Outline

This document contains all the plots/tables needed for the Faces-Slides that I can make in R. NOTE I’ve added a tab to the google-sheet with names matched and the slide-show reference, that’s where the names for these plots are from (i.e. Figure 3.a etc.).

I’ve also split this document into Figures and Tables in two seperate sections.

Tables

Summary Statistics

Summary Statistics - Population Overview
Variable Count-Train Count-Val Count-All
AvrAge 31.880 31.689 31.822
percentBlack 0.695 0.698 0.696
percentWhite 0.277 0.271 0.275
percentOther 0.028 0.031 0.029
percentMale 0.790 0.784 0.788
percentFemale 0.210 0.216 0.212
percentFelony 0.415 0.427 0.419
percentViolent 0.089 0.092 0.090
percentGunCrime 0.070 0.072 0.070
percentDrugCrime 0.179 0.183 0.180
percentPropertyCrime 0.271 0.275 0.272
percentMisdRecord 0.313 0.322 0.316
percentFelonyRecord 0.332 0.333 0.333
percentAnyRecord 0.460 0.463 0.461
percentReArrest 0.275 0.274 0.275
percentReleaseOutcome 0.770 0.762 0.767
Count 20020.000 8835.000 28855.000
Count_2017 7177.000 3165.000 10342.000
Count_2018 8549.000 3773.000 12322.000
Count_2019 4294.000 1897.000 6191.000 
## \captionsetup[table]{labelformat=empty,skip=1pt}
## \begin{longtable}{lrrr}
## \caption*{
## \large Summary Statistics - Population Overview\\ 
## \small \\ 
## } \\ 
## \toprule
## Variable & Count-Train & Count-Val & Count-All \\ 
## \midrule
## AvrAge & $31.880$ & $31.689$ & $31.822$ \\ 
## percentBlack & $0.695$ & $0.698$ & $0.696$ \\ 
## percentWhite & $0.277$ & $0.271$ & $0.275$ \\ 
## percentOther & $0.028$ & $0.031$ & $0.029$ \\ 
## percentMale & $0.790$ & $0.784$ & $0.788$ \\ 
## percentFemale & $0.210$ & $0.216$ & $0.212$ \\ 
## percentFelony & $0.415$ & $0.427$ & $0.419$ \\ 
## percentViolent & $0.089$ & $0.092$ & $0.090$ \\ 
## percentGunCrime & $0.070$ & $0.072$ & $0.070$ \\ 
## percentDrugCrime & $0.179$ & $0.183$ & $0.180$ \\ 
## percentPropertyCrime & $0.271$ & $0.275$ & $0.272$ \\ 
## percentMisdRecord & $0.313$ & $0.322$ & $0.316$ \\ 
## percentFelonyRecord & $0.332$ & $0.333$ & $0.333$ \\ 
## percentAnyRecord & $0.460$ & $0.463$ & $0.461$ \\ 
## percentReArrest & $0.275$ & $0.274$ & $0.275$ \\ 
## percentReleaseOutcome & $0.770$ & $0.762$ & $0.767$ \\ 
## Count & $20020.000$ & $8835.000$ & $28855.000$ \\ 
## Count\_2017 & $7177.000$ & $3165.000$ & $10342.000$ \\ 
## Count\_2018 & $8549.000$ & $3773.000$ & $12322.000$ \\ 
## Count\_2019 & $4294.000$ & $1897.000$ & $6191.000$ \\ 
## \bottomrule
## \end{longtable}
##              Df     Pillai approx F num Df den Df Pr(>F)
## train_val     1 0.00034146   1.6423      6  28848  0.131
## Residuals 28853
##  Response age_arrest :
##                Df  Sum Sq Mean Sq F value Pr(>F)
## train_val       1     225  224.63  1.6748 0.1956
## Residuals   28853 3870030  134.13               
## 
##  Response gender_male :
##                Df Sum Sq Mean Sq F value Pr(>F)
## train_val       1    0.2 0.20687  1.2372  0.266
## Residuals   28853 4824.7 0.16722               
## 
##  Response race_black :
##                Df Sum Sq  Mean Sq F value Pr(>F)
## train_val       1    0.0 0.043089  0.2037 0.6518
## Residuals   28853 6104.5 0.211571               
## 
##  Response race_other :
##                Df Sum Sq  Mean Sq F value Pr(>F)
## train_val       1   0.05 0.048104  1.7239 0.1892
## Residuals   28853 805.13 0.027905               
## 
##  Response misd_true :
##                Df Sum Sq Mean Sq F value Pr(>F)
## train_val       1    0.5 0.50217  2.3253 0.1273
## Residuals   28853 6231.1 0.21596               
## 
##  Response felon_true :
##                Df Sum Sq  Mean Sq F value Pr(>F)
## train_val       1    0.0 0.001698  0.0076 0.9303
## Residuals   28853 6405.4 0.222002

