[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
RE: st: xtivreg, fe: missing R-sq within interpretation.
"Guillaume Frechette" <email@example.com> wrote that he obtained a missing
value for R-sq after -xtivreg , fe-. Specifically,
> after performing a xtivreg, fe command, I noticed that the results
> were strange (very large coefficient estimates as compared to
> similar samples). While trying to determine if I had done anything
> wrong, I noticed that R-sq: within = . (it is set to missing). Can
> someone tell me how to interpret this. Does this mean there is a
> problem? If so, what can you tell me about it?
The R-sq after -xtivreg, fe- is subject to the same caveats of an R-sq after
any IV regression. On this point, I recommend that Guillaume see the FAQ
As the FAQ notes, a missing R-sq from an IV regression does not necessarily
indicate that there is a problem. The fact that the estimated coefficients
change by a large amount relative to other samples raises a flag to me as it
did for Guillaume. The first possible cause that pops into my mind is that
that the strength of the instruments varies significantly over the samples.
Here it is important to keep in mind that the demeaned instruments must have
sufficient correlation with the demeaned endogenous variables on the
right-hand-side to identify the coefficients.
If some of the instruments, or endogenous variables, have very little within
panel variation in some of the subsamples, this could cause the demeaned
instruments to lack sufficient correlation with the demeaned endogenous
variables to identify the coefficients in these samples.
If I were confronted with the puzzle that Guillaume faces, I would do three
things. First, I would use the -first- option to examine the first stage
regressions. Second, if this did not lead me to any specific conclusions, I
would use -xtsum- to look for variables that are nearly constant within
panel for some samples. Third, if I had further questions, I would demean
the instruments and the endogenous variables and examine their correlations
I hope that this helps.
* For searches and help try: