# RECORD 0462b0d87ab72b3c90375cd7f1b22600ca9530ef
# TITLE: Sorting or Steering: The Effects of Housing Discrimination on Neighborhood Choice
# YEAR: 2018.0  VENUE: nan  DOI: 10.3386/w24826
# AUTHORS: P. Christensen; Christopher Timmins
# PAGES: 70  CANDIDATE TABLE PAGES (1-based): [39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 59, 60, 61, 63, 65, 66, 67, 69, 70]
# ABSTRACT: Housing discrimination is illegal. However, paired-tester audit experiments have revealed evidence of discrimination in the interactions between potential buyers and real estate agents, raising concern about whether certain groups are systematically excluded from the beneficial effects of healthy neighborhoods. Using data from HUD’s most recent Housing Discrimination Study and micro-level data on key attributes of neighborhoods in 28 US cities, we find strong evidence of discrimination in the characteristics of neighborhoods towards which individuals are steered. Conditional upon the characteristics of the house suggested by the audit tester, minorities are significantly more likely to be steered towards neighborhoods with less economic opportunity and greater exposures to crime and local pollutants. We find that holding locational preferences or income constant, discriminatory steering alone can contribute substantially to the disproportionate number of minority households found in high poverty neighborhoods in the United States. The steering effect is also large enough to fully explain the differential in proximity to Superfund sites among African American mothers. These results have important implications for studies of “neighborhood effects” and confirm an important mechanism underlying observed correlations between race and pollution in the environmental justice literature. Our results also suggest that the basic utility maximization assumptions underlying hedonic and residential sorting models may often be violated, resulting in an important distortion in the provision of local public goods.


===== PAGE 1 (score 2, numeric-paren 0, table-words 0, var-words 2) =====

NBER WORKING PAPER SERIES
SORTING OR STEERING:
THE EFFECTS OF HOUSING DISCRIMINATION ON NEIGHBORHOOD CHOICE
Peter Christensen
Christopher Timmins
Working Paper 24826
http://www.nber.org/papers/w24826
NATIONAL BUREAU OF ECONOMIC RESEARCH
1050 Massachusetts Avenue
Cambridge, MA 02138
July 2018, Revised June 2021
Previously circulated as “Sorting or Steering: Experimental Evidence on the Economic Effects of 
Housing Discrimination.” We thank Spencer Banzhaf, Kelly Bishop, Don Fullerton, Ludovica 
Gazze, Lu Han, Kelly Maguire, David Molitor, Steve Ross, Kerry Smith, Matthew Turner, Randy 
Walsh, Ann Wolverton, John Yinger and seminar participants at the Wharton School, University 
of Arizona, University of Pittsburgh, Simon Fraser University, Washington University in St. 
Louis, 2019 meetings of the American Economic Association, the 2018 meetings of the 
Association of Urban Economics, the 2018 China Meetings of the Econometrics Society, and the 
2018 meetings of the Southern Economic Association for excellent comments. This project 
makes use of data from an experiment conducted by the Department of Housing and Urban 
Development and the Urban Institute in 2012 (Turner et al., 2013). All HDS 2012 data and 
documentation 
are 
publicly 
available 
on 
https://www.huduser.gov/portal/datasets/hsg 
discrimination.html. All data and programs used in this paper will be made available on: https://
github.com/peterchristensen/Sorting-or-Steering. Thanks to research assistants in the University 
of Illinois Big Data and Environmental Economics and Policy (BDEEP) Group and the Duke 
Environmental Justice Lab for excellent work. We acknowledge generous support from the 
National Center for Supercomputing Applications. All errors are our own.
NBER working papers are circulated for discussion and comment purposes. They have not been 
peer-reviewed or been subject to the review by the NBER Board of Directors that accompanies 
official NBER publications.
© 2018 by Peter Christensen and Christopher Timmins. All rights reserved. Short sections of text, 
not to exceed two paragraphs, may be quoted without explicit permission provided that full 
credit, including © notice, is given to the source.


===== PAGE 39 (score 23, numeric-paren 0, table-words 2, var-words 5) =====

Table 1. Descriptive Statistics for Tester Dataset
Tester Characteristics
(True Actors, Not Assigned)
Variable
All
White
African
Hispanic
Asian
Groups
American
Age
40.910
41.485
41.494
42.070
36.398
Percent Male
0.388
0.415
0.375
0.284
0.420
Percent Rented Home
0.588
0.580
0.630
0.516
0.614
Percent Owned Home
0.305
0.296
0.312
0.421
0.217
Personal Income
Under $10,000
0.244
0.275
0.241
0.125
0.249
$10,000 - $19,999
0.164
0.130
0.225
0.114
0.235
$20,000 - $29,999
0.237
0.293
0.136
0.249
0.184
$30,000 - $39,999
0.136
0.114
0.152
0.183
0.152
$40,000 - $49,999
0.096
0.086
0.105
0.117
0.090
$50,000 - $74,999
0.068
0.062
0.069
0.136
0.022
$75,000 - $99,999
0.009
0.013
0.008
0.000
0.004
$100,000 or more
0.003
0.002
0.004
0.011
0.000
Education
Attended High School
0.003
0.000
0.002
0.000
0.018
GED
0.012
0.021
0.004
0.000
0.000
High School Diploma
0.035
0.020
0.045
0.037
0.079
Attended Vocational School
0.005
0.005
0.000
0.018
0.000
Vocational School Diploma
0.022
0.032
0.006
0.007
0.029
Attended College
0.162
0.130
0.182
0.308
0.108
Associate’s Degree
0.078
0.059
0.101
0.136
0.054
Bachelor’s Degree
0.358
0.404
0.283
0.183
0.484
Attended Graduate School
0.052
0.074
0.020
0.048
0.025
Graduate Degree
0.243
0.230
0.308
0.253
0.170
Assigned Characteristics
Monthly Rent
1, 332
1, 334
1, 392
1, 277
1, 285
Percent Tester Went First
0.516
0.589
0.417
0.454
0.462
Percent Appointment in AM
0.416
0.410
0.425
0.428
0.422
Percent Car Owner
0.854
0.861
0.796
0.922
0.861
Length of Employment (Years)
4.091
3.345
4.936
5.203
4.507
Years at Residence
3.650
2.925
4.405
4.766
4.101
Lease Type
Month-to-Month
0.569
0.558
0.565
0.581
0.620
Lease
0.431
0.442
0.435
0.419
0.380
N = 2,260
N = 1,161
N = 512
N = 286
N = 294
Note: Table reports the mean values of the innate (actor) and the assigned (scripted) characteristics of testers in the
study. Column 1 reports the mean for all testers in the study and columns 2-4 report the mean for testers in a given race
group. Testers were matched based on approximately similar actor (not assigned) characteristics and then tester pairs
were matched to assigned roles. Assigned incomes vary by trial and were calculated as a function of the listing price of an
advertised home.
37


===== PAGE 40 (score 23, numeric-paren 0, table-words 2, var-words 10) =====

Table 2. Home and Neighborhood Characteristics
Characteristics of
Advertised Homes
Variable
All Groups
White
African American
Hispanic
Asian
Listing Price
378, 983
381, 064
344, 169
387, 648
408, 701
Building Type
Single family, detached
0.739
0.744
0.738
0.699
0.749
Duplex
0.017
0.017
0.013
0.018
0.018
Rowhouse or Townhouse
0.134
0.128
0.118
0.158
0.154
Multi-family structure
0.101
0.101
0.125
0.112
0.069
Mobile home
0.001
0.001
0
0.005
0
Neighborhood Characteristics
School-Specific Test Scores
Elem. School Test Score
0.278
0.279
0.206
0.344
0.340
Midd. School Test Score
0.173
0.164
0.185
0.116
0.247
Housing Search Platform
Assaults
93.336
88.352
96.150
95.346
104.878
Elem. School Quality
6.210
6.203
6.202
6.169
6.260
ACS Block Group Characteristics
Poverty Rate
0.089
0.088
0.092
0.090
0.086
Percent High Skill
0.467
0.467
0.467
0.463
0.472
Percent College Graduate
0.501
0.504
0.501
0.494
0.501
Single-Parent HH
0.146
0.146
0.141
0.155
0.147
Ownership Rate
0.731
0.735
0.731
0.715
0.728
ACS Block Group Racial Composition
Percent White
0.675
0.685
0.684
0.639
0.664
Percent African American
0.090
0.090
0.103
0.082
0.084
Percent Hispanic
0.148
0.140
0.134
0.185
0.151
Percent Asian
0.067
0.065
0.061
0.073
0.079
Pollution Measurements
Superfund Sites
0.327
0.311
0.337
0.353
0.352
Risk-Screening Environmental Indicators (RSEI)
6, 076.8
6, 252.5
7, 865.3
4, 283.2
5, 131.2
PM 25
9.283
9.378
9.253
9.056
9.293
N = 7,026
N = 3,607
N = 1,212
N = 1,028
N = 1,108
Note: Table reports the mean values of the advertised listings assigned to testers in the study, which are based on sampling
frames generated for each metropolitan market. The composition of trials within a given market is defined to match the
racial composition of the market, such that the characteristics of advertised listings can vary across the race groups (though
they will not vary within a trial). Column 1 reports the mean characteristics for all testers in the study and columns 2-4
report the mean for testers in a given race group.
38


