Adjustment when Covariates are Fallible

This chapter discusses adjustment when covariates are not perfectly reliable. It starts with reviewing a theoretical framework that applies to fallible and latent covariates. This framework allows for deriving conditions under which adjustment has to be based on the latent covariate and conditions under which adjustment has to be based on the fallible covariate. As most methodological research has considered the case in which adjustment has to be based on the latent covariates. The author presents analytic derivations, simulation studies, and empirical analyses on the biasing effect of measurement error in covariates for causal effect estimation. The chapter evaluates different approaches to adjust for latent covariates. It considers the role of further covariates for the biasing effect of measurement error in another covariate. For this, the author presents an empirical analysis and discusses results from simulation studies. He concludes the implications for adjustment in empirical applications when covariates are fallible.