Applied Multiple Imputation

About

Missing data are a pervasive problem in the social sciences. Data for a given unit may be missing entirely, for example, because a sampled respondent refused to participate in a survey (survey nonresponse). Alternatively, information may be missing only for a subset of variables (item nonresponse), for example, because a respondent refused to answer some of the questions in a survey. The traditional way of dealing with item nonresponse, referred to as “complete case analysis” (CCA) or “listwise deletion”, excludes every observation with missing information from the analysis. While easy to implement, complete case analysis is wasteful and can lead to biased estimates. Multiple imputation (MI) seeks to address these issues and provides more efficient and unbiased estimates when certain conditions are met. Over the past decades, it has therefore become a widely used alternative to CCA across the social sciences.

The goals of the course are to introduce you to the basic concepts and statistical foundations of missing data analysis and MI, and to enable you to use MI in your own work. The course puts heavy emphasis on the practical application of MI and on the complex decisions and challenges researchers are facing in its course. The focus is on MI using iterated chained equations (aka “fully conditional specification”) and its implementation in the software package Stata. You should have a good working knowledge of Stata to follow the applied parts of the course and to successfully master the exercises. If you are not familiar with Stata, you may still benefit from the course but will likely find the exercises quite challenging.

Course leader

Jan Paul Heisig is Professor of Sociology at Freie Universität Berlin, Germany. Ferdinand Geißler is a senior lecturer at the Chair of Social Research & Methods at the Humboldt-University Berlin, Germany.

Target group

You will find the course useful if:
- you use survey or other types of quantitative data and want to learn about MI as an alternative to CCA,
- are already using MI but want to gain a better understanding of the underlying assumptions, of current best practice recommendations, and/or of how to solve specific problems that arise in its application (e.g., imputation diagnostics, convergence problems, imputation of transformed variables such as interactions, imputation of hierarchical and longitudinal data).

Prerequisites:
- Experience in the analysis of quantitative data
- Good knowledge of regression analysis
- Good working knowledge of Stata
- Basic understanding of probability theory and sampling

Course aim

By the end of the course, you will:
- understand basic concepts of missing data analysis such as “missing at random”,
- be familiar with different approaches of how to handle item nonresponse and with their advantages and drawbacks,
- have a solid understanding of the main assumptions and statistical theory underlying MI and of the main steps of an analysis involving MI (imputation, diagnostics, and analysis),
- know how to implement MI using chained equations in Stata,
- know how to deal with various (Stata-specific and general) practical complications that arise in the application of MI using chained equations.

Fee info