AUCs

Approach 1: Fitting logit classifier on full validation set

AUC of all Models
Model ID	Model Name	Pre-Trained	Baseline	Just Photo	Photo and Covariates
1	inception-flat_decay	imagenet	0.585 (0.555-0.615)	0.669 (0.640-0.697)	0.672 (0.643-0.700)
2	inception-steep_decay	imagenet	0.585 (0.555-0.615)	0.656 (0.627-0.685)	0.660 (0.631-0.688)
3	resnet18-flat_decay	imagenet	0.585 (0.555-0.615)	0.685 (0.657-0.713)	0.689 (0.661-0.716)
4	resnet18-gentle_decay	imagenet	0.585 (0.555-0.615)	0.685 (0.657-0.713)	0.686 (0.658-0.714)
5	resnet18-steep_decay	imagenet	0.585 (0.555-0.615)	0.687 (0.658-0.715)	0.689 (0.661-0.717)
6	resnet50-flat_decay	imagenet	0.585 (0.555-0.615)	0.691 (0.664-0.719)	0.696 (0.668-0.723)
7	resnet50-gentle_decay	imagenet	0.585 (0.555-0.615)	0.683 (0.655-0.711)	0.687 (0.659-0.715)
8	resnet50-steep_decay	imagenet	0.585 (0.555-0.615)	0.692 (0.665-0.720)	0.693 (0.665-0.720)

Multihead_01 Summary Output

**Model 1 (ResNet-50) Classifier (full validation set)**

	Dependent variable:

	Election Outcome (did win)
	(1)	(2)	(3)

P-hat Baseline	1.030^***		0.536^***
	(0.726, 1.333)		(0.232, 0.840)

P-hat CNN		0.440^***	0.408^***
		(0.375, 0.504)	(0.342, 0.475)

Constant	-0.016	0.409^***	0.112
	(-0.192, 0.160)	(0.377, 0.441)	(-0.060, 0.283)


Observations	1,393	1,393	1,393
Adjusted R²	0.021	0.082	0.087
F Statistic	31.167^*** (df = 1; 1391)	125.498^*** (df = 1; 1391)	67.291^*** (df = 2; 1390)

Note:	p<0.1; p<0.05; p<0.01

Multihead_03 Summary Output

**Model 1 (ResNet-50) Classifier (full validation set)**

	Dependent variable:

	Election Outcome (did win)
	(1)	(2)	(3)

P-hat Baseline	1.030^***		0.415^**
	(0.726, 1.333)		(0.111, 0.718)

P-hat CNN		0.771^***	0.728^***
		(0.670, 0.871)	(0.623, 0.834)

Constant	-0.016	0.160^***	-0.056
	(-0.192, 0.160)	(0.102, 0.219)	(-0.225, 0.113)


Observations	1,393	1,393	1,393
Adjusted R²	0.021	0.101	0.104
F Statistic	31.167^*** (df = 1; 1391)	157.765^*** (df = 1; 1391)	81.635^*** (df = 2; 1390)

Note:	p<0.1; p<0.05; p<0.01

Multihead_04 Summary Output

**Model 1 (ResNet-50) Classifier (full validation set)**

	Dependent variable:

	Election Outcome (did win)
	(1)	(2)	(3)

P-hat Baseline	1.030^***		0.284
	(0.726, 1.333)		(-0.026, 0.595)

P-hat CNN		0.851^***	0.812^***
		(0.739, 0.962)	(0.694, 0.931)

Constant	-0.016	0.073^*	-0.069
	(-0.192, 0.160)	(0.003, 0.142)	(-0.238, 0.100)


Observations	1,393	1,393	1,393
Adjusted R²	0.021	0.101	0.102
F Statistic	31.167^*** (df = 1; 1391)	158.231^*** (df = 1; 1391)	80.325^*** (df = 2; 1390)

Note:	p<0.1; p<0.05; p<0.01

Multihead_05 Summary Output

**Model 1 (ResNet-50) Classifier (full validation set)**

	Dependent variable:

	Election Outcome (did win)
	(1)	(2)	(3)

P-hat Baseline	1.030^***		0.436^**
	(0.726, 1.333)		(0.133, 0.739)

P-hat CNN		0.569^***	0.537^***
		(0.494, 0.644)	(0.459, 0.615)

Constant	-0.016	0.335^***	0.097
	(-0.192, 0.160)	(0.297, 0.373)	(-0.072, 0.267)


Observations	1,393	1,393	1,393
Adjusted R²	0.021	0.100	0.103
F Statistic	31.167^*** (df = 1; 1391)	155.755^*** (df = 1; 1391)	80.941^*** (df = 2; 1390)

Note:	p<0.1; p<0.05; p<0.01

Multihead_06 Summary Output

**Model 4 (ResNet-50) Classifier (full validation set)**

	Dependent variable:

	Election Outcome (did win)
	(1)	(2)	(3)

P-hat Baseline	1.030^***		0.494^***
	(0.726, 1.333)		(0.195, 0.793)

P-hat CNN		0.655^***	0.619^***
		(0.570, 0.739)	(0.532, 0.706)

Constant	-0.016	0.304^***	0.034
	(-0.192, 0.160)	(0.263, 0.344)	(-0.135, 0.202)


Observations	1,393	1,393	1,393
Adjusted R²	0.021	0.104	0.108
F Statistic	31.167^*** (df = 1; 1391)	162.507^*** (df = 1; 1391)	85.317^*** (df = 2; 1390)

Note:	p<0.1; p<0.05; p<0.01

Multihead_07 Summary Output

**Model 1 (ResNet-50) Classifier (full validation set)**

	Dependent variable:

