The Spillovers of Charter Schools on Neighborhood Public School Students: Evidence from NYC.

THE SPILLOVERS OF CHARTER SCHOOLS
ON NEIGHBORHOOD PUBLIC SCHOOLS:
EVIDENCE FROM NEW YORK CITY
SARAH CORDES
IESP SUMMER SEMINAR
JUNE 17, 2014
*FINDINGS PREPARED FOR IESP SUMMER SEMINAR, PLEASE DO NOT CITE WITHOUT PERMISSION OF THE AUTHOR 1
MOTIVATION
 Historically, neighborhood strongly liked to schooling:

Levels of funding

Teacher characteristics

School quality
 Numerous reforms attempt to break this link: busing, school finance reform,
magnet schools, charter schools
 Charter schools designed to address neighborhood school-quality link in at least
two ways

Offering students choice to attend schools outside their designated attendance zone

Increasing performance in traditional public schools (TPSs) through competitive
pressures
 Majority of public school students still education in TPSs, important to consider
the spillovers of charter schools on these students
2
RESEARCH QUESTIONS
 Charter schools aimed at improving performance through competition,
innovation, or both
 Opponents argue that charter schools sap funds, resources, and the best
students from traditional public schools
 Given continuing expansion of charter schools, it is important to
understand the effect of charter schools on traditional public school
resources
 Spillovers of charters on neighborhood public schools are unclear a
priori

Positive or negative: if charters may draw students primarily from neighborhood schools

No spillovers if: charters draw students from private schools or public schools students from
other neighborhoods
 What spillover effects do charter schools have on neighborhood public
schools?
3
CONTRIBUTION
 Most recent literature on charter schools focuses on the effects of charter
schools on charter school students
 Long line of research examining effects of charter schools on public schools

Either conducted at district-level or using a large radius around TPSs (2.5-10 miles)

Problematic for identifying spillovers in an urban context to the extent that location
and enrollment in charter schools reflect local neighborhood characteristics
 Prior literature does not fully account for endogenous charter school location
within a city or district
 This paper:

Uses better data and methods to address endogenous location of charter schools

Examines spillovers at the neighborhood level where one might expect effects to be
largest

Explores mechanisms that might explain charter school impacts
4
RESEARCH QUESTIONS
 What are the spillover effects of charter schools on the outcomes of students
attending nearby TPSs?
 What school-level mechanisms might explain these effects?
5
PREVIEW OF FINDINGS
 Charter schools increase performance of students attending nearby TPSs by 0.03
to 0.06 SDs in both ELA and math
 Impacts are larger among TPS students in schools that are co-located with
charters
 Potential explanations for increased performance include lower pupil-teacher
rations and increased per-pupil expenditures in TPSs located near to charter
schools
6
ECONOMICS OF CHARTER SCHOOLS
 Consider a school-level production function

Y represents educational outcomes including achievement and attainment

Instruction, class size, etc. represent the level of specific inputs at a given school
 Charter school can affect student performance by:

Changing the level of inputs

Changing the value of the parameters (i.e. through changes in efficiency)
7
CHANGING INPUTS
 Changes in TPS student composition


If charter schools systematically attract specific types of students from nearby public schools

More advantaged students

Students who are low-cost relative to peers but eligible for additional funds

Higher achieving students
Evidence: changes in TPS student characteristics after charter entry
 Changes in TPS resources

Declining enrollments: teachers and categorical aid will be spread over fewer students

Redistribution of teachers

Evidence: changes in pupil-teacher ratios, per-pupil expenditures, or teacher characteristics
after charter entry
 Changes in TPS parent characteristics

Charter schools might attract more active or vocal parents

Charter entry may affect the participation of parents remaining in TPSs

Evidence: parent responses to student surveys
8
CHANGING EFFICIENCY
 TPSs may change practices in order to retain students
 Efficiency gains could be realized through:

