Marco Roelofs <
[hidden email]> wrote:
> I have dataset that I would normally analyse by means of logistic
> regression to compare the steepness of slopes between various groups, but
> due to a low sample size this is now impossible. Could anyone tell me
> whether there is a nonparametric alternative to this test, and if so,
> what test that is? I was thinking of Fisher's randomization tests but
> unfortunately I cannot find an SPSS script to do the job.
>
Why should small sample size prevent you from doing logistic regression?
Non-parametric tests, agreed with the exception of randomisation tests, are
less efficient that parametric ones, so effectively reducing even further
your sample size.
Of course with very small samples you can't check properly for goodness of
fit.
Ideally you will use "exact logistic regression", a method intended for
small samples. I don't think that it is available in SPSS, but it can be
found in SAS. Cytel's LogXact claims to be even more powerful.
David Hitchin
Locked