===== PAGE 41 (score 111, numeric-paren 84, table-words 2, var-words 6) =====

Table 3. Balance Statistics for Testers
Tester Characteristics
(True Actors, Not Assigned)
Variable
African
Hispanic
Asian
American
Age
-0.874
1.476
-2.060
(0.996)
(1.858)
(1.833)
Percent Male
0.009
0.000
0.000
(0.015)
(0.000)
(0.000)
Percent Rented Home
0.138
-0.044
0.083
(0.100)
(0.124)
(0.114)
Percent Owned Home
-0.153∗
0.033
-0.111
(0.093)
(0.113)
(0.110)
Personal Income
Under $10,000
-0.012
-0.107
-0.000
(0.097)
(0.116)
(0.125)
$10,000 - $19,999
0.118
0.012
0.235∗∗
(0.076)
(0.116)
( 0.102)
$20,000 - $29,999
-0.212∗∗
-0.066
-0.113
(0.086)
(0.151)
(0.107)
$30,000 - $39,999
0.126
0.147∗
0.008
(0.081)
(0.088)
(0.114)
$40,000 - $49,999
-0.002
0.011
-0.066
(0.063)
(0.079)
(0.077)
$50,000 - $74,999
-0.023
0.006
-0.090
(0.057)
(0.065)
(0.080)
$75,000 - $99,999
-0.028
-0.027
-0.012
(0.022)
(0.041)
(0.018)
$100,000 or more
0.000
0.030
0.000
(0.000)
(0.046)
(0.000)
Education
Attended High School
0.000
0.000
0.020
0.000
0.000
0.031
GED
0.011
-0.071
-0.028
(0.017)
(0.050)
(0.044)
High School Diploma
0.036
0.041
0.116∗
(0.035)
(0.037)
(0.068)
Attended Vocational School
-0.007
0.015
0.000
(0.011)
(0.034)
(0.000)
Vocational School Diploma
-0.014
-0.002
-0.012
(0.030)
(0.035)
(0.018)
Attended College
0.100
0.109
0.082
(0.080)
(0.138)
(0.103)
Associate’s Degree
-0.017
0.088
-0.016
(0.043)
(0.066)
(0.052)
Bachelor’s Degree
-0.161
-0.251∗
-0.032
(0.110)
(0.147)
(0.104)
Attended Graduate School
-0.062
-0.043
-0.036
(0.054)
(0.075)
(0.032)
Graduate Degree
0.106
0.114
-0.088
(0.107)
(0.120)
( 0.113)
Assigned Characteristics
Percent Tester Went First
-0.093
-0.231
-0.047
(0.165)
(0.200)
(0.209)
Percent Appointment in AM
0.002
-0.014
0.028
(0.041)
(0.038)
(0.038)
Percent Car Owner
0.010
0.000
-0.065
(0.015)
(0.000)
(0.058)
Length of Employment (Years)
1.559∗∗∗
2.050∗∗∗
1.444∗∗∗
(0.270)
(0.328)
(0.238)
Years at Residence
1.511∗∗∗
1.898∗∗∗
1.551∗∗∗
(0.288)
(0.301)
(0.203)
Lease Type
Month-to-Month
0.012
0.072
0.115
(0.047)
(0.091)
(0.076)
Lease
-0.012
-0.072
-0.115
(0.047)
(0.091)
(0.076)
N = 512
N = 286
N = 294
Note: Table reports the results of balance tests that compare tester characteristics from a given group in a trial to the
mean for the white (comparison) group. These tests are based on a simplified version of model 3 that includes a term
for race group and a trial fixed effect, but omits other controls (tester and advertised home attributes). The top panel
reports differences in innate (actor) characteristics, whereas the bottom panel reports differences in assigned (scripted)
characteristics. Significance levels are: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01.
39


===== PAGE 42 (score 86, numeric-paren 59, table-words 2, var-words 11) =====

Table 4. Balance Statistics for Homes
Characteristics of
Advertised Homes
Variable
African
Hispanic
Asian
American
Listing Price
-397.595
-388.358
-1, 866.292
(960.555)
(1,288.875)
(1,136.40)
Building Type
Single-family Detached
0.009
0.006
0.029***
(0.010)
(0.013)
(0.011)
Duplex
-0.001
-0.002
-0.005
(0.005)
(0.005)
(0.008)
Rowhouse or Townhouse
0.005
0.007
0.005
(0.010)
(0.011)
(0.015)
Multi-family Structure
-0.006
-0.006
-0.029**
(0.008)
(0.009)
(0.012)
Mobile Home
-0.001
-0.001
-0.00000
0.001
0.003
0.00000
Neighborhood Characteristics
School-Specific Test Scores
Elem. School Test Score
0.012**
0.009
-0.006
(0.006)
(0.007)
(0.006)
Midd. School Test Score
-0.010
-0.007
-0.005
(0.006)
(0.005)
(0.005)
Housing Search Platform
Assault
-0.040
0.109
-0.153
(0.174)
(1.069)
(0.173)
Elem. School Quality
0.001
0.019
0.025
(0.021)
(0.022)
(0.034)
ACS Block Group Characteristics
Poverty Rate
0.0005
0.001
-0.001
(0.002)
(0.002)
(0.001)
Percent High Skill
-0.005
-0.002
0.0004
(0.003)
(0.003)
(0.002)
Percent College Educated
-0.005**
-0.003
0.0002
(0.002)
(0.004)
(0.002)
Single-Parent HH
0.002
0.002
-0.001
(0.002)
(0.002)
(0.002)
ACS Block Group Racial Composition
Percent African American
0.003
0.004
-0.003
(0.002)
(0.003)
(0.002)
Percent Hispanic
-0.002
0.003
0.001
(0.001)
(0.002)
(0.001)
Percent Asian
-0.001
-0.0002
-0.001
(0.001)
(0.001)
(0.001)
Pollution Measurements
Superfund Sites
-0.0002
0.012
0.001
(0.005)
(0.011)
(0.006)
Risk-Screening Environmental Indicators (RSEI)
-30.108
-108.290
89.371
(51.410)
(102.645)
(141.098)
PM 25
-0.010
0.0001
0.015
(0.011)
(0.016)
(0.016)
Note: Table reports the results of balance tests that compare the mean values of home characteristics assigned to a given
race group to the mean for the white (comparison) group. These tests are based on a simplified version of model 3 that
includes a term for race group and a trial fixed effect, but omits other controls (tester and advertised home attributes).
The top panel reports differences in property characteristics within trials, which could occur if the advertised home was
removed from the market during a trial. The bottom panels report differences in neighborhood characteristics. Significance
levels are: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01.
40


===== PAGE 43 (score 37, numeric-paren 16, table-words 5, var-words 8) =====

Table 5. Differences in Recommendations and Availability of Advertised Properties
Dependent variable:
Number of Recommendations
Home Availability
Racial Minority
−0.1282
−0.1419
0.0048
0.0057
(0.1990)
(0.1987)
(0.0183)
(0.0183)
African American
−0.1608
−0.1690
−0.0097
−0.0087
(0.2713)
(0.2707)
(0.0220)
(0.0219)
Hispanic
−0.1340
−0.1304
−0.0090
−0.0077
(0.2465)
(0.2474)
(0.0255)
(0.0258)
Asian
0.1231
0.0833
0.0174
0.0178
(0.2477)
(0.2465)
(0.0227)
(0.0227)
ln(Price) Advert Home
N
Y
N
Y
Racial Comp Advert Home
N
Y
N
Y
Observations
6,580
6,555
6,588
6,562
Adjusted R2
-0.2192
-0.2348
-0.1687
-0.1792
Note: Coefficients report differences in the number of homes recommended and number of available homes recommended
to minority testers relative to a white tester (the omitted category). Regressions are estimated at the tester level using
data for the testers from the total sample of 6,962 that report these variables. Racial minority encompasses all individual
minority groups. All regression specifications include trial fixed effects and control for the full set of actor characteristics,
assigned characteristics, and search characteristics. Actor characteristics: tester income, tester household income, gender of
tester, age of tester, month of test, educational attainment of tester; Assigned characteristics: household members, current
home ownership status, current lease type, car ownership status, reason for moving, years in current residence, length of
employment at current job, reason tester can afford down payment, current lease assigned to tester; Search Characteristics:
month of test, sequence of tester appointments, time of the appointment (am/pm), type of recommended building. Standard
errors are clustered by trial, which is the level of randomization. Significance levels are: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01.
41


===== PAGE 44 (score 43, numeric-paren 20, table-words 5, var-words 8) =====

Table 6. Discriminatory Steering and Neighborhood Racial Composition
Dependent variable: White Household Share
I
II
III
IV
V
Racial Minority
−0.0089
−0.0089
−0.0091
−0.0096
−0.0096
(0.0112)
(0.0112)
(0.0112)
(0.0112)
(0.0112)
African American
−0.0399∗∗∗
−0.0398∗∗∗
−0.0403∗∗∗
−0.0409∗∗∗
−0.0410∗∗∗
(0.0148)
(0.0148)
(0.0148)
(0.0147)
(0.0147)
Hispanic
0.0028
0.0028
0.0028
0.0025
0.0025
(0.0152)
(0.0152)
(0.0152)
(0.0152)
(0.0152)
Asian
0.0182
0.0182
0.0181
0.0176
0.0176
(0.0136)
(0.0136)
(0.0136)
(0.0136)
(0.0136)
Comparison Mean (White)
0.69
0.69
0.69
0.69
0.69
Share White Advert Home
N
Y
Y
Y
Y
ln(Price) Advert Home
N
N
Y
Y
Y
Racial Comp Advert Home
N
N
N
Y
Y
Poverty Share Advert Home
N
N
N
N
Y
Observations
21,904
21,893
21,871
21,871
21,871
Adjusted R2
0.7772
0.7772
0.7766
0.7766
0.7765
Note: Coefficients report differences in the racial composition of neighborhoods recommended to minority testers relative
to a white tester (the omitted category).
Racial minority encompasses all individual minority groups.
All regression
specifications include trial fixed effects and control for the full set of actor characteristics, assigned characteristics, and
search characteristics. Actor characteristics: tester income, tester household income, gender of tester, age of tester, month
of test, educational attainment of tester; Assigned characteristics: household members, current home ownership status,
current lease type, car ownership status, reason for moving, years in current residence, length of employment at current
job, reason tester can afford down payment, current lease assigned to tester; Search characteristics: month of test, sequence
of tester appointments, and time of the appointment (am/pm). Standard errors are clustered by trial, which is the level of
randomization. Significance levels are: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01.
42