Changes in curricula

Increased professional development

Increased collaboration
 Evidence: teacher responses to NYC public school survey reporting variety of
course offerings, participation in professional development, teacher collaboration,
etc.
9
NEIGHBORHOOD MEASURES
 Distance between TPS and nearest charter school
 ½ or 1 mile radius of school
 Corresponds to walk distances in NYC
 50 percent of charter school students attend school within ½ mile of residence and
75 percent attend within one mile
 Planned robustness checks:

Vary size of radius from ¼ to 2 miles

Census tract

School zone

Community district
10
WHERE DO CHARTER STUDENTS LIVE?
11
CHARTER SCHOOL EXPOSURE
 ANY charter school in the neighborhood
 Indicator equal to 1 in all years that there is a charter within a ½ or 1 mile
radius of school
 Conducted as ITT: student is fixed in the first school he/she is observed
attending
 Distance to nearest charter school

Based on Euclidian distance between each public school and the closest charter school
 Added to baseline model of whether there is any charter school in the
neighborhood
 As charter school is closer to public school expect greater impact
12
STUDENT-LEVEL OUTCOMES
 Performance on ELA and math exams

Standardized to have mean 0 and standard deviation of 1 across all students in a given
grade and year
 Outcome measures for future analysis

Attendance

Grade retention

Probability of exit from NYC public school system
13
MECHANISMS
 Student composition

Percent free lunch eligible, reduced price lunch eligible, recent immigrants, LEP, black,
Hispanic, Asian
 School resources

Per-pupil expenditures: total and instructional

Pupil-teacher ratios

Teacher characteristics: percentage of teachers with master’s degree and percentage of
teachers with more than two years experience in current school
 Future work

Responses from parent surveys on school satisfaction

Responses from teacher surveys on course offerings, PD, etc.
14
IDENTIFICATION CHALLENGES
 Charter schools locate non-randomly throughout city
 Tend to locate in higher poverty neighborhoods
 Locate near less advantaged schools: lower percentages of teachers with
master’s degrees, lower baseline test scores, higher percentages of poor
students, etc.
 Simple cross-sectional model of the usual form
will likely yield downwardly biased estimates of
 Solution: difference-in-difference approach
15
16
Table 1. Baseline Descriptive Statistics, Schools in Schools with and without Charter
Schools within ½, 1, and 3 miles, AY 1998-1999
Within 1 Mile
Within ½ Mile
Never Charter
Total Spending PP
Ever Charter
$11,331
$12,203
Instruc. Spending PP
$5,983
$6,314
% Teachers with MA
81.2
73.8
% teachers w/ more than 2 years exp. in school
64.7
59.2
Pupil-teacher ratio
17.5
16.7
Enrollment per school
867
795
Reading z-score
0.11
-0.24
Math z-score
0.14
-0.26
71.5
86.7
8.0
3.3
Black
29.3
51.0
Hispanic
36.4
42.4
Asian
14.4
3.5
6.5
6.4
ESL
14.8
14.5
LEP
10.8
11.4
8.6
4.8
Number of Schools
406
220
Number of Students
352,195
174,988
Never Charter
Within 3 Miles
Ever Charter
Never Charter
Ever Charter
$11,286
$11,902
$11,263
$11,675
$5,947
$6,214
$5,908
$6,118
83.3
75.1
84.6
78.0
65.7
60.6
68.5
62.2
17.6
16.9
17.6
17.2
839
844
746
852
0.23
-0.19
0.30
-0.04
0.27
-0.21
0.39
-0.04
65.1
85.6
55.9
78.9
9.2
4.2
9.9
6.0
23.4
47.0
17.0
38.9
31.8
43.5
26.7
39.7
17.8
5.1
24.8
9.1
6.7
6.3
6.8
6.4
13.9
15.3
13.1
14.9
9.8
12.0
9.2
11.2
9.4
5.7
8.5
7.1
269
357
57
569
225,785
301,398
42,515
484,668
Percent
Free lunch
Reduced lunch
Special Ed.
Recent Immigrant
NOTES: “Never charter” schools are schools that never have a charter school within a given radius during the sample period.
“Ever charter" schools are schools that are located within a given radius for at least one year of the sample period. Bold indicates
that the differences between “never” and “ever” charter schools are significantly different at the 0.05 level.
17
MODEL 1 - BASELINE

