Carrico and colleagues* also wanted to know whether stimulant drug use by HIV-positive gay men and also by their HIV-positive romantic partners was associated with HIV medication adherence in each man in all the couples. As part of a larger study about relationship dynamics,

gay couples completed a computerized survey containing questions about relationship factors, alcohol and substance use, and HIV medication adherence. In the chapter, we demonstrated how stimulant use was associated with whether HIV viral load was detectable. The study also collected self-reports of HIV medication adherence in the past 30 days from each research participant. Because so many men reported 100 percent adherence to their medication regimens, the percent adherence was dichotomized as M (for medication adherence) = 0 for not 100 percent adherent and M = 1 for 100 percent adherence over the past 30 days. Fit the same logistic regression models to the data as the ones shown in Eqs. (12.20) and (12.21), substituting medication adherence, M, for HIV viral load detectability, V, as the dependent variable. Fit the model using whichever of the three methods for analyzing logistic regression models with clustered data are available in your favorite statistical analysis program: ordinary logistic regression with robust standard errors, generalized estimating equations (GEE), and generalized linear mixed models (GLMM). Compare and contrast the results. The data are in Table C-33, Appendix C.