Fixed effects model with non-panel and panelvariables

[Laila Abd]

Dear all,

Excuse my english, it is not my first language. I have very little knowledge about Stata, but I am trying to conduct a fixed effects model. My dataset includes three variables that is not considered panel data(the include size of a board, percantage of female board member and a dummy regarding age difference of a board), while five other variables include observations for 2020 and 2021.
By reshaping from wide to long format, the observations for my three non-panel variables were "doubled up" - the same value is recorded for 2020 and 2021. I then used -xtset- to define my paneldata. But when I try to use -xtreg var1 var2 ... var(n), fe-, the three non panel variables are omitted. It states the following:

"note: MEMBER omitted because of collinearity
note: AGE_DIV omitted because of collinearity
note: GENDER_DIV omitted because of collinearity"

Is there any solution to this? It was possible to conduct a random effects model, but the fixed effects model

Thank you in advance!

[Clyde Schechter]

Your "non-panel" variables do not vary over time within a given panel. That makes them colinear with the panel-level fixed effects in the model, and that, in turn, explains why they are omitted from the regression. It is mathematically impossible to estimate the effects of variables that do not vary within panel in a fixed-effects model.

The question now is why you wanted to include those variables in the first place. If estimating their effects is part of your research goal, then the conclusion is simple: you cannot use a fixed-effects model. Use something different. You've already noticed that a random effects model does not suffer from this limitation. You might also look into -xthybrid-. a user-written command available from SSC.

If, on the other hand, you are only including these three variables to adjust for their possibly confounding effects (i.e. omitted variable bias), then there is no problem. Variables that do not vary within panel are automatically adjusted for in fixed-effects analyses. It comes "for free" and does not require any mention of those variables. (As already noted, mention of those variables is not even possible.) In this situation, just move on to the next steps in your project.