Re: st: How to estimate Fixed and random effects for a long panel dataset.

[James Bailey <feanor1600@gmail.com>]

Herman,

First, I wonder what controls you are using- this could be the true cause
of the degrees of freedom issue.

If it really is a clustering issue, then Stata may be trying to stop you
for your own good. Mostly Harmless Econometrics section 8.2.3 discusses how
clustered standard errors are biased if there are fewer than 42 clusters,
and you are way under 42.
They discuss possible solutions, one of which is bootstrapped errors.

Finally, with such a large T and small N, it may make sense to abandon
panel models entirely and use a time series technique like Vector
Autoregression instead.

Best,
James Bailey
Temple University Department of Economics

On Sat, Aug 3, 2013 at 5:27 AM, Herman Haugland
<herman.haugland@gmail.com> wrote:
> Dear all,
>
> I think I have sent this e-mail before, but I don't know if it made it
> through Majordomo.
>
>
> I have a long panel dataset, meaning my N is much smaller than my T.
> I have N = 5, T = 61. I am trying to perform OLS, Fixed-effects and
> Random-effects analysis, using vce(cluster id).
>
> I tried to estimate my model using xtreg for FE and RE, but I get an
> error related to the fact that I do not have enough degrees of freedom
> for performing the estimation.
>
> This is what I get:
>
> Wald chi2(4)       =        .
> Prob > chi2        =        .
>
>
> Stata sends me here for help:  help j_robustsingular  // My case is
> explained under the title "Are you using a svy estimator or did you
> specify the vce(cluster clustvar) option?"
>
> So, after reading that, I have assumed that I cannot trust the output
> of that estimation, because the errors might be biased.
>
> First question: Am I right on thinking that?
>
> In addition, in the book "Microeconometrics using Stata", the author
> clearly indicates that the xtreg command, with the vce(cluster id)
> option for calculating robust errors, is mostly appropriate for short
> panels, which is not my case.
>
> An alternative is to use the command xtregar for estimating random and
> fixed effects, which is based on an AR(1) process for the errors.
> However, I tested using xtserial, and the errors do not show serial
> autocorrelation. However, the xtregar command has the option rhof(#),
> where # indicates the desired rho value (AR(rho)).
>
> My main questions are:
>
> 1) What is the right way to calculate Fixed and Random Effects for a
> long panel dataset, in which the number of variables is larger than
> the N?
>
> 2) Would specifying rho = 0 completely eliminate the AR(1) process for
> the errors, and leave me with an estimation that fits my data?
>
>
> Thank you for your answers.
>
> Best regards,
> Herman Haugland
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/faqs/resources/statalist-faq/
> *   http://www.ats.ucla.edu/stat/stata/
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/