I am currently learning Survival Analysis for a project and I struggle a bit with the notion of Time-Dependent Covariates.

Let's take the following example:

I would like to know what is the survival probability that a customer would valid his basket t days after visiting the website.

One way to modelize this would be to use a Cox Proportional Hazards Model which could take into account some variables available at the time of his visit (e.g. Number of products bought before, number of products cancelled before, age, weekday of visit.).

Yet, I can add other sources of information available between the moment he visited the website and the moment he effectively validated his basket, such has the number of times he came back on the website to browse, the number of emails sent to remind him that he has a current basket on the website, the fact that he actually opened the email, the number of other products he bought and their categories, ...

How can I transform the dataset in order to take into account such variables ?

Thanks !


1 Answer 1


From what I've seen so far, one way to do it is to use a start-stop format. For example :

enter image description here

Here, the third ID bought 3 other products three days after visiting the website. So a new line is created whith the variable updated. This line is considered left truncated.

Yet, let's say I have tens of thousands of customers. If I do this for every one of them, i would have one line per customer per day. Is it an issue ?



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.