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
Each of the labels Well-groomed, Happy, Threatening and Messy come from the latest one-sided MTurk survey
I include each label separately, and also group Well-groomed + Happy as well as Threatening + Messy (the thought was that these groups capture generally positive/negative perception)
We can see that adjusted r-squared increases (which is great) and that labels remain significant

**Table No.1 - Baseline Regression for new one-sided MTurk labels**

	Dependent variable:

	Arrest-Final-Outcome
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)

Well-groomed	0.0121^***		0.0092^***				0.0158^***		0.0095^*
	(0.0069, 0.0173)		(0.0038, 0.0145)				(0.0074, 0.0242)		(0.0012, 0.0178)

Happy		0.0186^***	0.0160^***				0.0116^***		0.0066
		(0.0120, 0.0252)	(0.0092, 0.0228)				(0.0045, 0.0188)		(-0.0005, 0.0137)

Threatening				-0.0218^***		-0.0208^***	-0.0176^***		-0.0061
				(-0.0289, -0.0147)		(-0.0280, -0.0137)	(-0.0250, -0.0101)		(-0.0135, 0.0014)

Messy					-0.0065^**	-0.0048^*	0.0078^*		0.0069
					(-0.0112, -0.0019)	(-0.0094, -0.0001)	(0.0004, 0.0153)		(-0.0005, 0.0142)

ML-Face								0.6937^***	0.6657^***
								(0.6260, 0.7613)	(0.5959, 0.7354)

full_model_pred										1.1289^***
										(0.8499, 1.4078)

Constant	0.6994^***	0.6942^***	0.6562^***	0.8298^***	0.7888^***	0.8464^***	0.6605^***	0.2448^***	0.1829^***	-0.0797
	(0.6712, 0.7275)	(0.6690, 0.7195)	(0.6227, 0.6898)	(0.8065, 0.8531)	(0.7684, 0.8091)	(0.8181, 0.8747)	(0.5813, 0.7397)	(0.1938, 0.2958)	(0.0902, 0.2757)	(-0.2879, 0.1284)


Observations	8,336	8,336	8,336	8,336	8,336	8,336	8,336	8,336	8,336	8,336
Adjusted R²	0.0016	0.0024	0.0033	0.0030	0.0005	0.0032	0.0051	0.0329	0.0336	0.0052
F Statistic	14.5025^*** (df = 1; 8334)	21.4600^*** (df = 1; 8334)	14.7345^*** (df = 2; 8333)	25.6885^*** (df = 1; 8334)	5.4676^** (df = 1; 8334)	14.2806^*** (df = 2; 8333)	11.6643^*** (df = 4; 8331)	284.6992^*** (df = 1; 8334)	58.9077^*** (df = 5; 8330)	44.2966^*** (df = 1; 8334)

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 - Including new one-sided MTurk labels**

	Dependent variable:

	Arrest-Final-Outcome
	(1)	(2)	(3)	(4)	(5)	(6)

SexM	-0.0549^***	-0.0032	-0.0054	-0.0030	-0.0018	-0.0046
	(-0.0732, -0.0366)	(-0.0230, 0.0166)	(-0.0253, 0.0145)	(-0.0228, 0.0168)	(-0.0218, 0.0182)	(-0.0248, 0.0155)

RaceB	0.0414	0.0436	0.0448	0.0436	0.0439	0.0439
	(-0.0410, 0.1238)	(-0.0383, 0.1254)	(-0.0371, 0.1266)	(-0.0382, 0.1255)	(-0.0380, 0.1257)	(-0.0380, 0.1257)

Age	-0.0012^**	-0.0005	-0.0004	-0.0005	-0.0005	-0.0005
	(-0.0021, -0.0002)	(-0.0015, 0.0004)	(-0.0013, 0.0006)	(-0.0015, 0.0005)	(-0.0015, 0.0004)	(-0.0014, 0.0005)

