Handling Missing Data in Administrative Studies:multiple imputation & inverse probability weighting

Description:

Course number: ADRCE Training050 Carpenter

Course places are limited and registration by 2 November 2017 is strongly recommended.

Summary of Course

This ADRC-E course will consider the issues raised by missing data (both item and unit non-response) in studies using routinely collected data, for example electronic health records. Following a review of the issues raised by missing data, we will focus on two methods of analysis: multiple imputation and inverse probability weighting. We will also discuss how they can be used together. The concepts will be illustrated with medical and social data examples.

Course Contents 

The course covers:

• Issues raised by missing data in the administrative setting: when is a complete records analysis sufficient?
• Shortcomings of ad-hoc methods
• Introduction to multiple imputation, including algorithms, common pitfalls, reporting and examples
• Introduction to inverse probability weighting for missing data, and its pros and cons viz-a-viz multiple imputation
• Combining inverse probability weighting and multiple imputation to improve robustness
• Strategies for large datasets, including the two-fold multiple imputation algorithm
• Discussion of participants’ data.

Presenter

Professor James Carpenter

James is Professor of Medical Statistics at the London School of Hygiene and Tropical Medicine, and Programme Leader in Methodology at the MRC Clinical Trials Unit. He has a long-standing interest in longitudinal data and the issues raised by missing observations. He co-authored Multiple Imputation and its Application (Wiley, 2013) with Mike Kenward.

Target Audience 

The course is aimed at quantitative researchers, who have an interest or experience in analysing administrative data. PhD students are also welcome. Detailed technical arguments will not be presented; instead the focus will be on concepts and examples, with participants encouraged to bring their own data for discussion.

Pre-requisites

This course includes computer workshops, using the statistical software package Stata. Full details of all commands will be given, so no previous experience with Stata is necessary, though it will inevitably be an advantage.

Practical experience using regression modelling (including survival data modelling) and preferably multilevel modelling.

Event Outline 

(Draft Programme, subject to minor changes)

Thursday 9th November

 9.00 – 9.30                   Registration

 9.30 – 9.45                   Welcome and Introduction

 9.45 – 10.45                 Lecture 1: Introduction to issues raised by missing data in routine electronically collected databases

10.45 – 11.00                Coffee break

11.00 – 12.00                Practical 1: Missing data mechanisms

12.00 – 13.00                Lecture 2: Shortcomings of ad-hoc methods and introduction to Multiple Imputation

13.00 – 14.00                Lunch

14.00 – 15.00                Practical 2: Illustration of shortcomings of ad-hoc methods, simple MI, 1958 National Childhood Development Study data

15.00 – 15.30                 Lecture 3a: More on MI: algorithms

15.30 – 15.45                 Tea break

15.45 – 16.15                 Lecture 3b: More on multiple imputation: pitfalls and reporting

16.15 – 17.30                 Practical 3: MI for the 1958 National Childhood Development Study

Friday 10th November

 8.30 – 9.00                   Review, and continuation of practical 3

 9.00 – 10.00                 Lecture 4: Inverse probability weighting for missing item and unit data

10.00 –11.00                 Practical 4: Inverse probability weighting for routinely collected health data

11.00 – 11.15                Coffee break

11.15 – 12.30                Lecture 5: Combining MI and IPW to improve robustness

12.30 – 13.30                Lunch

13.30 – 14.30                Practical 5: Examples of combining MI and IPW

14.30 – 15.00                Lecture 6: Strategies for large datasets and the two-fold multiple imputation

15.00 – 15.15                Tea break

15.15 – 16.15                Practical 6: Putting large data strategies into practice

16.15 – 16.45                Discussion (including discussion of participants’ data)

Podcast

Podcasts for some of our previous courses can be found at https://adrn.ac.uk/about/network/england/training-podcasts/

--

Our courses are very popular and are often oversubscribed. If you cannot attend a course you have registered for, it is essential to kindly notify us a minimum of 30 days in advance so that your place can be released for another attendee. Details of our cancellation policy are here: http://store.southampton.ac.uk/help/?HelpID=1 . Please see our full course list here: http://store.southampton.ac.uk/browse/product.asp?compid=1&modid=5&catid=113.

Cost:

The fee per day is:

1. £30 - For UK registered postgraduate students
2. £60 - For staff at UK academic institutions, Research Council UK funded researchers, UK public sector staff and staff at UK registered charity organisations
3. £220 - For all other participants
4. Free Place for ADRC-E/ADRN/ADS staff

All fees include event materials, lunch, morning and afternoon tea. They do not include travel and accommodation costs.

Website and registration:

Region:

South East

Keywords:

Nonresponse , Missing data, Imputation, Weighting, Regression Methods, Multiple imputation, inverse probability weighting, survey data, administrative data

Related publications and presentations from our eprints archive:

Nonresponse
Missing data
Imputation
Weighting
Regression Methods

Back to the training database