Tomorrow I’m giving a talk (slides here) at the Joint Statistical Meeting in Vancouver on some work I’ve been doing on combining bootstrapping with multiple imputation (MI), something I’ve written about here before. That post looked at a recent paper by Schomaker and Heumann (2018) on various ways of combining bootstrapping and MI. A more recent post discussed an arXiv paper by von Hippel (2018) on maximum likelihood multiple imputation, which also contains a nice proposal for combining bootstrap and MI. My talk this week is about how these perform when the imputation and analysis models are not congenial.
Missing data
Missing not at random sensitivity analysis with FCS multiple imputation
Daniel Tompsett and colleagues have recently published a paper (open access here) on performing missing not at random (MNAR) sensitivity analyses within the fully conditional specification (FCS) framework for multiple imputation (MI). A number of previous papers had explored versions of the approach, and Tompsett et al bring these together to formalise the basis for the approach (which they term NARFCS) and importantly how to choose values of the sensitivity parameters involved.
Multiple imputation when estimating relative risks
Sullivan and colleagues have recently published a nice paper exploring multiple imputation for missing covariates or outcome when one is interested in estimating relative risks. They performed simulations where missing covariates or outcomes were imputed either using multivariate normal imputation or using fully conditional specification imputation (FCS), and where the true outcome model is a log link binomial model. They concluded that multivariate normal imputation performed poorly, producing estimated coefficients which were biased towards the null. Fully conditional specification performed better, although estimates were still biased in certain situations.