Student with Disability -15.300 .284658 61740.480 -53.750 .000
Limited English Proficient -7.718 .312807 61744.969 -24.673 .000
Continuously Enrolled 1.43 .319400 61742.451 4.495 .000
Step 3: Estimates of ELA Fixed Effects
Parameter Estimate Std. Error df t Sig.
Intercept 658.240 .225317 655.476 2921.394 .000
Grade 3 ELA score 0.574 .003101 61571.089 184.999 .000
Cumulative Attendance 0.486 .019388 61578.373 25.084 .000
Minority Status -4.714 .300715 61575.478 -15.676 .000
Female 3.174 .196250 61573.902 16.173 .000
Free- or Reduced-Price Lunch
Eligible
-2.777 .344604 61574.028 -8.059 .000
Student with Disability -15.304 .285014 61575.458 -53.695 .000
Limited English Proficient -7.741 .313212 61610.984 -24.716 .000
Continuously Enrolled 1.435 .319686 61592.305 4.490 .000
Grade 3 ELA Mean Score 0.786 .020596 706.522 38.168 .000
Mean Attendance 0.871 .166821 667.559 5.221 .000
Percentage Minority -0.045 .011363 663.074 -3.996 .000
Teach er Turn over Rate -0.159 .030092 702.797 -5.296 .000
66
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
APPENDIX B: ELA MULTILEVEL MODEL
Step 4: Estimates of ELA Fixed Effects
Parameter Estimate Std. Error df t Sig.
Intercept 658.236295 .225320 656.369 2921.340 .000
Grade 3 ELA score .579627 .005066 762.161 114.419 .000
Cumulative Attendance .497796 .022391 724.557 22.232 .000
Minority Status -5.203857 .409285 571.067 -12.715 .000
Female 3.159457 .194059 61209.312 16.281 .000
Free- or Reduced-Price
Lunch Eligible
-3.099032 .344816 60077.988 -8.988 .000
Student with Disability -14.759274 .284736 61560.638 -51.835 .000
Limited English Proficient -7.320015 .316212 60747.472 -23.149 .000
Continuously Enrolled 1.559966 .317424 61151.041 4.914 .000
School Grade 3 ELA Mean Score .794325 .020405 714.878 38.928 .000
School Mean Attendance .802870 .164392 665.945 4.884 .000
Percentage Minority -.043270 .011191 659.535 -3.866 .000
Teach er Turn over Rate -.147979 .029665 701.510 -4.988 .000
Grade 3 ELA Student Score by
School Mean Score
-.003228 .000276 656.636 -11.696 .000
Student Attendance by
School Mean Attendance
.037628 .011273 557.826 3.338 .001
Student Minority Status by
School Percent Minority
-.054057 .016457 527.785 -3.285 .001
Student Minority Status by
Teach er Turn over Rate
-.108660 .052255 385.664 -2.079 .038
Student Minority Status by
School Grade 3 ELA Mean
-.143197 .027988 446.581 -5.116 .000
Information Criteria for Determining Differences in Model Fit between Stages
Stage
Number of
Parameters
-2 Restricted Log
Likelihood
Stage 1 3 624,485.257*
Stage 2 11 577,222.923*
Stage 3 15 574,358.615*
Stage 4 29 573,329.739*
*p < .001
67
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
APPENDIX C: MATHEMATICS MULTI-LEVEL MODEL
The model for predicting grade 4 mathematics performance includes eight student
variables. Three student variables are allowed random coefficients; that is, they are
allowed to vary among schools: grade 3 math scaled score (GR3MATH1), attendance in
grades 3 and 4 (ATTEND1), and identification as Black, Hispanic, or Native American
(MINOR1). Historically, Black, Hispanic, and Native American students have
demonstrated very similar performance in New York City. Therefore, for parsimony,
students in these three ethnic groups are coded 1. In addition, the model includes five
student-level variables with fixed coefficients. All have documented associations with
mathematics performance. These variables are gender (X
GENDER1
), eligibility for free and
reduced-price lunch (X
FRPL1
), identification as a student with disability (X
SwD1
),
identification as limited English proficient (X
LEP1
), and being continuously enrolled in a
school from October 31, 2006 until the end of the 2007-08 school year (X
CONT_ENROLL1
). Four
contextual variables are considered: school mean grade 3 mathematics scale score for
study cohort (X
3MATH2
); school mean attendance in grades 3 and 4 for the study cohort
(X
ATTEND2
); school percentage Black, Hispanic or Native American (X
MINOR2
); and
percentage of teachers with a master’s degree plus 30 hours of graduate credit (X
MA+30_2
).
