Clean-Cutness-Regressions

Regression Table 01 - Simple Clean-Cutness Regression

In this regression we include the new clean-cutness feature from the last MTurk survey.

ML Face includes the p_hat_cnn values for release outcomes
Tidy-Aggregate Labels includes the new MTurk feature. These are aggregated from ratings of the three different labels: “well_groomed” , “tidy”, and “clean-cut” for each individual image

**Table No.1 - Clean Cutness Regression**

	Dependent variable:

	Release Outcome
	(1)	(2)	(3)

Tidy-Aggregate Labels	0.0474		0.0138
	(-0.0039, 0.0986)		(-0.0368, 0.0644)

ML-Face		0.6989^***	0.6964^***
		(0.6024, 0.7954)	(0.5995, 0.7933)

Constant	0.7381^***	0.2402^***	0.2351^***
	(0.7102, 0.7660)	(0.1674, 0.3131)	(0.1599, 0.3103)


Observations	4,086	4,086	4,086
Adjusted R²	0.0003	0.0334	0.0332
F Statistic	2.3133 (df = 1; 4084)	141.9434^*** (df = 1; 4084)	71.0586^*** (df = 2; 4083)

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

Regression Table 02 - Full Feature Set

In this regression we include all the RHS variables used in our main model. The main ones are explained below:

Skin-Tone-Continuous: This is our MTurk skin-tone label encoded continuously from 0-1
Current Charge (p_hat): This is the p_hats from a glm trained on the charge-characteristics
Recidivism Risk (p_hat): This is the p_hats from the XgBoost model to predict recidivism

**Table No.2 - Full Feature Set**

	Dependent variable:

	Release Outcome
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)

SexM	-0.0493^***	-0.0478^***	0.0054	0.0063	0.0065	0.0063	0.0063	0.0063
	(-0.0752, -0.0233)	(-0.0740, -0.0216)	(-0.0227, 0.0336)	(-0.0220, 0.0346)	(-0.0219, 0.0348)	(-0.0220, 0.0346)	(-0.0220, 0.0346)	(-0.0220, 0.0346)

RaceB	0.1182^*	0.1195^*	0.1184^*	0.1190^*	0.1192^*	0.1191^*	0.1190^*	0.1189^*
	(0.0085, 0.2279)	(0.0098, 0.2293)	(0.0096, 0.2273)	(0.0100, 0.2279)	(0.0103, 0.2282)	(0.0101, 0.2281)	(0.0100, 0.2279)	(0.0100, 0.2279)

Age	-0.0002	-0.0003	0.0005	0.0004	0.0005	0.0004	0.0004	0.0004
	(-0.0016, 0.0011)	(-0.0016, 0.0011)	(-0.0009, 0.0019)	(-0.0009, 0.0018)	(-0.0009, 0.0019)	(-0.0009, 0.0018)	(-0.0009, 0.0018)	(-0.0009, 0.0018)

† skin-tone-cont	0.0504^*	0.0496^*	0.0391	0.0385	0.0386	0.0385	0.0384	0.0385
	(0.0054, 0.0955)	(0.0046, 0.0947)	(-0.0057, 0.0838)	(-0.0063, 0.0833)	(-0.0062, 0.0833)	(-0.0063, 0.0833)	(-0.0064, 0.0832)	(-0.0063, 0.0832)

Current Charge (p-hat)	1.0323^***	1.0308^***	0.9869^***	0.9848^***	0.9849^***	0.9848^***	0.9847^***	0.9848^***
	(0.9359, 1.1287)	(0.9344, 1.1272)	(0.8908, 1.0831)	(0.8886, 1.0809)	(0.8888, 1.0811)	(0.8886, 1.0810)	(0.8886, 1.0809)	(0.8886, 1.0809)

Recidivism Risk (p-hat)	-1.0004^***	-0.9977^***	-0.9379^***	-0.9370^***	-0.9359^***	-0.9368^***	-0.9373^***	-0.9370^***
	(-1.1503, -0.8505)	(-1.1478, -0.8476)	(-1.0873, -0.7886)	(-1.0865, -0.7875)	(-1.0854, -0.7864)	(-1.0864, -0.7873)	(-1.0869, -0.7877)	(-1.0866, -0.7875)

Attractiveness		-0.0559^*		-0.0601^**	-0.0602^**	-0.0602^**	-0.0602^**	-0.0601^**
		(-0.1055, -0.0062)		(-0.1094, -0.0109)	(-0.1095, -0.0109)	(-0.1095, -0.0109)	(-0.1095, -0.0109)	(-0.1094, -0.0108)

Competence		-0.0039		-0.0047	-0.0049	-0.0047	-0.0047	-0.0047
		(-0.0529, 0.0452)		(-0.0534, 0.0439)	(-0.0536, 0.0437)	(-0.0534, 0.0439)	(-0.0533, 0.0440)	(-0.0534, 0.0439)

