Sean P. Corcoran Lori Nathanson James Kemple New York University INVALSI – Rome, Italy – October 2012 THE IMPACT OF HIGH SCHOOL CHOICE ON MEDIATORS OF STUDENT SUCCESS
Dec 24, 2015
Sean P. CorcoranLori NathansonJames Kemple
New York University
INVALSI – Rome, Italy – October 2012
THE IMPACT OF HIGH SCHOOL CHOICE
ON MEDIATORS OF STUDENT SUCCESS
HIGH SCHOOL CHOICE IN NYC
80,000 rising 8th graders participate in high school choice each year
700 high school programs available
12 choices, students rank in order of preference
1 matched school for 86% of high school admissions process (HSAPS) participants(main round in 2008 and 2009)
Preliminary, not for quotation or distribution.2
HIGH SCHOOL CHOICE IN NYC
Source: http://www.nytimes.com/schoolbook/2011/09/30/high-school-admissions-choice-but-no-equity/
Preliminary, not for quotation or distribution.3
RATIONALES FOR CHOICE
To provide opportunities for students in low-performing, racially and/or economically isolated schools
To facilitate matches between students and schools
To engage students and their families in the process of choosing a school
To promote competition, encourage innovation, and incentivize high performance
Preliminary, not for quotation or distribution.4
LITERATURE ON SCHOOL CHOICE
Preliminary, not for quotation or distribution.
A large literature has examined the effects of choice in the context of open enrollment, charter schools, and voucher programs (e.g., Cullen, Jacob, & Levitt, 2006; Abdulkadiroglu et al., 2009; Zimmer et al., 2011; Hastings & Weinstein, 2008)
These are typically voluntary (“opt-out”) or targeted programs
5
LITERATURE ON SCHOOL CHOICE
Preliminary, not for quotation or distribution.
Most of these papers are interested in effects on achievement, and the focus is often on school quality differences (e.g., Deming et al., 2009; Lai, 2007; Lauen, 2009; Pop-Eleches & Urquiola, 2011; Clark, 2010)
There is a growing literature that examines effects on “non-cognitive” outcomes, and achievement not captured by test scores (Imberman, 2011; Booker et al., 2011; Pop-Eleches & Urquiola, 2011; Deming et al., 2009; Hastings et al. 2012)
6
THIS PAPER
Preliminary, not for quotation or distribution.
We examine whether admission to a desired school impacts mediators of student achievement – indicators of student engagement in four broad categories
7
STUDENT ENGAGEMENT WITH SCHOOL
Multiple dimensions (e.g., Fredricks et al., 2004) Behavioral: attendance, participation in extracurricular
activities
Cognitive: perceptions about value of school, investment in learning
Psychological/Emotional: connections with teachers, peer culture
Academic: completing required tasks (courses, exams)
Preliminary, not for quotation or distribution.8
STUDENT ENGAGEMENT WITH SCHOOL
Why might choice have implications for engagement? Differences in school quality, practices, proximity Better match between students and schools (Stage-
Environment Fit Theory, Eccles et al., 1993) Promotes autonomy and investment (e.g. Connell’s
self-system model, 1990)
There are potentially both school and match/fit effects of choice on engagement
Preliminary, not for quotation or distribution.9
THIS PAPER
Preliminary, not for quotation or distribution.
We provide an initial look at differences between the engagement of students who are successfully matched to their first choice school and students who are not
Additionally, we look at differences within schools in the engagement of students matched as 1st choice vs. other lower choices
10
THIS PAPER
Preliminary, not for quotation or distribution.
We find significant effects of receiving a first choice on many measures of engagement, using a subsample of students who were subject to random assignment
Within schools, however, the differences are much smaller, suggesting differences in engagement are attributed to different school assignments rather than match/fit
11
DATA
HSAPS data (about 80,000 students per year) matched to NYCDOE administrative data
2008 and 2009 cohorts of 8th graders (N=133,387) Matched through HSAPS Did not opt out or enroll in a specialized exam HS (e.g.
Stuyvesant) Enrolled in a public high school in fall of 9th grade
Preliminary, not for quotation or distribution.12
THREE SAMPLES
Preliminary, not for quotation or distribution.
1. All students meeting the above criteria2. All students whose first choice was an
oversubscribed, nonselective program (e.g., Bloom et al., 2010)
3. All students who were matched to an oversubscribed, nonselective program, but not necessarily their first choice
13
HOW STUDENTS ARE ASSIGNED TO SCHOOLS
Preliminary, not for quotation or distribution.14
HOW STUDENTS ARE ASSIGNED TO SCHOOLS
Preliminary, not for quotation or distribution.15
PROGRAMS BY ADMISSIONS METHOD
Admissions Method # (%) of Programs
2008 2009
Screened 193 (29%) 197 (29%)
Educational option 197 (29%) 192 (28%)
Limited unscreened 167 (25%) 179 (27%)
Audition 68 (10%) 63 ( 9%)
Zoned 32 ( 5%) 32 ( 5%)
Unscreened 16 ( 2%) 15 ( 2%)
All HS Programs 673
(100%)
678
(100%)
Note: each program has different admissions criteria that determine the matching process. Students can rank any combination of programs for their 12 choices.