	Election Outcome (did win)
	(1)	(2)	(3)

P-hat Baseline	1.030^***		0.417^**
	(0.726, 1.333)		(0.113, 0.722)

P-hat CNN		0.640^***	0.603^***
		(0.554, 0.725)	(0.514, 0.693)

Constant	-0.016	0.261^***	0.039
	(-0.192, 0.160)	(0.214, 0.308)	(-0.131, 0.208)


Observations	1,393	1,393	1,393
Adjusted R²	0.021	0.098	0.100
F Statistic	31.167^*** (df = 1; 1391)	151.442^*** (df = 1; 1391)	78.482^*** (df = 2; 1390)

Note:	p<0.1; p<0.05; p<0.01

Multihead_08 Summary Output

**Model 8 (ResNet-50) Classifier (full validation set)**

	Dependent variable:

	Election Outcome (did win)
	(1)	(2)	(3)

P-hat Baseline	1.030^***		0.459^**
	(0.726, 1.333)		(0.157, 0.760)

P-hat CNN		0.636^***	0.600^***
		(0.552, 0.720)	(0.513, 0.687)

Constant	-0.016	0.166^***	-0.075
	(-0.192, 0.160)	(0.108, 0.224)	(-0.244, 0.094)


Observations	1,393	1,393	1,393
Adjusted R²	0.021	0.100	0.104
F Statistic	31.167^*** (df = 1; 1391)	156.029^*** (df = 1; 1391)	81.431^*** (df = 2; 1390)

Note:	p<0.1; p<0.05; p<0.01

Plotting Model Coeficients:

Combining models

AUC of Combination Models
Model ID	Baseline	Combined Models
1-2	0.585 (0.555-0.615)	0.681 (0.653-0.709)
1-2-3-4	0.585 (0.555-0.615)	0.706 (0.678-0.733)
1-2-3-4-5-6-7-8	0.585 (0.555-0.615)	0.721 (0.695-0.748)
3-4	0.585 (0.555-0.615)	0.697 (0.669-0.724)
5-6	0.585 (0.555-0.615)	0.710 (0.683-0.737)
5-6-7-8	0.585 (0.555-0.615)	0.716 (0.689-0.743)
7-8	0.585 (0.555-0.615)	0.703 (0.676-0.731)

Decile Plots

Only for p_hat_cnn from model 04 (highest AUC)

Decile Plot 1 - p_hat_cnn

Decile Plot 2 - p_hat_baseline

Separated by Gender:

AUC of Models Split by Sex
Model ID	Model Name	Pretrained On	Female Baseline	Female Full Model	Male Baseline	Male Full Model
1	inception-flat_decay	imagenet	0.486 (0.432-0.540)	0.625 (0.574-0.677)	0.558 (0.521-0.595)	0.676 (0.641-0.711)
2	inception-steep_decay	imagenet	0.486 (0.432-0.540)	0.582 (0.529-0.635)	0.558 (0.521-0.595)	0.669 (0.634-0.704)
3	resnet18-flat_decay	imagenet	0.486 (0.432-0.540)	0.638 (0.587-0.689)	0.558 (0.521-0.595)	0.688 (0.654-0.723)
4	resnet18-gentle_decay	imagenet	0.486 (0.432-0.540)	0.653 (0.602-0.703)	0.558 (0.521-0.595)	0.680 (0.645-0.715)
5	resnet18-steep_decay	imagenet	0.486 (0.432-0.540)	0.632 (0.580-0.683)	0.558 (0.521-0.595)	0.698 (0.664-0.733)
6	resnet50-flat_decay	imagenet	0.486 (0.432-0.540)	0.658 (0.608-0.709)	0.558 (0.521-0.595)	0.702 (0.668-0.736)
7	resnet50-gentle_decay	imagenet	0.486 (0.432-0.540)	0.632 (0.580-0.683)	0.558 (0.521-0.595)	0.695 (0.661-0.730)
8	resnet50-steep_decay	imagenet	0.486 (0.432-0.540)	0.648 (0.598-0.699)	0.558 (0.521-0.595)	0.700 (0.666-0.734)
1-2-3-4-5-6-7-8			0.486 (0.432-0.540)	0.681 (0.632-0.730)	0.558 (0.521-0.595)	0.730 (0.697-0.762)

Regression Outputs - Model 04

**Male Models**

	Dependent variable:

	Election Outcome
	(1)	(2)	(3)

P-hat Baseline	0.889^***		0.180
	(0.476, 1.302)		(-0.233, 0.593)

P-hat CNN		0.859^***	0.840^***
		(0.721, 0.996)	(0.696, 0.984)

Constant	0.081	0.081	-0.017
	(-0.171, 0.332)	(-0.009, 0.170)	(-0.257, 0.224)


Observations	944	944	944
Adjusted R²	0.012	0.100	0.099
F Statistic	12.519^*** (df = 1; 942)	105.410^*** (df = 1; 942)	52.934^*** (df = 2; 941)

Note:	p<0.1; p<0.05; p<0.01

**Female Models**

	Dependent variable:

	Election Outcome
	(1)	(2)	(3)

P-hat Baseline	0.455		0.016
	(-0.433, 1.342)		(-0.851, 0.882)

P-hat CNN		0.742^***	0.742^***
		(0.532, 0.952)	(0.529, 0.954)

Constant	0.256	0.102	0.094
	(-0.204, 0.716)	(-0.014, 0.218)	(-0.353, 0.541)


Observations	448	448	448
Adjusted R²	-0.001	0.068	0.066
F Statistic	0.711 (df = 1; 446)	33.851^*** (df = 1; 446)	16.888^*** (df = 2; 445)

Note:	p<0.1; p<0.05; p<0.01