Back transform log transformed coeffcients

[Ionas Papassotiriou]

Hello there!

I have run a linear regression where the dependent variable has been log transformed due to skewness of the data. I have been asked to back transform the coefficients of the regression. I tried this with the command disp exp(X). However, the coefficients negative and as far as I know they cannot be exponentiated? How can I translate them into their non log value?

[Joseph Coveney]

. . the coefficients negative and as far as I know they cannot be exponentiated?

. di exp(-1)
.36787944

Two better options . . .

First choice:

help glm

especially note the link functions.

Alterrnative: tell the referee off.

[Roman Mostazir]

You did not show how you fitted the regression. If you have used 'regress' command, 'eform' gives you the exponentiated coefficients:

reg y x1 x2, eform("exp")

[Maarten Buis]

Beware that with log transforming the dependent variable combined with the eform() option you no longer model the conditional arithmetic ("normal") mean but the conditional geometric mean: https://doi.org/10.1177%2F1536867X0400300412

If you want to model the conditional arithmetic mean you need to use glm with the link(log) and vce(robust) options, or equivalently poisson with the vce(robust) options: https://blog.stata.com/2011/08/22/us...tell-a-friend/