This dissertation is organized into two studies investigating common modeling issues that occur when including a covariate in a latent class analysis (LCA) model. When estimating a conditional LCA model, applied researchers must make decisions about the estimation strategy (one-step vs. three-step), the handling of incomplete covariate data, and specification of covariate relationships. Study 1 examined the performance of different methods for handling incomplete covariate data when using a three-step approach to estimation. The simulation results found that Bayesian estimation with informative normal priors correctly centered on the regression coefficient population values produced the most consistent and accurate regression coefficient estimates, regardless of the covariate distribution, strength, and missing data pattern. However, informative priors centered on the wrong population values produced some of the most biased regression coefficients. In most modeling conditions, full information maximum likelihood (FIML) and multiple imputation (MI) still worked well. When estimating a conditional LCA model, applied researchers must also make decisions about how to specify the covariate relations with the LCA measurement model. Specifically, applied researcher must decide if the covariate only has an indirect effect on the observed indicators via the latent class variable or if the covariate is related to one or more of the observed indicator variables. The goal of Study 2 was to explore the utility of using small-variance priors to help evaluate covariate relationships. Specifically, small- variance normal priors centered on zero were specified for the direct effects between the covariate and the latent class indicators for a series of population models with varying covariate relationships. Results from the Study 2 simulation indicate small-variance priors can be a useful tool for detecting covariate misspecifications, depending on the number of direct effects, sample size, and class sizes. Overall, findings from Study 1 and Study 2 highlight how Bayesian estimation can be especially helpful for handling common modeling issues in conditional LCA models.