Longitudinal data provide an important resource for social scientists to investigate question of dynamics, growth and change. They also provide a source of data that can, under some circumstances and assumptions, provide more believable answers to causal questions. Longitudinal data allow modelling of the sequence of events and changes of state such as movements between employment and unemployment, movements on and off of income support, changes in marital status, and movement of children from residing with parents to living independently.
Dr. Nicholas Biddle is a Fellow at the Centre for Aboriginal Economic Policy Research (CAEPR) at the Australian National University (ANU) and Deputy Director of the ANU Centres for Social Research Methods. He has a Bachelor of Economics (Hons.) from the University of Sydney and a Master of Education from Monash University. He also has a PhD in Public Policy from the ANU where he wrote his thesis on the benefits of and participation in education of Indigenous Australians. He previously held a Senior Research Officer and Assistant Director position in the Methodology Division of the Australian Bureau of Statistics.
The purpose of the course will be to provide students with a range of skills for the analysis of longitudinal data. We will begin by talking about how longitudinal data can be organized and cleaned and how to create variables for analysis from longitudinal data. We will examine descriptive statistics and descriptive analysis using longitudinal data. Progress can be made on many interesting and complex questions using such descriptive approaches. We will examine regression estimators applied to panel data. We will compare and contrast the main approaches to panel data regression: pooled analysis; fixed effects analysis and random effects analysis. We will examine their application to both linear and non-linear models.
The course will be computer-based with lectures and student exercises. There will be scope for participants to do some simple analysis of variables and questions of their own choice. The course will use the STATA software package and will provide examples from a range of data sources including the Household Income and Labour Dynamics of Australian study (HILDA).
The course is relevant to those who use longitudinal data in their own analysis, are involved in the design of longitudinal surveys, or use analysis of longitudinal data carried out by others to make informed policy decisions.
Day 1 – Introduction and overview
Students will be introduced to the main concepts in longitudinal data analysis, as well as how to frame and answer causal questions. Students will then be introduced to the main data formats for longitudinal data and how to clean data (using STATA).
Day 2 – Longitudinal data in Australia and descriptive analysis
The second day of the course will begin with an overview of the main sources of longitudinal data in Australia. Students will be exposed to the strengths and weaknesses of cohort studies, panel surveys, longitudinal administrative datasets, and quasi-longitudinal or linked datasets. They will then create policy relevant descriptive statistics from HILDA or their own dataset.
Day 3 – Regression analysis using longitudinal data (Part I – Linear models)
After a brief revision of the main concepts in regression analysis (using cross-sectional data), students will be introduced to the different assumptions, challenges, and approaches used in longitudinal regression analysis. This will be applied to linear models.
Day 4 – Regression analysis using longitudinal data (Part II – non-linear models)
The discussion of regression analysis will continue with applications to non-linear models. This includes binary dependent variables (logit/probit), but also ordinal, multinomial and count data. The final part of the day will turn to how to present and discuss longitudinal data analysis in a way that is useful for policy makers whilst still conveying the complexity and detail of the model.
Day 5 – Application and presentations.
In this final day, students will apply the learning from the course to their own example and present and discuss results with the class.
This course is run in a computer lab. Those students who have their own longitudinal data that they would like to use in their examples are encouraged to bring the data along on a USB stick.
Participants are encouraged to complete the ACSPRI course Fundamentals of Longitudinal Research and Design (or have equivalent experience) prior to attending the course. Participants should be familiar with OLS regression and logistic/probit regression. There is no prior required knowledge of STATA required for the course. However, students with no familiarity whatsoever are encouraged to watch one of the introductory videos online (for example - https://www.youtube.com/watch?v=QaI_a_l2jqo).
Students who are unfamiliar with regression analysis or have not used it in some time may consider reading Mastering ‘Metrics by Angrist and Pischke (2015) or similar.
Q: Do I need any prior knowledge before taking this course?
A: You should be familiar with regression analysis for both linear and non-linear variables.
Good theoretical explanations help to understand strengths of methods and chose btw different models (Summer 2015)
The instructor's bound, book length course notes will serve as the course texts.
Stata is distributed in Australia and New Zealand by Survey Design and Analysis Services.