† skin-tone	0.0556^***	0.0425^**	0.0432^**	0.0424^**	0.0420^**	0.0429^**
	(0.0237, 0.0875)	(0.0108, 0.0742)	(0.0115, 0.0749)	(0.0107, 0.0741)	(0.0103, 0.0737)	(0.0112, 0.0746)

Current Charge (p-hat)	1.1136^***	1.0729^***	1.0739^***	1.0717^***	1.0725^***	1.0736^***
	(1.0455, 1.1817)	(1.0050, 1.1408)	(1.0060, 1.1419)	(1.0037, 1.1397)	(1.0046, 1.1405)	(1.0056, 1.1415)

Recidivism (p-hat)	-0.8199^***	-0.7664^***	-0.7598^***	-0.7654^***	-0.7652^***	-0.7640^***
	(-0.9263, -0.7136)	(-0.8724, -0.6604)	(-0.8660, -0.6536)	(-0.8714, -0.6594)	(-0.8712, -0.6592)	(-0.8701, -0.6578)

Attractiveness	-0.0151	-0.0166	-0.0176	-0.0168	-0.0167	-0.0168
	(-0.0481, 0.0179)	(-0.0493, 0.0162)	(-0.0503, 0.0152)	(-0.0495, 0.0160)	(-0.0494, 0.0161)	(-0.0496, 0.0159)

Competence	0.0080	0.0034	0.0027	0.0029	0.0033	0.0030
	(-0.0249, 0.0409)	(-0.0293, 0.0361)	(-0.0300, 0.0354)	(-0.0298, 0.0356)	(-0.0294, 0.0360)	(-0.0297, 0.0357)

Dominance	-0.0042	-0.0003	0.0001	0.0005	0.0009	-0.0001
	(-0.0305, 0.0220)	(-0.0263, 0.0258)	(-0.0260, 0.0262)	(-0.0257, 0.0266)	(-0.0253, 0.0271)	(-0.0262, 0.0260)

Trustworthiness	0.0279	0.0244	0.0241	0.0241	0.0238	0.0243
	(-0.0055, 0.0613)	(-0.0088, 0.0576)	(-0.0091, 0.0573)	(-0.0091, 0.0573)	(-0.0094, 0.0570)	(-0.0089, 0.0575)

Likely-release	0.0132	0.0134	0.0132	0.0133	0.0129	0.0132
	(-0.0198, 0.0462)	(-0.0193, 0.0462)	(-0.0196, 0.0459)	(-0.0195, 0.0460)	(-0.0198, 0.0457)	(-0.0195, 0.0460)

ML Face		0.4779^***	0.4677^***	0.4745^***	0.4748^***	0.4731^***
		(0.4052, 0.5505)	(0.3943, 0.5412)	(0.4015, 0.5475)	(0.4018, 0.5477)	(0.3994, 0.5467)

Well-groomed			0.0047
			(-0.0004, 0.0097)

Happy				0.0028
				(-0.0035, 0.0091)

Threatening					-0.0032
					(-0.0101, 0.0038)

Messy						-0.0018
						(-0.0063, 0.0028)

Constant	0.1654^**	-0.2335^***	-0.2573^***	-0.2416^***	-0.2226^***	-0.2252^***
	(0.0524, 0.2783)	(-0.3610, -0.1060)	(-0.3874, -0.1272)	(-0.3704, -0.1128)	(-0.3523, -0.0929)	(-0.3544, -0.0959)


Observations	8,336	8,336	8,336	8,336	8,336	8,336
Adjusted R²	0.1176	0.1298	0.1299	0.1297	0.1297	0.1297
F Statistic	80.3590^*** (df = 14; 8321)	83.8572^*** (df = 15; 8320)	78.7712^*** (df = 16; 8319)	78.6452^*** (df = 16; 8319)	78.6474^*** (df = 16; 8319)	78.6365^*** (df = 16; 8319)

