Missing Data Workshop
Very little data of interest to social, behavioral, and health scientists contains complete information. Instead, some missing data tends to be the rule rather than the exception in applied data analysis. Survey respondents may choose to skip sensitive questions; economic data may be harder to find for developing countries; certain types of respondents may be most likely to drop out of panel studies. Rarely is data “missing completely at random” — instead there tend to be systematic factors accounting for missing observations — factors that can bias results if not properly handled.
This workshop will focus on the most effective techniques for conducting quantitative analyses with missing data. In addition to covering the basics of missing data theory and showing the problems that can occur when ignoring missing data, we will cover in detail methods for: (1) multiply imputing missing data, (2) handling missing data with hotdeck imputation, and (3) full information maximum likelihood. Which method to use depends on many factors idiosyncratic to different analyses. The workshop focuses primarily on the methods rather than statistical software; however example code and resources will be provided for implementing the methods in Stata and R, along with some resources for handling missing data in SAS and SPSS.