Dominance		0.0023		0.0117	0.0124	0.0117	0.0115	0.0117
		(-0.0366, 0.0412)		(-0.0270, 0.0504)	(-0.0264, 0.0511)	(-0.0270, 0.0505)	(-0.0272, 0.0503)	(-0.0270, 0.0504)

Trustworthiness		0.0452		0.0401	0.0399	0.0401	0.0402	0.0401
		(-0.0048, 0.0952)		(-0.0095, 0.0897)	(-0.0097, 0.0895)	(-0.0095, 0.0897)	(-0.0095, 0.0898)	(-0.0096, 0.0897)

Likely-release		0.0427		0.0442	0.0439	0.0442	0.0442	0.0441
		(-0.0064, 0.0917)		(-0.0045, 0.0929)	(-0.0048, 0.0926)	(-0.0045, 0.0929)	(-0.0045, 0.0930)	(-0.0046, 0.0928)

ML Face			0.4977^***	0.4986^***	0.4968^***	0.4985^***	0.4989^***	0.4985^***
			(0.3939, 0.6015)	(0.3946, 0.6026)	(0.3927, 0.6009)	(0.3945, 0.6025)	(0.3949, 0.6030)	(0.3944, 0.6025)

Tidy-Aggregate Labels					0.0157
					(-0.0327, 0.0642)

Well-Groomed						0.0012
						(-0.0251, 0.0276)

Tidy							-0.0018
							(-0.0249, 0.0214)

Clean-Cut								0.0013
								(-0.0287, 0.0313)

Constant	0.1972^**	0.1818^*	-0.2230^**	-0.2364^**	-0.2451^**	-0.2373^**	-0.2354^**	-0.2370^**
	(0.0447, 0.3498)	(0.0269, 0.3368)	(-0.3979, -0.0481)	(-0.4132, -0.0596)	(-0.4240, -0.0663)	(-0.4152, -0.0594)	(-0.4127, -0.0581)	(-0.4144, -0.0596)


Observations	4,086	4,086	4,086	4,086	4,086	4,086	4,086	4,086
Adjusted R²	0.1173	0.1178	0.1304	0.1309	0.1308	0.1307	0.1307	0.1307
F Statistic	61.3209^*** (df = 9; 4076)	39.9789^*** (df = 14; 4071)	62.2366^*** (df = 10; 4075)	42.0234^*** (df = 15; 4070)	39.4079^*** (df = 16; 4069)	39.3876^*** (df = 16; 4069)	39.3884^*** (df = 16; 4069)	39.3876^*** (df = 16; 4069)

Note:	p<0.1; p<0.05; p<0.01
	† We encode skin-tone as a continuous variable increasing in brightness

Additional Regressions:

Excluding `ML-Face` from final column

**Table No.3 - Full Feature Set excluding ML Face**

	Dependent variable:

	Release Outcome
	(1)	(2)	(3)	(4)	(5)

SexM	-0.0493^***	-0.0478^***	0.0054	0.0063	-0.0472^***
	(-0.0752, -0.0233)	(-0.0740, -0.0216)	(-0.0227, 0.0336)	(-0.0220, 0.0346)	(-0.0733, -0.0210)

RaceB	0.1182^*	0.1195^*	0.1184^*	0.1190^*	0.1200^*
	(0.0085, 0.2279)	(0.0098, 0.2293)	(0.0096, 0.2273)	(0.0100, 0.2279)	(0.0102, 0.2297)

Age	-0.0002	-0.0003	0.0005	0.0004	-0.0002
	(-0.0016, 0.0011)	(-0.0016, 0.0011)	(-0.0009, 0.0019)	(-0.0009, 0.0018)	(-0.0016, 0.0012)

† skin-tone-cont	0.0504^*	0.0496^*	0.0391	0.0385	0.0497^*
	(0.0054, 0.0955)	(0.0046, 0.0947)	(-0.0057, 0.0838)	(-0.0063, 0.0833)	(0.0046, 0.0947)

Current Charge (p-hat)	1.0323^***	1.0308^***	0.9869^***	0.9848^***	1.0308^***
	(0.9359, 1.1287)	(0.9344, 1.1272)	(0.8908, 1.0831)	(0.8886, 1.0809)	(0.9344, 1.1271)

Recidivism Risk (p-hat)	-1.0004^***	-0.9977^***	-0.9379^***	-0.9370^***	-0.9954^***
	(-1.1503, -0.8505)	(-1.1478, -0.8476)	(-1.0873, -0.7886)	(-1.0865, -0.7875)	(-1.1455, -0.8452)

Attractiveness		-0.0559^*		-0.0601^**	-0.0560^*
		(-0.1055, -0.0062)		(-0.1094, -0.0109)	(-0.1057, -0.0064)