===== PAGE 45 (score 31, numeric-paren 12, table-words 4, var-words 8) =====

Table 7. Discriminatory Steering and Neighborhood Racial Composition by Income
Dep. Variable: White Household Share by Income
High Inc
Mid Inc
Low Inc
Racial Minority
−0.0265∗∗∗
−0.0036
0.0173∗∗∗
(0.0081)
(0.0076)
(0.0062)
African American
−0.0338∗∗∗
−0.0187∗
0.0099
(0.0104)
(0.0100)
(0.0081)
Hispanic
−0.0145
−0.0038
0.0201∗∗
(0.0108)
(0.0100)
(0.0084)
Asian
−0.0246∗∗∗
0.0155∗
0.0217∗∗∗
(0.0095)
(0.0092)
(0.0077)
Comparison Mean (White)
0.21
0.339
0.146
Share White Advert Home
N
Y
Y
ln(Price) Advert Home
N
N
Y
Racial Comp Advert Home
N
N
N
Poverty Share Advert Home
N
N
N
Observations
21,470
21,470
21,470
Adjusted R2
0.7178
0.7064
0.6665
Note: Coefficients report differences in the racial composition (by income level) of neighborhoods recommended to minority
testers relative to a white tester (the omitted category). Racial minority encompasses all individual minority groups. All
regression specifications include trial fixed effects and control for the full set of actor characteristics, assigned characteristics,
and search characteristics. Actor characteristics: tester income, tester household income, gender of tester, age of tester,
month of test, educational attainment of tester; Assigned characteristics: household members, current home ownership
status, current lease type, car ownership status, reason for moving, years in current residence, length of employment at
current job, reason tester can afford down payment, current lease assigned to tester; Search characteristics: month of test,
sequence of tester appointments, and time of the appointment (am/pm). Standard errors are clustered by trial, which is
the level of randomization. Significance levels are: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01.
43


===== PAGE 46 (score 88, numeric-paren 54, table-words 7, var-words 10) =====

Table 8. Discriminatory Steering and Neighborhood Effects
School-Specific Test Scores:
Housing Search Platform:
Elem School
Middle School
Assaults
Elem School
Racial Minority
-0.0447**
-0.0328*
8.3591
-0.3424*
(0.0227)
(0.0178)
(6.4744)
(0.1824)
African American
-0.0856**
-0.0891***
23.1146**
-0.4727**
(0.0424)
(0.0242)
(9.8939)
(0.2379)
Hispanic
-0.0056
0.0100
-0.4643
-0.2462
(0.0349)
(0.0294)
(8.0628)
(0.2423)
Asian
-0.0442
-0.0287
5.1582
-0.3756*
(0.0380)
(0.0237)
(7.6687)
(0.2195)
Comparison Mean (White)
0.29
0.18
80.85
6.43
ln(Price) Advert Home
Y
Y
Y
Y
Racial Comp Advert Home
Y
Y
Y
Y
Outcome Advertised Home
Y
Y
Y
Y
Observations
9,360
9,713
11,161
11,032
Adjusted R2
0.7625
0.9128
0.7971
0.7475
American Community Survey
Poverty Rate
High Skill
College
Single-Parent HH
Ownership Rate
Racial Minority
0.0116**
-0.0254**
-0.0182*
0.0077
-0.0170
(0.0054)
(0.0099)
(0.0101)
(0.0062)
(0.0142)
African American
0.0120
-0.0302**
-0.0340***
0.0167*
-0.0263
(0.0075)
(0.0130)
(0.0127)
(0.0087)
(0.0184)
Hispanic
0.0205***
-0.0387***
-0.0216*
0.0157*
0.0059
(0.0076)
(0.0130)
(0.0127)
(0.0084)
(0.0203)
Asian
-0.0009
-0.0072
0.0029
-0.0061
-0.0232
(0.0069)
(0.0128)
(0.0142)
(0.0085)
(0.0176)
Comparison Mean (White)
0.082
0.49
0.53
0.14
0.74
ln(Price) Advert Home
Y
Y
Y
Y
Y
Racial Comp Advert Home
Y
Y
Y
Y
Y
Outcome Advertised Home
Y
Y
Y
Y
Y
Observations
21,871
21,871
21,871
21,871
21,871
Adjusted R2
0.5205
0.6925
0.7548
0.5237
0.6024
p-values
0.01
0.033
0.073
0.21
0.23
hochberg q-values
0.052
0.13
0.22
0.23
0.23
Note: Coefficients report differences in the attributes of properties/neighborhoods recommended to minority testers relative
to a white tester (the omitted category).
Racial minority encompasses all individual minority groups.
All regression
specifications include trial fixed effects and control for the full set of actor characteristics, assigned characteristics, and
search characteristics. Actor characteristics: tester income, tester household income, gender of tester, age of tester, month
of test, educational attainment of tester; Assigned characteristics: household members, current home ownership status,
current lease type, car ownership status, reason for moving, years in current residence, length of employment at current job,
reason tester can afford down payment, current lease assigned to tester; Search Characteristics: month of test, sequence of
tester appointments, and time of the appointment (am/pm). Standard errors are clustered by trial, which is the level of
randomization. Significance levels are: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01.
44


===== PAGE 47 (score 52, numeric-paren 24, table-words 7, var-words 9) =====

Table 9. Discriminatory Steering and Local Pollution Exposures
Pollution
Superfund
Toxics
PM
Racial Minority
0.1010∗∗∗
765.9137∗∗
−0.0438
(0.0325)
(324.9800)
(0.0504)
African American
0.1256∗∗∗
872.9819∗∗
−0.0252
(0.0465)
(403.4626)
(0.0678)
Hispanic
0.0591
459.7283
0.0102
(0.0376)
(544.8236)
(0.0638)
Asian
0.1046∗∗
899.5148∗∗
−0.1243∗∗
(0.0484)
(424.3961)
(0.0617)
Comparison Mean (White)
0.36
5100
9.4
ln(Price) Advert Home
Y
Y
Y
Racial Comp Advert Home
Y
Y
Y
Outcome Advertised Home
Y
Y
Y
Observations
21,883
21,883
21,883
Adjusted R2
0.8744
0.6939
0.9634
Pollution
Superfund
Toxics
PM
Racial Minority
0.1444∗∗∗
1, 110.5330
0.0082
(0.0496)
(691.5594)
(0.1091)
African American
0.1978∗∗∗
2, 387.1960∗∗
0.0836
(0.0750)
(992.5587)
(0.1470)
Hispanic
0.0276
1, 361.6980
0.2007
(0.0677)
(853.3277)
(0.1223)
Asian
0.0808
509.9719
−0.0378
(0.0627)
(858.7148)
(0.1472)
Comparison Mean (White)
0.38
6300
9.3
ln(Price) Advert Home
Y
Y
Y
Racial Comp Advert Home
Y
Y
Y
Outcome Advertised Home
Y
Y
Y
Observations
8,016
8,016
8,016
Adjusted R2
0.8508
0.5968
0.9672
Note: Coefficients report differences in the pollution levels at the location of properties recommended to minority testers
relative to a white tester (the omitted category). Racial minority encompasses all individual minority groups. All regression
specifications include trial fixed effects and control for the full set of actor characteristics, assigned characteristics, and search
characteristics. Actor characteristics: tester income, tester household income, gender of tester, age of tester, month of test,
educational attainment of tester; Assigned characteristics: household members, current home ownership status, current
lease type, car ownership status, reason for moving, years in current residence, length of employment at current job,
reason tester can afford down payment, current lease assigned to tester; Search Characteristics: month of test, sequence of
tester appointments, and time of the appointment (am/pm). Standard errors are clustered by trial, which is the level of
randomization. Significance levels are: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01.
45


===== PAGE 48 (score 95, numeric-paren 63, table-words 7, var-words 9) =====

Table 10. Discriminatory Steering and Neighborhood Effects (Mothers)
School-Specific Test Scores:
Housing Search Platform:
Elem School
Middle School
Assaults
Elem School
Racial Minority
-0.1471***
-0.0979***
39.4893***
-3.0893***
(0.0483)
(0.0278)
(14.0657)
(0.9501)
African American
-0.2231**
-0.2289***
79.7645***
-3.0262***
(0.0954)
(0.0564)
(12.7280)
(0.7800)
Hispanic
-0.2374***
-0.1214***
29.3393
-1.9212**
(0.0729)
(0.0364)
(18.4929)
(0.9598)
Asian
-0.0543
-0.1027**
22.2020
-3.8178***
(0.1265)
(0.0452)
(25.4432)
(0.9020)
Comparison Mean (White)
0.33
0.28
84
6.5
ln(Price) Advert Home
Y
Y
Y
Y
Racial Comp Advert Home
Y
Y
Y
Y
Outcome Advertised Home
Y
Y
Y
Y
Observations
3,806
3,811
4,589
4,521
Adjusted R2
0.7624
0.8925
0.8344
0.6992
American Community Survey
Poverty Rate
High Skill
College
Single-Parent HH
Ownership Rate
Racial Minority
0.0453***
-0.0347**
-0.0424***
0.0001
-0.0950***
(0.0102)
(0.0144)
(0.0163)
(0.0125)
(0.0242)
African American
0.0510***
-0.0622***
-0.0769***
0.0264
-0.1659***
(0.0147)
(0.0210)
(0.0234)
(0.0164)
(0.0329)
Hispanic
0.0557***
-0.0530***
-0.0483***
0.0197
-0.1021***
(0.0121)
(0.0171)
(0.0179)
(0.0139)
(0.0323)
Asian
0.0356**
-0.0081
-0.0432*
-0.0163
-0.0389
(0.0155)
(0.0206)
(0.0232)
(0.0164)
(0.0313)
Comparison Mean (White)
0.08
0.48
0.52
0.14
0.77
ln(Price) Advert Home
Y
Y
Y
Y
Y
Racial Comp Advert Home
Y
Y
Y
Y
Y
Outcome Advertised Home
Y
Y
Y
Y
Y
Observations
8,015
8,015
8,015
8,015
8,015
Adjusted R2
0.4561
0.6605
0.7395
0.4826
0.4863
Note: Coefficients report differences in the attributes of properties/neighborhoods recommended to minority testers that are
assigned the role of a mother (female gender with children in household) relative to a white tester with the same assigned role
(the omitted category). Racial minority encompasses all individual minority groups. All regression specifications include
trial fixed effects and control for the full set of actor characteristics, assigned characteristics, and search characteristics.
Actor characteristics: tester income, tester household income, gender of tester, age of tester, month of test, educational
attainment of tester; Assigned characteristics: household members, current home ownership status, current lease type, car
ownership status, reason for moving, years in current residence, length of employment at current job, reason tester can afford
down payment, current lease assigned to tester; Search Characteristics: month of test, sequence of tester appointments, and
time of the appointment (am/pm). Standard errors are clustered by trial, which is the level of randomization. Significance
levels are: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01.
46


