0

Is it possible to get a larger coefficient (either a larger negative or a larger positive) when moving from OLS to a fixed effects regression?

Furthermore, is it possible/likely for a coefficient sign to flip when moving from OLS to fixed effect regression?

Could this be completely 'normal' or is it a sign of problems with the model / data?

KSL
  • 61
  • 7
  • Are you adding fixed effects terms to an existing OLS regression? – jbowman Nov 28 '13 at 22:18
  • I find the flip sign (or larger coefficient) regardless of whether I am simply adding indicator terms or estimating a fixed effects regression. – KSL Nov 29 '13 at 00:21
  • ... but with the "estimating a fixed effects regression", the fixed effects weren't in the original model as indicators, for example? If you add new variables to a model, be they explicitly as indicators or implicitly as fixed effects, you're easily able to get sign changes in the coefficients of the variables that were there before. – jbowman Nov 29 '13 at 03:50
  • Sorry for not being clear. Yes, I can either estimate a regression without the fixed effects, get coefficients (say both positive on A and B), and then add the fixed effects (either directly as indicators or by estimating within-regression), and see the coefficients change dramatically (A get larger positive, and B flip negative, for examples). Sounds like that is completely normal and common? If so, thanks. – KSL Nov 29 '13 at 04:42
  • You might also want to look at these two answers: http://stats.stackexchange.com/questions/31841/coefficients-change-signs/32237#32237 and http://stats.stackexchange.com/questions/25605/why-anova-regression-results-change-when-controlling-for-another-variable/25650#25650 – jbowman Nov 29 '13 at 15:45

0 Answers0