The core idea
Survival analysis studies time until a defined event. In retention work, the event might be a voluntary departure. People still employed at the end of observation are right-censored: their eventual departure time is not yet known. Kaplan–Meier estimation can describe the probability of remaining event-free over time without imposing a particular distribution. The term survival is statistical terminology; it is not a judgement about employees.
Source and attribution [1]Using it in practice
Specify the starting event, outcome, observation end and treatment of other departures. Keep joining and leaving dates accurate. Show the number at risk alongside a curve and interpret the tail cautiously when few people remain observed. Consider whether censoring is plausibly unrelated to future event risk, conditional on the analysis. Different event types may require a competing-risks approach.
An example, not a reported case
Worked example · illustrative
In an invented cohort of ten people, one leaves at month one, giving an estimated retention probability of 9/10. Two others reach the end of their observation at month two and are censored. One of the seven still at risk leaves at month three. The Kaplan–Meier estimate becomes 9/10 × 6/7, approximately 77.1%. Censored people are not counted as having left at their censoring date.
What to watch for
A survival curve does not explain why people leave. Treating every non-voluntary departure as harmless censoring can be inappropriate for the question. Hazard is an instantaneous conditional rate, not the same as a cumulative probability. Formal comparisons and models require attention to assumptions and uncertainty.
Choose the event carefully
All exits and voluntary resignation answer different questions. Define internal transfers, retirement and contract completion explicitly. If another event prevents the event of interest, seek an approach suited to competing outcomes rather than automatically applying a basic curve.
Read the late part cautiously
A curve may look stable simply because few people remain under observation. Report numbers at risk and uncertainty. Avoid comparing a well-observed early period with a sparse late period as if the evidence were equally strong.
Take it into your next conversation
Three useful questions.
- What question can this method answer, and what can it not establish?
- Are the comparison, observation period and assumptions defensible?
- What decision follows, and how will its consequences be reviewed?
Related terms
Go to the evidence
Sources & attribution
[1] NIST: Kaplan–Meier estimation with censored data ↗
The core idea is an original summary of the cited work. Application notes, examples and sketches are our interpretations, not quotations or reproductions of the authors’ figures. Publisher records may require access to read the full original work.
Published 2026-09-20 · Reviewed 2026-09-20. Editorial approach