MTurk Summary Statistics No.1

MTurk Summary Statistics No.1 - Skin Tone Data
skin_tone count PercentBlack PercentWhite PercentOther
#301e10 517 100.00 0.00 0.00
#623a17 1688 99.82 0.12 0.06
#76441f 846 99.76 0.00 0.24
#80492a 868 99.31 0.35 0.35
#885633 1350 99.41 0.52 0.07
#94623d 2022 99.26 0.49 0.25
#ab8b64 1435 95.68 3.07 1.25
#b26949 338 55.62 39.94 4.44
#cb9662 828 82.73 13.29 3.99
#d09e7d 338 54.73 37.28 7.99
#e7bc91 508 24.41 67.52 8.07
#e9cba7 422 4.98 91.71 3.32
#ecc083 457 49.89 36.76 13.35
#eed0b8 882 4.42 91.16 4.42
#efc088 302 12.25 76.49 11.26
#efc794 514 14.01 75.49 10.51
#f6e1aa 414 11.11 80.92 7.97
#f7ddc4 1088 3.95 93.01 3.03 
## \captionsetup[table]{labelformat=empty,skip=1pt}
## \begin{longtable}{lcrrr}
## \caption*{
## \large MTurk Summary Statistics No.1 - Skin Tone Data\\ 
## \small \\ 
## } \\ 
## \toprule
## skin\_tone & count & PercentBlack & PercentWhite & PercentOther \\ 
## \midrule
## \#301e10 & 517 & $100.00$ & $0.00$ & $0.00$ \\ 
## \#623a17 & 1688 & $99.82$ & $0.12$ & $0.06$ \\ 
## \#76441f & 846 & $99.76$ & $0.00$ & $0.24$ \\ 
## \#80492a & 868 & $99.31$ & $0.35$ & $0.35$ \\ 
## \#885633 & 1350 & $99.41$ & $0.52$ & $0.07$ \\ 
## \#94623d & 2022 & $99.26$ & $0.49$ & $0.25$ \\ 
## \#ab8b64 & 1435 & $95.68$ & $3.07$ & $1.25$ \\ 
## \#b26949 & 338 & $55.62$ & $39.94$ & $4.44$ \\ 
## \#cb9662 & 828 & $82.73$ & $13.29$ & $3.99$ \\ 
## \#d09e7d & 338 & $54.73$ & $37.28$ & $7.99$ \\ 
## \#e7bc91 & 508 & $24.41$ & $67.52$ & $8.07$ \\ 
## \#e9cba7 & 422 & $4.98$ & $91.71$ & $3.32$ \\ 
## \#ecc083 & 457 & $49.89$ & $36.76$ & $13.35$ \\ 
## \#eed0b8 & 882 & $4.42$ & $91.16$ & $4.42$ \\ 
## \#efc088 & 302 & $12.25$ & $76.49$ & $11.26$ \\ 
## \#efc794 & 514 & $14.01$ & $75.49$ & $10.51$ \\ 
## \#f6e1aa & 414 & $11.11$ & $80.92$ & $7.97$ \\ 
## \#f7ddc4 & 1088 & $3.95$ & $93.01$ & $3.03$ \\ 
## \bottomrule
## \end{longtable}

