Cohort Analysis

Definition

Cohort analysis groups items or people by a shared starting point, such as the month they were first seen, and follows each group over time. It compares like with like and avoids mixing newcomers and veterans in one overall average.

In plain terms

An overall average mixes ages: this month’s figure combines newly arrived items with old ones. If the mix shifts, the average moves although no group changed (a composition effect; see Panel Data). Cohorts fix this by comparing groups at the same age: how many of last month’s new items are still active one month later, versus those of the month before?

Recent cohorts have not had time to reach later ages, so they can only be compared with older ones up to the age they have reached. Showing only the survivors of a cohort brings back Survivorship Bias.

Example

Listings grouped by the month they first appeared, and the share still active after each month:

At the same age (one month), February retains worse than January: 85% against 90%. A single “share active this month” figure would blend the two cohorts and hide this difference.

Common misunderstandings

• “The overall average shows the trend.” It also moves when the mix of old and new items changes.
• “A cohort is a demographic group.” It is any group defined by a common starting event or period.
• “A recent cohort can be compared with an old one at any age.” Only at ages both have reached.