. Author manuscript; available in PMC: 2017 May 16.

Published in final edited form as: Psychol Test Assess Model. 2016;58(2):255–307.

Table 7. PROMIS applied cognition - general concerns item set: Differential item function (DIF) results. Race/ethnicity subgroup comparisons.

Item description	IRTPRO			lordif			Magnitude (NCDIF)			Effect Size T1
Item description	White vs. Black	White vs. Hisp.	White vs. NHAPI	White vs. Black	White vs. Hisp.	White vs. NHAPI	White vs. Black	White vs. Hisp.	White vs. NHAPI	White vs. Black	White vs. Hisp.	White vs. NHAPI
I have had trouble forming thoughts				U^; NU^		U^*; NU	0.0006	0.0003	0.0005	-0.0135	-0.0055	0.0185
My thinking has been slow				U^; NU^	U	U^*; NU	0.0004	0.0003	0.0012	0.0073	0.0064	-0.0204
My thinking has been foggy		U	NU	U^; NU^	U^*; NU	U^*	0.0021	0.0036	0.0019	-0.0360	0.0277	0.0306
I have had trouble concentrating			NU	U^; NU^		U^; NU^	0.0005	0.0018	0.0045	0.0026	-0.0335	0.0481
I have had to work really hard to pay attention or I would make a mistake				U^; NU^		U^*; NU	0.0007	0.0014	0.0011	-0.0004	-0.0245	-0.0180
It has seemed like my brain was not working as well as usual	U^*	U^*	U	U^; NU^	U^*	U^; NU^	0.0201	0.0205	0.0060	0.1084^†	0.1148^†	0.0565
I have had to work harder than usual to keep track of what I was doing				U^; NU^		U^; NU^	0.0012	0.0013	0.0031	0.0241	-0.0130	-0.0327
I have had trouble shifting back and forth between different activities that require thinking			U	U^; NU^		U^; NU^	0.0009	0.0054	0.0074	-0.0123	-0.0568	-0.0650

Asterisks indicate significance after adjustment for multiple comparisons. All non compensatory DIF (NCDIF) values were smaller than the threshold (0.0960)

^†

Indicates value above threshold of 0.10.

NU = Non-uniform DIF involving the discrimination parameters; U = Uniform DIF involving the location parameters.

For the lordif analyses, uniform and non-uniform DIF were determined using likelihood ratio chi-square tests. Uniform DIF is obtained by comparing the log likelihood values from models one and two. Non-uniform DIF is obtained by comparing the log likelihood values from models two and three. DIF of high magnitude was not detected using the pseudo R² measures of Cox & Snell (1989), Nagelkerke (1991), and McFadden (1973) or with the change in β criterion. The threshold for β change was ≥ 0.1; pseudo R² was ≥ 0.02.