You are on page 1of 75

Exam Schools, Ability, and the Eects of

Armative Action: Latent Factor Extrapolation in


the Regression Discontinuity Design
Miikka Rokkanen

Massachusetts Institute of Technology

Introduction

Latent Factor Modeling in a Sharp RD Design

Boston Exam Schools

Identication and Estimation

Extrapolation Results

Counterfactual Simulations

Placebo Experiments

Conclusions

Extrapolation Problem in the RD Design


(0), (1)

[(1)|]

[(1) (0)| = ]

[(0)|]

Extrapolation Problem in the RD Design


(0), (1)

[(1)| = 1 ]

[(1)|]

[(1)| = 0 ]

[(1) (0)| = 1 ]
[(1) (0)| = ]

[(1) (0)| = 0 ]

[(0)| = 1 ]

[(0)|]

[(0)| = 0 ]
0

Boston Exam Schools

I Three selective high schools that are seen as the agship of


the Boston public school system

I RD estimates show little evidence of eects for marginal


applicants (Abdulkadiroglu, Angrist & Pathak, forthcoming)

I Is the lack of eects generalizable for inframarginal applicants?


I Important question for discussion of armative action at
selective school admissions

This Paper
1. Develop a latent factor-based approach to the extrapolation of
treatment eects away from the cuto in RD
I

Nonparametric identication of treatment eects at any point


in the running variable distribution

2. Estimate eects of Boston exam school attendance for full


population of applicants
I

Achievement gains concentrated among lower-scoring


applicants

3. Simulate eects of introducing minority/socioeconomic


preferences into exam school admissions
I

Both reforms increase average achievement among aected


applicants

Related Literature
1.

Extrapolation of treatment eects in RD: Angrist &


Rokkanen (2013); Cook & Wing (2013); Dong & Lewbel
(2013)

2.

Latent factor and measurement error models:

Hu &

Schennach (2008); Cunha, Heckman & Schennach (2010);


Evdokimov & White (2012)
3.

(Semi-)nonparametric instrumental variables models:


Newey & Powell (2003); Blundell, Chen & Kristensen (2007);
Darolles, Fan, Florens & Renault (2011)

4.

Eects of attending selective middle/high schools:


Jackson (2010); Pop-Eleches & Urquiola (2013);
Abdulkadiroglu, Angrist & Pathak (forthcoming)

5.

Eects of armative action in school admissions:


Arcidiacono (2005); Card & Krueger (2005); Hinrichs (2012)

Introduction

Latent Factor Modeling in a Sharp RD Design

Boston Exam Schools

Identication and Estimation

Extrapolation Results

Counterfactual Simulations

Placebo Experiments

Conclusions

Latent Factor Framework


I Running variable

is a noisy measure of a latent factor

R = gR ( , R )
I Potential outcomes
the noise in

Y (0)

and

Y (1)

depend on

R:
(Y (0) , Y (1)) R |

I Example: selective school admission


I
I
I

R : entrance exam score


: latent ability
Y (0), Y (1): future achievement

but not on

Treatment Assignment in the Latent Factor Framework

Treatment Assignment in the Latent Factor Framework

Extrapolation in the Latent Factor Framework


(0), (1)

[(1)| = 1 ]

[(1)|]

[(1)| = 0 ]
= {[(1)| ]| = 0 }

[(1) (0)| = 1 ]
[(1) (0)| = ]

[(1) (0)| = 0 ]

[(0)| = 1 ]
= {[(0)| ]| = 1 }

[(0)|]

[(0)| = 0 ]
0

Object 1: Conditional Latent Factor Distribution


| (|)

| (|0 )

| (|1 )

Object 2: Latent Conditional Expectation Functions


(0), (1)

[(1)|]

[(0)|]

Identication Using Multiple Noisy Measures


I Suppose one has available three noisy measures of

M1 = gM1 ( , M1 )
M2 = gM2 ( , M2 )
M3 = gM3 ( , M3 )
where

M1 , M2

are continuous, and

M3

is potentially discrete

is a deterministic function of a subset of

I Example: selective school admissions


I
I

R = M1 : entrance exam score


M2 , M3 : baseline test scores

Identication with less than three measures

M = (M1 , M2 , M3 )

Nonparametric Identication: Roadmap


Goal:

E [Y (1) Y (0) | R] = E {E [Y (1) Y (0) | ] | R}

I Inputs:

Step 1:

f |R , E [Y (0) | ],

and

E [Y (1) | ]

Identication of f ,M (and consequently f |R )

I Input:

fM

I Literature: nonclassical measurement error models

Step 2:

Identication of

I Inputs:

E [Y (0) | ]

and

E [Y (1) | ]

E [Y | M, D] and f |M,D

I Literature: nonparametric instrumental variables models

Nonparametric Identication: Roadmap


Goal:

E [Y (1) Y (0) | R] = E {E [Y (1) Y (0) | ] | R}

I Inputs:

Step 1:

f |R , E [Y (0) | ],

and

E [Y (1) | ]

Identication of f ,M (and consequently f |R )

I Input:

fM

I Literature: nonclassical measurement error models

Step 2:

Identication of

I Inputs:

E [Y (0) | ]

and

E [Y (1) | ]

E [Y | M, D] and f |M,D

I Literature: nonparametric instrumental variables models

Parametric Illustration
I Suppose the measurement model takes the form

Mk

= Mk + Mk + Mk , k = 1, 2, 3

M1 = 0, M1 = 1, and

2
0

0 0 M
M1
1
N
,

0
M2
0 0
M3
0
0
0

where

2M
0

2M

I Then the unknown parameters can be obtained from the


means, variances, and covariances of

Details

Identication of f ,M
f ,M

identied from fM under the following assumptions (Hu &

Schennach, 2008):

1.

f ,M ( , m)

is bounded with respect to the product measure of

the Lebesgue measure on


measure

on

M3 .

M1 M2

and some dominating

All the corresponding marginal and

conditional densities are also bounded.


2.

M1 , M2 ,

and

M3

are jointly independent conditional on

H such that
.




0
00
0
00
, , fM3 | m3 | and fM3 | m3 |

3. There exists a known functional



H fM1 | ( | ) =
4. For all

for all

over a set of strictly positive probability whenever


5.

dier

6=

00

f |M1 ( | m1 ) and fM1 |M2 (m1 | m2 ) form (boundedly) complete


families of distributions indexed by m1 M1 and m2 M2 .

Nonparametric Identication: Roadmap


Goal:

E [Y (1) Y (0) | R] = E {E [Y (1) Y (0) | ] | R}

I Inputs:

Step 1:

f |R , E [Y (0) | ],

and

E [Y (1) | ]

Identication of f ,M (and consequently f |R )

I Input:

fM

I Literature: nonclassical measurement error models

Step 2:

Identication of

I Inputs:

E [Y (0) | ]

and

E [Y (1) | ]

E [Y | M, D] and f |M,D

I Literature: nonparametric instrumental variables models

Parametric Illustration

I Suppose the latent outcome model takes the form

E [Y (D) | ] = D + D , D = 0, 1
and

(Y (0) , Y (1)) M |
I Then the unknown parameters can be obtained from the
conditional means of
the cuto

Details

and

given

to the left and right of

Identication of E [Y (0) | ] and E [Y (1) | ]

E [Y (0) | ] and E [Y (1) | ] identied from E [Y | M, D] and


f |M,D under the following assumptions:
1.

(Y (0) , Y (1)) M | .
< P [D = 1 | ] < 1 a.s.


f |M,D | m0 , 0 and f |M,D | m1 , 1 form (boundedly)
0
0
complete families of distributions indexed by m M and
1
1
m M .

2. 0
3.

Nonparametric Identication: Roadmap


Goal:

E [Y (1) Y (0) | R] = E {E [Y (1) Y (0) | ] | R}

I Inputs:

Step 1:

f |R , E [Y (0) | ],

and

E [Y (1) | ]

Identication of f ,M (and consequently f |R )

I Input:

fM

I Literature: latent factor and measurement error models

Step 2:

Identication of

I Inputs:

E [Y (0) | ]

and

E [Y (1) | ]

E [Y | M, D] and f |M,D

I Literature: nonparametric instrumental variables models

Extensions

Extension 1: Latent Factor Modeling in a Fuzzy RD Design


Extension 2: Settings with Multiple Latent Factors

Introduction

Latent Factor Modeling in a Sharp RD Design

Boston Exam Schools

Identication and Estimation

Extrapolation Results

Counterfactual Simulations

Placebo Experiments

Conclusions

Boston Exam Schools


I Three selective public schools spanning grades 7-12
(new students admitted mainly for grades 7 and 9)
I
I
I

