In olden days, some "expert" with "experience" or "expertise" would say
something like "Oh, for such-and-such test n at least 30 is fine" or
"For Foobar's test you need 300 at least". Sometimes these judgements ex
cathedra were well based; sometimes they were not worth much.
Now, I have no idea about this test. But
1. You can decide what you mean by "meaningful results".
2. You can set up simulations with likely data generation processes and
see how they perform with small samples.
Stata 8 was a while ago but all the machinery you need is there.
Nick
[email protected]
Carlos Rodriguez
I have read that the Breusch-Godfrey test for serial correlation
should be used only for large samples. How large should they be to
run it in STATA8 and obtain meaningful results?
*
* 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/
