lillietest vs. jbtest

[tk]

I'm looking for some input on which of these two tests is better to
use for a test of normalcy. Matlab recommends using lillietest
instead of jbtest for small sample, but how small is small?? My
sample size will be around 40. Any advice would be appreciated.

Thanks

[Peter Perkins]

This does not directly answer your question, but you can pretty easily determine
the test properties yourself, using Monte-Carlo simulation.

If you want to compare the actual significance level of the tests, generate a
dataset of some desired size from a normal distribution, then perform both tests
at a specific significance level, and save the two results. Repeat this a bunch
of times, and see whether either or both tests reject the null hypothesis
(incorrectly) the appropriate percentage of times.

If you want to compare power, generate a dataset from whatever alternative
hypothesis you have in mind, then perform both tests, and save the two results.
Repeat this a bunch of times, and see which test rejects the null more often.

Hope this helps.

- Peter Perkins
The MathWorks,Inc.

[Marcelo]

Good Afternoon,

Can I use the lillietest to observed data series which size is greater than 1000? I have a sample which size is 9000. Can I change something in the lillietest code to use my entire series or is impossible to use a series like that?
Thank you

[Tom Lane]

>> x = randn(9000,1);
>> [h,p] = lillietest(x)
h =
0
p =
0.1512

Did this not work for you? Can you explain what went wrong?

-- Tom