Hi Dan
I wouldn't say that those odds ratios were anything to worry about. 5
is fine, it means the odds are 5 times higher in one group than
another group. I recently had odds ratios of 8000 - that was
something to worry about. If that's happening, it's something called
quasi-complete separation, and they still might be significant.
I'm not sure what you mean about cell counts. Can you use frequencies
or crosstabs?
Jeremy
2009/11/10 Dan <[log in to unmask]>:
> Dear all
>
> I am running a binary logistic regression with a set of dummy variables as
> the independent variables. Some of the odds ratios are coming out really
> high - 5.2, 3.4 etc. For my data, this doesn't make sense. I've read that
> when you get high odds ratios like this, they are probably artificially high
> because of low cell counts. In that case, why does the output still report
> them as significant? Also, it gives me no way to know whether a result is
> meaningful since then the significance levels don't mean much, right?
>
> I'm using SPSS by the way. Related to this point, how can I get SPSS to
> give me cell counts? It gives me the proportion who are either 1 or 0 for
> the binary dependent variable but not for the dummy variables.
>
> Thanks for any help
>
> I've attached the output, if that helps. The dummy variables relate to
> social classes.
>
--
Jeremy Miles
Psychology Research Methods Wiki: www.researchmethodsinpsychology.com
|