Panel regression - The industry variable coded 1-19 suffers collinearity and is being omitted from the xtreg. How to include?

[Alex Punshon]

Hello

I am trying to conduct an entity fixed effects panel regression.

The dataset is perfectly balanced and I am trying to control for entity fixed effects. When I run the regression, the variable INDUSTRY that I have coded 1-19 to reflect the various industries in the dataset suffers collinearity and as such is omitted from the regression.

The function is:

xtreg dependent_variable independence_variable control_variables i.INDUSTRY, fe robust

I had a look at an earlier post and it mentioned that perhaps I am asking Stata to make the INDUSTRY variable, but it didn't work when I tried to do the solution provided on that forum.

Please can you help me resolve this issue?

Thank you.

Alex.

[Andrew Musau]

Firms do not change industry in your sample, so the firm effects already capture the industry effects. No need to include industry dummies.

[Alex Punshon]

Andrew Musau thank you for your response. In that case, how will I determine the effect per industry, i.e., how do I replicate i.INDUSTRY?

[Andrew Musau]

In the majority of cases, you want to control for firm effects or industry effects. The coefficients are of no immediate interest (they are nuisance parameters). So when you state

how will I determine the effect per industry

what is your research question? Additionally, in large \(N\) small \(T\) panels, the firm effects are not estimated consistently even if you were to obtain the FE coefficients.

[Alex Punshon]

The research question is broadly 'Does ESG influence firm performance?' and I would like to control for industry effects by including the dummies coded 1-19. One of the hypotheses I will be testing is whether there is a difference between industries, i.e., is the relationship stronger in the Energy industry as opposed to the Retail industry etc. so I wanted to use i.INDUSTRY to compare.

[Andrew Musau]

You could do:

regress dependent_variable independence_variable control_variables i.INDUSTRY, cluster(firm)

This controls for industry effects, but if there is firm-level heterogeneity, the model leaves it unaccounted for.

[Jeff Wooldridge]

Alex: Your research question involves interacting ESG with the industry dummies, not estimating the coefficients on the dummies themselves. You can do that in an FE environment, assuming ESG changes over time.

xtset firmid
xtreg firmperformance c.ESG##(i.industry) control_variables i.year, fe vce(cluster firmid)

If you really want to see coefficients on the industry dummies while also using firm-level FE estimation, you can use the correlated random effects (Mundlak) approach.

[Andrew Musau]

deleted [explained in #5]

[Alex Punshon]

Thanks for your responses. I will see how I get on.