Consider a multinomial logit (mlogit) to retain the 3-level outcome. It's unusual for the continuous variable to be the one with collinearity problems. Make sure that the raw scores are not inadvertently also being included in the regression equation.

Thanks for your response, but looking at regular pass provisional pass which could mean the same, don't you think it might cause more harm?

Logit model use for samal sample. Moreover interpertation of its coefficients is not simple. I suggest you probit model as your sample size is large, so the variable follows normality.

And if you like to take third option provisional passed then use multnomial probit model.

Multicollinearity is no matter in categorical models . Best of luck

Hi

For the first part where you re-coded the three dependent variables as binary dependent variable, the binary probit or logit can be used.

Hi Joe,

As pointed earlier, a multinomial logit would be best. You could also do a logit with coding as you describe, but you would loose some information.

Collinearity means that some of your independent variables are highly related between each other. For example, Math scores may be correlated with English scores, which is possible if good students receive good marks for all subjects. It is also possible that seemingly unrelated variables are behaving similarly due to sample characteristics. For example, if all female students achieved high marks because the sample includes a small number of female students.

Before you run the model, I would suggest making a table comparing the frequencies and proportions of your dependent variable (pass/fail) with each of the independent variables.

For example:

. Pass | Fail | Total Sample

Female 3,551 (46%) | 394 (5%) | 3,946 (50%)

The table will help you find the problematic variables.

Good luck!

Thank you Sir.