0
$\begingroup$

i recently ran a regression with fixed effects. As expected, STATA removed one of the dummy-variables, as well as every time-invariant variable (educ). My question is now. Why does STATA also removes the Dummy from 1987? Whenever i remove 'jobexp' from the model, 1987 is included again? My guess is a strong (perfect) multicolinearity. But how do i test for it? Any ideas?

enter image description here

Improve this question
$\endgroup$

1 Answer 1

Reset to default
2
$\begingroup$

vif calculates the variance inflation factors, a common metric of multicollinearity. Usually multicollinearity is not a big deal, unless it is perfect multicollinearity. This seems to be the case here. If year ranges from 1981 to 1987, then yes, you have to drop one of the year dummies. Otherwise you would end up in the dummy variable trap. I have no idea how Stata chooses which one to drop (anybody else?), but note that you can also specify which one to drop by typing something like ib1981.state. Then the year 1981 will then be the "base category" and the other coefficients be relative to this.

Improve this answer
$\endgroup$
8
  • $\begingroup$ Hi, thanks for the reply. I should have been a little more specific in the data. The data is available from 1980-1987, so 1980 is automatically removed. Now the question is why 1987 is also removed. If you remove jobexp for example, then 1987 remains in the model. Does this mean that the variable "jobexp" and the year 1987 are perfectly correlated, i.e. that there is perfect multicollinearity? How can I observe this perfect multicollinearity in STATA? $\endgroup$
    – Karl Seidl
    Jun 20, 2022 at 20:48
  • $\begingroup$ What exactly then correlates with jobexp? The whole year 1987? How do you argue/justify this multicollinearity? $\endgroup$
    – Karl Seidl
    Jun 20, 2022 at 20:54
  • $\begingroup$ Welcome, @KarlSeidl You are observing perfect collinearity. Your results (some variables dropped) say so. Which variables cause it cannot be determined from the regression results. Careful (tedious) examination of the full data set is required. You could try regressing educ on i.state, on jobexp, and on i.state and jobexp, etc., jobexp on educ, on i.state, and on educ and i.state, etc., to have some ideas. $\endgroup$
    – chan1142
    Jun 21, 2022 at 4:19
  • $\begingroup$ @chan1142 Thanks. I will try that. What exactly do i have to Look for while regressing, to determine which variable causes the perfect multicollinearity? $\endgroup$
    – Karl Seidl
    Jun 21, 2022 at 7:34
  • $\begingroup$ Rsq = 1 means the lhs variable is perfectly explained by the rhs variables. $\endgroup$
    – chan1142
    Jun 21, 2022 at 7:36

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.