Hi Caroline,
1. Did you do a test of homogeneity of variance in regression? (This
is called White's test, and it's not available in SPSS automatically,
although it can be calculated - google it, and you should find it.)
2. Homogeneity of variance is not relevant if you have equal sample sizes.
3. You need a different transformation. The problem is not that SPSS
won't do analysis on transformed data, the problem is you're picking
an inappropriate transformation. It's also possible to do robust
analysis in SPSS that does not make the homogeneity of regression
assumption. http://www.jeremymiles.co.uk/regressionbook/2006/04/huber-white-estimates-in-spss.html
.
4. I don't understand why you want to do this anyway. Regression is
anova (and anova is regression). They are the same test, and make the
same assumptions.
Jeremy
2010/1/4 Caroline Wilson <[log in to unmask]>:
> Happy New Year all.
>
> I've been running a regression analysis of a whole sample and now want
> to examine the sample by group.
> I'm using Factorial Anova to assess why the groups differ. Until now the
> sample size was large and I had no issues with homogeneity of variance.
> Now the total sample of 198 is broken down into six groups of roughly
> equal sizes.
> Levene's Test is significant when doing Factorial ANOVA.
> The first advice of Andy Field on this is to go back and transform data.
> However SPSS (16) won't produce a Levene's Test with transformed data
> because it warns of 'non positive values'.
>
> This was understandable when I was using centred data (which I'd needed
> for earlier analysis and continued using without thinking) but not when
> I reverted to non centred data, none of which has any negative values in
> any of the columns.
>
> I've tried each form of transformation offered by SPSS without success.
>
> Has anyone got any advice over what is going on and what I might do
> next?
>
> Many thanks.
>
> Caroline Wilson
> PhD student
> De Montfort University
>
>
>
--
Jeremy Miles
Psychology Research Methods Wiki: www.researchmethodsinpsychology.com
|