Article Text

PDF
The fixed-effects model admittedly no quick fix, but still a step in the right direction and better than the suggested alternative
  1. Øystein Kravdal1,2
  1. 1Department of Economics, University of Oslo, Oslo, Norway
  2. 2Norwegian Institute of Public Health, Oslo, Norway
  1. Correspondence to Professor Øystein Kravdal, Department of Economics, University of Oslo, PO Box 1095 Blindern, 0317 Oslo, Norway; okravdal{at}econ.uio.no

Statistics from Altmetric.com

Alastair Leyland1 commented on a paper in the December 2010 issue2 about how individual mortality is influenced by the person's own education and the average education in the municipality of residence. The paper was based on a so-called ‘fixed-effects’ model, which is quite common in social science, but not in epidemiology. Leyland's intention was to explain the assumptions and limitations of the model to the readers, which is a highly laudable initiative. However, I am not sure the alternative model he suggests is very valuable.

This is the model under discussion (with new notation to improve readability):

log(pijt/(1pijt))=β1Xijt+β2Ejt+β3Tt+β4Fj(1)

where pijt is the probability that individual i in municipality j dies within year t (1980–2002), Xijt is a vector of individual characteristics, including the person's education, Ejt is the average education in the municipality, Tt are year dummies and Fj are municipality dummies (or ‘fixed effects’). The latter are included to control for constant unobserved municipality factors that …

View Full Text

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.