Economic preference measurement is the practice of quantifying a person's risk taking, patience, trust, altruism and reciprocity with standardized questions or incentivized choices. Falk, Becker, Dohmen, Huffman and Sunde (2023) validated a two-item survey module against incentivized experiments with 409 University of Bonn students in winter 2010/2011. Predicted values from the two survey items correlated 0.5861 with experimental time discounting and 0.4079 with experimental risk taking. The paper, "The Preference Survey Module: A Validated Instrument for Measuring Risk, Time, and Social Preferences," pairs two items with each of six preferences.
What the Study Found
For each preference, the authors tested roughly 30 candidate survey measures drawn from a battery of 199 survey items. In an independent pre-test sample of 80 subjects, mean squared prediction error was minimized by the two-item model for all four preference dimensions tested. Out of sample, predicted and actual choices correlated 0.2919 for risk preferences, 0.5868 for time discounting, 0.2629 for trust, and 0.4424 for positive reciprocity. The general risk question alone correlated 0.24 with a single lottery experiment, against 0.25 in Dohmen et al.'s (2011) representative sample of German adults. Replacing the multiple price lists with five-choice staircase questions cut explanatory power (R²) by only 0.02 for risk taking and 0.04 for time discounting.
"Compared to this benchmark, our survey module achieves high explanatory power."
Falk, Becker, Dohmen, Huffman and Sunde (2016), The Preference Survey Module: A Validated Instrument for Measuring Risk, Time, and Social Preferences, IZA DP No. 9674, p. 16.
Methodology
The dataset is a two-session validation study at the University of Bonn's Laboratory for Experimental Economics, with sessions held one week apart. It covered 409 students who had never joined an experiment, plus an 80-student pre-test and a 44-subject test-retest sample. The main sessions ran in winter 2010/2011. Survey and experiment for the same preference never shared a session, and the order of experiments and surveys was reversed for about half of subjects.
Key Statistics
| Metric | Finding | Context |
|---|---|---|
| Survey items in the battery | 199 | Roughly 30 candidate measures evaluated per preference |
| Average subject earnings | 64 Euros (approximately 83 US-dollars) | Both sessions, including a fixed fee of 10 Euros |
| Multiple switchers, lottery tasks | 86 individuals; 36 in both experiments | Risk-taking multiple price lists |
| Multiple switchers, discounting tasks | Around 7 percent of subjects | Time-discounting multiple price lists |
| Model favored by BIC | Two items for altruism, negative reciprocity and trust; three items for risk taking, time discounting and positive reciprocity | In-sample selection, before out-of-sample evaluation |
| Out-of-sample calibration | Intercept of zero and slope of one not rejected at the 10 percent level for all preferences except trust | 80-subject pre-test panel |
| Streamlined positive reciprocity | Adjusted R² of 0.19 | Experimental subject pool |
| International pilot | 22 countries, 10 to 15 people each, more than 220 interviews | Conducted with Gallup Europe |
| Survey module vs experiment, correlation | 0.4079 risk taking; 0.5861 time discounting; 0.6748 trust; 0.4235 altruism; 0.5771 positive reciprocity; 0.3729 negative reciprocity | Predicted value from the two items, 409-subject main sample |
| Test-retest benchmark (R²) | 0.3469 risk taking; 0.6715 discounting; 0.5986 trust; 0.4203 altruism; 0.4336 positive reciprocity; 0.4446 negative reciprocity | 44 subjects repeating identical experiments one week apart |
Why This Matters
Brokers and advisers who score client risk tolerance by questionnaire can borrow items already checked against choices made under real incentives. The module offers a validated, low-cost alternative: one self-rated risk item paired with one hypothetical lottery item. Researchers linking patience to trading or saving behavior can attach the same two time-preference items to account-level surveys. The authors also argue that single-item measures understate true relationships because of measurement error, which favors using both items.