Boston Latin School


Boston Latin Academy
John D. O'Bryant High School of Mathematics and Science

I Exam schools dier considerably from traditional BPS


I

Peers, curriculum, teachers, resources

I Each applicant receives at most one exam school oer


I

Most preferred school the applicant qualies for

I School-specic RD experiments for sharp samples


I
I

Running variable: rank based on GPA and ISEE scores


Admissions cuto: lowest rank among admitted students

DA Algorithm

Sharp Sample

Data and Sample


I Data sources:
I
I
I
I

Exam school application le


BPS enrollment le
MCAS le
ACS 5-year summary le

I Sample restrictions:
I
I
I

7th grade applicants in 2000-2004


Enrolled in BPS in 6th grade
Non-missing 4th grade MCAS scores and covariates

I Outcome: high school MCAS composite score


I

Standardized average of 10th grade MCAS scores in English


and Math

Descriptive Statistics

Eects at the Admissions Cutos: First Stage

20

10

0
10
Running Variable

20

.8
0

.2

Enrollment Probability
.4
.6

.8
Enrollment Probability
.4
.6
.2
0

.2

Enrollment Probability
.4
.6

.8

Latin School

Latin Academy

OBryant

20

10

0
10
Running Variable

20

20

10

0
10
Running Variable

20

Eects at the Admissions Cutos: Reduced Form

20

Estimates

10

0
10
Running Variable

20

1.5
.5

.5

Mean Score

1.5
1
Mean Score
.5
0
.5

.5

.5

Mean Score

1.5

Latin School

Latin Academy

OBryant

20

10

0
10
Running Variable

20

20

10

0
10
Running Variable

20

Introduction

Latent Factor Modeling in a Sharp RD Design

Boston Exam Schools

Identication and Estimation

Extrapolation Results

Counterfactual Simulations

Placebo Experiments

Conclusions

Setup
I Specify a model with two latent factors:
I
I

English ability
Math ability

I School-specic running variables are functions of ISEE scores


I
I

Reading and Verbal: Noisy measures of English ability


Math and Quantitive: Noisy measures of Math ability

I Data also contains 4th grade MCAS scores


I
I

English: Noisy measure of English ability


Math: Noisy measure of Math ability

I Control for a set of additional covariates


I

Application year, application preferences, GPA,


sociodemographic characteristics

Measurement Scatterplots

Measurement Correlations

Estimation
I Measurement model: Maximum Simulated Likelihood

E
M

!
E M
2E
N
E M
2M

2 

0
N M E X + M E E , exp M E + M E E
k
k
k
k


2 
0
N M M X + M M M , exp M M + M M M
"


|X

MkE | , X
MkM | , X

E X
0
M X

# 
,

I Latent outcome model: Method of Simulated Moments


0

E [Ds (z) | , X ] = Ds (z) X + DEs (z) E + DMs (z) M


0

E [Y (S (z)) | , X ] = Y (S(z)) X + YE (S(z)) E + YM(S(z)) M


I Inference: 5-step bootstrap

Introduction

Latent Factor Modeling in a Sharp RD Design

Boston Exam Schools

Identication and Estimation

Extrapolation Results

Counterfactual Simulations

Placebo Experiments

Conclusions

Preliminaries

Figure: Distribution of English and Math ability


Table: Factor loadings on ISEE and MCAS scores
Table: Dependence of outcomes on English and Math ability

Eects Away from the Admissions Cutos

80

40

Running Variable

Estimates

40

80

3
2
1

Mean Score

2
1

Mean Score

2
1
1

Mean Score

Latin School

Latin Academy

OBryant

80

40

Running Variable

40

80

80

40

Running Variable

40

80

Eects Away from the Admissions Cutos

80

40

Running Variable

Estimates

40

80

3
2
1

Mean Score

2
1

Mean Score

2
1
1

Mean Score

Latin School

Latin Academy

OBryant

80

40

Running Variable

40

80

80

40

Running Variable

40

80

Exam School vs Traditional BPS

I Previous results about incremental eects of attending a


better exam school

I Latent factor model provides counterfactual predictions for all


exam schools

I This allows one to study the eects of attending a given exam


school versus a traditional BPS

Exam Schoolextrapolation_all
vs Traditional BPS: All Applicants
O'Bryant
(1)
0.754***
(0.020)

Latin
Academy
(2)
0.948***
(0.013)

Latin
School
(3)
0.968***
(0.016)

