Kimberly,
Panel data analysis and cross-sectional time series are essentially the same. You will need to run a Hausman test to ascertain whether your random error is correlated with the individual effects. If they are, you will have to use the fixed effects model. In that event, you need to be concerned that you don't have too many cross-sections and dummy variables to test your effects. You want to run your LSDV model against the pooled regression to determine whether there is a significant improvement in the rsquare. In this way you can test your group or cross-sectional effects. You will need to also determine whether there is autocorrelation in the model as well.
As for the estimation procedure, xtgls allows for correlation across the cross-sections, ar(1) within the panels, and heteroskedasticity and no cross-panel correlation. Xtgee allows for a richer designation of within panel correlation as long as that correlation applies to all panels.
See Stata Press, "Cross Sectional Time Series, p. 86" for this reference.
- Bob Yaffee
Robert A. Yaffee, Ph.D.
Senior Research/Statistical Consultant
Statistics and Social Science Group
Information Technology Services
Academic Computing Services
New York University
251 Mercer Street
New York, New York, 10012
phone: (1)212-998-3402
fax: (1)212-995-4120
e-mail: [email protected]
[email protected]
Office location: 75 Third Avenue
Level C-3
New York, NY 10003
----- Original Message -----
From: Kimberley Tran <[email protected]>
Date: Saturday, September 6, 2003 11:14 am
Subject: st: cross-sectional time series data
> Greetings all,
>
> I would like to get suggestions concerning cross-sectional time
> series data.
> My estimation model consists of 10 individual units, each sharing
> the same
> variables and for the same 19-year time period. Where I am
> confused is whether
> to consider this as Panel data or cross-sectional time series
> data. Certain
> literature treats Panel and cross-sectional time series as one in
> the same,
> while others indicate that they are not.
> Further complication is knowing which Stata commands to use.
> Initially I ran this command
> . xtreg y x1 x2 x3 x4 x5 x6 x7, re
> to test to see whether GLS is necessary or simple OLS will do for
> the
> estimation.
>
> Secondly I ran
> . xttest0
> to discriminate against eitehr the Pool Model and the Random
> Effects Model and
> the results indicated pooled model was not appropriate.
>
> I then ran the Hausman test in order to determine betweem the
> Random Effect
> Model or Fixed effects model:
> . xthaus
>
> The results indicated that the Fixed effects is appropriate.
>
> Are these the right steps and commands to apply to cross-sectional
> time series
> data? Or should I use the xtgls or the xtgee commands?
>
> I would greatly appreciate your suggestions in clarifying this
> confusion.
> Thank you,
> Kimberley
>
> *
> * For searches and help try:
> * http://www.stata.com/support/faqs/res/findit.html
> * http://www.stata.com/support/statalist/faq
> * http://www.ats.ucla.edu/stat/stata/
>
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
