Description
A non-parametric estimate of the proportion of users who have not yet adopted (or churned) over time, correctly accounting for users still under observation.
The Kaplan-Meier estimator computes, at each event time, the conditional probability of surviving past it given survival so far, multiplying these to give a step curve. Plotted as one minus survival it becomes a cumulative adoption curve. Unlike naive cohort percentages, it uses censored users (those who joined recently or left the observation window) correctly, so adoption is not underestimated for recent cohorts. Curves for different groups can be compared with the log-rank test, and median time-to-adoption read directly from the curve. Rigour depends on a sensible time zero, non-informative censoring and showing numbers at risk under the plot.
Attributes
- Discipline
- Statistics & Modelling
- Loss type
- Unmigrated
- Stage
- Post-launch
- Data
- Quantitative
- Effort
- Days
Steps
- Define time zero and the adoption event for every user and record follow-up time plus an event or censoring indicator.
- Compute the Kaplan-Meier estimate overall and for each comparison group (e.g. exposed versus not exposed to a new entry point).
- Plot cumulative adoption (1 minus survival) with 95% confidence bands and a numbers-at-risk table at regular intervals.
- Read median time-to-adoption and adoption proportion at fixed horizons such as 7, 30 and 90 days.
- Compare groups with the log-rank test and, if covariate adjustment is needed, move on to a Cox model.
- Check that censoring is not informative, for example that users are not censored because they uninstalled.
Sources
- Causal Inference: What If (Miguel Hernán & James Robins)
- Forward-Thinking Playbook (Vivek Gopinath)
- Behavioral Drift Playbook (Vivek Gopinath)
- Kaplan & Meier (1958) Nonparametric estimation from incomplete observations
- Causal Inference (Hernán & Robins)
- Forward-Thinking Playbook (Gopinath)
- Behavioral Drift Playbook (Gopinath)
- The Inquiry Series, Ch 44 — Survival Analysis, Longitudinal Design & Behavioral Drift
- Practical Statistics for Data Scientists (Bruce & Bruce)
Also known as
- Kaplan-Meier estimator
- Product-limit estimator
- Survival curves
- Logrank test
- Generalised Wilcoxon test
Categories
- Analytics & Behaviour
- Statistics & Modelling