MTurk Summary Statistics No.2

MTurk Summary Statistics No.2 - Psych Features
ID mean_attractiveness mean_competence mean_dominance mean_trustworthiness mean_human_guess
Full Sample 3.710862 3.733321 4.172015 3.339878 0.5021089
A 3.594127 3.731913 3.770253 3.473995 0.5175032
B 3.720118 3.745884 4.232167 3.363610 0.4921472
I 4.600000 4.321569 4.188235 3.996078 0.5661765
U 4.083596 4.006712 4.237263 3.602571 0.4989524
W 3.656360 3.676762 4.026096 3.251584 0.5267044
above_34 3.320608 3.584982 4.205890 3.202959 0.4900435
below_25 4.052978 3.847832 4.090316 3.505100 0.5190686
between_25_34 3.801396 3.783532 4.217054 3.325672 0.4984155
0 3.643274 3.665749 4.209290 3.242405 0.4917898
1 3.731389 3.753845 4.160694 3.369482 0.5052430 
## \captionsetup[table]{labelformat=empty,skip=1pt}
## \begin{longtable}{lrrrrr}
## \caption*{
## \large MTurk Summary Statistics No.2 - Psych Features\\ 
## \small \\ 
## } \\ 
## \toprule
## ID & mean\_attractiveness & mean\_competence & mean\_dominance & mean\_trustworthiness & mean\_human\_guess \\ 
## \midrule
## Full Sample & 3.710862 & 3.733321 & 4.172015 & 3.339878 & 0.5021089 \\ 
## A & 3.594127 & 3.731913 & 3.770253 & 3.473995 & 0.5175032 \\ 
## B & 3.720118 & 3.745884 & 4.232167 & 3.363610 & 0.4921472 \\ 
## I & 4.600000 & 4.321569 & 4.188235 & 3.996078 & 0.5661765 \\ 
## U & 4.083596 & 4.006712 & 4.237263 & 3.602571 & 0.4989524 \\ 
## W & 3.656360 & 3.676762 & 4.026096 & 3.251584 & 0.5267044 \\ 
## above\_34 & 3.320608 & 3.584982 & 4.205890 & 3.202959 & 0.4900435 \\ 
## below\_25 & 4.052978 & 3.847832 & 4.090316 & 3.505100 & 0.5190686 \\ 
## between\_25\_34 & 3.801396 & 3.783532 & 4.217054 & 3.325672 & 0.4984155 \\ 
## 0 & 3.643274 & 3.665749 & 4.209290 & 3.242405 & 0.4917898 \\ 
## 1 & 3.731389 & 3.753845 & 4.160694 & 3.369482 & 0.5052430 \\ 
## \bottomrule
## \end{longtable}

Table 1 - Relative feature importance

Table 1 V1 - R-sqrt & Partial R-sqrt.

Table 01 - ML Fusion Feature Importance
Importance measures in adjusted R-sqrt
Feature R Squared Partial R Squared
Felony 0.5396 0.6208
Violent-Crime 0.1772 0.3572
Property-Crime 0.0970 0.0991
Race 0.0690 0.2747
Sex 0.0676 0.1564
Gun-Crime 0.0198 0.0046
Drug-Crime 0.0036 0.0086
Age 0.0010 0.0183

Table 1 V2 - Dropout & Variance Decomposition for ML-Face

## [1] "Age"
## [1] "Sex"
## [1] "Race"
## [1] "Drug-Crime"
## [1] "Felony-Crime"
## [1] "Gun-Crime"
## [1] "Violent-Crime"
## [1] "Property-Crime"
## [1] "Face"
Table 01 V1 - ML Fusion Feature Importance
Predictions include XgBoost Risk & ML-Face
Feature Var Difference Lower 95% CI Upper 95% CI
Face 0.0042 0.0035 0.0048
Violent-Crime 0.0037 0.0032 0.0043
Property-Crime 0.0034 0.0028 0.0039
Felony-Crime 0.0032 0.0025 0.0040
Drug-Crime 0.0027 0.0021 0.0033
Gun-Crime 0.0027 0.0020 0.0032
Race 0.0007 0.0000 0.0013
Age −0.0001 −0.0011 0.0009
Sex −0.0005 −0.0012 0.0001

Table 1 V3 - Using OLS predictions on Val set for partial dropout variance

Table 01 V1.2 - ML Fusion Feature Importance
Predictions include XgBoost Risk & ML-Face
Feature Var % Difference Lower 95% CI Upper 95% CI
Felony-Crime 0.3758 0.2930 0.4357
ML-Face 0.1135 0.0055 0.1892
Risk Prediction 0.1069 −0.0003 0.1903
Violent-Crime 0.0889 −0.0129 0.1685
Race 0.0394 −0.0682 0.1324
Drug-Crime 0.0326 −0.0765 0.1151
Age 0.0080 −0.0973 0.0918
Gun-Crime 0.0039 −0.1021 0.0832
Skin-Tone 0.0036 −0.1066 0.0894
Sex 0.0019 −0.1065 0.0800
Property-Crime 0.0016 −0.1104 0.0858

Table 1 V4 - Permutation Test:

