After my previous post about missing data, Kathy asked on Twitter whether two wrong models (the imputation model + the outcome model) would be better than one (the outcome model alone). Without doing any of the math, I’d guess the assumption of correctly spec the model also has a bigger impact in the CC analysis. You need correct spec in MI, twice, but trade off that potential bias for higher prec. This is a great question! I am going to investigate via a small simulation (so the answer cou...
