Is it possible to use Logistic Model?

I am attempting to investigate 7891 students who recently took an undergraduate entrance exams. I am interested in estimating this econometric model (i.e. Logit/Probit) to investigate factors affecting the result of students who took the exam. The results status has three factors, i.e., unsuccessful, regular pass and provisional pass. I have attempted in coding the three factors as binary. That is, I assigned "1" for both regular and provisional pass as "Passed"; and "0" for unsuccessful as "Failed". Please comment on this coding. Another thing is that, I estimated both probit and logit models taking my y binary variable on average test scores (mathscores+englishscores)/2) plus gender plus other categorical variables, but I get error message as highly collinearity for the average test score. I really need the variable as my key predictor variable, but it is not working. Can I ignore or drop it and use the raw scores for english and math separately in my models? Your views and references are highly needed. Thanks! Joe


No cake for spunky
Does it make substantive sense to combine the two forms of pass into one? That would be my comment on that issue. Its really not a methods decision. I don't understand why you would get the error for one of these and not another tied to collinearity if that is what you are saying. Substantively they are similar models, you can generate the results from one for the most part with the other fairly simply. Are you getting high collinearity or perfect collinearity (not full rank which is a much more serious issue than collinearity).