Interpreting coefficients for log-transformed data (OLS regression)


If I run an OLS regression on log-transformed data, and then I exponentiate the coefficient to make it interpretable, what do I call that exponentiated value? My DV is a count of shares on facebook, and the model fit was better for log-transformed OLS regression than other count models (e.g., poisson or NB).

For example, if it was a count model, the exponentiated value would be the Incident Rate Ratio.


Not a robit
If value is not backtransformed then it would be change in y given ??% change in x. I don't think you would backtransform if just the x is ln transformed. I can check my cheatsheet later this week.

OLS is acceptable when mean counts are higher, >8, and helps if you have a large sample.