The study includes 705 schools with a total cohort enrollment of 64,062. Because of
missing data, not all students were included in the multivariate analyses.
MATH LEVEL 1 MODEL
Y
4MATH
= ࢠ
01
+
ࢠ
1j
GR3MATH1
+ ࢠ
2j
ATTEND1 + ࢠ
3j
MINOR1 + a
40
X
GENDER1
+ a
50
X
FRPL1
+ a
60
X
SwD1
+ a
70
X
LEP1
+a
80
X
CONT_ENROLL1
+ (+
0j
+ +
1j
3MATH1 + +
2j
ATTEND1 + +
3j
MINOR1
+ e
IJ
)
MATH LEVEL 2 MODEL
ࢠ
0j =
a
00
+
a
01
X
3MATH2
+ a
02
X
ATTEND2
+ a
03
X
MINOR2
+ a
04
X
MA+30_2
+ +
0j
ࢠ
1j =
a
10
+
a
11
X
3MATH2
+ a
12
X
ATTEND2
+ a
13
X
MINOR2
+ a
14
X
MA+30_2
+ +
1j
ࢠ
2j =
a
20
+
a
21
X
3MATH2
+ a
22
X
ATTEND2
+ a
23
X
MINOR2
+ a
24
X
MA+30_2
+ +
2j
ࢠ
3j =
a
30
+
a
31
X
3MATH2
+ a
32
X
ATTEND2
+ a
33
X
MINOR2
+ a
34
X
MA+30_2
+ +
3j
Appendix C
Mathematics Multilevel Model
68
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
APPENDIX C: MATHEMATICS MULTI-LEVEL MODEL
MATH FULL EQUATION
Y
4MATH
= a
00
+ a
10
GR3MATH1
+ a
20
ATTEND1 + a
30
MINOR1
+ a
40
X
GENDER1
+ a
50
X
FRPL1
+ a
60
X
SwD1
+ a
70
X
LEP1
+a
80
X
CONT_ENROLL1
+ a
01
X
3MATH2
+ a
02
X
ATTEND2
+ a
03
X
MINOR2
+ a
04
X
MA+30_2
+a
11
X
3MATH2
* GR3MATH1 + a
12
X
ATTEND2
* GR3MATH1
+ a
13
X
MINOR2
* GR3MATH1 + a
14
X
MA+30_2
* GR3MATH1 + a
21
X
3MATH2
*ATTEND1
+ a
22
X
ATTEND2
*ATTEND1 + a
23
X
MINOR2
* ATTEND1 + a
24
X
MA+30_2
*ATTEND1
+ a
31
X
3MATH2
*MINOR1 + a
32
X
ATTEND2
*MINOR1 + a
33
X
MINOR2
* MINOR1
+ a
34
X
MA+30_2
* MINOR1 + +
0j
+ +
1j
3MATH1 + +
2j
ATTEND1 + +
3j
MINOR1 + e
IJ
MATH EMPIRICAL MULTILEVEL MODEL (STEP 4)
Using the SPSS Mixed Model procedure, I first determined the intraclass correlation for
the null model with a random intercept (0.208). Second, I entered the eight level 1
variables with fixed coefficients. Third, I added the school contextual variables, X
gr3MATH2,
X
ATTEND2,
X
MINOR2,
and X
MA+30_2,
with fixed coefficients. This step reduced the intraclass
correlation to 0.078. In the fourth iteration, three student variables, GR3MATH1,
ATTEND1, and MINOR1, were allowed random coefficients and three cross-level
interaction terms implied by the school contextual variables were entered. In each case,
there are statistically significant increases in -2 Log Likelihood using r
2
with degrees of
freedom equal to the change in number of parameters from the previous model. Nine
potential cross-level interactions were eliminated from the model because they did not
result in statistically significant increases in -2 Log Likelihood.
The methodology described above resulted in the following empirical model:
Y
4MATH1
= 680.124
+ 0.640GR3MATH1
+ 0.613ATTEND1 – 6.972MINOR1 – 2.674 X
GENDER1
(0.274) (0.005) (0.021) (0.404) (0.185)
– 2.011 X
FRPL1
– 11.146X
SwD1
– 4.340X
LEP1
+ 0.352X
CONT_ENROLL1
+ 0.785X
gr3MATH2
(0.329) (0.267) (0.285) (0.294) (0.027)
+ 0.991 X
ATTEND2
– 0.119X
MINOR2
– 0.041X
MA+30_2
– 0.002X
GR3MATH2
*GR3MATH1
(0.202) (0.015) (0.023) (0.0003)
+ 0.076 X
ATTEND2
*ATTEND1 – 0.088 X
GR3MATH2
*MINOR1
(0.010) (0.026)
Note. The standard error of each coefficient is shown in parenthesis.
