Question: Why Do These Two Stata Models Produce Different Results?

[Michael Lee]

I'm running two xtpcse models in Stata, but they are giving different results, and I'm trying to understand why. Here are the models:

 xtpcse y c.x1##c.x2 c.x2##i.x3 ctrlvars, corr(ar1) hetonly
xtpcse y c.x1##c.x2 c.x2#i.x3 x3 ctrlvars, corr(ar1) hetonly

The first model includes the interaction c.x2##i.x3, which should automatically include x2, i.x3, and their interaction. The second model explicitly includes x3 as a separate term in addition to the interaction c.x2#i.x3. I expected the first model to automatically omit any redundant variables, but the results differ between the two specifications. Can anyone explain why this is happening? Is it related to how Stata handles the interaction terms or the hetonly option?

[Joseph Coveney]

Try this.

hetonly xtpcse y c.x1##c.x2 c.x2#i.x3 i.x3 ctrlvars, corr(ar1) hetonly

[Michael Lee]

No. Same problem.
in fact adding x2 to the second model generates same results as the first model!
Which one is true?

[Joao Santos Silva]

Please show us your results.

[Michael Lee]

-