[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
Re: st:panel data manipulation
On Aug 14, 2004, at 2:20 AM, caglar wrote:
Thanks for your suggestion. Actually, I tried to create lags and leads
for my explanatory variables before. When I include all the leads and
lags in an equation like the one below, STATA does not even do the
estimation as there is no observation.
Not that I know of. And I'm not certain you can easily do what you
want -- nor can I think of a statistically valid reason to do it. But
since you asked, there is (conceptually, at least) a way to get Stata
to do what you say you want.
Y_i(t)= X_i(t) +L.X_i(t)+L2.X_i(t)....+F.X_i(t)+F2.X_i(t)..F8.X_i(t)
Even if the above equation were estimated, this equation would not
give me coefficients on the variables from the regression
Y_i(t) = X_i(t) + X_i1987 + X_i1988+....+X_i1997
i:1...N while t:1987..1997
I wonder if there is any command in STATA that will include all past
and future values of the explanatory variables for every cross section
over the 1987-1997 time series?
You need to create 11 new variables: Xi_1987 - Xi_1997. The value of
Xi_1987 for each year t in your sample should be set to the 1987 value
of Xi. That is, you repeat the 1987 values for the 1988 observations
on Xi_1987, and for the 1989 observations, and for the 1990
observations, and so on. Repeat similarly for Xi_1988 through Xi_1997.
Then run your proposed regression.
If I'm not mistaken, as you have written your proposed regression
X_i(t) will be perfectly collinear with the variables you just created.
I think Stata will just drop a regressor in that case.
Again, I cannot conceive of a situation in which one might want to do
this, but I presume you have. Almost certainly theory suggests using
relative dates, as Clive suggested with the L. and B. operators, rather
than the absolute dates you have requested. Best of luck.
* For searches and help try: