Binary DV and orthogonal contrasts

#1
PLEASE HELP: Binary Dependent Variable and orthogonal contrasts

HEEEEEEEEELLLLPPPPPP!!!!!
The data I am trying to analyse has a binary dependent variable (1=yes, 0=no) and tow factors. Factor 1 is a treatment variable (treatment 1, 2, 3, 4) and Factor 2 is a manipulation (1=low , 2=high).

I need answers to the focused questions such as Treatment 1 > Treatment 2 or Treatment 1 low > Treatment 4 low, or Treatment 3 low > Treatment 3 high.
To make my life easier I ask SPSS to generate all the pairwise comparisons. Otherwise I should specify the contrasts matrix....I tried but I wasn't able to understand clearly what I was doing so I decided to ask SPSS to generate all possible combination and look at the one I am theoretically interested in.

This is the code I am using:
GENLIN binary_dv (REFERENCE=LAST) BY treatment manipulation (ORDER=DESCENDING)
WITH covariate1
/MODEL treatment manipulation treatment*manipulation INTERCEPT=NO
DISTRIBUTION=BINOMIAL LINK=LOGIT
/EMMEANS TABLES=treatment COMPARE=treatment CONTRAST=PAIRWISE
/EMMEANS TABLES=manipulation COMPARE=manipulation CONTRAST=PAIRWISE
/EMMEANS TABLES=treatment*manipulation COMPARE=manipulation CONTRAST=PAIRWISE
/EMMEANS TABLES=treatment*manipulation COMPARE=treatment CONTRAST=PAIRWISE
/PRINT CPS DESCRIPTIVES MODELINFO FIT SUMMARY SOLUTION (EXPONENTIATED).


Comments on Syntax:
/EMMEANS TABLES=treatment COMPARE=treatment CONTRAST=PAIRWISE
This returns all the combination of treatments

/EMMEANS TABLES=manipulation COMPARE=manipulation CONTRAST=PAIRWISE
This returns the difference between high vs. low manipulation

/EMMEANS TABLES=treatment*manipulation COMPARE=manipulation CONTRAST=PAIRWISE
/EMMEANS TABLES=treatment*manipulation COMPARE=treatment CONTRAST=PAIRWISE
This are all possible interactions pairings


The results make sense to me but I am skeptical of the procedure. I wonder if somebody could tell me whether there is an alternative way to double check my results. Basically, how else would you do a planned contrasts analysis when your dependent variable is binary?

Thanks,
Matteo
 
Last edited: