Summary

In case you dozed off, this chapter addressed a fairly common problem in real-world data analysis—especially for data collected outside your control or organization: missing data.

We first learned how to visualize missing data patterns, and how to recognize different types of missing data. You saw a few unprincipled ways of tackling the problem, and learned why they were suboptimal solutions. Specifically, most of the naïve solutions produced biased estimates on at least some crucial statistics and, in particular, almost always underestimated the variance and would produce confidence intervals that were way too narrow.

Multiple imputation, so we learned, addresses the shortcomings of these approaches and, through its usage of several imputed datasets, correctly communicates our uncertainty surrounding the imputed values. We used mice to perform this procedure. We discussed the different imputation methods for different types of variables, like predictive mean matching for continuous...