I have a question about the fixed effects model.

I have 25 manufacturing sectors for a 12 year period (balanced panel data set) and I try to find out the effect of each sector's investment expenditure on the ratio of skilled/ unskilled workers for each sector, so my regression model is:

S/U= f(real investment expenditure) among other things. Because I'm interested in capturing the effect of investment on S/U for different sectors individually, I created dummy variables for each sector and interacted them with investment expenditures of each sector. When I run that regression with fixed effects, investment appears to be a significant factor for determining skill ratio only for 2 sectors. However, when I run the regression with fe, vce (robust), the relationship seems significant for almost all sectors.

I'm wondering how this is possible.


  • $\begingroup$ It’s not clear to me exactly what you’re doing in the two cases you’re comparing. $\endgroup$ – The Laconic Mar 16 at 3:25

Regarding your question and the Stata FAQ quoted above, note that when you use vce(robust) in [xtreg, fe], Stata interprets this as [xtreg, fe vce(cluster)].

If you have a negative correlation in either the regressor of interest or the errors within the cluster, you can have such effect.

Please look at this Stata FAQ: https://www.stata.com/support/faqs/statistics/standard-errors-and-vce-cluster-option/ or Cameron & Miller, 2015 (http://cameron.econ.ucdavis.edu/research/Cameron_Miller_JHR_2015_February.pdf , p17)


I think your question is more suitable to model in a generalized linear mixed model with a random and slope model specification, ie.

$$ \log(\frac{S}{U})\sim 1+ investment + (1+investment | sector),$$

It's a more powerful method that also take into account the random effect of the sectors


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.