I found out that my research data do not meet the requirements of canonical and discriminant analysis (the number of observations is very small while the number of independant variables is big and also there is no variable normality despite data transformations). I need to analyse the influance of a group of many independant variables on many dependant variables. I also plan to detect the independant variables which determine given classification (discriminant analysis would work well for this if meeting the assumptions). I have problems with finding a non-parametric method for this or overcoming the assumption problem.
I would much appreciate any suggestions.
Hi! If you don't have too many variables and they are categorical, you could try loglinear analysis. In SPSS you find it in Analyze/Loglinear/Model Selection. You can use maximum 10 variables per analysis (though you need a very good computer, otherwise it may take days...) It doesn't treat the variables as dependent and independent, but it does give you the flexibility interpret them as such.
Also, have you tried to use a data reduction method (Principal components analysis for example)?
- Badges
- Science topic
- Similar topics
- Medicine
- Physiology
- Biosignals
- Biomedical Signal Processing
- Statistical Analysis