===== PAGE 49 (score 38, numeric-paren 18, table-words 4, var-words 8) =====

Table 11. Discriminatory Steering: Low Poverty Neighborhoods
Dependent variable:
Low Poverty
Low Poverty: Families
Low Poverty: Moms
Low Pov/High Dad
Low Pov/High Dad: Families
Low Pov/High Dad: Moms
African American
−0.1179∗∗
−0.1730∗∗∗
−0.4318∗∗∗
−0.1175∗∗
−0.1705∗∗∗
−0.3958∗∗∗
(0.0464)
(0.0579)
(0.1268)
(0.0478)
(0.0588)
(0.1296)
Hispanic
−0.1585∗∗∗
−0.1986∗∗∗
−0.4283∗∗∗
−0.1543∗∗∗
−0.1657∗∗∗
−0.3358∗∗
(0.0425)
(0.0534)
(0.1114)
(0.0429)
(0.0518)
(0.1308)
Asian
−0.0437
−0.0770
−0.3101∗∗
−0.0419
−0.0757
−0.2618∗∗
(0.0445)
(0.0603)
(0.1282)
(0.0445)
(0.0610)
(0.1275)
Observations
7,566
4,643
2,794
7,524
4,639
2,790
Adjusted R2
0.3967
0.3959
0.4248
0.3954
0.3941
0.4205
Note: Coefficients report differences in the attributes of neighborhoods recommended to minority testers relative to a white tester (the omitted category). Racial minority encompasses all
individual minority groups. All regression specifications include trial fixed effects and control for the full set of actor characteristics, assigned characteristics, and search characteristics. Actor
characteristics: tester income, tester household income, gender of tester, age of tester, month of test, educational attainment of tester; Assigned characteristics: household members, current
home ownership status, current lease type, car ownership status, reason for moving, years in current residence, length of employment at current job, reason tester can afford down payment,
current lease assigned to tester; search Characteristics: month of test, sequence of tester appointments, and time of the appointment (am/pm). Standard errors are clustered by trial, which
is the level of randomization. Significance levels are: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01.
47


===== PAGE 50 (score 30, numeric-paren 9, table-words 5, var-words 9) =====

Table 12. Discriminatory Steering: Median Income in Neighborhood
Dependent variable: log(Median Income)
All Testers
Families
Moms
African American
−0.0729∗∗
−0.1085∗∗∗
−0.2485∗∗∗
(0.0335)
(0.0401)
(0.0738)
Hispanic
−0.0478
−0.0388
−0.0782
(0.0310)
(0.0376)
(0.0567)
Asian
−0.0126
−0.0511
−0.0773
(0.0271)
(0.0367)
(0.0583)
Comparison Mean (White)
11.405
11.46
11.45
Observations
21,457
6,369
3,772
Adjusted R2
0.6657
0.6756
0.6104
Note: Coefficients report differences in the median income of households in neighborhoods recommended to minority
testers relative to a white tester (the omitted category). Racial minority encompasses all individual minority groups. All
regression specifications include trial fixed effects and control for the full set of actor characteristics, assigned characteristics,
and search characteristics. Actor characteristics: tester income, tester household income, gender of tester, age of tester,
month of test, educational attainment of tester; Assigned characteristics: household members, current home ownership
status, current lease type, car ownership status, reason for moving, years in current residence, length of employment at
current job, reason tester can afford down payment, current lease assigned to tester; Search characteristics: month of test,
sequence of tester appointments, and time of the appointment (am/pm). Standard errors are clustered by trial, which is
the level of randomization. Significance levels are: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01.
48


===== PAGE 51 (score 92, numeric-paren 59, table-words 7, var-words 10) =====

Table 13. Discriminatory Steering by Implied Preferences for Neighborhood Attributes
School-Specific Test Scores:
Housing Search Platform:
Elem School
Middle School
Assaults
Elem School
Racial Minority
-0.0618***
-0.0326***
9.2013**
-0.2583***
(0.0131)
(0.0092)
(4.3081)
(0.0618)
African American
-0.0810***
-0.0210*
3.9675
-0.5920***
(0.0177)
(0.0122)
(5.7978)
(0.0821)
Hispanic
-0.0701***
-0.0124
12.9738**
-0.2014**
(0.0204)
(0.0138)
(6.3158)
(0.0922)
Asian
-0.0557***
-0.0510***
17.3702***
-0.0299
(0.0183)
(0.0128)
(5.9039)
(0.0826)
Comparison Mean (White)
0.02
0.016
-7.4
0.18
Observations
8,951
9,242
11,161
11,032
Adjusted R2
0.3187
0.4313
0.2657
0.3751
American Community Survey
Poverty Rate
High Skill
College
Single-Parent HH
Ownership Rate
Racial Minority
0.0012
-0.0070**
-0.0018
0.0042*
0.0042
(0.0022)
(0.0034)
(0.0035)
(0.0024)
(0.0052)
African American
-0.0085***
-0.0120***
-0.0068
0.0092***
0.0035
(0.0030)
(0.0045)
(0.0047)
(0.0032)
(0.0069)
Hispanic
0.0052
-0.0158***
0.0042
-0.0032
0.0146*
(0.0033)
(0.0049)
(0.0051)
(0.0035)
(0.0075)
Asian
0.0032
0.0134***
0.0028
0.0059*
0.0139*
(0.0031)
(0.0046)
(0.0049)
(0.0033)
(0.0072)
Comparison Mean (White)
7.6e-05
0.0025
0.0068
-0.0025
-0.0063
Observations
21,871
21,871
21,871
21,871
21,871
R2
0.3396
0.3992
0.3838
0.3643
0.3727
Adjusted R2
0.2778
0.3430
0.3262
0.3048
0.3140
Note: Coefficients report differences in the gap between implied preferences and homes recommended to minority testers
relative to a white tester (the omitted category). Dependent variable is the difference between the level of each attribute
in the advertised listing and the recommended property.
Racial minority encompasses all individual minority groups.
This specification omits trial fixed to allow identifying variation in the advertised home attribute, which is typically held
fixed within a trial (and otherwise balanced). All regression specifications control for the full set of actor characteristics,
assigned characteristics, and search characteristics. Actor characteristics: tester income, tester household income, gender of
tester, age of tester, month of test, educational attainment of tester; Assigned characteristics: household members, current
home ownership status, current lease type, car ownership status, reason for moving, years in current residence, length of
employment at current job, reason tester can afford down payment, current lease assigned to tester; Search Characteristics:
month of test, sequence of tester appointments, and time of the appointment (am/pm). Standard errors are clustered by
trial, which is the level of randomization. Significance levels are: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01.
49


===== PAGE 52 (score 56, numeric-paren 27, table-words 7, var-words 9) =====

Table 14. Discriminatory Steering and Later Transactions
Buyers Upon Sale
White
African American
Hispanic
Asian
Differentiability Score
68%
15%
87%
91%
Same Race Tester
-0.0064
-0.0019
0.0733*
0.1169**
(0.0481)
(0.0149)
(0.0442)
(0.0510)
Buyers(n)
7,598
217
1,224
1,014
Observations
9,428
9,428
9,428
9,428
Adjusted R2
0.3654
0.2035
0.3886
0.4163
Dependent variable: Logarithm of Price
I
II
III
IV
V
Racial Minority
-0.0117
-0.0116
-0.0107
-0.0106
-0.0540
(0.0336)
(0.0336)
(0.0336)
(0.0336)
(0.0371)
African American
0.0263
0.0262
0.0275
0.0276
-0.0363
(0.0404)
(0.0404)
(0.0404)
(0.0404)
(0.0445)
Hispanic
0.0307
0.0305
0.0362
0.0360
-0.0004
(0.0664)
(0.0665)
(0.0671)
(0.0674)
(0.0699)
Asian
-0.0899
-0.0897
-0.0896
-0.0892
-0.1101*
(0.0561)
(0.0561)
(0.0560)
(0.0561)
(0.0592)
Comparison Mean (White)
0.66
0.66
0.66
0.66
0.66
Share White Advert Home
N
Y
Y
Y
Y
ln(Price) Advert Home
N
N
Y
Y
Y
Racial Comp Advert Home
N
N
N
Y
Y
Poverty Share Advert Home
N
N
N
N
Y
Year
Y
Y
Y
Y
Y
Month of Year
Y
Y
Y
Y
Y
Observations
10,862
10,829
10,808
10,808
10,808
Adjusted R2
0.8624
0.8621
0.8599
0.8599
0.8689
Note: Coefficients report differences in the predicted race groups of home buyers and sale prices (recorded in ZTRAX
transactions data) of properties recommended to minority testers relative to a white tester (the omitted category). Top
panel reports estimates from a regression where the outcome is the predicted racial/ethnic identity associated with a
first/last name of a ZTRAX home buyer. Differentiability score is calculated as the difference in predicted probability of
the most likely buyer race group and the second most likely buyer race group. Buyers(n) reports the number of buyers from
each predicted group that are observed in the ZTRAX transactions data. Bottom panel reports estimates from a regression
where the outcome is the transaction price of the property. Racial minority encompasses all individual minority groups. All
regression specifications include trial fixed effects and control for the full set of actor characteristics, assigned characteristics,
and search characteristics. Actor characteristics: tester income, tester household income, gender of tester, age of tester,
month of test, educational attainment of tester; Assigned characteristics: household members, current home ownership
status, current lease type, car ownership status, reason for moving, years in current residence, length of employment at
current job, reason tester can afford down payment, current lease assigned to tester; Search characteristics: month of test,
sequence of tester appointments, and time of the appointment (am/pm). Standard errors are clustered by trial, which is
the level of randomization. Significance levels are: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01.
50


