Sequential factor variables create "extra" variables. How?

[Nils Enevoldsen]

. sysuse auto, clear
(1978 Automobile Data)

. fvexpand 1.foreign#0.rep78#c.price

. return list

macros:
              r(fvops) : "true"
            r(varlist) : "1o.foreign#0o.rep78#co.price"

. fvexpand  1.foreign#1.rep78

. return list

macros:
              r(fvops) : "true"
            r(varlist) : "1o.foreign#1o.rep78"

So far, so good. But then if I expand both together…

. fvexpand 1.foreign#0.rep78#c.price 1.foreign#1.rep78

. return list

macros:
              r(fvops) : "true"
            r(varlist) : "1o.foreign#0b.rep78#co.price 1o.foreign#1o.rep78#co.price 1o.foreign#0b.rep78 1o.foreign#1o.rep78"

By what logic did 1o.foreign#1o.rep78#co.price 1o.foreign#0b.rep78 get created?

[Phil Bromiley]

You didn't get a quick answer. Most of us don't use fvexpand since most estimators allow factor variables and using factor variables lets you use the power of margins after the estimation.

[Nils Enevoldsen]

I don't use fvexpand directly either. I discovered this behavior when troubleshooting some unintended coefficients popping up in a regression I was running.

[William Lisowski]

What version of Stata are you running? I'm not able to recreate your results.

. sysuse auto, clear
(1978 Automobile Data)

. fvexpand 1.foreign#0.rep78#c.price

. return list

macros:
              r(fvops) : "true"
            r(varlist) : "1o.foreign#0o.rep78#co.price"

. fvexpand  1.foreign#1.rep78

. return list

macros:
              r(fvops) : "true"
            r(varlist) : "1o.foreign#1o.rep78"

. fvexpand 1.foreign#0.rep78#c.price 1.foreign#1.rep78

. return list

macros:
              r(fvops) : "true"
            r(varlist) : "1o.foreign#0o.rep78#co.price 1o.foreign#1o.rep78"

. about

Stata/SE 15.1 for Mac (64-bit Intel)
Revision 15 Oct 2018

[Nils Enevoldsen]

William Lisowski I'm running Stata/SE 14.2 for Mac (64-bit Intel). It seems they've improved this behavior. Good to know! Thanks for the replication.

[Nils Enevoldsen]

Ah, it might be related to this change:

7. Factor variables now work more like you would expect they would

Consider fitting a model with the terms

. est_command ... i.a i(2 3).a#i.b ...

What should happen? What happens now should be more in line with your expectations. i.a adds main-effect coefficients for each level of a, and the interaction i(2 3).a#i.b is restricted to a’s levels 2 and 3.

What used to happen was rather more surprising. The entire RHS of the model was restricted to levels 2 and 3 of a.