Loss aversion measurement

Description

Estimates how much more heavily users weigh losses than equivalent gains in a product context, yielding a loss aversion coefficient to guide framing and change design.

Prospect theory proposes that outcomes are evaluated as gains or losses from a reference point, with losses weighted more heavily; Tversky and Kahneman estimated the coefficient lambda at about 2.25. Later work shows lambda varies by domain, stakes and person, and some studies find much smaller effects for routine amounts. Loss aversion measurement estimates lambda for your users and domain, typically through mixed gambles (accept or reject a 50/50 chance to win X or lose Y) or through paired gain and loss framings of product changes (losing a feature versus gaining it). The coefficient tells you how much benefit a change must deliver to offset what users lose, which is central to migrations, price changes and feature removals. Rigour requires incentive-compatible or realistic stakes and enough trials per person for stable estimates.

Attributes

Loss type
Unmigrated
Stage
Strategy, Design
Effort
Weeks

Steps

  1. Choose the domain of interest (money, time, features, status) and define the reference point users hold, for example their current plan or their current workflow.
  2. Build a mixed-gamble task with 20 to 50 trials varying gain and loss amounts (for example gains of 10 to 40 against losses of 5 to 20 units), or a set of paired product scenarios framed as equivalent gains and losses.
  3. Run the task with at least 50 to 100 participants from the target population, using real or realistic stakes where ethically feasible.
  4. Estimate lambda per participant by fitting the acceptance threshold (the gain-to-loss ratio at which acceptance reaches 50%) or with a logistic choice model, and summarise the median and spread.
  5. Translate lambda into design thresholds: a change that removes value worth L must provide perceived gains of roughly lambda times L to be accepted.
  6. Validate predictions against real behaviour, such as uptake of a migration or churn after a feature change, and update estimates.

Sources

  • Prospect Theory works (Kahneman/Tversky)
  • Nash Equilibrium in UX (Gopinath)
  • The Inquiry Series, Ch 23 — Reference Points, Prospect Theory & the Mechanics of Satisfaction

Also known as

  • Lambda estimation
  • Mixed-gamble loss aversion test
  • Loss aversion coefficient elicitation

Categories

  • Analytics & Behaviour
  • Behavioral Science & Economics