===== PAGE 53 (score 0, numeric-paren 0, table-words 0, var-words 0) =====

Figure 1. Markets in 2012 HUD Buyer Experiment
Note: Red points identify the location of 28 markets utilized in 2012 HDS study. See Turner et al. (2013) for detailed
discussion of sampling design and methodology.
51


===== PAGE 59 (score 66, numeric-paren 32, table-words 9, var-words 8) =====

C
Steering and Neighborhood Racial Composition
This section replicates estimates of steering by neighborhood racial composition reported
as white shares in Tables 6 and 7 using shares of each minority group.
Table C.1. Differential Steering and Neighborhood Racial Composition:
African American Share
Dependent variable: African American Household Share
I
II
III
IV
V
Racial Minority
0.0155∗∗
0.0156∗∗
0.0156∗∗
0.0157∗∗
0.0156∗∗
(0.0073)
(0.0073)
(0.0073)
(0.0073)
(0.0072)
African American
0.0388∗∗∗
0.0389∗∗∗
0.0390∗∗∗
0.0393∗∗∗
0.0396∗∗∗
(0.0107)
(0.0107)
(0.0107)
(0.0107)
(0.0107)
Hispanic
0.0126
0.0126
0.0126
0.0126
0.0126
(0.0086)
(0.0086)
(0.0086)
(0.0086)
(0.0086)
Asian
−0.0068
−0.0068
−0.0068
−0.0067
−0.0067
(0.0095)
(0.0095)
(0.0095)
(0.0095)
(0.0094)
Comparison Mean
0.09
0.09
0.09
0.09
0.09
Share Black Advert Home
N
Y
Y
Y
Y
ln(Price) Advert Home
N
N
Y
Y
Y
Racial Comp Advert Home
N
N
N
Y
Y
Poverty Share Advert Home
N
N
N
N
Y
Observations
21,904
21,893
21,871
21,871
21,871
Adjusted R2
0.7719
0.7720
0.7693
0.7693
0.7693
Table C.2. Differential Steering and Neighborhood Racial Composition by Income:
African American Share
Dep. Variable: African American Household Share by Income
High Inc
Mid Inc
Low Inc
Racial Minority
0.0019∗
0.0097∗∗∗
0.0050
(0.0011)
(0.0036)
(0.0041)
African American
0.0051∗∗∗
0.0175∗∗∗
0.0176∗∗∗
(0.0016)
(0.0061)
(0.0058)
Hispanic
0.0005
0.0068∗
0.0055
(0.0015)
(0.0041)
(0.0049)
Asian
−0.0000
0.0030
−0.0095∗
(0.0015)
(0.0042)
(0.0056)
Comparison Mean
0.01
0.04
0.04
Share White Advert Home
N
Y
Y
ln(Price) Advert Home
N
N
Y
Racial Comp Advert Home
N
N
N
Poverty Share Advert Home
N
N
N
Observations
21,470
21,470
21,470
Adjusted R2
0.7299
0.7625
0.6706
Note: Coefficients report differences in the racial composition (by income) of neighborhoods recommended to minority
testers relative to a white tester (the omitted category). Racial minority encompasses all individual minority groups. All
regression specifications control for the full set of actor characteristics, assigned characteristics, and search Characteristics.
Actor characteristics: tester income, tester household income, gender of tester, age of tester, month of test, educational
attainment of tester; Assigned Characteristics: household members, current home ownership status, current lease type, car
ownership status, reason for moving, years in current residence, length of employment at current job, reason tester can afford
down payment, current lease assigned to tester; Search Characteristics: month of test, sequence of tester appointments, time
of the appointment (am/pm), type of recommended building, total number of homes recommended to tester, availability
of advertised home as stated by agent. Standard errors are clustered by market for consistency with sampling design.
Significance levels are: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01.
Appendix–5


===== PAGE 60 (score 65, numeric-paren 32, table-words 9, var-words 7) =====

Table C.3. Differential Steering and Neighborhood Racial Composition:
Hispanic Share
Dependent variable: Hispanic Household Share
I
II
III
IV
V
Racial Minority
0.0032
0.0033
0.0034
0.0034
0.0034
(0.0070)
(0.0070)
(0.0070)
(0.0070)
(0.0070)
African American
0.0159
0.0160
0.0162
0.0161
0.0160
(0.0103)
(0.0103)
(0.0103)
(0.0103)
(0.0103)
Hispanic
−0.0016
−0.0014
−0.0013
−0.0013
−0.0013
(0.0091)
(0.0091)
(0.0091)
(0.0091)
(0.0091)
Asian
−0.0094
−0.0092
−0.0093
−0.0093
−0.0093
(0.0088)
(0.0088)
(0.0088)
(0.0088)
(0.0088)
Comparison Mean
0.13
0.13
0.13
0.13
0.13
Share Hispanic Advert Home
N
Y
Y
Y
Y
ln(Price) Advert Home
N
N
Y
Y
Y
Racial Comp Advert Home
N
N
N
Y
Y
Poverty Share Advert Home
N
N
N
N
Y
Observations
21,904
21,893
21,871
21,871
21,871
Adjusted R2
0.8437
0.8437
0.8437
0.8437
0.8437
Table C.4. Differential Steering and Neighborhood Racial Composition by Income:
Hispanic Share
Dep. Variable: Hispanic Household Share by Income
High Inc
Mid Inc
Low Inc
Racial Minority
0.0015
−0.0010
0.0030
(0.0017)
(0.0039)
(0.0043)
African American
0.0053∗∗
0.0044
0.0054
(0.0026)
(0.0048)
(0.0072)
Hispanic
−0.0016
−0.0048
0.0050
(0.0023)
(0.0056)
(0.0056)
Asian
−0.0012
−0.0042
−0.0023
(0.0024)
(0.0050)
(0.0047)
Comparison Mean
0.03
0.07
0.03
Share White Advert Home
N
Y
Y
ln(Price) Advert Home
N
N
Y
Racial Comp Advert Home
N
N
N
Poverty Share Advert Home
N
N
N
Observations
21,470
21,470
21,470
Adjusted R2
0.6521
0.8328
0.7115
Note: Coefficients report differences in the racial composition (by income) of neighborhoods recommended to minority
testers relative to a white tester (the omitted category). Racial minority encompasses all individual minority groups. All
regression specifications control for the full set of actor characteristics, assigned characteristics, and search Characteristics.
Actor characteristics: tester income, tester household income, gender of tester, age of tester, month of test, educational
attainment of tester; Assigned Characteristics: household members, current home ownership status, current lease type, car
ownership status, reason for moving, years in current residence, length of employment at current job, reason tester can afford
down payment, current lease assigned to tester; Search Characteristics: month of test, sequence of tester appointments, time
of the appointment (am/pm), type of recommended building, total number of homes recommended to tester, availability
of advertised home as stated by agent. Standard errors are clustered by market for consistency with sampling design.
Significance levels are: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01.
Appendix–6


===== PAGE 61 (score 65, numeric-paren 32, table-words 9, var-words 7) =====

Table C.5. Differential Steering and Neighborhood Racial Composition:
Asian Share
Dependent variable: Asian Household Share
I
II
III
IV
V
Racial Minority
−0.0097∗
−0.0098∗
−0.0098∗
−0.0095∗
−0.0097∗
(0.0053)
(0.0053)
(0.0053)
(0.0053)
(0.0053)
African American
−0.0150∗∗
−0.0149∗∗
−0.0149∗∗
−0.0147∗∗
−0.0148∗∗
(0.0069)
(0.0068)
(0.0068)
(0.0068)
(0.0068)
Hispanic
−0.0137∗
−0.0140∗
−0.0140∗
−0.0136∗
−0.0136∗
(0.0079)
(0.0078)
(0.0079)
(0.0078)
(0.0078)
Asian
−0.0014
−0.0013
−0.0013
−0.0011
−0.0011
(0.0065)
(0.0065)
(0.0065)
(0.0064)
(0.0064)
Comparison Mean
0.07
0.07
0.07
0.07
0.07
Share Asian Advert Home
N
Y
Y
Y
Y
ln(Price) Advert Home
N
N
Y
Y
Y
Racial Comp Advert Home
N
N
N
Y
Y
Poverty Share Advert Home
N
N
N
N
Y
Observations
21,904
21,893
21,871
21,871
21,871
Adjusted R2
0.7433
0.7433
0.7433
0.7433
0.7432
Table C.6. Differential Steering and Neighborhood Racial Composition by Income:
Asian Share
Dep. Variable: Asian Household Share by Income
High Inc
Mid Inc
Low Inc
Racial Minority
−0.0020
−0.0036
−0.0022
(0.0019)
(0.0027)
(0.0022)
African American
−0.0033
−0.0067∗∗
−0.0039
(0.0023)
(0.0033)
(0.0029)
Hispanic
−0.0034
−0.0046
−0.0055∗
(0.0030)
(0.0041)
(0.0028)
Asian
0.0013
−0.0002
0.0011
(0.0023)
(0.0033)
(0.0027)
Comparison Mean
0.01
0.03
0.02
Share White Advert Home
N
Y
Y
ln(Price) Advert Home
N
N
Y
Racial Comp Advert Home
N
N
N
Poverty Share Advert Home
N
N
N
Observations
21,470
21,470
21,470
Adjusted R2
0.6870
0.7235
0.6258
Note: Coefficients report differences in the racial composition (by income) of neighborhoods recommended to minority
testers relative to a white tester (the omitted category). Racial minority encompasses all individual minority groups. All
regression specifications control for the full set of actor characteristics, assigned characteristics, and search Characteristics.
Actor characteristics: tester income, tester household income, gender of tester, age of tester, month of test, educational
attainment of tester; Assigned Characteristics: household members, current home ownership status, current lease type, car
ownership status, reason for moving, years in current residence, length of employment at current job, reason tester can afford
down payment, current lease assigned to tester; Search characteristics: month of test, sequence of tester appointments, time
of the appointment (am/pm), type of recommended building, total number of homes recommended to tester, availability
of advertised home as stated by agent. Standard errors are clustered by market for consistency with sampling design.
Significance levels are: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01.
Appendix–7


===== PAGE 62 (score 6, numeric-paren 0, table-words 2, var-words 2) =====

D
Comparison with Specifications Used in HDS 2012
Report
This section examines differences in steering into high poverty neighborhoods reported
in the present study versus differences documented by Turner et al. (2013) in the initial
HDS 2012 report. Replication code is not available for the steering-into-poverty tests or
racial composition test from the HDS 2012 report. We therefore construct tests that are
as similar as possible to the description provided in Turner et al. (2013) using publicly
available data files that contain census tract measures of poverty rates and white shares
used by the authors.
Table D.1 reports estimates of differences for African Americans relative to white
testers using Census tract-level data (Columns 1-2) and tester averages (Columns 3-4)
from the prior report. We compare results on white shares and poverty rates using the
primary specification with block group measures used in the present study (Table 8) to
comparable specifications with/without controls that use census tract measures (Columns
1-2) and specifications that aggregate recommendations by tester and also use census tract
measures (Columns 3-4). The tests reported in the final two columns are the most similar
to those reported in Turner et al. (2013) though there are differences in the samples and
we stress that this is an approximate and not an exact replication. We note a substantial
number of null values in RHGEO dataset, which contains the census tract data used in
Turner et al. (2013). This reduces the sample size relative to the tester dataset that is
merged with block group measures. Both sets of results indicate that the power and the
magnitude of effects diminishes as a result of aggregation to the trial level and to the
census tract. The choice of controls also affects estimate magnitudes, but less so.
In both cases, estimates of differences become substantially smaller and tests less
powered when we measure outcomes at the census tract level and aggregate to the level of
a tester recommendation set (comparing differences in tester means). We find a difference
of 2.8% [0.87] for differences in white shares when the test is constructed in this way,
which is larger but not statistically different from the the 1.8% [0.8] difference reported
in Turner et al. (2013). Differences in poverty rates fall by an order of magnitude and
become statistically non-significant.
Using this specification, we estimate that homes
recommended to African American testers have a .27% [0.29] higher poverty share than
those recommended to a white tester. This is very similar to the difference of .2% [0.2]
that is documented in the HDS report. Variation in poverty rates between block groups
appears to be particularly diminished when using Census tract measures. For instance,
while the tract-level poverty rate (8.5%) is similar to the block group-level poverty rate
(8.6%) for the average home in the HDS sample, the standard deviation of neighborhood
poverty rates for the set of homes within a given trial is substantially smaller when
using the tract-level measure of poverty: 3.79% at the tract level vs. 5.02% at the block
group level. The standard deviation of block group-level poverty rates within the average
census tract where recommendations are made in the study also appears to be quite
large: 3%. When considering that the average neighborhood level poverty rate is 8.6%49,
it becomes clear that using tract-level measures may mask considerable heterogeneity in
the neighborhoods that testers are steered into.
49The average neighborhood level poverty rate is 8.5% when measured at the tract level.
Appendix–8


===== PAGE 63 (score 31, numeric-paren 8, table-words 7, var-words 7) =====

Table D.2. Differential Steering and White Share
Dep. Variable: White Share
Tract
Tract
Tester Mean
Tester Mean
African American
−0.0315∗
−0.0358∗∗
−0.0280∗∗∗
−0.0294∗∗
(0.0166)
(0.0164)
(0.0087)
(0.0135)
Poverty Rate Advert Home
N
Y
N
Y
Racial Comp Advert Home
N
Y
N
Y
ln(Price) Advert Home
N
Y
N
Y
Observations
15,270
14,342
498
480
Adjusted R2
0.8320
0.8293
0.7598
0.7492
Dep. Variable: Poverty Rate
Tract
Tract
Tester-Tract
Tester-Tract
African American
0.0093
0.0077
0.0027
0.0027
(0.0076)
(0.0075)
(0.0032)
(0.0029)
Poverty Rate Advert Home
N
Y
N
Y
Racial Comp Advert Home
N
Y
N
Y
ln(Price) Advert Home
N
Y
N
Y
Observations
15,269
14,341
498
480
Adjusted R2
0.5877
0.5926
0.6721
0.7071
Note: Coefficients report differences in the poverty rates in neighborhoods recommended to minority testers relative to a
white tester (the omitted category). Columns 1-2 report estimates from tests that measure poverty rates at the census
Tract level whereas columns 3-4 report estimates using tester averages (data come from (Turner et al., 2013). All regression
specifications control for the full set of actor characteristics, assigned characteristics, and search Characteristics. Actor
characteristics: tester income, tester household income, gender of tester, age of tester, month of test, educational attainment
of tester; Assigned Characteristics: household members, current home ownership status, current lease type, car ownership
status, reason for moving, years in current residence, length of employment at current job, reason tester can afford down
payment, current lease assigned to tester; Search characteristics: month of test, sequence of tester appointments, time
of the appointment (am/pm), type of recommended building, total number of homes recommended to tester, availability
of advertised home as stated by agent. Standard errors are clustered by market for consistency with sampling design.
Significance levels are: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01.
Appendix–9


===== PAGE 64 (score 3, numeric-paren 0, table-words 1, var-words 1) =====

E
Steering and Neighborhood Effects: Controls
This section compares three alternate specifications for the main tests reported in Table
8. The first column under each attribute reports estimates from specifications that drop
controls for the attributes of advertised homes (buyer preferences). Column 2 reports
estimates from the preferred specification in the paper. Estimates do not change when
the implied preferences of buyers are added as controls, indicating that these do not
appear to affect differences in recommendations made to minority buyers relative to a
white counterpart. Column 3 adds a control for the average level of the particular neigh-
borhood attribute being considered among recommended homes within a tight interval
of price (within intervals of $20,000). Reported estimates of differences do not change
with the addition of this control, suggesting that minorities are steered into homes with
worse attributes than others at a similar price point and that the neighborhood steer-
ing behavior documented in this study is not explained by differential steering of buyers
into neighborhoods that are simply more affordable. This evidence is consistent with the
findings reported in Appendix F.
Appendix–10