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

In this table I exclude ML-Face (i.e. the CNN signal) from the regression. We can see that the label for Well-Groomed then appear significant, even with the inclusion of all other covariates and features (i.e. recidivism and current-charge)

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

	Dependent variable:

	Arrest-Final-Outcome
	(1)	(2)	(3)	(4)	(5)	(6)

SexM	-0.0032	-0.0571^***	-0.0535^***	-0.0509^***	-0.0583^***	-0.0499^***
	(-0.0230, 0.0166)	(-0.0755, -0.0388)	(-0.0719, -0.0352)	(-0.0696, -0.0322)	(-0.0768, -0.0399)	(-0.0684, -0.0314)

RaceB	0.0436	0.0439	0.0416	0.0421	0.0426	0.0436
	(-0.0383, 0.1254)	(-0.0384, 0.1263)	(-0.0408, 0.1240)	(-0.0403, 0.1245)	(-0.0398, 0.1250)	(-0.0388, 0.1259)

Age	-0.0005	-0.0008	-0.0011^*	-0.0011^*	-0.0009	-0.0009
	(-0.0015, 0.0004)	(-0.0018, 0.0002)	(-0.0020, -0.0001)	(-0.0021, -0.0002)	(-0.0019, 0.0001)	(-0.0019, 0.00004)

† skin-tone	0.0425^**	0.0565^***	0.0551^***	0.0541^***	0.0565^***	0.0538^***
	(0.0108, 0.0742)	(0.0246, 0.0883)	(0.0232, 0.0869)	(0.0222, 0.0860)	(0.0247, 0.0884)	(0.0220, 0.0857)

Current Charge (p-hat)	1.0729^***	1.1140^***	1.1098^***	1.1121^***	1.1146^***	1.1097^***
	(1.0050, 1.1408)	(1.0459, 1.1820)	(1.0417, 1.1780)	(1.0440, 1.1802)	(1.0465, 1.1827)	(1.0416, 1.1778)

Recidivism Risk (p-hat)	-0.7664^***	-0.8042^***	-0.8165^***	-0.8164^***	-0.8089^***	-0.8067^***
	(-0.8724, -0.6604)	(-0.9108, -0.6975)	(-0.9229, -0.7101)	(-0.9228, -0.7100)	(-0.9155, -0.7023)	(-0.9133, -0.7002)

Attractiveness	-0.0166	-0.0172	-0.0157	-0.0154	-0.0161	-0.0168
	(-0.0493, 0.0162)	(-0.0502, 0.0158)	(-0.0487, 0.0173)	(-0.0483, 0.0176)	(-0.0491, 0.0169)	(-0.0497, 0.0162)

Competence	0.0034	0.0063	0.0066	0.0077	0.0062	0.0062
	(-0.0293, 0.0361)	(-0.0266, 0.0392)	(-0.0264, 0.0395)	(-0.0252, 0.0406)	(-0.0267, 0.0391)	(-0.0267, 0.0390)

Dominance	-0.0003	-0.0033	-0.0023	-0.0015	-0.0035	0.0006
	(-0.0263, 0.0258)	(-0.0296, 0.0229)	(-0.0286, 0.0240)	(-0.0279, 0.0249)	(-0.0297, 0.0228)	(-0.0258, 0.0269)

Trustworthiness	0.0244	0.0271	0.0270	0.0264	0.0274	0.0252
	(-0.0088, 0.0576)	(-0.0063, 0.0605)	(-0.0064, 0.0604)	(-0.0070, 0.0599)	(-0.0059, 0.0608)	(-0.0082, 0.0586)

Likely-release	0.0134	0.0127	0.0128	0.0120	0.0124	0.0114
	(-0.0193, 0.0462)	(-0.0203, 0.0456)	(-0.0202, 0.0457)	(-0.0210, 0.0450)	(-0.0205, 0.0454)	(-0.0216, 0.0443)

ML-Face	0.4779^***
	(0.4052, 0.5505)

Well-groomed		0.0095^***
		(0.0045, 0.0146)

