Problem with insufficient observations

[Emanuel Guerra Leandro]

Hi!

I'm with a doubt that i already saw the answer on the forum, but on my code is not working.

I'm trying to estimate a model with first difference, so im doing:

A example of my base:

ano	cod_munic	pib	prop_serv
2002	2800100	5509.195	.7160816
2007	2800100	10907.59	.7550546
2002	2800209	44256.39	.7147408
2007	2800209	86428.34	.689031

What i'm doing:

xtset cod_munic ano, delta(5)
xtreg pib prop_serv , fe
xtreg d.pib d.prop_serv

But the stata is telling me that i have insufficient observations, but i know that i have the same number of observations to the two years.

The function xtreg works with just two years or i have to try another way??

Thanks!!

[Clyde Schechter]

If you have only two observations for each cod_munic, one for 2002 and one for 2007, then d.pib and d.prop_serv are only defined for one observation, as it is the difference between two consecutive observations. So for your second regression, you have just one observation per cod_munic. But singleton panels don't contribute any information to a fixed effects regression. So, you have no information for the regression, and Stata appropriately tells you that.

Added: I didn't need to use your example data in Stata to answer this question. But if I had, it would have been time-consuming and difficult to get it into Stata from the form in which you posted it. In the future, all data examples should be posted using the -dataex- command, which you can get by running -ssc install dataex-. Read -help dataex- for the simple instructions. Using -dataex- makes it quick and easy for those who want to help you to create a faithful replica of your data example in their own Stata.

[Emanuel Guerra Leandro]

Thanks, Clyde!

So what is the right way to estimate this model with first difference on the stata? I know what i have to do, but i dont know how the stata does that.

[Emanuel Guerra Leandro]

http://www.statalist.org/forums/foru...nce-regression

I'm trying to do the same thing that was said in this link

[Clyde Schechter]

Well, that thread never did close with a statement that everything had been worked out. I suspect it had not. You simply can't combine -xtreg, fe-, the difference operator and only two observations per panel and get results. It's a mathematical impossibility, and I suspect that if follow-up on that thread were available we would learn that they were not successful.

With only two observations per municipality, you are more or less stuck. What you can do is abandon -xtreg, fe- and just do

regress d.pib d.prop_serve

You still have only one observation per municipality for this regression, but because it is not a fixed effects regression, that will be sufficient to get results. Whether results based on only two years of data per municipality are meaningful depends on your context and your research goals, so I can't comment on that. It might well be quite adequate, or it might not.