===== PAGE 65 (score 283, numeric-paren 211, table-words 13, var-words 11) =====

Table E.1. Discriminatory Steering and Neighborhood Effects: Robustness
Elem School Test Scores
Middle School Test Scores
Assaults
Racial Minority
-0.0358*
-0.044**
-0.049**
-0.043**
-0.033*
-0.036**
7.7983
8.3591
3.4936
(0.0210)
(0.0227)
(0.0222)
(0.0179)
(0.0178)
(0.0176)
(6.0232)
(6.4744)
(6.5735)
African American
-0.0625
-0.086**
-0.095**
-0.01***
-0.09***
-0.09***
19.213**
23.115**
17.600*
(0.0385)
(0.0424)
(0.0401)
(0.0239)
(0.0242)
(0.0238)
(7.7954)
(9.8939)
(10.0867)
Hispanic
0.0013
-0.0056
-0.0121
-0.0010
0.0100
0.0096
0.3826
-0.4643
-4.7086
(0.0338)
(0.0349)
(0.0353)
(0.0291)
(0.0294)
(0.0291)
(8.0997)
(8.0628)
(8.3707)
Asian
-0.0409
-0.0442
-0.0448
-0.0312
-0.0287
-0.0336
2.3480
5.1582
0.2606
(0.0374)
(0.0380)
(0.0387)
(0.0240)
(0.0237)
(0.0236)
(8.3593)
(7.6687)
(7.9009)
Comparison Mean (White)
0.29
0.29
0.29
0.18
0.18
0.18
80.85
80.85
80.85
Outcome Advert Home
N
Y
Y
N
Y
Y
N
Y
Y
Racial Comp Advert Home
N
Y
Y
N
Y
Y
N
Y
Y
ln(Price) Advert Home
N
Y
Y
N
Y
Y
N
Y
Y
Outcome by Price Bin
N
N
Y
N
N
Y
N
N
Y
Observations
10,004
9,360
9,297
10,258
9,713
9,652
14,385
11,161
11,101
Adjusted R2
0.7690
0.7625
0.7647
0.9139
0.9128
0.9149
0.7763
0.7971
0.8000
Elem School
Poverty Rate
High Skill
Racial Minority
-0.2266
-0.3424*
-0.3465*
0.0115**
0.0116**
0.0125**
-0.025**
-0.025**
-0.02**
(0.1622)
(0.1824)
(0.1802)
(0.0055)
(0.0054)
(0.0054)
(0.0100)
(0.0099)
(0.0094)
African American
-0.43**
-0.47**
-0.54**
0.0114
0.0120
0.0123
-0.03**
-0.03**
-0.025**
(0.2035)
(0.2379)
(0.2322)
(0.0076)
(0.0075)
(0.0075)
(0.0131)
(0.0130)
(0.0124)
Hispanic
-0.1012
-0.2462
-0.2282
0.02***
0.02***
0.02***
-0.04***
-0.04***
-0.03**
(0.2246)
(0.2423)
(0.2462)
(0.0076)
(0.0076)
(0.0076)
(0.0132)
(0.0130)
(0.0127)
Asian
-0.0679
-0.38*
-0.3707
-0.0005
-0.0009
0.0006
-0.0070
-0.0072
-0.0042
(0.1962)
(0.2195)
(0.2255)
(0.0070)
(0.0069)
(0.0069)
(0.0129)
(0.0128)
(0.0120)
Comparison Mean (White)
6.43
6.43
6.43
0.082
0.082
0.082
0.49
0.49
0.49
Outcome Advert Home
N
Y
Y
N
Y
Y
N
Y
Y
Racial Comp Advert Home
N
Y
Y
N
Y
Y
N
Y
Y
ln(Price) Advert Home
N
Y
Y
N
Y
Y
N
Y
Y
Outcome by Price Bin
N
N
Y
N
N
Y
N
N
Y
Observations
14,401
11,032
10,984
21,904
21,871
21,744
21,904
21,871
21,744
Adjusted R2
0.7409
0.7475
0.7495
0.5200
0.5205
0.5302
0.6927
0.6925
0.7013
College
Single-Parent HH
Ownership Ratel
Racial Minority
-0.02*
-0.02*
-0.02*
0.007
0.0077
0.0086
-0.0159
-0.0170
-0.0186
(0.0101)
(0.0101)
(0.0099)
(0.0062)
(0.0062)
(0.0062)
(0.0142)
(0.0142)
(0.0139)
African American
-0.04***
-0.03***
-0.04***
0.015*
0.017*
0.017*
-0.0255
-0.0263
-0.0251
(0.0127)
(0.0127)
(0.0126)
(0.0088)
(0.0087)
(0.0088)
(0.0184)
(0.0184)
(0.0178)
Hispanic
-0.0207
-0.0212*
-0.023*
0.0141
0.016*
0.018**
0.0085
0.0059
0.0039
(0.0126)
(0.0127)
(0.0124)
(0.0086)
(0.0084)
(0.0087)
(0.0202)
(0.0203)
(0.0202)
Asian
0.0027
0.0029
0.0024
-0.0065
-0.0061
-0.0056
-0.0234
-0.0232
-0.0259
(0.0142)
(0.0142)
(0.0138)
(0.0086)
(0.0085)
(0.0085)
(0.0177)
(0.0176)
(0.0174)
Comparison Mean (White)
0.53
0.53
0.53
0.14
0.14
0.14
0.74
0.74
0.74
Outcome Advert Home
N
Y
Y
N
Y
Y
N
Y
Y
Racial Comp Advert Home
N
Y
Y
N
Y
Y
N
Y
Y
ln(Price) Advert Home
N
Y
Y
N
Y
Y
N
Y
Y
Outcome by Price Bin
N
N
Y
N
N
Y
N
N
Y
Observations
21,904
21,871
21,744
21,904
21,871
21,744
21,904
21,871
21,744
Adjusted R2
0.7550
0.7548
0.7610
0.5236
0.5237
0.5310
0.6022
0.6024
0.6077
Superfund
Toxics
PM
Racial Minority
0.10***
0.10***
0.11***
779.06**
765.91**
682.7*
-0.0438
-0.0438
-0.0264
(0.0326)
(0.0325)
(0.0332)
(326.3205)
(324.9800)
(359.3615)
(0.0504)
(0.0504)
(0.0531)
African American
0.13***
0.13***
0.13***
891.8**
872.9**
690.97
-0.025
-0.025
-0.022
(0.0467)
(0.0465)
(0.0457)
(409.0238)
(403.4626)
(468.2560)
(0.0678)
(0.0678)
(0.0690)
Hispanic
0.0578
0.0591
0.0603
471.3270
459.7283
456.4245
0.0063
0.0102
0.0457
(0.0381)
(0.0376)
(0.0392)
(546.7939)
(544.8235)
(559.8852)
(0.0639)
(0.0638)
(0.0684)
Asian
0.10**
0.11**
0.09*
909.50**
899.51**
885.51*
-0.12**
-0.12**
-0.11
(0.0484)
(0.0484)
(0.0480)
(429.1031)
(424.3961)
(473.3210)
(0.0618)
(0.0617)
(0.0649)
Comparison Mean (White)
0.36
0.36
0.36
5100
5100
5100
9.4
9.4
9.4
Outcome Advert Home
N
Y
Y
N
Y
Y
N
Y
Y
Racial Comp Advert Home
N
Y
Y
N
Y
Y
N
Y
Y
ln(Price) Advert Home
N
Y
Y
N
Y
Y
N
Y
Y
Outcome by Price Bin
N
N
Y
N
N
Y
N
N
Y
Observations
21,916
21,883
21,756
21,916
21,883
21,756
21,916
21,883
21,756
Adjusted R2
0.8745
0.8744
0.8777
0.6941
0.6939
0.6945
0.9635
0.9634
0.9641
Note: Coefficients report differences in the attributes of neighborhoods recommended to minority testers relative to a
white tester (the omitted category) using different sets of controls.
The 3rd specification for each variable contains a
fixed effect using tight bins of list price ($20,000) for the recommended home. Racial minority encompasses all individual
minority groups.
All regression specifications control for the full set of actor characteristics, assigned characteristics,
and search Characteristics. Actor characteristics: tester income, tester household income, gender of tester, age of tester,
month of test, educational attainment of tester; Assigned Characteristics: household members, current home ownership
status, current lease type, car ownership status, reason for moving, years in current residence, length of employment at
current job, reason tester can afford down payment, current lease assigned to tester; Search Characteristics: month of test,
sequence of tester appointments, time of the appointment (am/pm), type of recommended building, total number of homes
recommended to tester, availability of advertised home as stated by agent. Standard errors are clustered by market for
consistency with sampling design. Significance levels are: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01.
Appendix–11


===== PAGE 66 (score 39, numeric-paren 12, table-words 8, var-words 8) =====

F
Discriminatory Steering and Home Price
This section provides a test of differences in the prices of homes recommended to minority
testers relative to a white counterpart. We do not find strong evidence that minority
testers receive recommendations for lower priced homes relative to their white paired
testers, though we do find a marginally significant price reduction for Hispanic/LatinX
testers. Looking at point estimates alone, it appears that homes recommended to African
American and Asian buyers may have somewhat higher list prices on average. Along with
the tests of differences in transaction prices reported in Table 14 and within tight intervals
of list price reported in Column 3 of Table E.1, we take these findings as evidence that
the primary results on steering that are documented in this paper are not explained by
steering of minority buyers into more affordable neighborhoods. There is no evidence
that this is the case for African American testers in particular, who are subject to the
strongest steering effects.
Table F.1. Discriminatory Steering and Home Price
Dependent variable: log(Price)
1
2
3
Racial Minority
0.0024
0.0034
0.0120
(0.1047)
(0.1055)
(0.0994)
African American
0.0776
0.0764
0.0937
(0.1633)
(0.1634)
(0.1583)
Hispanic
−0.1562
−0.1526
−0.1504∗
(0.0988)
(0.0996)
(0.0908)
Asian
0.0642
0.0642
0.0665
(0.0966)
(0.0970)
(0.0904)
Comparison Mean (White)
12.40
12.40
12.40
Racial Comp Advert Home
N
Y
Y
ln(Price) Advert Home
N
N
Y
Observations
21,787
21,776
21,763
Adjusted R2
0.7223
0.7222
0.7241
Note: Coefficients report differences in the list prices of homes recommended to minority testers relative to a white tester
(the omitted category). Racial minority encompasses all individual minority groups. All regression specifications control
for the full set of actor characteristics, assigned characteristics, and search Characteristics. Actor characteristics: tester
income, tester household income, gender of tester, age of tester, month of test, educational attainment of tester; Assigned
Characteristics: household members, current home ownership status, current lease type, car ownership status, reason for
moving, years in current residence, length of employment at current job, reason tester can afford down payment, current
lease assigned to tester; Search Characteristics: month of test, sequence of tester appointments, time of the appointment
(am/pm), type of recommended building, total number of homes recommended to tester, availability of advertised home
as stated by agent. Standard errors are clustered by market for consistency with sampling design. Significance levels are:
∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01.
Appendix–12


===== PAGE 67 (score 96, numeric-paren 57, table-words 8, var-words 14) =====

G
Discriminatory Steering and Implied Preferences
This section explores interactions between the implied preferences of buyers and dis-
criminatory constraints facing buyers. Table G.1 tests for differences within the sample
of buyers who present an advertised listing that falls in a census block group with an
above-median white share. The results suggest stronger constraints for African American
testers, in particular, that search in these neighborhoods. Homes recommended to these
buyers are located in attendance zones with considerably lower elementary school test
scores and neighborhoods with greater numbers of assaults, higher poverty rates, lower
skill levels, lower college attendance rates, higher shares of single-parent households, lower
home ownership rates, and higher exposures to toxic pollution.
Table G.1. Discriminatory Steering: White Neighborhoods
School-Specific Test Scores:
Housing Search Platform:
Elem School
Middle School
Assaults
Elem School
African American
-0.4317***
-0.0670**
47.1246***
-2.0678
(0.0418)
(0.0304)
(9.2629)
(1.7155)
Hispanic
-0.1327*
0.0197
-45.7471***
1.0544
(0.0697)
(0.0309)
(11.7732)
(0.9374)
Asian
-0.1533**
-0.0081
18.5930***
1.0308
(0.0685)
(0.0274)
(6.4357)
(1.3990)
Comparison Mean (White)
0.37
0.28
51
6.70
Observations
5,021
5,217
5,574
5,415
Adjusted R2
0.8419
0.9240
0.9106
0.7569
American Community Survey
Poverty Rate
High Skill
College
Single-Parent HH
Ownership Rate
African American
0.0293***
-0.0550***
-0.0434***
0.0203*
-0.0479**
(0.0083)
(0.0149)
(0.0149)
(0.0106)
(0.0216)
Hispanic
0.0329***
-0.0215
-0.0108
0.0154*
-0.0372
(0.0077)
(0.0173)
(0.0186)
(0.0087)
(0.0235)
Asian
0.0040
-0.0149
-0.0621***
0.0278*
-0.0639**
(0.0108)
(0.0199)
(0.0213)
(0.0146)
(0.0298)
Comparison Mean (White)
0.07
0.50
0.54
0.12
0.77
Observations
10,975
10,975
10,975
10,975
10,975
Adjusted R2
0.4631
0.7111
0.7806
0.4797
0.5685
Pollution
Superfund
Toxics
PM
African American
0.0315
1,537.3840***
0.0500
(0.0400)
(480.2213)
(0.0727)
Hispanic
0.1098**
779.8187
0.0828
(0.0437)
(524.6936)
(0.0875)
Asian
0.1709***
328.7771
-0.2565**
(0.0622)
(519.5311)
(0.1008)
Comparison Mean (White)
0.30
5900
9.80
Observations
10,977
10,977
10,977
Adjusted R2
0.8642
0.5916
0.9438
Note: Coefficients report differences in the attributes of properties/neighborhoods recommended to minority testers relative
to a white tester (the omitted category) for the subset of trials where the advertised home falls above the 50th percentile
in the block group share of white households. Racial minority encompasses all individual minority groups. All regression
specifications include trial fixed effects and control for the full set of actor characteristics, assigned characteristics, and
search characteristics. Actor characteristics: tester income, tester household income, gender of tester, age of tester, month
of test, educational attainment of tester; Assigned characteristics: household members, current home ownership status,
current lease type, car ownership status, reason for moving, years in current residence, length of employment at current job,
reason tester can afford down payment, current lease assigned to tester; Search Characteristics: month of test, sequence of
tester appointments, and time of the appointment (am/pm). Standard errors are clustered by trial, which is the level of
randomization. Significance levels are: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01.
Appendix–13


===== PAGE 68 (score 4, numeric-paren 0, table-words 1, var-words 2) =====

Table G.2 tests for differences within the sample of buyers who present an advertised
listing that falls above the median price in the audit ($245,000). These results suggest
that the differences in elementary and middle school test scores, poverty rates, skill,
college attainment, single-parent households, and exposure to toxics may be amplified for
African Americans who present these preferences.
Appendix–14


===== PAGE 69 (score 111, numeric-paren 67, table-words 10, var-words 13) =====

Table G.2. Discriminatory Steering and Neighborhood Effects: Higher-Priced Homes
School-Specific Test Scores:
Housing Search Platform:
Elem School
Middle School
Assaults
Elem School
Racial Minority
-0.0285
0.1056***
-10.9046*
-0.0134
(0.0320)
(0.0222)
(6.3792)
(0.2832)
African American
-0.1731***
-0.0219
-22.2033**
0.1497
(0.0668)
(0.0285)
(10.8133)
(0.3350)
Hispanic
0.0420
0.1119***
-49.4148***
-0.0965
(0.0378)
(0.0190)
(10.2786)
(0.5100)
Asian
-0.2061***
0.0704**
2.2784
-0.4517
(0.0622)
(0.0357)
(7.7459)
(0.5991)
Comparison Mean (White)
0.4
0.36
80.9
6.94
ln(Price) Advert Home
Y
Y
Y
Y
Racial Comp Advert Home
Y
Y
Y
Y
Outcome Advertised Home
Y
Y
Y
Y
Observations
3,799
3,831
5,293
4,906
Adjusted R2
0.8379
0.9283
0.8393
0.7440
American Community Survey
Poverty Rate
High Skill
College
Single-Parent HH
Ownership Rate
Racial Minority
0.0146**
-0.0208*
-0.0332***
0.0106
-0.0118
(0.0062)
(0.0118)
(0.0111)
(0.0068)
(0.0169)
African American
0.0254***
-0.0390**
-0.0599***
0.0157*
-0.0223
(0.0078)
(0.0171)
(0.0150)
(0.0095)
(0.0239)
Hispanic
0.0104
-0.0240
-0.0224
-0.0024
0.0182
(0.0083)
(0.0172)
(0.0156)
(0.0086)
(0.0233)
Asian
0.0063
-0.0003
-0.0156
0.0133
-0.0141
(0.0080)
(0.0158)
(0.0151)
(0.0084)
(0.0207)
Comparison Mean (White)
0.071
0.54
0.6
0.12
0.73
ln(Price) Advert Home
Y
Y
Y
Y
Y
Racial Comp Advert Home
Y
Y
Y
Y
Y
Outcome Advertised Home
Y
Y
Y
Y
Y
Observations
10,850
10,850
10,850
10,850
10,850
Adjusted R2
0.5589
0.6829
0.7236
0.5453
0.6698
Pollution
Superfund
Toxics
PM
Racial Minority
0.0651
1,360.2900**
0.0179
(0.0554)
(589.0348)
(0.0670)
African American
0.0758
994.0474*
0.0708
(0.0831)
(578.0719)
(0.0868)
Hispanic
0.0247
2,239.0900**
0.1133
(0.0716)
(907.8452)
(0.0889)
Asian
0.0578
1,272.7980*
-0.0896
(0.0648)
(744.7305)
(0.0939)
Comparison Mean (White)
0.45
4900
9.5
ln(Price) Advert Home
Y
Y
Y
Racial Comp Advert Home
Y
Y
Y
Outcome Advertised Home
Y
Y
Y
Observations
10,862
10,862
10,862
Adjusted R2
0.8758
0.6063
0.9599
Note: Coefficients report differences in the attributes of properties/neighborhoods recommended to minority testers relative
to a white tester (the omitted category) for the subset of trials where the advertised home falls above the 50th percentile
in the block group share of the price distribution (above $245,000). All regression specifications include trial fixed effects
and control for the full set of actor characteristics, assigned characteristics, and search characteristics. Actor character-
istics: tester income, tester household income, gender of tester, age of tester, month of test, educational attainment of
tester; Assigned characteristics: household members, current home ownership status, current lease type, car ownership
status, reason for moving, years in current residence, length of employment at current job, reason tester can afford down
payment, current lease assigned to tester; Search Characteristics: month of test, sequence of tester appointments, time of
the appointment (am/pm), type of recommended building. Standard errors are clustered by trial, which is the level of
randomization. Standard errors are clustered by trial, which is the level of randomization. Significance levels are: ∗p<0.1;
∗∗p<0.05; ∗∗∗p<0.01.
Appendix–15


===== PAGE 70 (score 59, numeric-paren 28, table-words 6, var-words 12) =====

H
Sensitivity Analysis: Superfund Proximity
This section explores the sensitivity of results to different distance-based measures of
proximity to Superfund sites.
We rely on proximity as our measure of exposure and
evaluate the trade-offbetween different distances and statistical power. Table H.1 reports
estimates using a range of distances between 3-6km, which is slightly wider than the 2-
3 mile range used in prior work (Gamper-Rabindran and Timmins, 2011, Greenstone
and Gallagher, 2008). The number of sites for the comparison mean recommendation
gets mechanically smaller with a tighter radius and so do our estimates of differences
(in levels) for the minority group as a whole. Estimates do not decrease monotonically
within each tester group, but are consistent in magnitude across the range. We do not
have sufficient power to detect differences within 3km.
Table H.1. Discriminatory Steering and Superfund Proximity by Distance
Dependent variable: Superfund
3 Km
3.5 Km
4 Km
4.5 Km
5 Km
5.5 Km
6 Km
Racial Minority
0.0392
0.0782∗∗
0.0773∗∗
0.0813∗∗∗
0.1010∗∗∗
0.1010∗∗∗
0.1031∗∗∗
(0.0280)
(0.0316)
(0.0313)
(0.0306)
(0.0325)
(0.0365)
(0.0387)
African American
0.0595
0.0789∗
0.0805∗
0.1041∗∗
0.1256∗∗∗
0.1118∗∗
0.1008∗∗
(0.0364)
(0.0406)
(0.0431)
(0.0435)
(0.0465)
(0.0491)
(0.0488)
Hispanic
−0.0217
0.0425
0.0593
0.0516
0.0591
0.0343
0.0382
(0.0378)
(0.0406)
(0.0405)
(0.0380)
(0.0376)
(0.0446)
(0.0490)
Asian
0.0820∗∗
0.1231∗∗∗
0.0963∗∗
0.0849∗
0.1046∗∗
0.1339∗∗
0.1528∗∗∗
(0.0385)
(0.0427)
(0.0425)
(0.0443)
(0.0484)
(0.0527)
(0.0578)
Comparison Mean (White)
0.13
0.17
0.22
0.29
0.35
0.41
0.48
ln(Price) Advert Home
N
Y
N
Y
Y
Y
Y
Racial Comp Advert Home
N
Y
N
Y
Y
Y
Y
Observations
21,883
21,883
21,883
21,883
21,883
21,883
21,883
Adjusted R2
0.8028
0.8412
0.8439
0.8597
0.8744
0.8868
0.9015
Note:
∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01
Note:
Coefficients report differences in the number of Superfund sites within close proximity (3-6km) to proper-
ties/neighborhoods recommended to minority testers relative to a white tester (the omitted category).
All regression
specifications include trial fixed effects and control for the full set of actor characteristics, assigned characteristics, and
search characteristics. Actor characteristics: tester income, tester household income, gender of tester, age of tester, month
of test, educational attainment of tester; Assigned characteristics: household members, current home ownership status,
current lease type, car ownership status, reason for moving, years in current residence, length of employment at current job,
reason tester can afford down payment, current lease assigned to tester; Search Characteristics: month of test, sequence
of tester appointments, time of the appointment (am/pm), type of recommended building. Standard errors are clustered
by trial, which is the level of randomization. Standard errors are clustered by trial, which is the level of randomization.
Significance levels are: ∗p<0.1; ∗∗p<0.05; ∗∗∗p<0.01.
Appendix–16