Stochastic transitivity test

Description

Build personal option sets around each person's threshold for negligible differences, then run replicated paired choices to test whether preferences cycle.

Tversky built options in which one attribute improved slightly as another worsened, such as gambles where the chance of winning rose as the payoff fell, shown as pie slices. People who ignored small probability differences chose by payoff on adjacent pairs but by probability on the extreme pair, producing cycles. After a screening session, eight participants took five weekly sessions with fillers and real play; weak stochastic transitivity was rejected for five. A second study used applicant profiles, estimated each person's threshold and built personal sets; intransitive triples occurred at about the rate a lexicographic semiorder predicts. Nobody noticed their own cycles. The test is rigorous because it is per person, replicated and compared against a random-responding baseline.

Attributes

Loss type
Unmigrated
Effort
Weeks

Steps

  1. Choose two attributes with a trade-off and design options in small steps along both.
  2. Run a screening session with adjacent and extreme pairs plus filler pairs to identify people who use a lexicographic rule.
  3. Estimate each person's threshold for treating a difference as negligible.
  4. Build individual option sets around that threshold.
  5. Collect replicated paired choices over several sessions (the original used five weekly sessions).
  6. Test weak stochastic transitivity per person by likelihood ratio, comparing with the rate expected from random responding.

Sources

  • Preference, Belief, and Similarity (Tversky, 2004)
  • Tversky, A. (1969). Intransitivity of preferences. Psychological Review, 76
  • Foundations of Measurement, Vol. 2: Geometrical, Threshold, and Probabilistic Representations (Suppes et al., 1989)

Also known as

  • Stochastic transitivity test for multi-attribute choices
  • Intransitivity test
  • Lexicographic semiorder test

Categories

  • Analytics & Behaviour
  • Behavioral Science & Economics