Table 3.

Graded Response Model with Black Respondents Only (PrEP@Home Enrollees from Sites in 3 Regions of the United States were Cisgender Men who have Sex with Men Ages 18–49)

	Item-category intercepts			Loadings.	SSVS-DIFa
	${\mathrm{\tau }}_1$	${\mathrm{\tau }}_2$	${\mathrm{\tau }}_3$	${\beta }_j$	$\mathrm{Pr}({\omega }_j)$	${\delta }_j$
Feel ashamed	− 0.05 [ −0.44, 0.32]	1.91 [1.47, 2.41]	3.41 [2.78, 4.06]	2.06 [1.60, 2.61]	0.11	0.06 [ −0.08, 0.55]
Hiding pills	−0.45 [−0.73, −0.15]	0.89 [0.56, 1.18]	2.53 [2.08, 2.99]	1.06 [0.75, 1.38]	0.16	−0.07 [ −0.55, 0.08]
Negatively judged	−1.99 [ −2.39, −1.61]	−0.3 [ −0.64, 0.01]	1.25 [0.88, 1.61]	1.71 [1.33, 2.07]	0.04	0 [ −0.08, 0.07]
Would have sex with PrEP user*	0.98 [0.67, 1.30]	3.27 [2.75, 3.83]	4.02 [3.33, 4.75]	0.82 [0.52, 1.15]	1.00	−1.26 [−1.77,−0.69]
Seen as ‘slutty’ taking PrEP	−1.21 [ −1.56, −0.82]	0.89 [0.53, 1.27]	2.51 [2.05, 3.03]	1.9 [1.49, 2.35]	0.05	0.01 [ −0.08, 0.08]
Praised for taking PrEP*	−1.23 [ −1.56, −0.89]	1.1 [0.75, 1.41]	2.81 [2.35, 3.27]	1.06 [0.79, 1.36]	1.00	−1.29 [−1.72, −0.81]
Friends support	0.25 [ −0.11, 0.58]	2.66 [2.18, 3.15]	4.43 [3.60, 5.25]	1.57 [1.18, 2.00]	0.10	−0.04 [ −0.35, 0.11]
Treated unfairly	−1.12 [ −1.45, −0.81]	0.85 [0.51, 1.15]	3.1 [2.53, 3.62]	1.2 [0.88, 1.49]	0.05	0.02 [ −0.07, 0.08]
Partner problems*	−0.97 [ −1.30, −0.64]	1.1 [0.75, 1.44]	3.18 [2.61, 3.70]	1.23 [0.95, 1.56]	0.97	−0.86 [−1.47, −0.35]
Feel proud	−0.63 [ −0.97, −0.34]	1.54 [1.18, 1.95]	3.88 [3.18, 4.60]	1.41 [1.08, 1.79]	0.04	0.01 [ −0.07, 0.07]
Harassment	−1.5 [ −1.90, −1.08]	0.94 [0.58, 1.34]	3.26 [2.67, 3.82]	1.77 [1.38, 2.17]	0.27	−0.15 [ −0.75, 0.08]
Taking care*	0.63 [0.33, 0.95]	3.21 [2.70, 3.74]	4.36 [3.54, 5.15]	0.93 [0.60, 1.23]	1.00	−1.36 [−1.88, −0.85]
Family support	−1.32 [ −1.68, −0.98]	0.15 [ −0.17, 0.46]	1.64 [1.26, 2.01]	1.23 [0.93, 1.54]	0.21	0.1 [ −0.09, 0.69]

Graded response is an item response theoretical model that deals with ordered polytomous categories. Thresholds report the probability of a particular response category by the latent trait (e.g., latent PrEP stigma). Entries are estimates (posterior means) and 95% credible intervals are added in brackets

Bold values indicate the 4 items on the HIV PrEP Stigma Scale that showed presence of differentialitem functioning (DIF) in our item response theory (IRT) model

Items with an * denote presence of DIF in IRT models

^aStochastic search variable selection for Differential Item Functioning (SSVS-DIF) is a Bayesian variable selection method, the goal of which is to estimate a model to include only those predictors supported by data. This is achieved by adding a hierarchical prior to the model, where the relevance of a variable is assessed in each step of the sampling algorithm. This method is based on 10,000 Markov chain Monte Carlo samples. Column Pr(ω) lists DIF parameter inclusion probabilities estimated using stochastic search variable selection. Columnδlists DIF parameter estimates