Hmmmm... maybe. You might consider some sort of exact logistic
regression. i know that can be done in SAS, I'm not sure about other
programs.
Jeremy
2009/11/19 Dan <[log in to unmask]>:
> Hi Jeremy
>
> Thanks for your reply
>
> I've been trying to find the website where I read that high odds ratios
> might just be because of low cell counts but, as is usually the case when
> you find random things through google, I can't re-find it.
>
>
> By cell counts I just meant that, for some variables, only a small number of
> people meet the condition (say, 15 out of 200). This 15 is distributed
> amongst my 7 dummy groups (of unequal sizes). Thus in some groups on only 2
> or 3 people meet the condition. It seems spurious to get significant
> results based on so few people. the website inferred binary logistic
> regression can produce such spurious results..
>
> Cheers
> Dan
>
--
Jeremy Miles
Psychology Research Methods Wiki: www.researchmethodsinpsychology.com
|