2
$\begingroup$

I have an employee churn problem where I have data for every three months of employees in a company ranging from 2015-2019. Does it make sense to model this problem as a time-to-failure/survival analysis problem?

I've seen some studies where survival analysis/time-to-failure has been applied to customer churn, however I am wondering if it also applies to employee churn.

Furthermore:

  • An assumption in time-to-failure is that the observations have a degrading aspect with relation to time, which I don't think it is quite applicable to churn. I.e. a customer/employee who has stayed for 20 years with the company is less likely to leave it, not more likely as time-to-failure analysis would expect it to. Am interpreting this wrong?

  • If these types of analysis fit churn prediction, should only the churning employees/customers be considered in the training set, or should all employees/customers be considered?

Thank you!

Improve this question
$\endgroup$

1 Answer 1

Reset to default
5
$\begingroup$

Yes it makes sense, and all employees should be considered (up to the time they leave)

I think you are slightly confusing the time to failure analysis: Probability of failing in each time period is independent of what happened before. However, a basic model with no time dependence fitted to eg up to 5 year employees will sooner or later churn, so your 20 year employee is unlikely to occur... But there is no reason that you can't add eg time dependence in your model ( eg probability of churning in each 3 month period decays linearly with time period), and you can add arbitrary nonlinear functions of time too.

Treat as discrete time survival analysis Use a probabilistic classifier such as logistic regression or xgboost to predict churn in each 3 month period for each employee (add all the variables you want, time or employee.. ) Then you have 2 small issues.

A) generate person period data set B) generate survival time profile by chaining survival probability of each period. IE prob survival in 9 months = product of probability of survival of each 3 month period.

https://stats.idre.ucla.edu/r/faq/how-can-i-convert-from-person-level-to-person-period/

Improve this answer
$\endgroup$
3
  • $\begingroup$ Thanks for the explanation! In the link you provided it's mentioned person-period and person-level approaches, is person-period always a preferred approach to survival analysis since it may provide additional information about each observations "time-series"? $\endgroup$
    – lsfischer
    Jan 6, 2020 at 10:18
  • $\begingroup$ are survival models designed for making predictions on the individual level? or are they only meant for groups? E..g churn model for all men with biology degrees or churn models for john smith who has a degree in biology? $\endgroup$
    – stats_noob
    Aug 7, 2021 at 20:45
  • $\begingroup$ @stats555 this is a general statistical modelling question.How do I predict your eg buying behaviour? I come up with a model that predicts buying behaviour of people like you. How detailed you make the model is up to you... Put another way, how would I build statistics of employee churn on an individual level? $\endgroup$
    – seanv507
    Aug 8, 2021 at 9:41

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.