Reduced
Form

0.021
(0.037)

0.049
(0.058)

0.024
(0.064)

LATE

0.027
(0.049)

0.052
(0.062)

0.025
(0.066)

First
Stage

N
3,704
* significant at 10%, ** significant at 5%, ***
significant at 1%

Exam School vs Traditional BPS: By Oer Status


extrapolation_atu_att

First
Stage

No Exam School Offer


Latin
Latin
O'Bryant Academy
School
(1)
(2)
(3)
0.887*** 0.987*** 0.955***
(0.049)
(0.030)
(0.026)

Exam School Offer


Latin
Latin
O'Bryant Academy
School
(4)
(5)
(6)
0.758*** 0.925*** 0.987***
(0.017)
(0.010)
(0.005)

Reduced
Form

0.334***
(0.088)

0.429***
(0.105)

0.428***
(0.094)

-0.268***
(0.066)

-0.300***
(0.048)

-0.348***
(0.052)

LATE

0.376***
(0.102)

0.435***
(0.110)

0.448***
(0.098)

-0.353***
(0.087)

-0.325***
(0.053)

-0.353***
(0.053)

N
1,777
* significant at 10%, ** significant at 5%, *** significant at 1%
Updated: 2014-01-13

1,927

Exam School vs Traditional BPS: By Oer Status


extrapolation_atu_att

First
Stage

No Exam School Offer


Latin
Latin
O'Bryant Academy
School
(1)
(2)
(3)
0.887*** 0.987*** 0.955***
(0.049)
(0.030)
(0.026)

Exam School Offer


Latin
Latin
O'Bryant Academy
School
(4)
(5)
(6)
0.758*** 0.925*** 0.987***
(0.017)
(0.010)
(0.005)

Reduced
Form

0.334***
(0.088)

0.429***
(0.105)

0.428***
(0.094)

-0.268***
(0.066)

-0.300***
(0.048)

-0.348***
(0.052)

LATE

0.376***
(0.102)

0.435***
(0.110)

0.448***
(0.098)

-0.353***
(0.087)

-0.325***
(0.053)

-0.353***
(0.053)

N
1,777
* significant at 10%, ** significant at 5%, *** significant at 1%
Updated: 2014-01-13

1,927

Exam School vs Traditional BPS: By Oer Status


extrapolation_atu_att

First
Stage

No Exam School Offer


Latin
Latin
O'Bryant Academy
School
(1)
(2)
(3)
0.887*** 0.987*** 0.955***
(0.049)
(0.030)
(0.026)

Exam School Offer


Latin
Latin
O'Bryant Academy
School
(4)
(5)
(6)
0.758*** 0.925*** 0.987***
(0.017)
(0.010)
(0.005)

Reduced
Form

0.334***
(0.088)

0.429***
(0.105)

0.428***
(0.094)

-0.268***
(0.066)

-0.300***
(0.048)

-0.348***
(0.052)

LATE

0.376***
(0.102)

0.435***
(0.110)

0.448***
(0.098)

-0.353***
(0.087)

-0.325***
(0.053)

-0.353***
(0.053)

N
1,777
* significant at 10%, ** significant at 5%, *** significant at 1%
Updated: 2014-01-13

1,927

Introduction

Latent Factor Modeling in a Sharp RD Design

Boston Exam Schools

Identication and Estimation

Extrapolation Results

Counterfactual Simulations

Placebo Experiments

Conclusions

Counterfactual Simulations
I Simulate eects of two admissions reforms:

1.

Introducing minority preferences


I
I

2.

(Boston 1975-1998)

65% of seats assigned among all applicants


35% of seats assigned among blacks and Hispanics

Introducing socioeconomic preferences


I
I

(Chicago 2010-)

30% of seats assigned among all applicants


70% of seats assigned within four SES tiers

I Exam school assignment of 27-35% of applicants aected


I
I
I

Table: Actual and counterfactual assignments


Table: Counterfactual admissions cutos
Table: Descriptives by counterfactual assignment

I Caveats (outside the scope of this paper):


I

Application behavior, teacher behavior, peer eects

Eect of Amative Action on Average Achievement


cfsim12_effects

All
Applicants

Minority Preferences
Applicant Group
All
NonApplicants Minority
Minority
(1)
(2)
(3)
0.023*** 0.027***
0.017*
(0.004)
(0.007)
(0.009)
3,704
2,086
1,618