Y is a school or student outcome measure for student i, in school s, in neighborhood n,
at time t

POSTCS is an indicator equal to one in the period after a charter school opens in the
neighborhood of school s

δ are grade effects, φ are school effects, and µ are year effects

is an error term with the usual properties
 Coefficient of interest is β, which captures the effect of charter schools TPS
student performance

Identified by variation in timing of charter entry

Can be interpreted as causal if, conditional on student-level covariates, grade, school,
and year effects, timing of charter school entry is random
18
MODEL 2 – DISTANCE TO NEAREST CHARTER
 CSDIST measures the distance to the nearest charter schools within that radius
 Coefficients of interest are β1 and β2, which capture the effect of charter
schools TPS student performance
 Remaining concerns:

Timing of charter entry is correlated with pre-existing trends in TPS student
performance

To address this, re-estimate models 1 and 2, adding non-parametric time trends:
indicators for one, two, or three plus years before charter entry

Same strategy employed by Figlio and Hart (2014) in analysis of competitive effects of
vouchers on public school student performance
19
DATA
 NYCDOE

Student-level administrative data on student demographics, program information,
performance, and attendance

School report cards (SRC): school-level information on enrollment, pupil-teacher
ratios, and teacher characteristics

School Based Expenditure Reports: school-level expenditure data
 Common Core of Data (CCD)

Charter school opening year

Latitude and longitude of all charter and public schools

School level addresses – used to identify co-located schools
20
SAMPLE
 Students attending public elementary schools from AY 1996-97 to AY 2009-10

Elementary school defined as any TPS with a fourth grade
 Exclude:

Middle schools and high schools because almost all charter during this time served
elementary grades

Staten Island – only one charter school in operation there during this time period