Happy			0.0071^*
			(0.0007, 0.0134)

Threatening				-0.0074^*
				(-0.0143, -0.0004)

Messy					-0.0066^**
					(-0.0111, -0.0021)

full_model_pred						0.5269^***
						(0.2525, 0.8013)

Constant	-0.2335^***	0.0997	0.1380^**	0.1846^***	0.1815^***	-0.2403^*
	(-0.3610, -0.1060)	(-0.0185, 0.2178)	(0.0224, 0.2535)	(0.0702, 0.2990)	(0.0680, 0.2949)	(-0.4798, -0.0007)


Observations	8,336	8,336	8,336	8,336	8,336	8,336
Adjusted R²	0.1298	0.1185	0.1179	0.1178	0.1181	0.1186
F Statistic (df = 15; 8320)	83.8572^***	75.7192^***	75.2487^***	75.2234^***	75.4330^***	75.7476^***

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 `Well-Groomed, Happy, Threatening, Messy`

Here I’m checking how much of the CNN signal we can capture with the labels
The combinations of positive and negative labels get above 0.40 on adjusted R-sqrt which is very promising

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

	Dependent variable:

	P-hat-cnn
	(1)	(2)	(3)	(4)	(5)	(6)	(7)

Well-Groomed	0.0106^***		0.0083^***				0.0094^***
	(0.0092, 0.0119)		(0.0069, 0.0097)				(0.0072, 0.0115)

Happy		0.0148^***	0.0124^***				0.0076^***
		(0.0131, 0.0166)	(0.0107, 0.0142)				(0.0057, 0.0094)

Threatening				-0.0207^***		-0.0194^***	-0.0173^***
				(-0.0225, -0.0189)		(-0.0213, -0.0176)	(-0.0192, -0.0154)

Messy					-0.0078^***	-0.0061^***	0.0015
					(-0.0090, -0.0066)	(-0.0073, -0.0049)	(-0.0004, 0.0034)

Constant	0.6909^***	0.6917^***	0.6573^***	0.8100^***	0.7774^***	0.8311^***	0.7174^***
	(0.6836, 0.6982)	(0.6851, 0.6982)	(0.6486, 0.6659)	(0.8040, 0.8159)	(0.7721, 0.7827)	(0.8239, 0.8384)	(0.6972, 0.7376)


Observations	8,336	8,336	8,336	8,336	8,336	8,336	8,336
Adjusted R²	0.0192	0.0236	0.0349	0.0401	0.0133	0.0482	0.0600
F Statistic	164.4825^*** (df = 1; 8334)	202.5415^*** (df = 1; 8334)	151.6477^*** (df = 2; 8333)	349.5588^*** (df = 1; 8334)	113.5397^*** (df = 1; 8334)	212.1070^*** (df = 2; 8333)	133.9688^*** (df = 4; 8331)

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

Summary Regression

This is the table we discussed in our Monday meeting. It has three base regressions (column 1 - 3) and then a variation of these with our known MTurk features (attractiveness, dominance, trustworthiness, competence) and ML-Face, and both sets of controlls together (see the last three columns for this)

**Table No.5 - Summary Table**

	Dependent variable:

	Arrest final-outcome
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)

well_groomed	0.0121^***		0.0048	0.0115^***		0.0047	0.0077^**		0.0022
	(0.0069, 0.0173)		(-0.0003, 0.0100)	(0.0063, 0.0168)		(-0.0005, 0.0099)	(0.0025, 0.0129)		(-0.0030, 0.0073)

p_hat_cnn		0.6937^***	0.6848^***		0.6905^***	0.6823^***		0.5978^***	0.5942^***
		(0.6260, 0.7613)	(0.6165, 0.7531)		(0.6225, 0.7585)	(0.6137, 0.7509)		(0.5298, 0.6657)	(0.5257, 0.6626)

