Modeling quesion, linear regression or survival analysis

Question

I am analyzing a dataset where the outcome variable is "teenagers' recreational drug use in weeks." This is continuous and highly skewed outcome variable. While the range spans from 2 weeks to 100 weeks, most teenagers reported experimenting with drugs for durations between 2 and 25 weeks, with very few values beyond 25 weeks.

This dataset comprises around 2000 children, each with a single row of data. There are no repeated measures; each child enters the study at a different time.

The exposure of interest is a single event: the abolition of a drug counselor position by the federal government in 2020.

The goal of the analysis is to examine changes in drug use duration before 2020, prior to the abolition of drug counselors, compared to after 2020, following the abolition.

My question is what is a appropriate model to analyze such outcome. Should I apply simple linear model or survival analysis ? I welcome any advice on this topic. Thanks in advance.

Answer 1

If there are no repeated measures, I suppose it means that you are not following the children in time, i.e. weeks of recreational drug use is an endpoint measure. A survival analysis would not be appropriate, because your outcome variable is not the occurrence of an event. It would work if it were time to drug use for example.

I would first check that the time spent in the study does not differ between the 2 conditions (before the suppression of the counselor position and after). Then I would analyse it with a GLM, probably for a binomial distribution since it is skewed count data, but this depends how skewed the distribution of the outcome variable is.