Table S6.

Study 2 primary analyses for treatment recommendation accuracy with different statistical models (target race × beliefs interaction)

Model	β	SE	F (χ2)	P
GLM (with repeated factor)	—	—	5.68	0.018
Mixed effects	−0.11	0.06	4.70	0.031
Generalized linear models
Logit link	−0.62	0.32	3.55	0.061
Probit link	−0.38	0.20	3.76	0.054
Ordered logistic regressions
Logit link	0.40	0.17	(5.89)	0.015
Probit link	0.22	0.09	(5.64)	0.018

The GLM is the model reported in the main text. A difference score (white accuracy minus black accuracy) was used for ordered logistic regressions, yielding three potential outcomes: −1, 0, 1; higher, positive estimates reflect greater racial bias, and, thus, the estimates for these models are positive. The dashes indicate that there are no beta and standard error estimates for this statistical model.