Applied
Statistical Analysis with Missing Data, Aarhus 2016
Preliminary
programme
Teachers: Morten Frydenberg and Henrik Støvring
(Section of Biostatistics, Department of Public Health)
Last
revision Henrik Støvring: November 28, 2016
The programme below will be continuously updated during the course with links to the lecture notes, exercises and datasets.
Day
1: Monday, December
12, 2016
The
first day will focus on learning the basic principles and how to
conduct simple MI analyses in Stata
Slides and datasets (Updated 14/12 2016): Missingdata_MI.pdf, Wooddata1.dta, ess2e03_scand.dta (version 13: Wooddata_13.dta)
Exercise 2 link: https://docs.google.com/spreadsheets/d/11GxmvTPvH2_xbj93lGqgrBuwh13rnTR3JKzuyrv_PLY/edit?usp=sharing
Exercise 5 link:
https://docs.google.com/spreadsheets/d/1CwqP7mRs2-rrZfazrsNdL3tN9o5LdGcSnsgBIkJb410/edit?usp=sharing
|
9.00 - 10.00 |
Introduction – the ordinary analysis and its shortcomings with a brief outlook to sensitivity analyses and an example of a simple multiple imputation based analysis. |
|
10.00 -12.00 |
Understanding the missing data mechanism (MCAR, MAR and MNAR) – what can we learn from analyzing the data? |
|
12.00 - 13.00 |
Lunch break |
|
13.00 - 14.00 |
Our first imputation model – understanding the concept of random data, prediction and filling in missing observations |
|
14.00 -15.00 |
Multiple Imputation by Chained Equations (MICE) – an iterative procedure for imputing missing values based on regression analyses (black box version) |
|
15.00 -16.00 |
MICE continued – how to analyze multiply imputed data |
Day
2: Wednesday, December 14, 2013
The
second day will focus on how to conduct more advanced MI analyses
with Stata
Slides and datasets: Missingdata_MI_II.pdf, Wooddata2.dta, Wooddata2_13.dta, Wooddata2_11.dta
Google document for checking assumptions: Exercise 4 - assumptions checking
Do-file example with log: day2.do, day2.log
|
9.00 - 10.00 |
A closer look at how Stata handles imputed datasets – the -mi-commands in Stata; MI data types; adding and extracting imputed datasets; examining missing and imputed values |
|
10.00 - 12.00 |
Tailor-made regression equations with MICE – how to exploit insights about the missing data mechanism and associations between variables when imputing missing values |
|
12.00 - 13.00 |
Lunch break |
|
13.00 - 14.00 |
Advanced concepts in MI – passive variables, choosing a regression type for categorical imputed variables (logit, ologit, mlogit), debugging strategies when your -mi impute- will not run |
|
14.00 - 16.00 |
Examples based student’s own projects |
Day
3: Friday November 29 2013
The
third day will focus on the methodological background of the methods
|
9.00 - 12.00 |
Statistical inference – fundamental principles, likelihood, estimates,
uncertainty and bias. Slides
|
|
|
Lunch break |
|
12.30 - 13.30 |
Sensitivity analysis and other methods for handling missing data problems.Slides |
|
13.30 - 15.00 |
A case study.Slides |
|
15.00 - 15.30 |
Final remarks and course evaluation |
Link to homepage on missing data