Preliminary, not for quotation or distribution.16
LIMITED UNSCREENED PROGRAMS
Priority to students demonstrating interest in the school: attending a school’s Information Session or Open
House visiting the school’s exhibit at a High School Fair
When oversubscribed, HSAPS awards seats at random
Preliminary, not for quotation or distribution.17
STUDENTS’ CHOICES AND MATCHES
Students made an average of 7 choices. 52% of students received their first choice. 77% of students received one of their top three choices.
Preliminary, not for quotation or distribution.
0.2
.4.6
.81
1 2 3 4 5 6 7 8 9 10 11 12
1st 2nd 3rd 4th 5th 6-8th
9-12th Supp 1st Supp 2nd Supp 3-12th Unmatch
18
OUTCOME MEASURES
Preliminary, not for quotation or distribution.
Behavioral engagement:
Attendance rate (9th grade)Extracurricular activity participation
Cognitive engagement:
Single scale based on 12 survey items
Psychological engagement –relationships with teachers:
Single scale based on 4 survey items
Psychological engagement – perceptions of peer culture:
Single scale based on 4 survey items
Academic engagement:
Credits attempted and earnedRegents exams attempted and passed
20
BEHAVIORAL ENGAGEMENT
Preliminary, not for quotation or distribution.
Mean
SD
Percent of days in attendance, 8th grade 90.8 9.8Percent of days in attendance, 9th grade 89.1 15.7Percent of days in attendance, 10th grade (2008 only)
86.2 19.3
Change in attendance 8th-9th grade -1.65 10.9Change in attendance 8th-10th grade (2008 only)
-4.50 14.5
Number of extracurricular activities 1.48 1.63Participated in any extracurricular activity 0.65 0.48Participated in two or more extracurricular activities
0.40 0.49
21
COGNITIVE ENGAGEMENT
Cognitive Engagement “I need to work hard to
get good grades at my school.”
“My teachers connect what I am learning to life outside the classroom.”
Cognitive Engagement Measure
Number of items
12
Alpha .88
Mean 3.06
SD .51Note: Based on 4-pt Likert scale
Preliminary, not for quotation or distribution.
22
PSYCHOLOGICAL ENGAGEMENT
Relationships with Teachers “How comfortable are
you talking to teachers and other adults at your school about a problem you are having in class?”
Peer Culture “Most students in my
school help and care about each other.”
Relationships with Teachers Measure
Number of items
4
Alpha .79
Mean 2.64
SD .77
Peer Culture Measure
Number of items
4
Alpha .78
Mean 2.46
SD .67
Note: Based on 4-pt rating scale, 1=uncomfortable to 4=comfortable, or 1=unavailable to 4=available.
Note: Based on 4-pt Likert scalePreliminary, not for quotation or distribution.
Preliminary, not for quotation or distribution.
23
ACADEMIC ENGAGEMENT
Students attempted an average of 15 credits and earned approximately 12 credits in 9th grade.
The average student passed one Regents exam by the end of 9th grade.
Preliminary, not for quotation or distribution.
Mean
SD
Total credits attempted 14.5 4.4Total credits earned 11.6 5.4Total Regents exams attempted 1.5 1.2Total Regents exams passed (65+) 1.0 1.0
24
EMPIRICAL MODEL
Preliminary, not for quotation or distribution.
We are interested in the causal impact of assignment to one’s nth choice high school (e.g. 1st), relative to that if one was assigned to a lower choice
(1) Yi = β0 + γZ1i + βXi + ui
E.g. Yi = engagement, Z1i = assignment to first choice, Xi = student covariates Problem: consider how Z1i might be assigned
25
EMPIRICAL MODEL: SELECTION PROBLEMS
Preliminary, not for quotation or distribution.26
EMPIRICAL MODEL
Preliminary, not for quotation or distribution.
One approach: contrast students with same first choice
(2) Yi = β0 + α1i + γZ1i + βXi + ui
α1i = first choice effects Problem: generally, the likelihood of attending one’s
first choice can still be correlated with school quality, unobserved characteristics of students
27
WHAT WE DO IN THIS PAPER
Preliminary, not for quotation or distribution.
