My nemesis: multinomial logistic regression

Dear all,

For the final article of my PhD-thesis, I need to analyze various categorical results.
My goal is to find whether a certain genotype will give a predisposition to no, few or many side-effects after taking medication.
To ascertain this, I used multinomial logistic regression in SPSS, with the number of side-effects as output/dependent variable. The genotype I set up as a scale variable, as we work with an allele dosage model. Because of this, I entered the genotype as a covariate rather than a factor, as I was taught that categorical data belongs in the "factor box".
However, when running the analysis, I get no significance for my model fit. How can I tackle this problem? Is there a better way to perform this analysis than multinomial?
I already looked at Pearson's Chi Square, but I have more than 2 groups in both variables.

If anyone could help me, that person would be a real life-saver!
Thanks very much in advance!


Fortran must die
I think this probably belongs in the SPSS forum as you are asking a question about how SPSS analyzes data, but I am not entirely certain what your question is. You might well have poor fit because your model does not fit the data well. Changing the variables, changing how they are structured (for example collapsing or expanding categories of a given variable) and so on can be used to address this. At least for binary logistic regression, I am not sure about multinomial, there are critics of goodness of fit tests as they produce uncertain or possibly wrong results.

You need to provide more details of what exactly your are doing and what your problem is (is this a software issues, a problem with the method, a problem with your findings, etc).