Table 01 V2 - Permutation Test
Predictions include XgBoost Risk & ML-Face
Feature AUC Difference Lower 95% CI Upper 95% CI
ML-Face 0.2243 0.2115 0.2290
Violent-Crime 0.1263 0.1147 0.1377
Property-Crime 0.1253 0.1138 0.1370
Drug-Crime 0.1243 0.1134 0.1367
Gun-Crime 0.1233 0.1119 0.1354
Felony-Crime 0.1033 0.0923 0.1143
Race 0.0083 −0.0023 0.0185
Sex 0.0013 −0.0088 0.0122
Age 0.0003 −0.0104 0.0101
Table 01 - ML Fusion Feature Importance
Feature Drop Out & Variance Permutation & AUC
Var % Difference1 Lower 95% CI Upper 95% CI AUC Difference2 Lower 95% CI Upper 95% CI
Felony-Crime 0.3758 0.2930 0.4357 0.1033 0.0923 0.1143
ML-Face 0.1135 0.0055 0.1892 0.2243 0.2115 0.2290
Violent-Crime 0.0889 −0.0129 0.1685 0.1263 0.1147 0.1377
Race 0.0394 −0.0682 0.1324 0.0083 −0.0023 0.0185
Drug-Crime 0.0326 −0.0765 0.1151 0.1243 0.1134 0.1367
Age 0.0080 −0.0973 0.0918 0.0003 −0.0104 0.0101
Gun-Crime 0.0039 −0.1021 0.0832 0.1233 0.1119 0.1354
Sex 0.0019 −0.1065 0.0800 0.0013 −0.0088 0.0122
Property-Crime 0.0016 −0.1104 0.0858 0.1253 0.1138 0.1370

1 From OLS: Var(Full Predictions) - Var(Prediction w. constant feature) / Var(Full Predictions)

2 AUC(Full Predictions) - AUC(Prediction w. permuted feature)

Table 2 - Release regressions with ML Fusion and ML Face

Table No.2 - Rediscovering Demographic Discrimination
Dependent variable:
Release-Final-Outcome
(1) (2) (3) (4) (5) (6) (7)
ML-Face Fusion 1.009***
(0.033)
ML-Face CNN 0.692*** 0.652*** 0.647*** 0.541*** 0.597*** 0.522***
(0.040) (0.045) (0.045) (0.045) (0.043) (0.043)
SexM -0.029** -0.028** 0.001 -0.009 0.019
(0.012) (0.012) (0.012) (0.012) (0.012)
Age 0.001 0.001 -0.0005 -0.001 -0.001***
(0.0004) (0.0004) (0.0004) (0.0004) (0.0004)
RaceB -0.043 -0.025 0.030 0.004 0.046
(0.049) (0.050) (0.050) (0.048) (0.048)
Skin-tone 0.266** 0.267** 0.260** 0.268**
(0.132) (0.131) (0.127) (0.126)
Risk Prediction -0.677*** -0.624***
(0.048) (0.054)
Felony-Dummie -0.207*** -0.208***
(0.010) (0.010)
Gun-Crime-Dummie 0.038** 0.038**
(0.017) (0.017)
Drug-Crime-Dummie 0.039*** 0.078***
(0.012) (0.012)
Violent-Crime-Dummie -0.160*** -0.161***
(0.015) (0.015)
Property-Crime-Dummie -0.042*** 0.016
(0.010) (0.011)
Constant -0.009 0.245*** 0.326*** 0.300*** 0.551*** 0.447*** 0.638***
(0.025) (0.030) (0.066) (0.067) (0.069) (0.065) (0.066)
ROC-AUC 0.708 0.622 0.623 0.624 0.662 0.72 0.736
Observations 8,821 8,821 8,821 8,821 8,821 8,821 8,821
Adjusted R2 0.098 0.032 0.033 0.033 0.054 0.111 0.124
F Statistic 958.772*** (df = 1; 8819) 296.642*** (df = 1; 8819) 43.944*** (df = 7; 8813) 38.969*** (df = 8; 8812) 57.268*** (df = 9; 8811) 85.787*** (df = 13; 8807) 90.560*** (df = 14; 8806)
Note: p<0.1; p<0.05; p<0.01
We encode skin-tone as a continuous variable increasing in brightness

Table 2.1 - ML-Face + Demographics + Risk

Table No.2.1 - Rediscovering Demographic Discrimination
Dependent variable:
Judge Release Final Outcome
(1) (2) (3) (4) (5) (6)
ML-Face CNN 0.692*** 0.652*** 0.562*** 0.546***
(0.040) (0.045) (0.041) (0.045)
Male -0.106*** -0.029** -0.059*** -0.001
(0.011) (0.012) (0.011) (0.012)
Age -0.00003 0.001 -0.001** -0.0005
(0.0005) (0.0004) (0.0004) (0.0004)
Black -0.058 -0.043 0.008 0.012
(0.050) (0.049) (0.049) (0.049)
Risk Prediction -0.596*** -0.775*** -0.677***
(0.045) (0.048) (0.048)
Constant 0.245*** 0.908*** 0.326*** 0.519*** 1.087*** 0.577***
(0.030) (0.053) (0.066) (0.036) (0.053) (0.067)
ROC-AUC 0.622 0.55 0.623 0.655 0.637 0.661
Observations 8,821 8,821 8,821 8,821 8,821 8,821
Adjusted R2 0.032 0.010 0.033 0.051 0.038 0.054
F Statistic 296.642*** (df = 1; 8819) 15.843*** (df = 6; 8814) 43.944*** (df = 7; 8813) 238.904*** (df = 2; 8818) 51.194*** (df = 7; 8813) 63.886*** (df = 8; 8812)
Note: p<0.1; p<0.05; p<0.01
We encode skin-tone as a continuous variable increasing in brightness

