When estimating an average treatment effect with observational data, it's possible to get an unbiased estimate of the causal effect if all confounding variables are observed and reliably measured. In education, confounding variables are often latent constructs. Covariate selection methods used in causal inference applications assume that all predictors are measured without error; in contrast, confirmatory factor analytic models account for measurement error. The effectiveness of using CFA to remove selection bias and reduce covariate set dimension is not yet understood well. In this paper, we will investigate how confirmatory factor analysis could be used for controlling for selection bias caused by a latent confounding variable. We compare the effectiveness of different approaches in a Monte Carlo simulation study.