I must apologize if this seems a very basic question. I am fairly new to survival analysis. Could some please enlighten me on how survival data is collected? I am very much aware of some of the techniques used to analyze survival data.
Added after comment
For instance, suppose I want to model the data using a piecewise exponential model, how would I go about collecting the data?
For a basic Survival-analysis (no time-dependent covariates, no competing risks), you only need
The baseline-hazard/baseline survival describes the event-rates over time. That could e.g. be very low in the first year and higher afterwards, which may be modeled reasonably well with a piecewise exponential. Notice that this concerns the outcome (which you know/collected over time) and doesn't need intermediate measurements of covariates: your analysis is condional on baseline-covariates.
Fundamentally, you should collect data in a way that justifies the assumptions of your desired model of the process / allows you to answer questions of your population of interest. For example, data that justifies assumptions of iid, is representative of the population, etc.
Note that survival analysis can deal with censored data, you do not need to exclude observations for which the event(s) of interest have not yet occurred.
You need to have data on how much time has elapsed since treatment application, last activity, or days as a customer before churn, for example. You also need to know if the event under consideration has occurred yet or not (churn, death, cancelled subscription). If you are comparing two groups of subjects, you also need a variable which identifies the groups and can be used to plot different Kaplan-Meier survival curves.
For a churn setting example, I would have days as customer, churn status, and additional predictors such as demographic data and product type.
