# st: areg vs xi reg vs xtreg vs what else?

 From Dana Chandler To statalist@hsphsun2.harvard.edu Subject st: areg vs xi reg vs xtreg vs what else? Date Thu, 17 Sep 2009 08:39:39 -0500

```Hello,

I'm curious if someone could kindly explain the difference between the
abovementioned stata commands. Below is a specifically empirical
problem and a case where the commands do not seem to be generating
what I want. Finally, I ask what the benefit of xtreg is.

I'm running a large fixed effects model where each observation is a
test score for every grade level, year, and state. My outcome variable
is a test score and the focal independent variable is the % of HH in
state that have cable television. The only other variable I have is
log_population for the first year of my study at the state level which
I also want to interact with each combination of grade-year.

First I would generate a variable representing each grade-year and log
pop interaction:
renpfix _yeaxln AY_pop
** Hence there are now many AY_pop* variables which represent
individual grade-year interactions with log population for the first
year of my study.

Using areg (which sometimes seems faster), I would type:
xi: areg score perc_HH i.grade*i.state i.state*i.year AY_pop*, a(gradeyear)
** After doing this regression with areg, I get a message that the
matrix is highly symmetric and no standard errors are calculated. Why
is this happening? Is it a stata rounding error?

So next, I try to not use areg and instead list the gradeyear
interactions with an xi:
xi: reg score perc_HH i.grade*i.year i.grade*i.state i.state*i.year AY_pop*
** Now, my focal independent variable has a standard error, but many
of the other dummy variables in my model don't and some have standard
errors that are extremely large making me suspicious of the standard
error on my focal independent variable - % of HH that have cable.

Can anyone shed light onto this problem? I am not familiar with xtreg,
but does anyone know if this might work in xtreg which i know is
another command frequently used for FE regressions?