xpcse vs xtreg

[Kenulina Schneider]

Dear all,

my panel data consist if 9 countries and 12 years.For estimation I was running the following commands:
xtreg depvar indepvar, fe robust
estimates store fixed
xtreg depvar indep var, re robust
estimates store random
xtoverid
(p<o.o5, meaning that fixed effect is appropriate)

However, in one paper I noticed that the author uses Prais-Winsten AR (1) Regressions , which I think is xpcse depvar indepvar, core (ar1) command.

1)Does the robust cluster(id) option eliminate serial correlation and heterosk. problems? Because i read somewhere that cluster(id) option works only when N>T, which is not my case.
2)Can anyone explain the difference between xpcse and xtreg commands?
P.S. Data has serial correlation and heterosk. problem. Also with xpcse command i get results that almost consistent with the literature. Im so confused between commands.
Any suggestion would be appreciated!

Best,Kenulina

[Carlo Lazzaro]

Kenulina:
1) cluster(id) option is not available with -xtpcse-;
2) The differences between -xtreg- and -xtpcse- are covered in Cameron CA, Trivedi PK. Micreoeconometrics Using Stata. Revised Edition. College Station, TX: Stata Press, 2010: page 271-8,

[Kenulina Schneider]

Thank you Carlo!
My problem is that i get coefficients with negative sign, while on theory it is expected to be positive and significant. I run all possible commands but cannot fix the problem. I dont know maybe it is due to data structure or model is not good constructed.

Best, Kenulina

[Carlo Lazzaro]

Kenulina:
sign-flipping coefficients deserve some scrutiny. A possible approach can be starting out with one (or a limited handful) of predictors, adding the remaining one by one and see when the unexpected (if truly so) beahaviour start to come alive.
As far as the model construction is concerned, the usual golden rule is to take a look at what other researchers did in dealing with the same research topic.