Article Text

Download PDFPDF
No quick fix: understanding the difference between fixed and random effect models
  1. Alastair H Leyland
  1. Correspondence to Professor Alastair H Leyland, MRC/CSO Social and Public Health Sciences Unit, 4 Lilybank Gardens, Glasgow G12 8RZ, UK; a.leyland{at}sphsu.mrc.ac.uk

Statistics from Altmetric.com

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

In this issue, Kravdal considers the impact of municipality level education on individual mortality (see page 1029).1 The educational level of each municipality in Norway was calculated annually from 1981–2002 by averaging the number of years of education of individuals living in the municipality. (Municipality education is considered an indicator of socioeconomic resources at the area level.) The typically high correlations between area-level measures of socioeconomic status means the choice of variable is probably less critical than ensuring adequate control for individual socioeconomic characteristics2 3; Kravdal also uses education as an individual-level control. While studies of contextual effects are commonly cross-sectional, this report uses data with a panel design,4 with observations (death or survival) on individuals made annually in each municipality. This commentary considers the major assumption underlying the chosen fixed effects model, details how the alternative random effects approach would allow testing of this assumption and reflects on the need for a widespread understanding of the importance of the choice of model.

Kravdal uses two types of logistic regression model to estimate the impact of municipality level education. In the first model, the clustering within municipalities (and hence the correlation in outcomes between individuals from the same municipality) is ignored; the probability pijt that individual i is alive in municipality j at the beginning of year t is given bylog(pijt1pijt)=β0+β1x1ijt+β2x2ijt+β3x3jt+β4T4t(1)where x1ijt and x2ijt …

View Full Text

Footnotes

  • Linked articles 081034.

  • Funding Medical Research Council, London, UK; Chief Scientist Office, Scottish Government Health Directorate, Edinburgh, UK.

  • Competing interests None.

  • Provenance and peer review Commissioned; externally peer reviewed.

Linked Articles