I am using panel data set and want to know which model I should use, fixed effect or random effect. On which basis is this selection done? When I asked my teacher he told me that model is selected on the base of Hausman specification test .

I am using the panel data for measuring the bilateral trade between two countries and to find out the effect of distance I am using panel data se otherwise I was using time series only. But only time series was not able to find out the impact of distance on bilateral trade between two countries..? I have read all the chapter of panel data of Gujarati book and still confused about model selection. I was thinking to run OLS on all the panel data set ignoring the cross sectional and time series...by doing this I will be able to get the impact of distance on bilateral trade also about other variables..

  • 2
    $\begingroup$ Welcome to the site. I fixed some of the spelling and grammar in your post, but it is still a bit unclear. Can you tell us what data you have and what you are trying to find out? $\endgroup$
    – Peter Flom
    Dec 29, 2013 at 13:44
  • 1
    $\begingroup$ As with @PeterFlom's comment: more small edits, notably Hausman for Houseman, but it's still not clear. You're presupposing applied economics/econometrics readership; you may be better off in a forum directly aimed at such people. $\endgroup$
    – Nick Cox
    Dec 29, 2013 at 13:48
  • $\begingroup$ If you're interested in the impact of distance, random effects is the only option since the transformations that are used for estimating fixed effects models do not allow you to estimate parameters on variables that don't change over time (such as distance). $\endgroup$
    – dimitriy
    Jan 28, 2014 at 22:14

1 Answer 1


As you are dealing with countries, most likely you will have to use fixed effects.

Intuitively, you can see random effects as follows: you draw the observations of a random sample, for example you have time series of the wear and tear of shoes of the same brand and type. It is unlikely that all shoes have the same wear and tear over time, however one wear and tear profile is not really a property of a particular shoe.

On the other hand, GDP and other parameters are country specific. If you model GDP over time per country, even if you take a couple of good predictors per country, it is very unlikely that you will capture all of the variation of GDP. You explanatory variables will be correlated with the error terms, for example if one of your explanatory variable is unemployment, low unemployment will lead to a high GDP, but unemployment is not truly exogenous, high GDP might cause also low unemployment. In that case you use fixed effects.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.