About unbalanced data panel

[Lenny Tran]

I have a data set , 12 countries and a period of 10 years
which of them have different year period, like country A data from year 1-10, B data from year 2-10, and 4-10
and none of them have data like this: year 1, missing 2,3, year4-10
so my data is called unbalanced data right?
and i can use all the command : xtreg fe, xtreg re,xtreg fe robust, xtreg re robust, xtserial, collin, hausman, hausman sigmamore, xttest3, xttest0,...
for my unbalanced data, right?

[Carlo Lazzaro]

Lenny:
yes, you can.
However, since your T dimension is almost equal to your N dimension, you may want to consider -xtregar- if you have a continuos regressand.

[Lenny Tran]

Lenny:
yes, you can.
However, since your T dimension is almost equal to your N dimension, you may want to consider -xtregar- if you have a continuos regressand.

Thanks

1. What if I have 9 or 10 countries and a year of 11, I should use xtregar, right ? And if i dont, does the result still useable ?

2. Can you define "almost equal", like 0.9 < T/N < 1.1 or something ?

3. If I have 9 or 10 countries and I try the quarter data for 10 years (which means 10-40 quarter depend on country) , then I should use the xtreg fe, xtreg re but not xtregar, correct ?

[Carlo Lazzaro]

Lenny:
as far as I know, there's no fast and hard rule to split short from long panel datasets.
As far as the T dimension becomes larger, issues like AR(1) disturbance increase their importance: if you omit to take them into account, your results may be biased.

[Lenny Tran]

Lenny:
as far as I know, there's no fast and hard rule to split short from long panel datasets.
As far as the T dimension becomes larger, issues like AR(1) disturbance increase their importance: if you omit to take them into account, your results may be biased.

If I have 9 or 10 countries and I try the quarter data for 10 years (which means 10-40 quarter depend on country) , then I should use the xtreg fe, xtreg re but not xtregar, correct ?

Can you help me with this, i just edited my post, sorry.

[Carlo Lazzaro]

Lenny:
if you have N=9 or 10 and T=ranging between 10 and 40 quarters, on average T>N: hence, you should go -xtrergar-.
Take also a look at -xtgls-,

[Lenny Tran]

Lenny:
if you have N=9 or 10 and T=ranging between 10 and 40 quarters, on average T>N: hence, you should go -xtrergar-.
Take also a look at -xtgls-,

Thanks alot, it helps me much !!!

[Lenny Tran]

Lenny:
if you have N=9 or 10 and T=ranging between 10 and 40 quarters, on average T>N: hence, you should go -xtrergar-.
Take also a look at -xtgls-,

Anyway, xtgls or xtrergar is a different model from xtreg fe and xtreg re, right ?

thanks

[Carlo Lazzaro]

Lenny:
yes.
Just to wrap up:
-xtreg- has been devised for N>T panel datasets with a continuous regressand;
-xtgls- and -xtregar- have been devised for T>N panel datasets with a continuous regressand;

[Lenny Tran]

Lenny:
yes.
Just to wrap up:
-xtreg- has been devised for N>T panel datasets with a continuous regressand;
-xtgls- and -xtregar- have been devised for T>N panel datasets with a continuous regressand;

As my knowledge
N>T :
reg : ols
xtreg fe: fixed
xtref re: random

N< T
xtgls: ols
xtregar fe: fixed
xtregar re: random

if I use xtgls and xtregar , does it have command for me to check if my model have collinear, hetero, or correlation, ... ?
Or like hausman test to choose the model ( xtgls, xtregar fe, xtregar re)
sorry for asking a lot of question, but i think i shouldnot make another topic about this

[Lenny Tran]

Lenny:
yes.
Just to wrap up:
-xtreg- has been devised for N>T panel datasets with a continuous regressand;
-xtgls- and -xtregar- have been devised for T>N panel datasets with a continuous regressand;

I have tried the xtgls and xtregar fe, xtregar re
and the results doesn't good (as I exptected) then the xtreg fe robust
I sitll prefer the results xtreg fe robust, the variable make sense in economics, and the same result as other paper I read
the xtregar and xtgls : the results is non sense and ridiculous
what should I do ?

[Carlo Lazzaro]

Lenny:
it is difficult to advise without knowing your data, nor seeing your results.
In addition to skimming the literature of your research field, you can seek advice from a colleague or teacher/professor/supervisor.

[Lenny Tran]

Lenny:
it is difficult to advise without knowing your data, nor seeing your results.
In addition to skimming the literature of your research field, you can seek advice from a colleague or teacher/professor/supervisor.

Thanks, but any command like xtreg fe robust for xtregar
like xtregar fe robust ?
and "hausman test" for xtregar fe and xtregar re

[Carlo Lazzaro]

Lenny:
-hausman- supports the comparison between -fe- vs -re- under -xtregar- with default standard errors.
Conversely, I'm not aware of any (official or community-contributed) command for -xtregar- that works as -xtoverid- does for -xtreg- with non-default standard errors.

[Lenny Tran]

Lenny:
-hausman- supports the comparison between -fe- vs -re- under -xtregar- with default standard errors.
Conversely, I'm not aware of any (official or community-contributed) command for -xtregar- that works as -xtoverid- does for -xtreg- with non-default standard errors.

Thanks, but I dont understand what you said.
It just a homework for the subject, I think I just stick to the xtreg fe robust.