Mean Cumulative Function for Recurrent Events

Description

Estimate the average cumulative number or cost of repeated events per user by age, correctly handling users observed for different lengths of time.

Many UX events recur: support contacts, errors, purchases, password resets. The mean cumulative function is the population average of each unit's running count of events as a function of age, and its slope is the recurrence rate. Averaging raw counts is wrong when units have been observed for different durations. Nelson's estimator orders all event times and, at each, divides the number of events by the number of units still under observation, then accumulates these per-unit increments, giving a step function with pointwise confidence intervals. It needs few assumptions: units need not follow a Poisson process, but the end of observation must not depend on a unit's event history. Two groups are compared through the difference of their MCFs with a confidence band. The same method accumulates cost instead of counts. Curve shape is informative: straight means a steady rate, upward-bending a worsening one.

Attributes

Loss type
Unmigrated
Effort
Days

Steps

  1. Draw an event plot with one horizontal line per unit spanning its observation window and a mark at each event.
  2. List all distinct event ages across units.
  3. At each age, count the events and the number of units still under observation, and divide to get the mean number of events per unit at risk.
  4. Accumulate these increments to obtain the MCF and compute standard errors and pointwise confidence intervals.
  5. For two groups, estimate the difference in MCFs with its confidence band; a band excluding zero over a range of ages indicates a real difference.
  6. Interpret the shape (linear: constant rate; convex: increasing rate; concave: improvement) and note where few units remain, since late estimates are unstable.

Sources

  • Statistical Methods for Reliability Data (Meeker & Escobar, 1998)

Also known as

  • MCF
  • Nelson MCF estimator
  • Mean cumulative number of recurrences
  • Mean cumulative cost
  • Recurrence data analysis
  • Mean cumulative function (MCF) for recurrent events across a population of units, with two-sample comparison

Categories

  • Analytics & Behaviour
  • Statistics & Modelling