Koen Vermeer <[EMAIL PROTECTED]> wrote in message
asneph$s5l$[EMAIL PROTECTED]">news:asneph$s5l$[EMAIL PROTECTED]...
> Hi,
>
> I want to test whether a set is drawn from a normal distribution. With a
> Kolmogorov-Smirnov test, I can do this for known mean and variance. The
> Lilliefors test is essentially the same, but for unknown mean and variance
> (thus estimated from the data).
> Now, in my case, I have a known mean (zero) and unknown variance, meaning
> that my situation is somewhere in between Kolmogorov-Smirnov and
> Lilliefors. Is there a separate test for this? I haven't checked the
> Lilliefors paper yet, so maybe I am able to partly follow the paper and
> come up with a similar result for known mean and unknown variance, but if
> someone else has already done it, I'd rather use those results.

I don't know of it being done, but Lilliefors just used simulation.
You could do the same if you wanted. (Another possibility:
the Lilliefors and the K-S statistics would provide bounds on your
p-value. If the p-value was bounded well away from your nominal
significance level, would you particularly need to know it exactly?)

You may want to look at the Anderson-Darling test. In the goodness of fit book
by D'Agostino and Stephens, (I think, or it may have a reference to a paper in
which it can be found) they look at the approximate asymptotic distribution of
the A-D under estimation of 0, 1 and 2 parameters, if I am remembering
correctly (this is going back about 13 years since I looked at it so I may be
misremembering some details). The dependence on the original distribution
is apparently not strong, so the test is approximately distribution-free.

The asymptotics kick in very rapidly (I think they suggest n=3 is sufficient).
They give tables that are based on a function of n and the usual A-D statistic.

Glen


.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
.                  http://jse.stat.ncsu.edu/                    .
=================================================================

Reply via email to