on 24/11/2003 14:25, Richard Williams at [email protected] wrote:
This is discussed on p. 357 of the Stata 8 Users guide (see point #2). I
can't say that I fully understand the explanation, but the key phrase seems
to be "Using z rather than t statistics only yields a nontrivial difference
when there is a small number of clusters (<50)."
That strikes me as not an entirely unmeaningless statement. Does this mean
that the difference is important in any case when the number of clusters
goes above fifty? It could be clearer, so.
Incidentally, I live with someone who occasionally uses modified triple
negatives in conversation. Makes life colourful.
My head was spinning a bit too. My interpretation of the complete
paragraph is that (1) t statistics are actually what is technically correct
even when z statistics are reported by Stata (2) when the number of
clusters is small, the difference between t and z is trivial (3) when the
number of clusters is large (>50) the difference between t and z is not
trivial (or at least need not be trivial). If that is an incorrect
interpretation, maybe somebody who understands the svy commands can explain
it better.
