Further control variables in a state-year-fixed effect model

[Korhan Zeyrek]

Dear statalists,

I am somehow confused as I find no information regarding this issue, although a collegue pointed me in a (probably) wrong direction.

I am currently conducting a panel regression with firm data in the United States. More concretely, I analyse the influence of financial markets on the entry rate of new firms. I use state-fixed and year-fixed effects, to control for differences between states, and effects due to business cycles, etc.

Considering this, does it make any sense to use any control variables in the model (such as GDP, GDP growth, etc.)? I am asking because I thought by using both state- and year-fixed-effects I basically control for every potential effect.

Your help is much appreciated!
Best
Korhan

[Clyde Schechter]

By using state and year fixed effects you have adjusted (not "controlled") for every potential effect that is constant within each state over the years, or is constant across all states within each year. But any effect that varies on both of these dimensions remains unadjusted for. Think about it, if that weren't the case, it wouldn't even make sense to analyze your main variables because, by your reasoning, those effects are already adjusted for!

Now, if by GDP you are referring to national GDP, which would presumably not vary among the states within any years, that is, indeed, already taken care of, and were you to include it, Stata would identify its colinearity with the state effects and omit it from the model anyway. But you have to examine each of your potential covariates individually to determine whether they vary on both the state and time dimensions. If you had state-level economic growth statistics, for example, they would certainly be includable.

IN truth, you don't need to worry much about this issue anyway. If you mistakenly include a variable that is already covered by the state and year effects, Stata will recognize this fact and omit it for you anyway. No harm done.

[Korhan Zeyrek]

Dear Clyde,

Thank you very much for the quick reply and comprehensive information. It is very clear to me now. Though from a more practical point of view it seems very complicated to me to actually find data which may vary simulatiously across states and years. In this case, I do not even have states, but counties as my regional component.

However, if I consider it possible that the initial share of firms in a region (as compared to all firms in the US) may be a proxy for the development stage of that region, and that high developed regions to some extent grow more slowely than those at an earlier stage of development, would this qualify for a potential control? Following this reasoning, firm shares of regions would be different across states and its growth potentially vary across time. If this argument would be correct, it is probably the only practical type of control variable I see here. I am very open to further possibilities/thoughts.

Thank you again!

Best
Korhan

[Clyde Schechter]

I'm not an economist, so I don't know if that idea makes substantive sense. But from a statistical point of view, that would be an example of something that varies in both the spatial and time dimensions and could be added to the model. Whether it should be added to the model is an economics question, for which I cannot advise you.

[Korhan Zeyrek]

Okay, I understand.

Thank you very much for the help which I highly appreciate!

Best Korhan

PS: As I justed stumbled upon a paper about convergence/devergence of GDP by region, I want to share it here in case anyone finds it interesting: https://repository.up.ac.za/bitstrea...pdf?sequence=1