0

I have a dataset with time-to-event data (so this is a continuous outcome). The data, however, is such that I do know that when the event did not occur - then it did not happen.

What could be an appropriate model approach for this scenario? I was thinking about a hurdle model, but are there other approaches?

Thanks!

clog14
  • 221
  • Survival model? What models / frameworks have you looked into so far? I m a bit confused on your stress on "however" : isn't that the default behavior when you don't observe it then it didn't happen? Did you mean to say " however when the event is not observed it could have happened [with some probability]"? Or are you referring to no censorship problem? – Georg M. Goerg Nov 02 '22 at 09:59
  • hm... maybe i am totally confused here. But is it not the case that in survival models for those that did not die you actually are not sure what happened to them whereas in my scenario if you do not observe the event you know it did not happen. I guess I have thinking into the direction of hurdle models - model whether the event occurs or not with a binary and then if it occurs i guess how long it took. – clog14 Nov 02 '22 at 16:03
  • Oh I see , yes that makes sense . This sounds like a customer lifetime value (clv) might be useful. See https://towardsdatascience.com/the-paper-a-deep-probabilistic-model-for-customer-lifetime-value-prediction-eb5d61a83ecd ( with link to GitHub). This is a hurdle model with log normal if positive. I think you can repurpose this for your use case. Alternatively you seem to have a survival model with 0 censorship ( which would reduce to simply predicting distributions -> that's where this CLV paper comes in) – Georg M. Goerg Nov 03 '22 at 01:54
  • Btw what is your application and what business / domain problem are you trying to solve? – Georg M. Goerg Nov 03 '22 at 01:54
  • hey thanks - interesting article/paper. regarding the application: it is more of a time-to-checkout within a session problem (so i know that the event did not happen with the session). I am also more interested from a causal perspective than from a predictive perspective (i.e. I care about certain parameter estimates). – clog14 Nov 03 '22 at 07:31

0 Answers0