Binary or not, autocorrelations are still defined.
I would expect the standard graphs still to make much
sense, but P-values to make less sense, although my
recollection is that the standard errors of autocorrelations
are not that sensitive to the marginal distribution of the
process. A plot of the response against
lagged response will just be four blobs unless you -jitter()-.
But a 2 X 2 table might then be a good summary.
Nick
n.j.cox@durham.ac.uk
Johan Hellström
> I have an unbalanced time-series-cross-sectional dataset with a binary
> dependent variable. I would like to find a way for detecting
> autocorrelation
> in the model. However, as the sample is rather small (about
> 160) and the
> time-series unit is about 10-15 it would probably be unwise to use
> statistical tests on time dummy variables to detect temporal
> dependence (as
> the loss of degrees of freedom would affect the efficiency of
> these tests).
>
> Does anyone have an alternative solution for this problem
> (i.e. detecting
> temporal dependence in an unbalanced TSCS-data set with a
> limited dependent
> variable)?
*
* 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/