Distribution plot for p_hat_cnn

I plot the difference in distribution between the old model without the minority-class-sampler and the new model with an updated data-loader.

Decile Plot - p_hat_cnn

Repeating the decile plots for the new p_hat_cnn we notice that we don’t see the high degree of non-linearity.

Baseline Regression - Increase MTurk Detail

We repeat our baseline regressions with the new CNN model.

Multihead(ResNet50)
Dependent variable:
Release Outcome
(1) (2) (3) (4) (5)
risk_pred_prob -1.105*** -1.107*** -1.106*** -0.778*** -0.696***
(-1.211, -0.998) (-1.215, -1.000) (-1.214, -0.999) (-0.883, -0.673) (-0.801, -0.591)
skin_tonenumber_f7ddc4 0.002 0.002 -0.027 -0.025
(-0.033, 0.037) (-0.033, 0.038) (-0.061, 0.007) (-0.059, 0.009)
age -0.0003 0.0003 0.001
(-0.001, 0.001) (-0.001, 0.001) (-0.0002, 0.002)
attractiveness -0.002 0.001 0.003
(-0.012, 0.008) (-0.009, 0.010) (-0.007, 0.013)
competence 0.003 -0.001 -0.004
(-0.009, 0.015) (-0.013, 0.010) (-0.015, 0.007)
dominance 0.001 0.005 0.006
(-0.008, 0.009) (-0.003, 0.012) (-0.002, 0.014)
trustworthiness 0.001 0.001 -0.0002
(-0.010, 0.011) (-0.009, 0.011) (-0.010, 0.010)
p_hat_covariates 1.083*** 1.006***
(1.018, 1.148) (0.940, 1.071)
p_hat_cnn 0.380***
(0.319, 0.441)
Constant 1.102*** 1.088*** 1.088*** 0.114** -0.110*
(1.069, 1.136) (1.045, 1.131) (1.017, 1.158) (0.024, 0.203) (-0.206, -0.015)
Observations 8,479 8,479 8,479 8,479 8,479
Adjusted R2 0.033 0.033 0.033 0.112 0.123
F Statistic 291.841*** (df = 1; 8477) 17.216*** (df = 18; 8460) 13.490*** (df = 23; 8455) 45.641*** (df = 24; 8454) 48.550*** (df = 25; 8453)
Note: p<0.1; p<0.05; p<0.01

Gender Split

Female Subsample:

Multihead(ResNet50)
Dependent variable:
Release Outcome
(1) (2) (3) (4) (5)
risk_pred_prob -0.966*** -0.946*** -0.916*** -0.679*** -0.569***
(-1.232, -0.700) (-1.214, -0.678) (-1.185, -0.647) (-0.942, -0.416) (-0.830, -0.308)
skin_tonenumber_f7ddc4 -0.071 -0.077* -0.085* -0.045
(-0.144, 0.003) (-0.150, -0.003) (-0.157, -0.013) (-0.116, 0.027)
age -0.002 -0.001 0.00004
(-0.004, 0.0002) (-0.003, 0.001) (-0.002, 0.002)
attractiveness 0.012 0.012 0.013
(-0.006, 0.031) (-0.006, 0.030) (-0.005, 0.031)
competence -0.012 -0.014 -0.019
(-0.034, 0.009) (-0.035, 0.007) (-0.040, 0.002)
dominance 0.019** 0.016* 0.017*
(0.003, 0.034) (0.001, 0.031) (0.002, 0.032)
trustworthiness 0.004 0.005 0.004
(-0.015, 0.024) (-0.014, 0.024) (-0.015, 0.022)
p_hat_covariates 1.140*** 1.092***
(0.969, 1.311) (0.923, 1.261)
p_hat_cnn 0.509***
(0.391, 0.627)
Constant 1.118*** 1.145*** 1.095*** 0.043 -0.344**
(1.041, 1.195) (1.046, 1.244) (0.951, 1.239) (-0.168, 0.254) (-0.570, -0.117)
Observations 1,833 1,833 1,833 1,833 1,833
Adjusted R2 0.019 0.022 0.026 0.086 0.110
F Statistic 35.610*** (df = 1; 1831) 3.239*** (df = 18; 1814) 3.113*** (df = 23; 1809) 8.192*** (df = 24; 1808) 10.083*** (df = 25; 1807)
Note: p<0.1; p<0.05; p<0.01

Male Subsample:

Multihead(ResNet50)
Dependent variable:
Release Outcome
(1) (2) (3) (4) (5)
risk_pred_prob -1.039*** -1.043*** -1.048*** -0.791*** -0.732***
(-1.158, -0.920) (-1.163, -0.923) (-1.168, -0.928) (-0.908, -0.674) (-0.848, -0.615)
skin_tonenumber_f7ddc4 0.001 0.003 -0.013 -0.013
(-0.040, 0.043) (-0.039, 0.045) (-0.053, 0.027) (-0.053, 0.027)
age 0.001 0.001 0.001
(-0.001, 0.002) (-0.001, 0.002) (-0.0003, 0.002)
attractiveness -0.006 -0.002 0.001
(-0.018, 0.005) (-0.013, 0.010) (-0.010, 0.012)
competence 0.005 0.001 -0.001
(-0.008, 0.019) (-0.012, 0.015) (-0.014, 0.013)
dominance 0.002 0.0004 -0.001
(-0.008, 0.012) (-0.009, 0.010) (-0.010, 0.009)
trustworthiness -0.003 -0.0005 -0.001
(-0.015, 0.010) (-0.012, 0.011) (-0.013, 0.011)
p_hat_covariates 1.086*** 1.040***
(1.011, 1.160) (0.966, 1.115)
p_hat_cnn 0.396***
(0.321, 0.471)
Constant 1.066*** 1.053*** 1.041*** 0.130** -0.107
(1.028, 1.105) (1.005, 1.101) (0.959, 1.124) (0.029, 0.231) (-0.216, 0.003)
Observations 6,646 6,646 6,646 6,646 6,646
Adjusted R2 0.030 0.030 0.029 0.107 0.117
F Statistic 206.526*** (df = 1; 6644) 12.312*** (df = 18; 6627) 9.751*** (df = 23; 6622) 34.077*** (df = 24; 6621) 36.110*** (df = 25; 6620)
Note: p<0.1; p<0.05; p<0.01