[R] what is the difference between survival analysis and logistic regression with a timing variable?

[Stephen Weigand]

On 3/27/07, zhongmiao wang <zhongmiao at gmail.com> wrote:
> Hello:
>
> If the question is how likely an event will occur at a give time point, can
> we use logistic regression with time t as a predictor variable? For example,
> if the data is
> ID   Gender  Tenure     Churn
> 1       M        17             0
> 2       M        3               1
> 3       M        6               0
> 4       F         10             1
> 5       F         9               0
> 6       F         20             1
>
> We want to predict the likelihood that an insurance policy holder will churn
> at a given tenure, can we build the model as:
>
> logit (churn=1)=b0+b1*Gender+b2*tenure?
>
> or we have to use survival analysis for discrete time? Thank you.
>
> Best Regards
> Zhongmiao Wang
> Senior Analyst
> RMG Connect
>
>

I'd guess you've got censored data so survival analysis
is more appropriate.

If you've followed a customer for only four months
and she hasn't churned (switched companies?) yet,
you only know her time to churning is > 4 months,
but not exactly how long it is. So you need survival
(time to event) methods that account for this partial
information.

Hope this helps,

Stephen
Rochester, MN