Spss Help!!!


New Member


My research is a psych experimental 2 independent between subjects design. My independent variable is (task type- maping or prose). I am looking at whether the task type produces an sig effect in my dependent variables (accuracy) which is measured in percentage and (sorting time) which is measured in minutes. The levenes test for normal distribution is not sig for accuracy- so I assume its normally distributed and doesn't violate assumptions but (sorting time) does seem to violate assumption of normality.

Another variable (education) which I thought would probably have some influence but I didn't control for it- rather it was just randomly distributed depending on the random allocation of particpants

My concern is I originally was going the conduct an independent T-test for group*accuracy and because the violation of normality I was going to use a non-parametric independent T test- both were non sig >.05. However than I found that their was a sig correlation between education and accuracy so I thought that there may be some causal relationship so to assess my first DV (accuracy) I used a univariate GLM and put (accuracy) in Dependent variable box, (task) in the fixed factor box and I put (education) as a random factor box because I didn't actually assign particpants on the basis of this variable.
Results turned out to be significant the effect of task was <.05 and education sig <.001. There was no sig interaction. What I want to know is which test was giving me what I was looking for? Is their really a sig effect for task on accuracy or not? I rather like the results that the GLM gave me! :)

In answering this- should I stay with the non-parametric t-test for soting time since it wasn't normally distributed or if the GLM was correct to make education a random effect should I do something similar with this DV?

If the GLM is appropriate- How do I report the results if I do keep education as a random factor (in the same way as if it were another fixed factor?)

Thanks ?????


i will tell you a general thing ,that basis of design of expriments is normality and if ei was not normal (i.e : ei have not ~NID(0,sigma))then you can use transformations( e.g cox -box transform )then if your data have been not normal again, you should use non parametric test .