Table 2.2 - ML-Face + Demographics + Skin-tone + Risk

Table No.2.2 - Adding skin-tone feature
Dependent variable:
Judge Release Final Outcome
(1) (2) (3) (4)
ML-Face CNN 0.652*** 0.647*** 0.546*** 0.541***
(0.045) (0.045) (0.045) (0.045)
SexM -0.029** -0.028** -0.001 0.001
(0.012) (0.012) (0.012) (0.012)
Age 0.001 0.001 -0.0005 -0.0005
(0.0004) (0.0004) (0.0004) (0.0004)
RaceB -0.043 -0.025 0.012 0.030
(0.049) (0.050) (0.049) (0.050)
Skin-tone 0.266** 0.267**
(0.132) (0.131)
Risk Prediction -0.677*** -0.677***
(0.048) (0.048)
Constant 0.326*** 0.300*** 0.577*** 0.551***
(0.066) (0.067) (0.067) (0.069)
ROC-AUC 0.623 0.624 0.661 0.662
Observations 8,821 8,821 8,821 8,821
Adjusted R2 0.033 0.033 0.054 0.054
F Statistic 43.944*** (df = 7; 8813) 38.969*** (df = 8; 8812) 63.886*** (df = 8; 8812) 57.268*** (df = 9; 8811)
Note: p<0.1; p<0.05; p<0.01
We encode skin-tone as a continuous variable increasing in brightness

Table 2.3 - ML-Face + Demographics + Skin-tone + Risk + Charge

Table No.2.3 - Adding charge dummies
Dependent variable:
Judge Release Final Outcome
(1) (2) (3) (4)
ML-Face CNN 0.647*** 0.597*** 0.541*** 0.522***
(0.045) (0.043) (0.045) (0.043)
SexM -0.028** -0.009 0.001 0.019
(0.012) (0.012) (0.012) (0.012)
Age 0.001 -0.001 -0.0005 -0.001***
(0.0004) (0.0004) (0.0004) (0.0004)
RaceB -0.025 0.004 0.030 0.046
(0.050) (0.048) (0.050) (0.048)
Skin-tone 0.266** 0.260** 0.267** 0.268**
(0.132) (0.127) (0.131) (0.126)
Felony-Dummie -0.207*** -0.208***
(0.010) (0.010)
Gun-Crime-Dummie 0.038** 0.038**
(0.017) (0.017)
Drug-Crime-Dummie 0.039*** 0.078***
(0.012) (0.012)
Violent-Crime-Dummie -0.160*** -0.161***
(0.015) (0.015)
Property-Crime-Dummie -0.042*** 0.016
(0.010) (0.011)
Risk Prediction -0.677*** -0.624***
(0.048) (0.054)
Constant 0.300*** 0.447*** 0.551*** 0.638***
(0.067) (0.065) (0.069) (0.066)
ROC-AUC 0.624 0.72 0.662 0.736
Observations 8,821 8,821 8,821 8,821
Adjusted R2 0.033 0.111 0.054 0.124
F Statistic 38.969*** (df = 8; 8812) 85.787*** (df = 13; 8807) 57.268*** (df = 9; 8811) 90.560*** (df = 14; 8806)
Note: p<0.1; p<0.05; p<0.01
We encode skin-tone as a continuous variable increasing in brightness

