2

What I learned from the post Can I treat a counting response variable as a continuous variable and run OLS? is if the $\lambda$ is large, It is OK to run linear regression on count variable.

My question this time is: Can we run linear regression and treat survival time as a continuous variable instead of survival analysis? In what cases, it would be OK or wrong?

Community
  • 1
Haitao Du
  • 32,885
  • 17
  • 118
  • 213
  • 1
    Censored cases would make the biggest difference (besides the obvious differences in the form of the hazard function when using a linear model of time). – Andy W Mar 16 '17 at 14:07
  • @AndyW Thanks!, if we do not have any censored cases, is it OK to run linear regression? – Haitao Du Mar 16 '17 at 14:13
  • At that point you are just talking about different functional forms for the hazard - so it is just curve fitting whether you should use OLS or some other model. That will be data specific though. One could probably nitpick in terms of covariates an OLS model can result in negative predictions of time if you don't transform the outcome, but that should only happen if you extrapolate beyond the data. – Andy W Mar 16 '17 at 14:18
  • @hxd1011, I have the same question today. Would you mind writing an answer if you have got one. My question is "If we ignore censored cases, time is a continuous variable and if it can be explained relatively well by other variables in the data, can we/should we use linear regression to model it. If no, then why?" – Gaurav Singhal Jun 28 '18 at 09:43

0 Answers0