attractiveness				-0.0150	-0.0257	-0.0271	-0.0222	-0.0312	-0.0318
				(-0.0489, 0.0188)	(-0.0590, 0.0076)	(-0.0604, 0.0062)	(-0.0555, 0.0111)	(-0.0641, 0.0017)	(-0.0647, 0.0011)

competence				0.0105	0.0093	0.0083	0.0105	0.0091	0.0086
				(-0.0236, 0.0446)	(-0.0243, 0.0428)	(-0.0252, 0.0419)	(-0.0230, 0.0440)	(-0.0240, 0.0422)	(-0.0245, 0.0418)

dominance				-0.0212	-0.0007	-0.0004	-0.0053	0.0108	0.0109
				(-0.0487, 0.0063)	(-0.0278, 0.0265)	(-0.0275, 0.0267)	(-0.0324, 0.0218)	(-0.0160, 0.0376)	(-0.0159, 0.0377)

trustworthiness				0.0481^**	0.0386^*	0.0383^*	0.0408^**	0.0333^*	0.0332^*
				(0.0144, 0.0817)	(0.0055, 0.0717)	(0.0052, 0.0714)	(0.0077, 0.0739)	(0.0006, 0.0660)	(0.0005, 0.0659)

risk_pred_prob							-1.0928^***	-0.9598^***	-0.9573^***
							(-1.2007, -0.9849)	(-1.0673, -0.8522)	(-1.0650, -0.8496)

Constant	0.6994^***	0.2448^***	0.2263^***	0.6909^***	0.2364^***	0.2194^***	1.0471^***	0.6012^***	0.5923^***
	(0.6712, 0.7275)	(0.1938, 0.2958)	(0.1716, 0.2810)	(0.6590, 0.7229)	(0.1827, 0.2900)	(0.1625, 0.2763)	(0.9999, 1.0942)	(0.5343, 0.6681)	(0.5222, 0.6624)


Observations	8,336	8,336	8,336	8,336	8,336	8,336	8,336	8,336	8,336
Adjusted R²	0.0016	0.0329	0.0331	0.0022	0.0331	0.0332	0.0343	0.0573	0.0573
F Statistic	14.5025^*** (df = 1; 8334)	284.6992^*** (df = 1; 8334)	143.5541^*** (df = 2; 8333)	4.7311^*** (df = 5; 8330)	57.9927^*** (df = 5; 8330)	48.6959^*** (df = 6; 8329)	50.3509^*** (df = 6; 8329)	85.4760^*** (df = 6; 8329)	73.3309^*** (df = 7; 8328)

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

Repeating table 5 with `Well-groomed + Messy`

This just repeats the regression in table 5 but with Messy + Well-Groomed

**Table No.6 - Summary Table Repeated**

	Dependent variable:

	Arrest final-outcome
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)

well_groomed	0.0162^***		0.0097^*	0.0158^***		0.0097^*	0.0103^**		0.0057
	(0.0078, 0.0246)		(0.0014, 0.0180)	(0.0074, 0.0242)		(0.0014, 0.0180)	(0.0021, 0.0186)		(-0.0025, 0.0139)

messy	0.0047		0.0055	0.0048		0.0057	0.0030		0.0040
	(-0.0027, 0.0121)		(-0.0018, 0.0128)	(-0.0026, 0.0122)		(-0.0016, 0.0130)	(-0.0043, 0.0103)		(-0.0032, 0.0112)

p_hat_cnn		0.6937^***	0.6854^***		0.6905^***	0.6830^***		0.5978^***	0.5948^***
		(0.6260, 0.7613)	(0.6171, 0.7536)		(0.6225, 0.7585)	(0.6144, 0.7516)		(0.5298, 0.6657)	(0.5263, 0.6632)

attractiveness				-0.0154	-0.0257	-0.0275	-0.0224	-0.0312	-0.0321
				(-0.0493, 0.0184)	(-0.0590, 0.0076)	(-0.0609, 0.0058)	(-0.0557, 0.0109)	(-0.0641, 0.0017)	(-0.0650, 0.0008)