Table 3 - Release regressions with ML Face and Psych Features

Table No.3 - is Model Rediscovering Psychological Features
Dependent variable:
Release-Final-Outcome
(1) (2) (3) (4) (5) (6) (7) (8)
Attractiveness 0.004 -0.001 0.002 0.004 0.003
(0.006) (0.005) (0.006) (0.006) (0.006)
Competence 0.004 0.0004 0.005 0.004 -0.001
(0.006) (0.006) (0.006) (0.006) (0.006)
Dominance -0.012*** -0.004 -0.007 -0.012*** -0.003
(0.004) (0.004) (0.004) (0.004) (0.004)
Trustworthiness 0.019*** 0.015*** 0.018*** 0.019*** 0.009*
(0.006) (0.006) (0.006) (0.006) (0.006)
ML-Face CNN 0.696*** 0.677*** 0.694*** 0.510***
(0.040) (0.041) (0.040) (0.044)
Male -0.099*** 0.019
(0.011) (0.012)
Age 0.0004 -0.001***
(0.0005) (0.0005)
Black -0.022 0.059
(0.051) (0.048)
Skin-tone 0.371*** 0.277**
(0.134) (0.126)
Risk Prediction -0.629***
(0.054)
Felony-Dummie -0.207***
(0.010)
Gun-Crime-Dummie 0.039**
(0.017)
Drug-Crime-Dummie 0.079***
(0.012)
Violent-Crime-Dummie -0.165***
(0.015)
Property-Crime-Dummie 0.017
(0.011)
Human Guess 0.030** 0.024 0.013 0.013
(0.015) (0.015) (0.015) (0.014)
Constant 0.718*** 0.243*** 0.228*** 0.782*** 0.747*** 0.706*** 0.238*** 0.598***
(0.023) (0.030) (0.037) (0.060) (0.009) (0.024) (0.031) (0.070)
ROC-AUC 0.542 0.623 0.626 0.576 0.515 0.544 0.624 0.739
Observations 8,722 8,722 8,722 8,722 8,722 8,722 8,722 8,722
Adjusted R2 0.004 0.033 0.034 0.014 0.0003 0.004 0.033 0.126
F Statistic 9.036*** (df = 4; 8717) 297.870*** (df = 1; 8720) 62.075*** (df = 5; 8716) 11.912*** (df = 11; 8710) 4.008** (df = 1; 8720) 7.751*** (df = 5; 8716) 149.318*** (df = 2; 8719) 67.422*** (df = 19; 8702)
Note: p<0.1; p<0.05; p<0.01
We encode skin-tone as a continuous variable increasing in brightness

Table 3.1 - Release regressions with ML Face and Psych Features

Table No.3 - is Model Rediscovering Psychological Features
Dependent variable:
Judge Release Final Outcome
(1) (2) (3) (4) (5) (6) (7) (8)
ML-Face CNN 0.696*** 0.632*** 0.538*** 0.511***
(0.040) (0.046) (0.040) (0.044)
Attractiveness 0.004 0.002 0.001 0.011** 0.004 0.003
(0.006) (0.006) (0.006) (0.005) (0.006) (0.006)
Competence 0.004 0.005 0.0004 0.0003 0.003 -0.001
(0.006) (0.006) (0.006) (0.006) (0.006) (0.006)
Dominance -0.012*** -0.007 -0.004 -0.007 -0.005 -0.003
(0.004) (0.004) (0.004) (0.004) (0.004) (0.004)
Trustworthiness 0.019*** 0.018*** 0.014** 0.013** 0.012** 0.009*
(0.006) (0.006) (0.006) (0.006) (0.006) (0.006)
Male -0.099*** -0.027** -0.034*** 0.019
(0.011) (0.012) (0.011) (0.012)
Age 0.0004 0.001* -0.002*** -0.001***
(0.0005) (0.0005) (0.0005) (0.0005)
Black -0.022 -0.016 0.061 0.058
(0.051) (0.051) (0.049) (0.048)
Skin-tone 0.371*** 0.276** 0.355*** 0.280**
(0.134) (0.133) (0.127) (0.126)
Risk Prediction -0.480*** -0.631*** -0.720*** -0.629***
(0.050) (0.049) (0.054) (0.054)
Felony-Dummie -0.203*** -0.206*** -0.207*** -0.207***
(0.010) (0.010) (0.010) (0.010)
Gun-Crime-Dummie 0.051*** 0.049*** 0.042** 0.039**
(0.017) (0.017) (0.017) (0.017)
Drug-Crime-Dummie 0.069*** 0.073*** 0.078*** 0.079***
(0.012) (0.012) (0.012) (0.012)
Violent-Crime-Dummie -0.157*** -0.162*** -0.172*** -0.165***
(0.015) (0.015) (0.015) (0.015)
Property-Crime-Dummie 0.007 0.016 0.019 0.017
(0.011) (0.011) (0.011) (0.011)
Constant 0.243*** 0.718*** 0.782*** 0.257*** 0.586*** 0.973*** 1.052*** 0.604***
(0.030) (0.023) (0.060) (0.071) (0.036) (0.025) (0.059) (0.070)
ROC-AUC 0.623 0.542 0.576 0.628 0.73 0.713 0.725 0.739
Observations 8,722 8,722 8,722 8,722 8,722 8,722 8,722 8,722
Adjusted R2 0.033 0.004 0.014 0.035 0.120 0.104 0.113 0.126
F Statistic 297.870*** (df = 1; 8720) 9.036*** (df = 4; 8717) 11.912*** (df = 11; 8710) 27.225*** (df = 12; 8709) 170.599*** (df = 7; 8714) 102.119*** (df = 10; 8711) 66.334*** (df = 17; 8704) 71.122*** (df = 18; 8703)
Note: p<0.1; p<0.05; p<0.01
We encode skin-tone as a continuous variable increasing in brightness

