A “CACE” in point: Estimating causal effects via a latent class approach in RCTs with noncompliance using Stata

In randomized control trials (RCT), intention-to-treat (ITT) analysis is customarily used to estimate the effect of the trial; however, in the presence of noncompliance, this can often lead to biased estimates because ITT completely ignores varying levels of actual treatment received. This is a known issue that can be overcome by adopting the complier average causal effect (CACE) approach, which estimates the effect the trial had on the individuals who complied with the protocol. This can be obtained via a latent class specification when compliance is unobserved in the control group, under certain reasonable assumptions, for example, randomization, exclusion restriction, and ignorable missingness. This model is fit as a mixture model for the outcome of interest with two latent classes: a) compliers and b) noncompliers. This presentation will briefly introduce the issues around noncompliance and the assumptions of the CACE model. It will then illustrate the use of the gsem command in Stata 15 onward to estimate this effect with open access data and compare across other commonly used software packages. Finally, results using this approach in the context of a recent school-based RCT in England, the Good Behaviour Game (GBG), will be discussed.