Affected
Applicants

All
Applicants
(4)
0.015***
(0.005)
3,704

0.062*** 0.074***
0.046*
0.050***
(0.011)
(0.019)
(0.025)
(0.016)
1,367
754
613
1,100
* significant at 10%, ** significant at 5%, *** significant at 1%
Updated: 2014-01-13

Socioeconomic Preferences
Applicant Group
SES
SES
SES
Tier 1
Tier 2
Tier 3
(5)
(6)
(7)
0.024**
0.011*
0.006
(0.011)
(0.006)
(0.005)
737
876
901

SES
Tier 4
(8)
0.018*
(0.010)
1,190

0.068**
(0.031)
265

0.054*
(0.029)
404

0.044*
(0.023)
223

0.026
(0.024)
208

Eect of Amative Action on Average Achievement


cfsim12_effects

All
Applicants

Minority Preferences
Applicant Group
All
NonApplicants Minority
Minority
(1)
(2)
(3)
0.023*** 0.027***
0.017*
(0.004)
(0.007)
(0.009)
3,704
2,086
1,618

Affected
Applicants

All
Applicants
(4)
0.015***
(0.005)
3,704

0.062*** 0.074***
0.046*
0.050***
(0.011)
(0.019)
(0.025)
(0.016)
1,367
754
613
1,100
* significant at 10%, ** significant at 5%, *** significant at 1%
Updated: 2014-01-13

Socioeconomic Preferences
Applicant Group
SES
SES
SES
Tier 1
Tier 2
Tier 3
(5)
(6)
(7)
0.024**
0.011*
0.006
(0.011)
(0.006)
(0.005)
737
876
901

SES
Tier 4
(8)
0.018*
(0.010)
1,190

0.068**
(0.031)
265

0.054*
(0.029)
404

0.044*
(0.023)
223

0.026
(0.024)
208

Eect of Amative Action on Average Achievement


cfsim12_effects

All
Applicants

Minority Preferences
Applicant Group
All
NonApplicants Minority
Minority
(1)
(2)
(3)
0.023*** 0.027***
0.017*
(0.004)
(0.007)
(0.009)
3,704
2,086
1,618

Affected
Applicants

All
Applicants
(4)
0.015***
(0.005)
3,704

0.062*** 0.074***
0.046*
0.050***
(0.011)
(0.019)
(0.025)
(0.016)
1,367
754
613
1,100
* significant at 10%, ** significant at 5%, *** significant at 1%
Updated: 2014-01-13

Socioeconomic Preferences
Applicant Group
SES
SES
SES
Tier 1
Tier 2
Tier 3
(5)
(6)
(7)
0.024**
0.011*
0.006
(0.011)
(0.006)
(0.005)
737
876
901

SES
Tier 4
(8)
0.018*
(0.010)
1,190

0.068**
(0.031)
265

0.054*
(0.029)
404

0.044*
(0.023)
223

0.026
(0.024)
208

Introduction

Latent Factor Modeling in a Sharp RD Design

Boston Exam Schools

Identication and Estimation

Extrapolation Results

Counterfactual Simulations

Placebo Experiments

Conclusions

Placebo Experiments

Strategy:
1. Split data on applicants with assignment

Z = 0, 1, 2, 3

in half

based on the median of the running variable


2. Re-estimate latent outcome models to the left and right of the
placebo cutos
3. Repeat reduced form extrapolations to the left and right of the
placebo cutos

Placebo Extrapolation: Figures

2
0

25

50

50

25

Running Variable

Running Variable

Latin Academy

Latin School

25

50

25

50

2
1
0
1

Mean Score

25

50

Mean Score

Mean Score

1
1

Mean Score

OBryant

No Offer

50

25

25

50

Running Variable

50

25

Running Variable

Fitted

Extrapolated

Placebo Extrapolation:
Estimates
placebo

All
Applicants

Below
Placebo
Cutoff

No
Offer
(1)
-0.057
(0.064)
1,777

O'Bryant
(2)
0.080
(0.068)
563

Latin
Academy
(3)
-0.041
(0.062)
625

Latin
School
(4)
-0.008
(0.045)
793

-0.133
(0.099)
887

0.027
(0.061)
280

-0.028
(0.055)
311

0.027
(0.049)
368

Above
0.020
0.133
-0.054
-0.044
Placebo
(0.056)
(0.102)
(0.097)
(0.068)
Cutoff
890
283
314
371
* significant at 10%, ** significant at 5%, *** significant at 1%