competence				0.0112	0.0093	0.0092	0.0110	0.0091	0.0092
				(-0.0229, 0.0453)	(-0.0243, 0.0428)	(-0.0244, 0.0428)	(-0.0226, 0.0445)	(-0.0240, 0.0422)	(-0.0239, 0.0424)

dominance				-0.0210	-0.0007	-0.0001	-0.0051	0.0108	0.0111
				(-0.0485, 0.0065)	(-0.0278, 0.0265)	(-0.0272, 0.0271)	(-0.0322, 0.0220)	(-0.0160, 0.0376)	(-0.0157, 0.0379)

trustworthiness				0.0481^**	0.0386^*	0.0383^*	0.0408^**	0.0333^*	0.0332^*
				(0.0144, 0.0817)	(0.0055, 0.0717)	(0.0052, 0.0714)	(0.0077, 0.0739)	(0.0006, 0.0660)	(0.0005, 0.0659)

risk_pred_prob							-1.0917^***	-0.9598^***	-0.9557^***
							(-1.1996, -0.9838)	(-1.0673, -0.8522)	(-1.0635, -0.8479)

Constant	0.6589^***	0.2448^***	0.1782^***	0.6488^***	0.2364^***	0.1691^***	1.0205^***	0.6012^***	0.5565^***
	(0.5886, 0.7291)	(0.1938, 0.2958)	(0.0941, 0.2623)	(0.5765, 0.7210)	(0.1827, 0.2900)	(0.0832, 0.2550)	(0.9405, 1.1006)	(0.5343, 0.6681)	(0.4611, 0.6519)


F(Messy + Groomed)	3.25 (0.039)	X	0.9932 (0.3706)	2.924 (0.05401)	X	0.9322 (0.3939)	2.1326 (0.1189)	X	0.6793 (0.5071)
Observations	8,336	8,336	8,336	8,336	8,336	8,336	8,336	8,336	8,336
Adjusted R²	0.0016	0.0329	0.0331	0.0023	0.0331	0.0333	0.0342	0.0573	0.0572
F Statistic	7.7871^*** (df = 2; 8333)	284.6992^*** (df = 1; 8334)	96.2201^*** (df = 3; 8332)	4.1337^*** (df = 6; 8329)	57.9927^*** (df = 5; 8330)	41.9785^*** (df = 7; 8328)	43.2201^*** (df = 7; 8328)	85.4760^*** (df = 6; 8329)	64.2664^*** (df = 8; 8327)

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

Correlants to Well-Groomed

**Checking correlation to Well-Groomed labels**

	Dependent variable:

	Well-Groomed
	(1)	(2)	(3)

Sex	0.1198^***
	(0.0554, 0.1841)

raceB		-0.3003^*
		(-0.5965, -0.0041)

raceI		0.6109
		(-0.0776, 1.2995)

raceU		0.2793
		(-0.0673, 0.6259)

raceW		-0.4778^***
		(-0.7766, -0.1789)

skin_tone_cat_light_skin			-0.1216^***
			(-0.1806, -0.0625)

skin_tone_cat_medium_skin			0.0032
			(-0.0696, 0.0759)

Constant	5.0962^***	5.5224^***	5.2341^***
	(5.0392, 5.1532)	(5.2279, 5.8169)	(5.1944, 5.2739)


Observations	8,336	8,336	8,336
Adjusted R²	0.0010	0.0073	0.0014
F Statistic	9.3618^*** (df = 1; 8334)	16.2590^*** (df = 4; 8331)	6.7323^*** (df = 2; 8333)

