logistic regression or survival analysis?


I have two main questions that I hope you help me answer. Let me give you the scenario first. Let’s say that some students in a school are assigned to 3 different programs after being involved in a fight. The aim of the program is to prevent students from fighting with each other. My goal is to evaluate the effectiveness of the programs and whether there is a difference between them. My dependent variable is where or not students are involved in a fight (yes/no) after being in one of the special programs. The problem is that time after the program varies across students. Some students stay in the school after the special program for 5 days while other for 100 days. Of course if a student only stays 2 days after the program, he/she will be less like to engage in a bad behavior -because there are fewer opportunities- than a student that have 100 days.

1.****** Can I just simply include a postdays (number of days after being in the special program) co-variable as a control? Is this a valid approach?

2.****** Or should I use survival analysis? Is a survival analysis model appropriate here? Approximately half of the student were involved in the event. I have never worked with survival analysis models but it seems appropriate.

I would appreciate any ideas.

Thank you so much,



TS Contributor
both seem to be workable. With the logistic regression and post-days as an independent variable the risk is that length of stay might be the only variable with a significant impact.
The survival analysis, where you can consider shorter stays as censored data is imho better adapted to your case.


Less is more. Stay pure. Stay poor.
Student group assignment randomized?

Any hypothesized relationship between time and group?

Can students get in multiple post-intervention fights?
Thank you so much to both of you.
@ rogojel dayspost is a significant predictor indeed but also type of program. Students in Program1 are more likely to be involved in fights. So it is valid to just include the post days?
1. Students were not randomly assigned to the groups. We didn't design the study. I just work with data we already have- unfortunately.
2. No, I don't think so.
3. Yes, some students have more than 1 fight after being in the special program. However, since most of them had only 1 fight, I want to keep it simple and use a binary outcome.

I guess it depends on what questions/hypothesis I want to answer. 1. What would be the hypothesis if I use logistic regression instead of survival? If you have any other ideas I would be happy to hear them.

On the other hand, if I just want to assess the effectiveness of the program. 2. What is the appropriate test to see if the number of fights decrease from pre to post and whether there is a difference among groups. I guess it has to be a count data test since we are dealing with counts of fights. I researched this int he past but I couldn't fight a concrete answer- due to the nature of the DV (count data). Perhaps A negative binomial/Poisson model?

Thank you in advance for any ideas.


Less is more. Stay pure. Stay poor.
Yes, not knowing your data I would imagine that you could use zero-inflated Poisson or Recurrent event survival analysis.

Without randomization, you have to control for baseline difference. One major threat could be confounding by indication, or the worst kids were funneled into the assumed intervention with the great effect.