Introduction

Latent Factor Modeling in a Sharp RD Design

Boston Exam Schools

Identication and Estimation

Extrapolation Results

Counterfactual Simulations

Placebo Experiments

Conclusions

Conclusions

I Develop a latent factor-based approach to the extrapolation of


treatment eects away from the cuto in RD

I Achievement gains from exam school attendance concentrated


among lower-scoring applicants

I Armative action predicted to increase average achievement


among aected applicants

I Latent factor-based extrapolation likely to be a promising


approach also in various other RD designs

Identication with Less than Three Measures

I One noisy measure enough for the identication of f ,M if


I
I
I

M1 = + M1
and M1 independent
either f or fM1 known

(Carroll & Hall, 1988; Carrasco & Florens, 2011)


I Two noisy measures enough for the identication of f ,M if
I
I

M1 = + M1 , M2 = + M2
, M1 , and M2 jointly independent

(Kotlarski, 1967; Evdokimov & White, 2012)


Back

Parametric Illustration: Details

E [M1 ] =
E [Mk ] = Mk + Mk , k = 2, 3
Cov [M1 , Mk ] = Mk 2 , k = 2, 3
Cov [M2 , M3 ] = M2 M3 2
Var [M1 ] = 2 + 2M
2

Var [Mk ] = Mk + 2M , k = 2, 3
k

Back

Parametric Illustration: Details





E Y | M = m10 , D = 0 = 0 + 0 E | M = m10 , D = 0




E Y | M = m20 , D = 0 = 0 + 0 E | M = m20 , D = 0




E Y | M = m11 , D = 1 = 1 + 1 E | M = m11 , D = 1




E Y | M = m21 , D = 1 = 1 + 1 E | M = m21 , D = 1
Back

Extrapolation of LATE in Fuzzy RD


Suppose
1.

(Y (0) , Y (1) , D (0) , D (1)) R |

2.

P [D (1) D (0) | ] = 1

a.s.

3.

P [D (1) > D (0) | ] > 0

a.s.

Then

E [Y (1) Y (0) | D (1) > D (0) , R = r ]


E {E [Y (D (1)) Y (D (0)) | ] | R = r }
=
E {E [D (1) D (0) | ] | R = r }
for all

Back

r R

Settings with Multiple Latent Factors


I Suppose the data contains 2 K
factors

+1

noisy measures of latent

1 , . . . , K :
M1k
M2k



= gM k k , M k , k = 1, . . . , K
1
1


= gM k k , M k , k = 1, . . . , K
2

M3 = gW (1 , . . . , K , M3 )
M1k
I

M2k , k = 1, . . . , K ,

M3

potentially discrete

is a deterministic function of a subset of

M = (M1 , M2 , M3 )

and

continuous,

I Extending all of the identication results to this setting


requires only slight modications

Back

Deferred Acceptance Algorithm


I Round 1: Applicants are considered for a seat in their most
preferred exam school. Each exam schools rejects the
lowest-ranking applicants in excess of its capacity. The rest of
the applicants are provisionally admitted.

I Round

k > 1:

Applicants rejected in Round

k 1

are

considered for a seat in their next most preferred exam school.


Each exam schools considers these applicants together with
the provisionally admitted applicants from Round

k 1

and

rejects the lowest-ranking students in excess of its capacity.


The rest of the students are provisionally admitted.

I The algorithm terminates once either each applicants are


assigned an oer from one of the exam schools or all
unmatched applicant are rejected by every exam school in their
preference ordering.

Back

Sharp Sample
I Three ways for an applicant to be admitted to exam school

(given the school-specic cutos)

1. Exam school s is the applicant's 1st choice, and she clears the
admissions cuto.
2. The applicant does not clear the admissions cuto for her 1st
choice, exam school s is her 2nd choice, and she clears the
admissions cuto.
3. The applicant does not clear the admissions cuto for her 1st
or 2nd choice, exam school s is her 3rd choice, and she clears
the admissions cuto.
I This forms the basis for the denition of a sharp sample for
each exam school
I

Back

Applicants who obtain an oer from exam school s i they


rank higher than the school-specic cuto
A given applicant can be in multiple sharp samples

Descriptive Statistics
descriptives
Exam School Assignment
All

All

No

BPS

Applicants

Offer

O'Bryant

Latin

Latin

Academy

School

(1)