Competence		-0.0039		-0.0047	-0.0042
		(-0.0529, 0.0452)		(-0.0534, 0.0439)	(-0.0532, 0.0448)

Dominance		0.0023		0.0117	0.0036
		(-0.0366, 0.0412)		(-0.0270, 0.0504)	(-0.0354, 0.0426)

Trustworthiness		0.0452		0.0401	0.0448
		(-0.0048, 0.0952)		(-0.0095, 0.0897)	(-0.0051, 0.0948)

Likely-release		0.0427		0.0442	0.0422
		(-0.0064, 0.0917)		(-0.0045, 0.0929)	(-0.0069, 0.0912)

ML-Face			0.4977^***	0.4986^***
			(0.3939, 0.6015)	(0.3946, 0.6026)

Tidy-Aggregate Labels					0.0278
					(-0.0209, 0.0765)

Constant	0.1972^**	0.1818^*	-0.2230^**	-0.2364^**	0.1637^*
	(0.0447, 0.3498)	(0.0269, 0.3368)	(-0.3979, -0.0481)	(-0.4132, -0.0596)	(0.0056, 0.3219)


Observations	4,086	4,086	4,086	4,086	4,086
Adjusted R²	0.1173	0.1178	0.1304	0.1309	0.1178
F Statistic	61.3209^*** (df = 9; 4076)	39.9789^*** (df = 14; 4071)	62.2366^*** (df = 10; 4075)	42.0234^*** (df = 15; 4070)	37.3714^*** (df = 15; 4070)

Note:	p<0.1; p<0.05; p<0.01
	† We encode skin-tone as a continuous variable increasing in brightness

Regressing `ML-Face` on `Tidy-Aggregate`

**Table No.4 - P-hat-release vs. Groomed Labeles**

	Dependent variable:

	P-Hat-CNN
	(1)	(2)	(3)	(4)

Tidy-Aggregate Labels	0.0482^***
	(0.0348, 0.0616)

Well-Groomed		0.0117^***
		(0.0044, 0.0191)

Tidy			0.0151^***
			(0.0087, 0.0216)

Clean-Cut				0.0205^***
				(0.0122, 0.0289)

Constant	0.7222^***	0.7405^***	0.7387^***	0.7361^***
	(0.7149, 0.7295)	(0.7358, 0.7452)	(0.7344, 0.7431)	(0.7310, 0.7412)


Observations	4,086	4,086	4,086	4,086
Adjusted R²	0.0083	0.0014	0.0034	0.0037
F Statistic (df = 1; 4084)	35.0852^***	6.8766^***	14.8505^***	16.2739^***

Note:	p<0.1; p<0.05; p<0.01
	† We encode skin-tone as a continuous variable increasing in brightness

Regressing `Tidy-Aggregate` on Mturk Labels

**Table No.5 - Tidy Aggregate**

	Dependent variable:

	Tidy-Aggregate
	(1)	(2)	(3)	(4)	(5)	(6)	(7)

attractiveness	0.0284^**
	(0.0094, 0.0474)

competence		0.0242^**
		(0.0050, 0.0433)

dominance			-0.0356^***
			(-0.0545, -0.0166)

trustworthiness				0.0305^***
				(0.0115, 0.0495)

sexM					-0.0370^***
					(-0.0504, -0.0236)

raceB						-0.0264
						(-0.0838, 0.0311)

age_arrest							-0.0016^***
							(-0.0021, -0.0011)

Constant	0.4875^***	0.4895^***	0.5194^***	0.4864^***	0.5307^***	0.5213^***	0.5525^***
	(0.4765, 0.4984)	(0.4785, 0.5006)	(0.5085, 0.5303)	(0.4755, 0.4974)	(0.5189, 0.5426)	(0.4642, 0.5783)	(0.5364, 0.5685)


Observations	4,086	4,086	4,086	4,086	4,086	4,086	4,086
Adjusted R²	0.0012	0.0008	0.0021	0.0015	0.0048	0.0020	0.0072
F Statistic	6.0305^** (df = 1; 4084)	4.3166^** (df = 1; 4084)	9.5553^*** (df = 1; 4084)	6.9928^*** (df = 1; 4084)	20.7169^*** (df = 1; 4084)	3.0454^** (df = 4; 4081)	30.7715^*** (df = 1; 4084)

Note:	p<0.1; p<0.05; p<0.01
	† We encode skin-tone as a continuous variable increasing in brightness

Clean-Cutness-Regressions

Regression Table 01 - Simple Clean-Cutness Regression

Regression Table 02 - Full Feature Set

Additional Regressions:

Excluding ML-Face from final column

Regressing ML-Face on Tidy-Aggregate

Regressing Tidy-Aggregate on Mturk Labels

Excluding `ML-Face` from final column

Regressing `ML-Face` on `Tidy-Aggregate`

Regressing `Tidy-Aggregate` on Mturk Labels