6
$\begingroup$

What is the connection (if any) and difference between logistic regression and survival analysis?

If we build two models one using logistic regression to predict if the instances will survive or not. Another is using survival analysis. Can we compare two model's coefficients?

Improve this question
$\endgroup$
1
  • 1
    $\begingroup$ In discrete survival analysis, you can use logistic regression to predict survival in one period given survived up to previous period. Then survival iupnto time n is product of survival at time i for I≤n $\endgroup$
    – seanv507
    Commented Mar 16, 2017 at 0:05

2 Answers 2

Reset to default
9
$\begingroup$

They are different categories of things - survival analysis is the analysis of data where time to a given event is the dependent variable. The given event may include death, failure of a machine, a criminal's time to (re)offending or becoming ill, for example. It uses a number of techniques to analyse this data, including certain generalised linear models, including sometimes for the purpose of analysing whether specific variables influence the probability of an event, logistic regression. Overall, survival analysis encompasses many techniques and methods to achieve different subordinate objectives, including tools for exploratory data analysis, distribution fitting and methods for designing experiments. Hence, I don't think you can meaningfully say 'build a model using survival analysis' rather than, perhaps 'build a model using Weibull regression/ Cox regression', as examples of tools closely associated with survival analysis.

Logistic regression, on the other, is a regression technique for analysing binary data. Hence, it is a single tool (but still a very powerful tool useful in many contexts) rather than an overall category of analysis. It could possibly be described as one of the central tools of categorical data analysis, where categorical data analysis belongs to the same category as survival analysis.

Improve this answer
$\endgroup$
1
  • $\begingroup$ Disagree they are fundamentally different things. One thing you get from survival analysis is the hazard rate, and from the hazard rate you can get the cumulative hazard, and from the cumulative hazard you can get the cumulative incidence function, and every point along that function is basically what you would get out of a logistic regression that estimates the chance the event will happen within that amount of time. $\endgroup$
    – Brash Equilibrium
    Commented Sep 12, 2022 at 6:26
5
$\begingroup$

They have different dependent variables (for logistic 1/0 and for survival time to event).

So the short answer is no - you cannot compare the coefficients.

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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