e(sample) giving more observations than were used in last estimation command

[Arman Azedi]

Hi all,

I'm trying to get a dichotomous variable that codes only the observations used in my last regression command as "1" (ultimately because I need descriptive statistics on these).

I run my regression command:

xtregar logprot logtotalevents logpop elecdem10 polcorrupt10 loggdppc urbanpop unemployment antigov findissat gincdif polinterest antigovXunemp findissatXunemp gincdifXunemp polintXunemp if person > 200, fe

Then do the following command:

gen samp = e(sample)

However, the regression output itself says that 260 observations were included in the model, but after using the e(sample) technique the ensuing samp variable has 293 observations coded as "1."

What am i missing here? Any help is greatly appreciated!

[Andrew Musau]

In xtregar -fe-, you lose a cross-section for each panel due to lagging. But the lost observations are still part of the initial sample. So the command reports the number of observations after lagging and not the number of initial sample observations.