Table 3.2 - Release regressions with ML Face and Psych Features

Table No.3 - Is the Model Rediscovering Psychological Features
Dependent variable:
Judge Release Final Outcome
(1) (2) (3) (4) (5)
ML-Face CNN 0.694*** 0.511*** 0.510***
(0.040) (0.044) (0.044)
Human Guess 0.030** 0.024 0.013 0.013
(0.015) (0.015) (0.015) (0.014)
Male 0.019 0.019
(0.012) (0.012)
Age -0.001*** -0.001***
(0.0005) (0.0005)
Black 0.058 0.059
(0.048) (0.048)
Skin-tone 0.280** 0.277**
(0.126) (0.126)
Risk Prediction -0.629*** -0.629***
(0.054) (0.054)
Felony-Dummie -0.207*** -0.207***
(0.010) (0.010)
Gun-Crime-Dummie 0.039** 0.039**
(0.017) (0.017)
Drug-Crime-Dummie 0.079*** 0.079***
(0.012) (0.012)
Violent-Crime-Dummie -0.165*** -0.165***
(0.015) (0.015)
Property-Crime-Dummie 0.017 0.017
(0.011) (0.011)
Attractiveness 0.004 0.003 0.003
(0.006) (0.006) (0.006)
Competence 0.004 -0.001 -0.001
(0.006) (0.006) (0.006)
Dominance -0.012*** -0.003 -0.003
(0.004) (0.004) (0.004)
Trustworthiness 0.019*** 0.009* 0.009*
(0.006) (0.006) (0.006)
Constant 0.747*** 0.706*** 0.238*** 0.604*** 0.598***
(0.009) (0.024) (0.031) (0.070) (0.070)
ROC-AUC 0.515 0.544 0.624 0.739 0.739
Observations 8,722 8,722 8,722 8,722 8,722
Adjusted R2 0.0003 0.004 0.033 0.126 0.126
F Statistic 4.008** (df = 1; 8720) 7.751*** (df = 5; 8716) 149.318*** (df = 2; 8719) 71.122*** (df = 18; 8703) 67.422*** (df = 19; 8702)
Note: p<0.1; p<0.05; p<0.01
We encode skin-tone as a continuous variable increasing in brightness

Table 4 - Correlates with ML-Face Predictions

Correlates with ML-Face
Dependent variable:
ML-Face Prediction
(1) (2) (3) (4) (5) (6)
Male -0.115*** -0.114*** -0.117*** -0.116***
(0.003) (0.003) (0.003) (0.003)
Age -0.001*** -0.001*** -0.0005*** -0.0005***
(0.0001) (0.0001) (0.0001) (0.0001)
Black -0.013 -0.011 -0.007 -0.006
(0.012) (0.012) (0.012) (0.012)
Skin-tone 0.163*** 0.153*** 0.167*** 0.165***
(0.031) (0.031) (0.031) (0.031)
Attractiveness 0.007*** 0.001 -0.003*
(0.001) (0.001) (0.001)
Competence 0.005*** 0.007*** 0.005***
(0.002) (0.001) (0.001)
Dominance -0.011*** -0.004*** -0.004***
(0.001) (0.001) (0.001)
Trustworthiness 0.006*** 0.005*** 0.003**
(0.002) (0.001) (0.001)
Well-Groomed 0.016*** 0.017*** 0.016***
(0.001) (0.001) (0.001)
Constant 0.874*** 0.725*** 0.832*** 0.667*** 0.772*** 0.773***
(0.013) (0.006) (0.014) (0.005) (0.014) (0.014)
Observations 8,821 8,821 8,821 8,821 8,821 8,821
Adjusted R2 0.204 0.028 0.214 0.025 0.229 0.232
F Statistic 323.060*** (df = 7; 8813) 65.473*** (df = 4; 8816) 219.821*** (df = 11; 8809) 229.145*** (df = 1; 8819) 328.819*** (df = 8; 8812) 223.109*** (df = 12; 8808)
Note: p<0.1; p<0.05; p<0.01

