Willingness to Pay and Welfare Measures from Choice Models

Description

Convert choice-model coefficients into money: attribute willingness to pay as a ratio to the price coefficient, its distribution under random coefficients, and welfare change from the log-sum.

Once a conjoint or discrete choice model is fitted with price as a numeric predictor, coefficients can be expressed in currency. The average willingness to pay for an attribute level is its coefficient divided by minus the price coefficient, scaled to the price unit; with time and cost coefficients the same ratio gives the value of time. With random coefficients WTP becomes a distribution: a ratio of lognormal coefficients is itself lognormal, so median and mean can differ greatly. For whole choice sets, expected consumer surplus under a logit model with income-linear utility is the log of the logit denominator (the log-sum) divided by the marginal utility of income; computing it before and after a change in alternatives or attributes gives the welfare change. Chapman and Feit warn that 'willingness to pay' is a misnomer: it is the price change at which the average respondent is indifferent, not an amount customers would actually pay, so share simulations often communicate better.

Attributes

Loss type
Siloed
Effort
Weeks

Steps

  1. Fit a choice model with price as a single numeric coefficient, checking with a likelihood ratio test that this fits as well as price coded as a factor.
  2. For each attribute level, divide its coefficient by minus the price coefficient and multiply by the price unit to get its money equivalent relative to the base level.
  3. If coefficients are random (mixed logit), simulate the ratio across the coefficient distributions and report median, mean and spread of WTP rather than a single value.
  4. For a change to the choice set or its attributes, compute the log-sum divided by the marginal utility of income before and after, and take the difference as the change in expected consumer surplus.
  5. Report results as indifference prices from stated choices, alongside market-share simulations, with uncertainty intervals.

Sources

  • Discrete Choice Methods with Simulation (Train, 2009)
  • R for Marketing Research and Analytics (Chapman & Feit, 2019)

Also known as

  • Willingness to pay from a choice model
  • WTP from conjoint part-worths
  • Price-equivalent of an attribute level
  • Coefficient-ratio WTP
  • Value of time
  • WTP distributions from random coefficients
  • Consumer surplus (log-sum)
  • Log-sum welfare change

Categories

  • Analytics & Behaviour
  • Behavioral Science & Economics