Comparing coefficients across subgroups with non-linear regression

[Daniela Vidart]

Hi all,

I am currently trying to compare the coefficients obtained from a non-linear model for two different subgroups. For each subgroup separately, I run a regression like:

Y = b0 + b1*X^b2 + b3*Z^b2 + b4*W

Where Y, X, Z and W are variables, and b0, b1, b2, b3 and b4 are regression coefficients.

I want to test whether b1, b2 and b3 are statistically the same across two subgroups. If this was a linear model, I would either use SUEST or create interaction terms for the variables I am interested in and repeat the above regression including the interaction terms. However, since I have non-linear terms, I am unsure as to what the correct way to proceed is, and whether Stata has something similar to SUEST for non-linear models.

Thank you very much.

[Maarten Buis]

Interactions are enough for both linear and non-linear regression. There is no need for suest when you want to compare coefficients across groups.

Whether such a comparison makes sense for non-linear models is a matter of debate. My position is here: http://maartenbuis.nl/wp/oddsratio.html , and that also contains plenty of reference for the opposite position.