Here's a statement I read from the method section in a paper: "One disadvantage of the fixed-effects approach is that the results obtained are conditional on the data used to estimate them; that is, results cannot be generalized to other years or microregions not included in the study."

I think I haven't read a similar comment on fixed effects model elsewhere so want to know if that argument is true. If so, is it a disadvantage of fixed effects versus random effects model?

  • 1
    $\begingroup$ I also would not know what would make FE results have any more or less external validity than those of RE or any other empirical technique. $\endgroup$ – Christoph Hanck May 23 '15 at 21:25

Well the statement is true for all statistics, since a statistic is (but) a function of the data. Converserly, I know of no statistical model/method that overcome the problem.

It a generally a big no-no in statistics to generalize to the entire population, unless you can somehow argue that your sample is "very" good.

On a side note; I have seen some reserachers, within machine learning, who leave out some of the data when they estimate. Then they try to predict the outcome of these (leftout) observations. As a way of testing the the predictive power.

| cite | improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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