Table 5 - Well Groomed Predicting Judge Decision

Table No.5 - Does Well-Groomed Predict Judge Decisions ?
Dependent variable:
Release-Final-Outcome
(1) (2) (3) (4)
Well-Groomed 0.019*** 0.015*** 0.012*** 0.004
(0.004) (0.004) (0.005) (0.005)
Male -0.038*** -0.035*** 0.018
(0.011) (0.011) (0.012)
Age -0.002*** -0.001*** -0.001**
(0.0004) (0.0005) (0.0005)
Black 0.051 0.053 0.049
(0.048) (0.048) (0.048)
Risk Prediction -0.706*** -0.702*** -0.618***
(0.053) (0.053) (0.054)
Skin-tone 0.355*** 0.349*** 0.269**
(0.127) (0.127) (0.126)
Felony-Dummie -0.208*** -0.207*** -0.208***
(0.010) (0.010) (0.010)
Gun-Crime-Dummie 0.041** 0.041** 0.038**
(0.017) (0.017) (0.017)
Drug-Crime-Dummie 0.078*** 0.079*** 0.079***
(0.012) (0.012) (0.012)
Violent-Crime-Dummie -0.167*** -0.167*** -0.161***
(0.015) (0.015) (0.015)
Property-Crime-Dummie 0.017 0.017 0.016
(0.011) (0.011) (0.011)
Attractiveness 0.001 0.002
(0.006) (0.006)
Competence 0.001 -0.001
(0.006) (0.006)
Dominance -0.005 -0.003
(0.004) (0.004)
Trustworthiness 0.010* 0.009
(0.006) (0.006)
ML-Face CNN 0.504***
(0.044)
Constant 0.670*** 1.021*** 1.007*** 0.594***
(0.021) (0.059) (0.061) (0.070)
ROC-AUC 0.531 0.723 0.724 0.737
Observations 8,821 8,821 8,821 8,821
Adjusted R2 0.002 0.112 0.112 0.125
F Statistic 20.197*** (df = 1; 8819) 80.123*** (df = 14; 8806) 62.793*** (df = 18; 8802) 67.206*** (df = 19; 8801)
Note: p<0.1; p<0.05; p<0.01
We encode skin-tone as a continuous variable increasing in brightness

Other Tables:

(A) ML Fusion and Risk Prediction
ML-Fusion and Risk-predictions
Dependent variable:
Final Judge Release Decision
(1) (2) (3)
ML-Fusion 1.009*** 0.928***
(0.033) (0.033)
Risk Prediction -0.743*** -0.467***
(0.044) (0.043)
Constant -0.009 0.983*** 0.193***
(0.025) (0.014) (0.031)
ROC-AUC 0.708 0.624 0.722
Observations 8,821 8,821 8,821
Adjusted R2 0.098 0.031 0.110
F Statistic 958.772*** (df = 1; 8819) 283.490*** (df = 1; 8819) 543.372*** (df = 2; 8818)
Note: p<0.1; p<0.05; p<0.01
We encode skin-tone as a continuous variable increasing in brightness
(B) - ML Fusion. ML Face, and Risk Prediction
ML-Fusion and ML-Face
Dependent variable:
Release-Final-Outcome
(1) (2) (3) (4)
ML-Fusion 1.009*** 0.928***
(0.033) (0.033)
ML-Face 0.692*** 0.562***
(0.040) (0.041)
Risk Prediction -0.467*** -0.596***
(0.043) (0.045)
Constant -0.009 0.245*** 0.193*** 0.519***
(0.025) (0.030) (0.031) (0.036)
ROC-AUC 0.708 0.622 0.722 0.655
Observations 8,821 8,821 8,821 8,821
Adjusted R2 0.098 0.032 0.110 0.051
F Statistic 958.772*** (df = 1; 8819) 296.642*** (df = 1; 8819) 543.372*** (df = 2; 8818) 238.904*** (df = 2; 8818)
Note: p<0.1; p<0.05; p<0.01
(C) New Table (B)
ML-Fusion, ML-Face, and Risk Predictions
Dependent variable:
Release-Final-Outcome
(1) (2) (3) (4) (5)
ML-Fusion 1.009*** 0.928***
(0.033) (0.033)
Risk Prediction -0.743*** -0.467*** -0.596***
(0.044) (0.043) (0.045)
ML-Face 0.692*** 0.562***
(0.040) (0.041)
Constant -0.009 0.983*** 0.193*** 0.245*** 0.519***
(0.025) (0.014) (0.031) (0.030) (0.036)
ROC-AUC 0.708 0.624 0.722 0.622 0.655
Observations 8,821 8,821 8,821 8,821 8,821
Adjusted R2 0.098 0.031 0.110 0.032 0.051
F Statistic 958.772*** (df = 1; 8819) 283.490*** (df = 1; 8819) 543.372*** (df = 2; 8818) 296.642*** (df = 1; 8819) 238.904*** (df = 2; 8818)
Note: p<0.1; p<0.05; p<0.01