3 dimensions in panel data, how to solve the problem

[Olivia santers]

Dear all , I have panel data for M firms over 15 years, where each firm has i Ki directors with their own characteristics. The problem arises when I attempt panel regression, as I can only accommodate a cross-sectional dimension and a time dimension. This means that in order to conduct my panel regression, I need to somehow condense my data. I have considered averaging certain variables, but here lies my dilemma: I am unsure how to proceed, as I ultimately wish to examine the effects of the share of female directors on firm performance.
Would anyone have any solution or any possible advice on how to tackle this problem while being able to estimate the proportion of female directors per firm per year?
I highly appreciate any help or advice. Thank you.

[Clyde Schechter]

Sounds like a job for -collapse-, to aggregate the data at the firm level. You can get the proportion of female directors as -(mean) sex- assuming sex is coded as 0 = male, 1 = female. As for your other variables, whether it is appropriate to average them, or take the median, or the max, or the min, etc., depends on exactly what those variables are and what combination of director-specific values is the best representative of the firm as a whole. That's a substantive question, not a statistical one.