69
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
APPENDIX C: MATHEMATICS MULTI-LEVEL MODEL
MATH MULTILEVEL MODEL RESULTS
Table 19 Expansion: Model Predicting
Grade 4 Mathematics Scaled Score
Step I: Estimates of Math Fixed Effects
Parameter Estimate Std. Error df t Sig.
Intercept 680.243 .685692 697.268 992.053 .000
Step 2: Estimates of Math Fixed Effects
Parameter Estimate Std. Error df t Sig.
Intercept 680.194 .688795 698.648 987.512 .000
Grade 3 Math Score 0.639 .003064 62998.055 208.418 .000
Cumulative Attendance 0.569 .018679 62998.761 30.467 .000
Minority Status -7.561 .285269 62998.013 -26.506 .000
Female -2.700 .185893 62998.003 -14.523 .000
Free- or Reduced-Price Lunch Eligible -2.095 .327134 62997.998 -6.405 .000
Student with Disability -11.448 .266895 62998.150 -42.895 .000
Limited English Proficient -4.256 .283521 62998.126 -15.012 .000
Continuously Enrolled 0.205 .295035 62998.218 .695 .487
Step 3: Estimates of Math Fixed Effects
Parameter Estimate Std. Error df t Sig.
Intercept 680.100 .273795 662.568 2483.979 .000
Grade 3 Math Score 0.639 .003064 62973.462 208.390 .000
Cumulative Attendance 0.569 .018681 62977.560 30.484 .000
Minority Status -7.561 .285303 62973.142 -26.500 .000
Female -2.698 .185915 62973.119 -14.515 .000
Free- or Reduced-Price Lunch Eligible -2.094 .327172 62973.079 -6.400 .000
Student with Disability -11.449 .266926 62974.069 -42.892 .000
Limited English Proficient -4.256 .283554 62973.816 -15.009 .000
Continuously Enrolled 0.206 .295069 62974.405 .698 .485
Grade 3 Math Mean Score 0.779 .027630 696.805 28.208 .000
Mean Attendance 1.008 .203940 673.184 4.945 .000
Percentage Minority -0.123 .014853 667.103 -8.258 .000
Teach ers wi th Mas ter’s + 30 C red its -0.043 .023670 684.087 -1.820 .069
70
Taking Attendance Seriously: How School Absences Undermine Student and School Performance in New York City
APPENDIX C: MATHEMATICS MULTI-LEVEL MODEL
Step 4: Estimates of Math Fixed Effects
Parameter Estimate Std. Error df t Sig.
Intercept 680.124 .274105 661.457 2481.255 .000
Grade 3 Math Score 0.640 .004768 739.766 134.306 .000
Cumulative Attendance 0.613 .020772 699.476 29.502 .000
Minority Status -6.972 .404143 497.109 -17.251 .000
Female -2.674 .184647 62568.378 -14.482 .000
Free- or Reduced-Price Lunch Eligible -2.011 .328620 61188.184 -6.120 .000
Student with Disability -11.146 .266871 62827.836 -41.765 .000
Limited English Proficient -4.340 .284995 62067.750 -15.228 .000
Continuously Enrolled 0.352 .293910 62393.793 1.196 .232
School Grade 3 Math Mean Score 0.785 .027483 700.951 28.571 .000
School Mean Attendance 0.991 .202388 673.181 4.895 .000
Percentage Minority -0.119 .014723 664.818 -8.065 .000
Teach ers wi th Mas ter’s + 30 C red its -0.041 .023437 677.843 -1.733 .083
Grade 3 Math Student Score by
School Mean Score
-0.002 .000295 716.543 -7.251 .000
Student Attendance by
School Mean Attendance
0.076 .010481 550.466 7.207 .000
Student Minority Status by
School Grade 3 Math Mean
-0.088 .025593 626.261 -3.433 .001
Information Criteria for Determining Differences in Model Fit between Stages
Stage
Number of
Parameters
-2 Restricted Log
Likelihood
Stage 1 3 636,772.746*
Stage 2 11 583,532.713*
Stage 3 15 582,258.873*
Stage 4 27 580,907.294*
*p < .001
Report prepared by:
The Campaign for Fiscal Equity, Inc.
May 2011
