08-08 Analysis of Longitudinal Data: A Multilevel Approach

Aim of the course

This course addresses recent developments in the analysis of longitudinal data. For multilevel designs (e.g. first level of observations are nested within a randomly selected second level sample of subjects), standard regression techniques lead to biased model-parameter estimates and incorrect standard errors. Biased regression parameter estimates typically occur if the subjects were not measured at the same time points (unbalanced data) and if missing observations are involved. Due to the multilevel structure, more sophisticated models should therefore be used to account for the correlation between observations. Moreover, correlated data often occur due to memory effects when subjects are measured repeatedly across time. This type of correlation is called serial correlation. In this course, the main emphasis is on linear models with random effects and serial correlations. Special attention will be given to the analysis of longitudinal intervention-studies, where the objective is to evaluate some treatment effect. A distinction will be made between statistical methods to analyze longitudinal data from randomized experiments and those from life-events studies (non randomized).

Course objectives

  1. The participant learns how to model and analyse specific longitudinal data.
  2. The participant learns about the possibilities of SPSS procedures regarding the analysis of longitudinal data.
  3. The participant learns in practice how to analyse longitudinal data using SPSS procedures.

Lecturers

The course coordinator and lecturer is:

  • Dr. Frans Tan
    Department of Methodology and Statistics
    University of Maastricht.
    P.O. Box 616, 6200 MD Maastricht.
    Phone: +31(0)433882278
    E-mail frans.tan@stat.unimaas.nl

SPSS trainer:

  • Mickey Chenault M.A
    Department of Methodology and Statistics
    University of Maastricht.
    P.O. Box 616, 6200 MD Maastricht.
    Phone: +31(0)433884294/3875597
    E-mail mickey.chenault@stat.unimaas.nl

Internet and intervision discussion:

  • Dr. Ir. Arno Muijtjens
    Department of Educational Development and Research
    P.O. Box 616, 6200 MD Maastricht
    Phone: +31(0)433881243
    E-mail a.muijtjens@educ.unimaas.nl

Requirements

Participants are required to be present at each of the course meetings. The course requires a total time investment of 100 hours, of which approximately 30 hours are for attending the meetings, and another 70 hours to work on (groups) assignments.

Dates and location

The course starts on Friday, April 4, 2008 and takes place each following Friday until April 25, 2008.

All meetings are planned from 10.00-17.00 hours and take place in Utrecht: meeting centre (vergadercentrum), Vredenburg 19.http://www.vergadercentrumvredenburg.nl

It is important that you take a laptop to the meetings with SPSS version 15 installed. If you do not have a laptop, then please contact Willemien Sanders.

Literature

All course study-materials will be made available at the Open University, NL Blackboard website for the course. This course is for PhD students and researchers involved in studies that consider longitudinal data and multilevel data in epidemiological, social, educational and behavioural research. For a fruitful participation in the course, participants should be familiar with linear regression analysis at an intermediate level (see e.g. Kleinbaum, Kupper, Muller, Nizam (1998). Applied regression analysis and multivariable methods. Duxbury press: London. 3rd edition, chapters 4, 5, 6.1-6.5, 8, 11, 14.1-14.6).

Experience with linear regression analysis with SPSS is also required.

Organization

The topics are discussed on four separate days. The program includes lectures, SPSS training and discussions of the assignments. There will be ample opportunity to discuss ones own research study. The following topics are included:

  1. Rationale for using methods for analysis of longitudinal data.
  2. Random-effects models (with serial correlations) and marginal models to account for correlated data.
  3. Basic guidelines for model selection.
  4. Analysis of longitudinal intervention-studies (randomized and non-randomized).
  5. How to deal with missing observations.
  6. Brief review of other generalized linear mixed models and estimation methods.
  7. Getting to know how to analyse longitudinal data with SPSS 11.5 and higher versions.

Outline of the assignments

  • Before first meeting
    Work individually on the assignments about OLS linear regression (will be announced later on blackboard) with SPSS.
    Be sure that your knowledge about linear regression (simple and multiple) is up to date.
  • After the first meeting
    Work individually on the assignments about random intercept models
    Internet discussion: Give a short description of your research study.
  • After the second meeting
    Work individually on the assignments about random slope models, marginal models and model selection.
    Internet discussion: Group discussion. Choose one research study. Work out in greater detail.
  • After the third meeting
    Work individually on the assignments about intervention studies.
    Internet discussion: Group discussion. Write a report using the obtained conclusions in your group.
  • These conclusions will be presented during the fourth meeting.

Additional literature

  • Hox, Joop. (2002). Multilevel analysis. Techniques and applications. Lawrence Erlbaum: New Jersey.
  • Landau, S., Everitt, B.S. (2004). A handbook of statistical analysis using SPSS, Chapter 8. Chapman & Hall: London.
  • Singer, J.D., Willett, J.B. (2003). Applied longitudinal data analysis: modelling change and event occurrence. University Press: Oxford.
  • Snijders T., Bosker R. (1999). Multilevel analysis. An introduction to basic and advanced multilevel modeling. Sage: London.
  • Tabachnick, B.G., Fidell, L.S. (2007). Using multivariate statistics, chapter 15. Pearson and Ab: Boston.
  • Twisk, Jos W.R.(2003). Applied longitudinal data analysis for epidemiology. University Press: Cambridge.
  • Verbeke & Molenberghs (2000). Linear mixed models for longitudinal data. Springer-Verlag: New York. (advanced)

Language

The course will be given in English or in Dutch depending on the participants.

Registration

To register, please fill out the registration form. The deadline for registration is 4th of March 2008. Students who want to register after the deadline are recommended to contact the course coordinator:frans.tan@stat.unimaas.nl.

Comments