RE: st: Best test to detect trends in panel data

[Gordon Hughes <G.A.Hughes@ed.ac.uk>]

As you have noted, the different approaches test different 
hypotheses.  However, panel unit root tests are never likely to be 
informative when you have such a small number of observations for 
each panel unit.  People differ on the minimum number of time periods 
required when unit root tests are used to identify trends but a 
maximum of 10 - and an average of 7.4 - is certainly too small.

Unfortunately, the same is probably true for -xtreg-.  In this case, 
you are testing whether there is the *same* time trend for each panel 
unit, so that even if every individual panel had a time trend you 
might still find that the coefficient on period in your model was not 
significantly different because of variation across panel units.

There is a further problem - the missing values in your data - since 
some more sophisticated specifications required balanced panel 
data.  You could try estimating the random coefficient model -xtrc- 
which will give you a better sense of the mean value of the time 
trends across panel units.  Of course, the limited number of time 
periods is likely to mean that the standard error of the mean time 
trend will be large.

Gordon Hughes
g.a.hughes@ed.ac.uk


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/