(2)

(3)

(4)

(5)

(6)

Female

0.489

0.545

0.516

0.579

0.581

0.577

Black

0.516

0.399

0.523

0.396

0.259

0.123

Hispanic

0.265

0.189

0.223

0.196

0.180

0.081

FRPL

0.755

0.749

0.822

0.788

0.716

0.499

LEP

0.116

0.073

0.109

0.064

0.033

0.004

Bilingual

0.315

0.387

0.353

0.420

0.451

0.412

SPED

0.227

0.043

0.073

0.009

0.006

0.009

English 4

0.000

0.749

0.251

0.870

1.212

1.858

Math 4

0.000

0.776

0.206

0.870

1.275

2.114

21,094

5,179

2,791

755

790

843

Updated: 2013-11-08
Back

Measurement Scatterplots

3
2
1
1
2
2

MCAS English 4

Math

2
1
2

ISEE Quantitative

2
1

ISEE Math

2
1
0
1

ISEE Quantitative

MCAS English 4

.
3

ISEE Verbal

2
2

ISEE Verbal

2
1
2

ISEE Reading

2
1
0

ISEE Reading

1
2

ISEE Math

English

MCAS Math 4

MCAS Math 4

2
0
2

MCAS English 4

Measurement Scatterplots (cont'd)

MCAS Math 4

Back

Measurement Correlations
measurement_correlations
ISEE

MCAS

Reading

Verbal

Math

Quantitative

English 4

Math 4

(1)

(2)

(3)

(4)

(5)

(6)

Panel A: ISEE
Reading

0.735

0.631

0.621

0.670

0.581

Verbal

0.735

0.619

0.617

0.655

0.587

Math

0.631

0.619

0.845

0.598

0.740

Quantitative

0.621

0.617

0.845

0.570

0.718

English 4

0.670

0.655

0.598

0.570

0.713

Math 4

0.581

0.587

0.740

0.718

0.713

1.000

Panel B: MCAS

N
Updated: 2013-11-08
Back

5,179

Eects at therd_estimates_all
Admissions Cutos: Estimates
O'Bryant

Latin

Latin

Academy

School

(1)

(2)

(3)

First

0.781***

0.955***

0.964***

Stage

(0.034)

(0.018)

(0.018)

Reduced

0.047

-0.021

-0.086

Form

(0.055)

(0.044)

(0.052)

LATE

0.060

-0.022

-0.089

(0.070)

(0.046)

(0.054)

1,475

1,999

907

* significant at 10%; ** significant at 5%; ***


significant at 1%
Back

Marginal Density
.2
.4

.6

Marginal Distribution of English Ability

2
English Ability

.1

Marginal Density
.2
.3

.4

.5

Marginal Distribution of Math Ability

2
Math Ability

English Ability
1
2

Scatterplot of English and Math Ability

2
Math Ability

Back

Factor Loadings on ISEE and MCAS Scores


factor_loadings
ISEE

Reading
(1)
E

1.160***
(0.033)

0.081***
(0.020)

MCAS
Math
Quantitative
English 4
Math 4
(3)
(4)
(5)
(6)
Panel A: Factor Loading on Mean
1.180***
1
(0.035)
1.135***
1.119***
1
(0.027)
(0.028)
Verbal
(2)

Panel B: Factor Loading on (Log) Standard Deviation


0.152***
0.016
(0.018)
(0.016)
0.019
-0.013
(0.018)
(0.017)

N
5,179
* significant at 10%, ** significant at 5%, *** significant at 1%
Updated: 2014-01-13
Back

0.124***
(0.015)

Dependencelatent_fs_rf
of Outcomes on English and Math Ability
No
Offer
(1)
E
M

E
M

-0.001
(0.011)
-0.012
(0.008)

Latin
O'Bryant Academy
(2)
(3)
Panel A: First Stage
-0.025
-0.102*
(0.076)
(0.053)
-0.059
0.012
(0.061)
(0.030)

Latin
School
(4)
0.003
(0.009)
0.027**
(0.012)

0.446***
(0.062)
0.604***
(0.056)

Panel B: Reduced Form


0.282*** 0.217***
(0.067)
(0.066)
0.346*** 0.267***
(0.070)
(0.057)

0.239***
(0.046)
0.158***
(0.044)

563
N
1,777
625
793
* significant at 10%, ** significant at 5%, *** significant at 1%
Back

Estimates: Allrd_extrapolation_all
Applicants

First
Stage

O'Bryant
(1)
0.858***
(0.038)