Estimate model (2) for all students in our sample, all programs represented: the “naïve” (non-causal) model
Estimate model (2) for all students whose first choice was a non-selective oversubscribed school (and were subject to random assignment)
Estimate model (2) with school fixed effects—comparing students matched to same schools but at different preference. All students, and those matched to an oversubscribed non-selective school.
28
STUDENT CHARACTERISTICS BY SAMPLE
Preliminary, not for quotation or distribution.
Sample 1:Base sample
Sample 2:Oversubscribe
dLimited
Unscreened1st choice
Sample 3:Matched to
Oversubscribed
LimitedUnscreened
Applicant from private school
4.1 4.1 4.3
Female 49.6 49.8 50.8Asian 12.1 3.7 3.7Black 34.1 41.4 44.0Hispanic 41.5 48.2 46.8White 11.6 6.0 4.7Special education 6.6 7.5 7.5English language learner 10.2 8.6 9.0Foreign born 19.0 12.6 13.3Eligible for free/reduced lunch
75.9 80.7 81.2
Attendance rate (% of days on roll)
90.8 89.9 89.8Note: Student characteristics are based on 8th grade data. All numbers represent percentages.
29
SAMPLE MEANS: 8TH GRADE STANDARDIZED TESTS
Preliminary, not for quotation or distribution.30
REGRESSION RESULTS – SAMPLE 1 ALL SCHOOL TYPES
Preliminary, not for quotation or distribution.
Outcome 1st choice, w/o school effects
1st choice, with school effects
N
Attendance in grade 9 0.95*** 0.48*** 120,200Any activities 0.03*** 0.01 92,397Two or more activities 0.04*** 0.01 92,397Number of activities 0.13*** 0.02 92,397Cognitive engagement 0.05*** 0.02*** 93,371Psychological engagement (1)
0.04*** 0.02** 93,772
Psychological engagement (2)
0.11*** 0.004 93,888
Credits attempted 0.11*** 0.11*** 120,615Credits earned 0.22*** 0.24*** 120,615Regents attempted -0.06*** 0.01 120,615
Note: * p<0.05, ** p<0.01, *** p<0.001Each cell represents a coefficient estimate from a separate regression, where the indicated outcome variable is used. Models include student-level covariates, cohort dummies, and program type effects.
31
REGRESSION RESULTS – SAMPLE 2:OVERSUBSCRIBED LIMITED UNSCREENED SCHOOLS, 1ST CHOICE
Outcome 1st choice, w/o school effects
1st choice, with school effects
N
Attendance in grade 9 1.16*** 0.57 19,065Any activities 0.03* 0.02 14,481Two or more activities 0.02 0.01 14,481Number of activities 0.08 0.01 14,481Cognitive engagement 0.05*** -0.004 14,691Psychological engagement (1)
0.05* -0.01 14,764
Psychological engagement (2)
0.09*** -0.01 14,791
Credits attempted 0.39*** 0.17 19,143Credits earned 0.95*** 0.44** 19,143Regents attempted 0.14*** 0.04 19,143
Preliminary, not for quotation or distribution.
Note: * p<0.05, ** p<0.01, *** p<0.001 Each cell represents a coefficient estimate from a separate regression, where the indicated outcome variable is used. Models include student-level covariates and cohort dummies.
32
REGRESSION RESULTS –SAMPLE 3: MATCHED TO OVERSUBSCRIBED LIMITED UNSCREENED
SCHOOLSOutcome 4th-12th choice,
w/ school effects
4th-12th choice, w/ school
effects, cntl
N
Attendance in grade 9 -0.55* -0.80* 23,599Any activities -0.02 -0.02 17,937Two or more activities -0.02 0.001 17,937Number of activities -0.02 -0.01 17,937Cognitive engagement -0.03* 0.001 18,223Psychological engagement (1)
-0.001 -0.01 18,300
Psychological engagement (2)
0.01 0.01 18,347
Credits attempted -0.35*** -0.42*** 23,707Credits earned -0.42*** -0.62*** 23,707Regents attempted -0.01 -0.04 23,707
Preliminary, not for quotation or distribution.
Note: * p<0.05, ** p<0.01, *** p<0.001 Each cell represents a coefficient estimate from a separate regression, where the indicated outcome variable is used. Models include student-level covariates and cohort dummies.
33
LIMITATIONS
Preliminary, not for quotation or distribution.
Agnostic about the motivation behind students’ rankings of schools; have yet to investigate mechanisms
Outcomes: smaller sample with School Survey data may represent restricted range of engagement
Random assignment prevalent in limited unscreened schools, but different assignment mechanisms require different analytic approach
34
NEXT STEPS
Preliminary, not for quotation or distribution.
Expand analyses by school type
Examine results for patterns across school type
35