Deep Expected Average Age CNN - Summary Tables

Basic Summary Plots:

Stratified Summary Plots by Gender:

Stratified Summary Plots by Race:

Regression Plots:

Here I’m regressing p_hat_age_arrest on the true label age_arrest and a host of other features.

DEX-CNN ResNet-50
Dependent variable:
P-Hat Age-Arrest
(1) (2) (3) (4) (5) (6) (7)
age_arrest 0.756*** 0.750***
(0.749, 0.764) (0.742, 0.757)
sexM 0.617** 0.989*** 0.718*** 0.341***
(0.182, 1.052) (0.556, 1.422) (0.283, 1.152) (0.130, 0.552)
raceB 0.753 0.751 0.729 1.412**
(-1.240, 2.747) (-1.241, 2.742) (-1.250, 2.709) (0.451, 2.373)
raceI -4.058 -3.854 -3.356 0.375
(-8.692, 0.576) (-8.486, 0.777) (-7.958, 1.246) (-1.859, 2.609)
raceU -1.823 -1.896 -1.770 0.829
(-4.153, 0.508) (-4.225, 0.433) (-4.084, 0.544) (-0.294, 1.953)
raceW 3.760*** 3.851*** 3.948*** 2.757***
(1.749, 5.771) (1.841, 5.860) (1.951, 5.945) (1.787, 3.726)
attractiveness -4.134*** -4.317*** -0.499**
(-4.918, -3.350) (-5.095, -3.540) (-0.878, -0.119)
competence -0.254 -0.193 0.050
(-1.045, 0.536) (-0.974, 0.589) (-0.330, 0.429)
dominance 1.567*** 1.761*** 0.108
(0.928, 2.206) (1.123, 2.399) (-0.202, 0.418)
trustworthiness 0.538 0.612 -0.166
(-0.242, 1.319) (-0.160, 1.384) (-0.541, 0.209)
Constant 7.103*** 30.557*** 29.538*** 28.741*** 32.185*** 30.010*** 5.551***
(6.851, 7.354) (30.171, 30.942) (27.556, 31.520) (26.730, 30.752) (31.750, 32.620) (27.980, 32.041) (4.536, 6.567)
Observations 8,370 8,370 8,370 8,370 8,370 8,370 8,370
Adjusted R2 0.769 0.001 0.020 0.022 0.012 0.035 0.773
F Statistic 27,802.050*** (df = 1; 8368) 5.437** (df = 1; 8368) 44.777*** (df = 4; 8365) 38.698*** (df = 5; 8364) 27.234*** (df = 4; 8365) 34.864*** (df = 9; 8360) 2,844.760*** (df = 10; 8359)
Note: p<0.1; p<0.05; p<0.01
DEX-CNN ResNet-50
Dependent variable:
True Age-Arrest
(1) (2) (3) (4) (5) (6) (7)
age_arrest_pred 1.016*** 1.020***
(1.006, 1.026) (1.010, 1.030)
sexM 0.533* 0.820*** 0.502 -0.229
(0.029, 1.038) (0.315, 1.326) (-0.004, 1.009) (-0.475, 0.017)
raceB -0.866 -0.868 -0.911 -1.654**
(-3.190, 1.458) (-3.191, 1.455) (-3.220, 1.398) (-2.775, -0.533)
raceI -5.689* -5.520* -4.978 -1.556
(-11.091, -0.286) (-10.921, -0.118) (-10.345, 0.389) (-4.161, 1.050)
raceU -3.550** -3.610** -3.468** -1.663**
(-6.267, -0.832) (-6.327, -0.894) (-6.167, -0.769) (-2.973, -0.352)
raceW 1.410 1.486 1.589 -2.437***
(-0.934, 3.755) (-0.858, 3.830) (-0.740, 3.918) (-3.568, -1.306)
attractiveness -4.958*** -5.095*** -0.692**
(-5.867, -4.050) (-6.002, -4.188) (-1.134, -0.249)
competence -0.363 -0.323 -0.127
(-1.279, 0.553) (-1.235, 0.588) (-0.570, 0.316)
dominance 2.040*** 2.205*** 0.409*
(1.300, 2.780) (1.461, 2.949) (0.047, 0.771)
trustworthiness 0.985* 1.037* 0.414
(0.081, 1.890) (0.137, 1.938) (-0.024, 0.851)
Constant 0.104 31.232*** 31.955*** 31.294*** 32.801*** 32.631*** 2.028***
(-0.223, 0.430) (30.785, 31.679) (29.645, 34.266) (28.949, 33.639) (32.297, 33.305) (30.263, 35.000) (0.838, 3.218)
Observations 8,370 8,370 8,370 8,370 8,370 8,370 8,370
Adjusted R2 0.769 0.0002 0.009 0.010 0.013 0.023 0.770
F Statistic 27,802.050*** (df = 1; 8368) 3.020* (df = 1; 8368) 20.454*** (df = 4; 8365) 17.802*** (df = 5; 8364) 27.990*** (df = 4; 8365) 23.048*** (df = 9; 8360) 2,799.631*** (df = 10; 8359)
Note: p<0.1; p<0.05; p<0.01