search

JMSLab / xtevent

Stata package -xtevent-
MIT License
43 stars 12 forks source link

Improve explanation of the coefficients for trend(, method(ols)) vs. trend(, method(gmm)) #192

Closed jorpppp closed 3 months ago

jorpppp commented 3 months ago

The explanation on the help file about the different methods to extrapolate the trend is a bit incomplete. It does not explain that the coefficients on the negative event-times will differ between the two methods and have a different interpretation. We should explain this better in the section on trend in the help file.

jorpppp commented 3 months ago

This discussion will be helpful to improve the explanation:

In the regression approach, when we include the trend we omit the event-time coefficients used to estimate it from the regression. So if event-time -2 is used to fit the trend, at event-time -2, y = fixed effects + controls + trend + error.

Instead, in the GMM aproach, we adjust the coefficients to be trend plus adjusted coefficient, so at event-time -2, y = fixed effects + controls + trend + adjusted coefficient + error. That is, this approach does not force the adjusted coefficient to be 0. In the regression approach, the adjusted coefficient is loaded in the error. Both approaches return the same slope for the trend though.

jorpppp commented 3 months ago

This issue will be subsumed by #210 . Closing now.