Reporting coefficients of log-linearized models with estout and tranform()

[Simon Heß]

People often estimate models with logarithmized dependent variables to interpret coefficients as approximations for semi-elasticites. All issues one might have with that aside, why is it not common to report transformed coefficients which correct for the imprecise elasticity approximation? The same applies (I think) to Poisson models?

With estout/esttab, applying the transformation should be straightforward:

reg lny x
esttab ., transform(exp(@)-1)

The sole paper paper that I found which does that seems to be: https://www.pnas.org/doi/10.1073/pnas.2024399118

My question in short: Why is this not common, since it's so easy? Is there something wrong with this?

[Andrew Musau]

The correct way of doing it is shown in the following link, along with the recommendation to fit a Poisson regression with robust standard errors instead.

https://blog.stata.com/2011/08/22/us...tell-a-friend/

[Simon Heß]

Thanks Andrew, I have reasons for which I am unable to use poisson. But thanks for the link.