How to retain dummy variables in a fixed effects (FE) model?

[Javier SV]

Some years ago, I came across a method to retain time-invariant dummy variables (those that remain constant for each observation over the entire period) in a fixed-effects regression. This method involves applying the following transformation:

obs year duco(original_dummy) ducoa(Transformed_dummy)
1 2002 1 1
1 2003 1 2
1 2004 1 3
1 2005 1 4
1 2006 1 5
1 2007 1 6
2 2002 1 1
2 2003 1 2
2 2004 1 3 ...


Have you ever heard of this method? would it be correct? Apparently, when applying the within-groups (WG) transformation —which includes, among other steps, taking differences— these values are transformed into a constant (1). This allows the OLS regression of the WG-transformed variables to retain the time-invariant dummies. I would appreciate any opinions or insights about this transformation.

Thanks for any help/advice

[George Ford]

isn't Transformed_dummy just a time fixed effect, or a trend?

[George Ford]

Mundlak regression might work.

[George Ford]

I suppose the Transformed_dummy is telling you how Y changes over time for the orig_dummy==1 group relative to the ==0 group. A slope not an intercept.

[Carlo Lazzaro]

Javier:
welcoem to this forum.
Following George's helpful hint, you may want to take a look at The Stata Blog » Fixed effects or random effects: The Mundlak approach