I’m embarking on a project to understand why students drop out of college. I think a survival analysis makes sense to use. But I’ve never done one. Before asking my institution for data I want to be sure I’m asking for it in the right form. So my questions are these: (1) can I use a mix of nominal (sex) and continuous (GPA) variables as predictors and (2) if I have these for each semester for ten years would I be able to predict the withdrawal (event) for each semester or should I do it for one specific time period at the end?
Hello Dear Dale J. Dwyer
It' will be a very intrested study and ,personnaly, I wish to join you if you do not mind .In any case, and as answers to your questions dear Professor,
For Survival analysis (or duration models, or time to event data...) the variable of interest is : time to event ( survival time), in your project: is definded as : the difference between begining of semester and time of drop out ( or date of end of semester) ;
And the event of interest it depends on the studied phenomenon ( death, remission, sucess, faillure,...) in your case is " drop out of college.
1- Yes you can use a mix of variables
2- The second question is mainly depend on heterogenity of sample over year; because as you know dear Professor, the statistical unit here is " Student". So, I see for this question, you have a lot of statistical techniques that you can deal with: ( as an example: Time series Analysis for % of drop out over ten year for each semester..., ANOVA, Manova, Logistique Regression,...).
I indicated for statistical unit , beacause for one student you can find multiple events ( so a multivariate survival analysis is required)
3- At the end you can make a comparaison of factors effects betwen 1th semester and 2th semestre, and also you can check if there is a dynmic change of theses effects over time (10 years).
Best Wishes
I've attached a weird example but I hope you get the idea. Best, D. Booth
Unfortunately, the article didn't have any measures described. So, no, I really don't get the idea.
Say what? The data are ALWAYS times and the associated censoring. Survival analysis is about time to an event. There are good references in the paper. I think you should start by looking up survival analysis. Perhaps that is not what you want. David Booth BTW every university has someone(usually in Institutional research). that tracks what you asked about. Socrates probably had an IR guy tracking data like that. That is not what survival analysis is about. You will notice my example has covariates. That is what you were talking about. if i knew the best covariates to predict student success I would publish it myself. A possible research topic is to find the covariates that predict student success and under what conditions.. If you want to do that then perhaps survival analysis makes sense. certain covariates lengthen time in college others shorten it .My paper was about what variables influence time to say getting a patent. That's why i sent you that paper. Nobody does other people's research and gives it to them.