13 August 2021
Statistical Analysis of Incomplete Dataonline course
This course provides an introduction to the theory and application of Multiple Imputation (MI) (Rubin 1987) which has become a very popular way for handling missing data, because it allows for correct statistical inference in the presence of missing data. With the advent of MI algorithms implemented in statistical standard software (R, SAS, Stata, SPSS,…), the method has become more accessible to data analysts. For didactic purposes, we start by introducing some naive ways of handling missing data, and we use the examination of their weaknesses to create an understanding of the framework of Multiple Imputation. The first half of this course is of a somewhat theoretical nature, but we believe that a fundamental understanding of the MI principle helps to adapt to a wider range of practical problems than focusing on a few select situations. However, at the latter stages of the course, frequent problems like regression with missing data will be addressed, and further typical situations will be covered by the lab session throughout the course, which is predominantly based on the statistical language R. We recommend basic R skills for this course, but it is possible to understand the course contents without prior knowledge in R, as the main MI algorithms are almost identical across all major software packages.
Therefore, by the end of the course, we hope that people will be able to use MI algorithms in general (irrespective of the software package) and to apply Rubin's combining rules correctly, as well as to explain to readers of their work how and why they used the method.
Dr. Florian Meinfelder is a senior lecturer at the Department for Statistics and Econometrics at the University of Bamberg, Germany.
Angelina Hammon is a PhD student in Statistics at the University of Bamberg, Germany.
Participants will find the course useful if they:
- are survey methodologists working with incomplete data;
- are researchers who want to learn more about the analysis of incomplete data in general;
- are already aware of MI and its benefits, but feel uncomfortable about the available parameter settings in MI algorithms implemented in their preferred statistical software.
- profound understanding of sampling theory;
- an advanced understanding of the (generalized) linear model;
- familiarity with statistical distributions;
- basic knowledge of matrix algebra;
- solid skills in either R, SPSS, or Stata (recommended for exercises).
Participants are encouraged to bring a laptop computer with an installed web browser for performing the practical exercises for this course. R (www.r-project.org) and RStudio (www.rstudio.com) can be downloaded and installed free of charge. During exercises, participants will also have access to SPSS, Stata, and R in a PC lab.
By the end of the course participants will:
- be familiar with the theoretical implications of the MI framework and will be aware of the explicit and implicit assumptions (e.g. will be able to explain within an article why MAR was assumed, etc.);
- know when to use MI (and when not);
- be aware how to specify a "good" imputation model and how to use diagnostics;
- be familiar with the availability of the various MI algorithms;
- be able to not only replicate situations akin to the case studies covered in the course, but also know how to handle incomplete data in general.
- Certificate of attendance issued upon completion.
- 4 ECTS points via the University of Mannheim for regular attendance and satisfactory work on daily assignments and for submitting a paper/report of about 5000 words to the lecturer(s) up to 4 weeks after the end of the summer school (EUR 50).
EUR 400: Student/PhD student rate.
EUR 600: Academic/non-profit rate.
The rates include the tuition fee and the course materials.
10 DAAD scholarships are available via the Institute of Sociology and Social Psychology (ISS) of the University of Cologne.