xtpcse instead of xtreg

[Chris Man]

Hi,
I used a dataset from a lab experiment
xtreg dep i1 if i2==0 & i3==0
and xtreg dep i1 i2 i3
The coefficient of id remained the same but the standard error changed so there must be some kind of problem. Adding “,robust” did not help, standard errors were still (slightly) different.

When I use xtpcse for the dataset (with the option “hetonly”) standard errors are the same in the two regressions (coefficients are exactly as for xtreg).

Is that the solution? Just say the panel is heteroskedastic and thus I need to use xtpcse?

[Richard Williams]

It could very much help if you showed exact code and output and explained what the variables are. What are these i variables? Are they all dummies created from a single categorical variable? You make reference to the coefficient for id but id is not in the model, at least by that name (do you mean i1?)

[Chris Man]

id1 id2 id3 are dummy variables for treatments 2-4, so they are 0 or 1. (problem remains when I use a larger dataset with more id-variables)
dep variable ranges from 0-10

The code looks as follows

case 1

xtreg dep id1 if id2==0 & id3==0, robust

Random-effects GLS regression
Group variable: subject

(Std. Err. adjusted for 24 clusters in subject)
------------------------------------------------------------------------------
| Robust
dep | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
id1 | a 1.559803 . 0.001 .
_cons | .
-------------+----------------------------------------------------------------

case 2

xtreg dep id1 id2 id3, robust

Random-effects GLS regression
Group variable: subject


(Std. Err. adjusted for 48 clusters in subject)
------------------------------------------------------------------------------
| Robust
dep | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
id1 | a 1.514323 . 0.001
id2 | .
id3 | .
_cons | .
-------------+----------------------------------------------------------------
Hope, it's ok that I deleted some information from the output

Coefficient for id1 (lets call the coeff. a) is identical in both regressions

Std. Err. is 1.559803 and 1.514323 but it should be identical?

Using reg instead of xtreg (with same paramates as above, including ,robust) produces identical SE
xtpcse with the option “hetonly” output has also identical SE

[Richard Williams]

I should have added that you should use the code formatting features. When writing message, click on the A (next to the smiley) and click on #. This will create two code tags; enter your code and output between them. As it is your output is very hard to read. And, I am not sure why you have censored things like the actual value of the coefficient.

Even if the coefficients are the same, the Ns will differ as well as the Ks (# of variables). I suspect that accounts for the small differences in the standard errors. I can't really comment on the other things you tried because you do not show the actual code and output.