Latin
Academy
(2)
0.962***
(0.021)

Latin
School
(3)
0.950***
(0.018)

Reduced
Form

0.252***
(0.068)

0.279***
(0.075)

0.199***
(0.061)

LATE

0.293***
(0.081)

0.290***
(0.079)

0.209***
(0.064)

N
2,240
2,760
3,340
* significant at 10%, ** significant at 5%, ***
significant at 1%

Estimates: Below/Above Admissions Cutos


rd_extrapolation_cutoff

First
Stage

Below Admissions Cutoff


Latin
Latin
O'Bryant Academy
School
(1)
(2)
(3)
0.892*** 0.984*** 0.959***
(0.050)
(0.027)
(0.021)

Above Admissions Cutoff


Latin
Latin
O'Bryant Academy
School
(4)
(5)
(6)
0.749*** 0.891*** 0.906***
(0.016)
(0.018)
(0.031)

Reduced
Form

0.343***
(0.089)

0.362***
(0.092)

0.281***
(0.075)

-0.021
(0.034)

-0.005
(0.046)

-0.091*
(0.054)

LATE

0.385***
(0.102)

0.367***
(0.097)

0.293***
(0.079)

-0.028
(0.045)

-0.006
(0.053)

-0.099*
(0.059)

625

739

N
1,677
2,135
2,601
563
* significant at 10%, ** significant at 5%, *** significant at 1%
Updated: 2014-01-13
Back

Socioeconomic Tiers
I Census tracts are given a socioeconomic index based on

1.
2.
3.
4.

Median family income


Percent of households occupied by the owner
Percent of families headed by a single parent
Percent of households where a language other than English is
spoken
5. Educational attainment score
I Each component is turned into a percentile and added up to
get the socioeconomic index

I BPS students divided into four tiers based on the quartiles of


the socioeconomic index distribution

Back

Observedcfsim12_assignments
and Counterfactual Assignments
Actual Assignment
No
Counterfactual
Assignment

Offer
(1)

Latin

Latin

O'Bryant

Academy

School

(2)

(3)

(4)

Panel A: Minority Preferences


No Offer

2418

221

113

39

O'Bryant

280

129

133

213

Latin Academy

88

389

268

45

Latin School

16

276

546

Panel B: Socioeconomic Preferences


No Offer

2579

159

39

O'Bryant

14

203

319

146

87

Latin Academy

106

403

272

Latin School

171

202

470

Back Updated:

2013-11-08

cfsim12_cutoffs

Counterfactual Admissions Cutos


Latin

Latin

O'Bryant

Academy

School

(1)

(2)

(3)

Panel A: Minority Preferences


Minority

-14.1

-20.6

-31.9

Non-Minority

15.8

12.4

7.8

Panel B: Socioeconomic Preferences


SES Tier 1

-20.4

-26.4

-32.9

SES Tier 2

-6.6

-11.7

-16.1

SES Tier 3

-2.5

-7.2

-17.1

SES Tier 4

8.0

2.1

-5.1

Notes: This table reports the differences between the actual


Back

Descriptives by Counterfactual Assignment


cfsim12_descriptives
No
Offer
(1)

Latin

Latin

O'Bryant

Academy

School

(2)

(3)

(4)

Panel A: Minority Preferences


Female

0.502

0.588

0.629

0.572

Black

0.430

0.440

0.386

0.274

Hispanic

0.182

0.220

0.203

0.172

FRPL

0.810

0.771

0.715

0.556

LEP

0.116

0.030

0.022

0.018

Bilingual

0.386

0.396

0.380

0.391

SPED

0.073

0.009

0.013

0.005

English 4

0.277

0.981

1.215

1.668

Math 4

0.277

0.963

1.289

1.781

Female

0.514

0.572

0.597

0.575

Black

0.499

0.360

0.295

0.203

Hispanic

0.223

0.191

0.143

0.120

FRPL

0.813

0.728

0.673

0.624

LEP

0.107

0.058

0.030

0.017

Bilingual

0.359

0.423

0.391

0.445

SPED

0.073

0.012

0.006

0.006

English 4

0.261

1.067

1.309

1.556

Math 4

0.217

1.146

1.365

1.744

2,791

755

790

843

Panel B: Minority Preferences

Back

Updated: 2013-11-08

You might also like