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

AUC and R-squared comparison

Table 01 - Arrest Regressions
Fit measured in adjusted R squared and AUC
Table 1 Column Labels:	Male-Female Combined		Male Subsample		Female Subsample		Felony Subsample		Non-Felony Subsample
Table 1 Column Labels:	Adjusted R Squared	ROC AUC	Adjusted R Squared	ROC AUC	Adjusted R Squared	ROC AUC	Adjusted R Squared	ROC AUC	Adjusted R Squared	ROC AUC
Well Groomed	0.0016	0.5266	0.0006	0.5161	0.0138	0.5956	0.0014	0.5241	0.0022	0.5373
Lower 95% C.I.	0.0004	0.5122	−0.0001	0.5001	0.0063	0.5604	−0.0001	0.5047	0.0003	0.5135
Upper 95% C.I.	0.0034	0.5411	0.0021	0.5320	0.0244	0.6308	0.0042	0.5436	0.0053	0.5611
Happy	0.0024	0.5298	0.0017	0.5251	0.0010	0.5331	−0.0002	0.5031	0.0033	0.5477
	0.0011	0.5158	0.0004	0.5099	−0.0005	0.4961	−0.0003	0.4841	0.0012	0.5251
	0.0045	0.5437	0.0037	0.5404	0.0055	0.5702	0.0009	0.5222	0.0065	0.5703
Well Groomed + Happy	0.0033	0.5366	0.0019	0.5281	0.0133	0.5950	0.0011	0.5246	0.0044	0.5535
	0.0016	0.5222	0.0005	0.5122	0.0053	0.5595	−0.0003	0.5049	0.0020	0.5301
	0.0056	0.5511	0.0043	0.5440	0.0242	0.6304	0.0045	0.5442	0.0084	0.5768
Threatening	0.0030	0.5378	0.0006	0.5175	0.0023	0.5442	0.0008	0.5216	0.0018	0.5316
	0.0013	0.5239	−0.0001	0.5020	−0.0003	0.5108	−0.0002	0.5026	0.0002	0.5090
	0.0052	0.5517	0.0022	0.5329	0.0084	0.5777	0.0033	0.5405	0.0045	0.5542
Messy	0.0005	0.5141	0.0003	0.5092	0.0107	0.5826	0.0003	0.5120	0.0014	0.5277
	−0.0001	0.4995	−0.0001	0.4932	0.0039	0.5468	−0.0003	0.4924	0.0001	0.5039
	0.0018	0.5286	0.0017	0.5252	0.0206	0.6183	0.0025	0.5316	0.0042	0.5516
Threatening + Messy	0.0032	0.5381	0.0007	0.5169	0.0111	0.5848	0.0009	0.5213	0.0028	0.5391
	0.0014	0.5237	−0.0001	0.5009	0.0041	0.5490	−0.0003	0.5016	0.0008	0.5158
	0.0058	0.5525	0.0027	0.5330	0.0223	0.6206	0.0043	0.5410	0.0066	0.5625
Well Groomed + Happy + Threatening + Messy	0.0051	0.5491	0.0019	0.5312	0.0135	0.5978	0.0019	0.5336	0.0047	0.5549
	0.0032	0.5348	0.0007	0.5153	0.0065	0.5619	0.0002	0.5140	0.0021	0.5317
	0.0084	0.5634	0.0047	0.5470	0.0268	0.6337	0.0064	0.5532	0.0093	0.5782
ML Face	0.0329	0.6233	0.0222	0.5982	0.0293	0.6568	0.0255	0.5930	0.0351	0.6552
	0.0268	0.6097	0.0165	0.5827	0.0178	0.6246	0.0181	0.5739	0.0279	0.6340
	0.0396	0.6369	0.0283	0.6136	0.0440	0.6889	0.0342	0.6121	0.0439	0.6764
Well Groomed + Happy + Threatening + Messy + ML Face	0.0336	0.6252	0.0226	0.6008	0.0341	0.6677	0.0256	0.5946	0.0354	0.6586
	0.0279	0.6116	0.0171	0.5854	0.0233	0.6357	0.0184	0.5755	0.0280	0.6376
	0.0413	0.6388	0.0301	0.6162	0.0511	0.6996	0.0362	0.6136	0.0456	0.6796