Transcript Do Judges Vary in their Treatment of Race?

Do Judges Vary in Their
Treatment of Race?
David Abrams (U of Chicago)
Marianne Bertrand (U of Chicago)
Sendhil Mullainathan (Harvard)
June 5, 2007
Research Questions

Does the legal system discriminate?

Are African-Americans more likely to be
incarcerated?

Do they receive longer sentences?
Standard Approach
sentenceijt = α + βraceijt + Xijt + εijt
jailijt = α + βraceijt + Xijt + εijt

Problem: Race is not randomly assigned, so
betas may be biased due to unobservables!
Our Approach
Use random assignment of cases to judges to
answer a related question:
Do judges vary in their treatment of race?
sentenceijt = α + βraceijt + Xijt + ΣδjDj +
ΣγjDj*raceijt + mot + εijt
Test for equality of the γj
Why is this interesting?

Large variance of sentencing disparities may have
negative implications for perceptions of fairness of
judicial system

Could also help explain different findings in different
studies

Legally important

Would such variation violate constitutional rights?
No State shall…deny to any person within its jurisdiction the
equal protection of the laws. (14th Amendment)
Objectives

Test that cases are randomly assigned to judges

Establish counterfactual where judges don’t
vary in treatment of race

For Both: Use Monte Carlo Simulation


Allows for small cell sizes
Allows for skewed Bernoulli variables
Monte Carlo Simulation




Use for both test of random assignment and
heterogeneity in racial gap
Create cells at the month level
Simulate each observation 500 times, draw
simulated data from same cell, with replacement.
For inter-judge heterogeneity in racial gap in
sentencing:

Create cells at the month-race level
Data Description-Chicago Data

Circuit Court of Cook County



Assignment procedure:




Largest unified court system in the country
Main Chicago location handles 85% of cases
Daily assignment of cases uses random number generator
Exceptions include drugs, murder, some sex crimes
Suburban court locations perform their own random
assignment
Data includes all felony cases from 1985-2004



Over 500,000 cases
Includes charge(s), judge(s), defendant characteristics, plea,
disposition, sentence
We use small subset of the data
Case Characteristics
Standard
Mean Deviation Minimum Median Maximum
Sample: Dataset only with African Americans and Whites
Number of Charges
Disposition (Guilty =1)
Plea
Length of Incarceration (Months)
Length of Incarceration (Conditional on Non-Zero)
Fraction African American
Fraction Female
Age
Probation
Incarceration (Dummy)
Judges
Cases
2.4
0.92
0.69
20
42
0.86
0.17
29
0.25
0.49
80
34298
5.1
0.27
0.46
36
42
0.35
0.38
10
0.44
0.5
1
0
0
0
0.032
0
0
16
0
0
1
1
1
0
36
1
0
27
0
1
266
1
1
720
720
1
1
89
1
1
Case types and outcomes
Standard
Mean Deviation Minimum Median Maximum
By Type of Charge
Drugs
Violent Crime
Embezzlement, Fraud, Theft
Other
Sentence Length by Type of Charge
Drugs
Violent Crime
EFT
Other
Sentence Length by Race
African American
White
Sentence Length on Non-Zero by Race
African American
White
Incarceration by Race
African American
White
Cases/Judge
Judges
Cases
0.39
0.16
0.19
0.26
0.49
0.37
0.39
0.44
0
0
0
0
0
0
0
0
1
1
1
1
15
24
23
24
22
43
31
48
0
0
0
0
0
0
3
0
360
480
360
720
21
16
36
33
0
0
0.19
0
720
420
42
42
41
43
0.032
0.032
36
36
720
420
0.51
0.38
432
0.5
0.48
419
0
0
10
1
0
281
1
1
1308
80
34298
Bootstrapping (I)
Testing for Random Assignment
Judge
Wapner
Judy
Dredd
Real Data
Case # Date
Race
1001 1/1/2000 Black
1414 1/15/2000 White
…
3141 1/5/2000 Black
6789 3/12/2000 White
…
2718 1/20/2000 Black
8765 2/29/2000 Black
Simulation 1
Race
Black
Black
Simulation …
Race
White
Black
Black
White
Black
Black
White
Black
Black
White
25%
25%
75%
75%
Random Assignment Checks

Race
Gender
Age
Total Number of Charges
Charge Type

Also use 10%-90% and 5%-95% ranges




Bootstrapping (II)
Testing for Racial gap Heterogeneity
Judge
Wapner
Judy
Dredd
Case #
1001
1414
…
3141
6789
…
2718
8765
Date
Race
1/1/2000 Black
1/15/2000 White
Real Data Simulation 1 Simulation …
Sent. Length Sent. Length Sent. Length
666
30
0
365
1/5/2000 Black
3/12/2000 White
30
3650
7300
0
1/20/2000 Black
2/29/2000 Black
7300
10500
1095
0
Numerical Implications

How much does the sentencing gap between black
and white defendants vary across judges?
Judge Percentile
shift
25%-75%
Change in
Change in
Incarceration Gap Sentence Length
(months)
.11
3
10%-90%
.18
10
Robustness Check

Perhaps race is a proxy for other characteristics, (such
as charge) that receive heterogeneous treatment by
judges.

Address this concern by looking at subsets of the data




Drug crimes
EFT
Violent
Other
Interpretations
Evidence of “too much heterogeneity” in
Chicago incarceration data need not imply
discrimination against Blacks.



Suppose the “appropriate” gap is 25%
Variation in judges between 15 and 25% would be
a sign of reverse discrimination
Possible approaches to deal with this issue in
the future:

Recidivism?
Focus on Restricted Sample

Eliminate “drug” judges



Keep only central location


Can expand to other locations, but central location accounts
for 85% of cases
Restrict to initial appearance of defendant


8 judges receive only drug cases
Overflow drug cases are randomized among remaining
judges
Subsequent appearances often assigned to the same judge
Restrict to judges known to have been regular judges
under current judicial administration (through
consultation with presiding judge’s office)