survival analysis of data with three possible outcomes

[#1]

Hello,

I'm analyzing retention in a STEM educational program, with two treatment groups (experimental vs case control). For these students, there are three possible outcomes: persisting but not yet graduated, graduated, and dropped out (or changed major). I understand that 'dropped out' would normally be censored but really this is as interesting an outcome as graduated or persisted.

Is there a strategy to analyze survival time to graduate versus drop out simultaneously? Or do I have to run a separate analysis of likelihood of persisting, and then only include retained students in the survival analysis?

Thank you

LE Higgins

Version of JMP: Pro 14 on a Mac.

[#2]

Maybe look at this way. You have three labels: "student", "graduated", "dropout". Do not treat "dropout" as censored, but a sure event. You have two events here: "graduate" and "dropout". It is "student" that is censored. Because you don't know whether a "student" will graduate or dropout, and when a "student" will graduate or will dropout. Following such a direction, the problem fits into "Competing Cause" problems. The competing causes are: "graduate" and "dropout", which ever happens first. From there, there may be different approaches to model, separately or simultaneously, depending on assumptions and objectives.