Schools serving exclusively special education students

Students with only one test score in ELA or math
 Final sample includes 626 unique schools serving 1,250,600 unique students
21
RESULTS – BASELINE MODEL
Table 2: Effects of charter schools on public school students’ test scores, any charter within 1 mile, AY
1997-2010
Post charter w/n 1 mile
Student Char.
Lagged test scores
School Effects
Student Effects
Observations
R-squared
(1)
Math
(2)
(5)
ELA
(6)
(3)
(4)
(7)
(8)
-0.227***
(0.009)
-0.034***
(0.004)
0.017***
(0.004)
-0.011**
(0.005)
-0.223***
(0.009)
-0.043***
(0.004)
0.002
(0.004)
-0.017***
(0.004)
N
N
N
N
Y
Y
N
N
Y
Y
Y
N
Y
N
Y
Y
N
N
N
N
Y
Y
N
N
Y
Y
Y
N
Y
N
Y
Y
2,797,449
2,797,449
2,797,449
2,797,449
2,669,382
2,669,382
2,669,382
2,669,382
0.041
0.469
0.480
0.833
0.039
0.416
0.429
0.807
Robust standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1
Notes: Post charter is equal to 1 for a student in the period after the public school we first observe him/her attending is located within
1 mile of a charter school. All models contain controls for residence borough, grade, and year. Models in columns 2, 3, 5, and 6 contain
controls for race, gender, free lunch eligibility, special education, LEP, recent immigrants, speaking a language other than English at
home, and lagged test scores. Standard errors are clustered at the school-year level.
22
TPS STUDENT PERFORMANCE BY DISTANCE TO CHARTER
Table 3: Effects of charter schools on public school students’ scores, any charter within 1 mile with pretrends, AY 1997-2010
Math
Post charter w/n 1 mile
ELA
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
0.029***
0.065***
0.028**
-0.006
0.021**
0.062***
0.006
-0.000
(0.011)
(0.015)
(0.014)
(0.019)
(0.010)
(0.014)
(0.013)
(0.017)
Dist. to charter
Dist. to charter sq.
-0.147***
0.060
-0.141***
-0.017
(0.044)
(0.052)
(0.039)
(0.046)
0.122***
-0.014
0.102***
0.034
(0.041)
(0.047)
(0.037)
(0.040)
Lagged test scores
Y
Y
N
N
Y
Y
N
N
Student FX
N
N
Y
Y
N
N
Y
Y
2,797,449
2,797,449
2,797,449
2,797,449
2,669,382
2,669,382
2,669,382
2,669,382
0.480
0.480
0.833
0.833
0.429
0.429
0.807
0.807
Observations
R-squared
Robust standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1
Notes: Post charter within 1 mile is equal to 1 for a student in the period after the public school we first observe him/her attending is located
within 1 mile of a charter school. Distance to charter is the distance (in miles) between a public school and the nearest charter school located
within 1 mile. A negative coefficient on distance to charter indicates that the effect of charter schools is decreasing with distance, or conversely,
that the effect increases with proximity. All models contain controls for free lunch eligibility, special education, LEP, recent immigrants, speaking
a language other than English at home, residence borough, grade, year, school effects, and indicators for one, two, or three plus years23prior to
charter entry. Models in columns 1, 2, 5, and 6 also include controls for race, gender, and lagged test scores. Standard errors are clustered at the
school-year level.
Table 4: Effects of co-located charter schools on public school students’ scores, AY 1997-2010
(1)
Post co-located charter
Post charter w/n ½ mile
Post charter w/n 1 mile
(3)
(4)
(5)
ELA
(6)
(7)
(8)
0.069***
(0.022)
0.015***
(0.006)
0.022*
(0.011)
0.069***
(0.022)
0.028***
(0.010)
0.018
(0.020)
-0.070
(0.047)
0.095**
(0.042)
-0.028
(0.021)
0.018*
(0.010)
0.027
(0.026)
0.071***
(0.022)
0.027
(0.026)
0.018*
(0.010)
-0.026
(0.024)
0.063
(0.057)
0.007
(0.047)
0.066***
(0.024)
0.014***
(0.005)
0.013
(0.011)
0.066***
(0.024)
0.007
(0.009)
0.037**
(0.019)
-0.084*
(0.043)
0.065*
(0.037)
-0.017
(0.020)
0.005
(0.009)
0.014
(0.023)
0.027
(0.019)
0.015
(0.023)
0.004
(0.009)
-0.009
(0.022)
-0.008
(0.051)
0.031
(0.041)
2,797,449
0.480
2,797,449
0.480
2,797,449
0.833
2,797,449
0.833
2,669,382
0.429
2,669,382
0.429
2,669,382
0.807
2,669,382
0.807
Dist. to charter
Dist. to charter sq.
Observations
R-squared
Math
(2)
Robust standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1
Notes: Post charter within ½ mile is equal to 1 for a student in the period after the public school we first observe him/her attending is
located within ½ mile of a charter school. Post char5er within 1 mile is measured analogously. Distance to charter is the distance (in miles)
between a public school and the nearest charter school located within 1 mile. A negative coefficient on distance to charter indicates that the
effect of charter schools is decreasing with distance, or conversely, that the effect increases with proximity. All models contain individuallevel controls for race, gender, free lunch eligibility, special education, LEP, recent immigrants, speaking a language other than English at
home, lagged test scores, residence borough, grade, year, and school effects. All models also contain controls for one, two, or three plus
years prior to charter entry. Standard errors are clustered at the school-year level.
24
Table 5: Effects of charter schools on public school characteristics, any charter within 1 mile with pretrends, AY 1997-2010
Dependent Variable:
Enrollment
SPED Enroll
% Free lunch
% Red. Lunch
% SPED
% LEP
% Rec. Imm.
Post charter w/n 1 mile
-12.376
(16.504)
-2.662
(3.409)
-0.225
(0.755)
-0.083
(0.316)
0.007
(0.005)
1.416***
(0.463)
0.406
(0.369)
Post charter w/n ½ mile
4.990
-2.461
0.656
0.283
-0.001
-0.418
0.071
(9.075)
-54.277***
(17.613)
-27.954
(1.875)
-15.159***
(3.639)
1.280
(0.415)
0.811
(0.805)
-0.499
(0.174)
0.175
(0.337)
-0.144
(0.003)
0.005
(0.006)
0.016***
(0.255)
0.228
(0.494)
0.475
(0.203)
0.908**
(0.394)
0.914**
(17.744)
(3.666)
(0.811)
(0.340)
(0.006)
(0.498)
(0.397)
9,113
0.912
9,113
0.825
9,113
0.974
9,113
0.908
9,113
0.986
9,113
0.938
9,113
0.836
Charter co-located
Dist. to charter
Observations
R-squared
Robust standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1
Notes: Post charter within ½ mile is equal to 1 for a student in the period after the public school we first observe him/her attending is
located within ½ mile of a charter school. Post char5er within 1 mile is measured analogously. Distance to charter is the distance (in miles)
between a public school and the nearest charter school located within 1 mile. A negative coefficient on distance to charter indicates that the
effect of charter schools is decreasing with distance, or conversely, that the effect increases with proximity. All models contain individuallevel controls for race, gender, free lunch eligibility, special education, LEP, recent immigrants, speaking a language other than English at
home, lagged test scores, residence borough, grade, year, and school effects. All models also contain controls for one, two, or three plus
years prior to charter entry. Standard errors are clustered at the school-year level.
25
Table 6: Effects of charter schools on public school resources, any charter within 1 mile with pre-trends, AY 1997-2010
Dependent Variable:
% Teach w/ MA
% Teach w/ 2+ years in school
PTR
Log pp total exp.
Log pp instruct. exp.
2.888**
-1.518
-1.598
0.031**
0.046***
(1.307)
(1.774)
(1.003)
(0.014)
(0.016)
-1.495**
-0.309
0.912*
0.011
0.005
(0.709)
(0.961)
(0.549)
(0.008)
(0.009)
-2.473*
-1.563
-0.436
0.061***
0.061***
(1.393)
(1.889)
(1.136)
(0.015)
(0.017)
2.262
1.380
-2.498**
0.025*
0.049***
(1.386)
(1.879)
(1.069)
(0.015)
(0.017)
Observations
8,948
8,906
8,235
8,972
8,972
R-squared
0.916
0.668
0.664
0.931
0.897
Post charter w/n 1 mile
Post charter w/n ½ mile
Charter co-located
Dist. to charter
Robust standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1
Notes: Post charter within ½ mile is equal to 1 for a student in the period after the public school we first observe him/her attending is
located within ½ mile of a charter school. Post char5er within 1 mile is measured analogously. Distance to charter is the distance (in miles)
between a public school and the nearest charter school located within 1 mile. A negative coefficient on distance to charter indicates that the
effect of charter schools is decreasing with distance, or conversely, that the effect increases with proximity. All models contain individuallevel controls for race, gender, free lunch eligibility, special education, LEP, recent immigrants, speaking a language other than English at
home, lagged test scores, residence borough, grade, year, and school effects. All models also contain controls for one, two, or three plus
26
years prior to charter entry. Standard errors are clustered at the school-year level.
SUMMARY OF RESULTS & NEXT STEPS
 Charter schools have small positive spillovers on public school performance in
both math and ELA ranging from 0.021 to 0.065 SDs
 Positive spillovers on performance potentially explained by smaller class sizes
and high per pupil expenditures
 Next steps

Explore other outcomes: attendance, grade retention, exit from NYC public schools

Re-estimate models limiting analysis to schools that have overlapping grade spans with
nearby charter schools

Use different neighborhood measures: census tracts, school zones, etc.

Add parent and teacher survey data to explore additional mechanisms
27