WNAR* short course, Multiple Imputation and Survey Analysis in Stata 11
Presented by Roberto G. Gutierrez, StataCorp’s Director of Statistics
as part of the WNAR annual meeting in Seattle, Washington
Sunday, June 20
Multiple imputation (three hours):
In practice, datasets often contain missing data where some observations do
not contain data for all the variables involved in an analysis. The most
direct method for dealing with missing data is listwise deletion, where by the
entire observation is deleted if any of the variables contain missing
values. A modern alternative to this approach is multiple imputation (MI),
which is supported in Stata 11. In this course we cover how to perform multiple
imputation in Stata in three simple steps:
- imputation of missing values to form multiple complete datasets
according to a chosen impuation model
- complete-data analysis of the multiple created datasets
- pooling the results of these complete-data analyses using Rubin's
combination rules
We will apply these steps to several examples and will discuss the required
syntax and how to interpret the output in each case
Survey analysis (three hours):
Most of Stata’s estimation commands are equipped to automatically handle
data from complex surveys. So long as we declare the survey aspects of our
data, parameter estimates and their standard errors are adjusted for pre-
and post-stratification, multilevel sampling (clustering), and weighted
sampling. We will cover the basic concepts of complex survey data, how to
declare your survey design within Stata, and the resulting estimation.
We will consider survey estimation for means and proportions and for two-way
tables. We will also consider survey estimation for linear and logistic
regression models and regression for survival data.
General comments:
This short course will focus on building the tools necessary for performing
these types of analyses using Stata. Stata commands and output will be
provided, and they also can be reproduced later within a working copy of Stata 11
that is web-aware.
The only prerequisite for this course is a working knowledge of standard
regression models for simple random sampling and for complete data. Familiarity
with Stata software, while helpful, is not required.
For more details or to register, visit
http://www.biostat.washington.edu/wnar2010/registration.
*WNAR is the Western North American Region
of the International Biometric Society.
|
