Lorenz Curve and Gini Coefficient for Engagement Concentration

Description

Plots and quantifies how unequally activity, value or exposure is distributed across users, creators or items.

The Lorenz curve plots the cumulative share of a quantity (sessions, revenue, views) against the cumulative share of units ranked from smallest to largest; the Gini coefficient is twice the area between the curve and the line of equality, from 0 (perfect equality) to 1. Concentration measures such as the Herfindahl index complement it. For products, these measures reveal dependence on a few power users, inequality of exposure among creators or sellers, and fairness of recommendation systems. Tracking Gini over time shows whether changes broaden or concentrate engagement, which raw averages hide.

Attributes

Loss type
Unmigrated
Effort
Hours

Steps

  1. Choose the quantity and units (e.g. views per creator, sessions per user, revenue per account).
  2. Sort units by the quantity and compute cumulative shares.
  3. Plot the Lorenz curve and compute the Gini coefficient with bootstrap confidence intervals.
  4. Compute complementary measures (top 1% share, Herfindahl index, Palma ratio).
  5. Compare across segments, time or experiment arms.
  6. Interpret implications for risk (dependence on few users), fairness and design.

Sources

  • Lorenz, Methods of measuring the concentration of wealth (1905), Publications of the American Statistical Association
  • Cowell, Measuring Inequality (3rd ed., 2011)

Also known as

  • Lorenz curve
  • Gini coefficient
  • Concentration analysis
  • Herfindahl index

Categories

  • Analytics & Behaviour
  • Product Analytics & Strategy