in alcohol use across the baseline and follow-up assessments, r(3736) = 0.69, p < 0.01. Covariates included sex and two dummy-coded race variables. Owing to the fact that our binary relationship status variable was a mediator in our model, we employed a robust weighted least squares estimator (WLSMV) and THETA parameterization. Indirect effects in Mplus are estimated using the product of coefficients approach. The assumption that this product is normally distributed is often violated (MacKinnon, Lockwood, & Williams, 2004). Thus, in keeping with best practices proposed by MacKinnon et al. (2004), we used a bias-corrected bootstrap (n = 5000) in order to calculate 95% confidence intervals for the indirect effects.
