That raises some fairly tricky issues - I'd send a message to the
impute mailing list if I were you. Details:
http://lists.utsouthwestern.edu/mailman/listinfo/impute
Jeremy
2010/1/9 [log in to unmask] <[log in to unmask]>:
> Dear All,
>
> I want to replace missing data using expectation maximisation (EM) in
> distributions that are skewed. SPSS EM has a function ('Mixed Normal') for
> imputing missing values for skewed distributions, but to use this I need to
> specify the "mixture proportion" and "standard deviation ratio". Does anyone
> know how to calculate the mixture proportion and standard deviation ratio?
>
> My fallback option was to transform the skewed distributions and then use EM
> for normally distributed data. However one of the distributions is very
> positively (left) skewed and about 60% of the scores are zero. I have tried
> weighting the zero scores, but I can't seem to get the distribution anywhere
> near normal despite using various combinations of transformations.
>
> Does anyone have any suggestions, either for calculating the mixture
> proportion and standard deviation ratio, or for normalising the extremely
> skewed distribution?
>
> Thanks for any advice.
>
> Regards,
>
> John.
>
--
Jeremy Miles
Psychology Research Methods Wiki: www.researchmethodsinpsychology.com
|