Evaluation of bias and gender/racial concordance based on sentiment analysis of narrative evaluations of clinical clerkships using natural language processing

Table 6 Likelihood ratios [LR] of token words by demographic groups

More Likely for Male		More Likely for Female
Word	LR	Word	LR
smart	0.36	willing	1.71
much	0.37	active	1.642558
fantastic	0.43	how	1.49
confident	0.57	most	1.43
short	0.60	professional	1.40
wonderful	0.61	oral	1.38
complicated	0.62	patient	1.34
nice	0.62	clear	1.32
future	0.67	comfortable	1.30
complex	0.67	solid	1.30
More Likely for Hispanic or Latino		More Likely for Not Hispanic or Latino
Word	LR	Word	LR
interested	0.24	easy	4.07
motivated	0.28	high	3.66
concise	0.39	confident	2.95
most	0.40	insightful	2.85
clinical	0.41	how	2.72
other	0.41	3rd	2.52
physical	0.47	actively	2.38
when	0.48	hard	2.21
oral	0.49	nice	2.19
own	0.50	pleasant	2.09
More Likely for Asian		More Likely for Not Asian
Word	LR	Word	LR
limited	0.70	sure	2.07
3rd	0.72	only	1.86
future	0.73	important	1.82
kind	0.73	especially	1.81
complete	0.76	helpful	1.71
patient	0.77	impressed	1.57
clear	0.78	so	1.54
active	0.80	short	1.52
attentive	0.80	early	1.51
compassionate	0.80	new	1.46
More Likely for White		More Likely for Not White
Word	LR	Word	LR
sure	0.36	future	1.72
insightful	0.39	complete	1.68
only	0.55	limited	1.58
easy	0.56	concise	1.47
important	0.57	strong	1.39
many	0.57	accurate	1.34
high	0.65	confident	1.33
impressed	0.68	just	1.33
helpful	0.69	third	1.33
efficient	0.70	patient	1.30

ISSN: 1472-6920