Interaction effects, distributions not normal- is ANOVA justified??

[jpkelley]

There are always hypotheses. Your hypothesis for a three-way interaction model is: "There is a joint effect of variable1, variable2, and variable3".

[jpkelley]

"There are always hypotheses." Isn't this everyone's mantra?

[knedlica]

since your both so helpfull I have another question. when I'm doing logistic regression and the independant variable is ordinal (duration of therapy ; 1-few weeks, 2-less than 3 months, 4-less than six months, 5-more than six months) do i have to make that one variable into 4 dummy variables like it's done with categorical variables or can i just leave it like this and interpert it's coefficients like i would for any interval variable? i can't seem to find any answer to this online.

[GretaGarbo]

Aha, ok it is not cia work, not a matter of life and death for these patients but rather a

“kind of exercise for work to see if they will hire me O.o ”

OK!
I guess the question is what model should you use. Or what is the problem here? Formulation of the problem. This is very late to ask this fundamental question.

Paste in the printout from the anova table. Then click on “Go Advanced” down to the right. Highlight the inserted anova table. Click on the # button (to create code taggs)
and click on submit reply (down to the right).

In the model there should at least be included the main effects A and B and C.

[GretaGarbo]

do i have to make that one variable into 4 dummy variables like it's done with categorical variables or can i just leave it like this and interpert it's coefficients like i would for any interval variable? i can't seem to find any answer to this online.

You can leave it as it is but declare it as a factor just as you have done here with the anova. Then spss will create the dummy variables for you. Then you run that as a logit or logistic regression (it is the same thing) or what ever it is called in spss. A good model start is to use the same model (with main effects and interactions) as this anova, since it would not be surprising if the two response variables (this anova variable and that 0/1 variable) are correlated.


No you cant interpret the coefficient as you would for a regression variable.

[knedlica]

I ended up using linear regression and it turned up okay. Thanks everybody!