First-hand information for everyone
This paper looks at how what workers believe about salaries changes the tradeoffs they make between money and on‑the‑job perks such as the option to work from home or having a permanent contract. The authors ran a multi‑stage, incentivized survey experiment with about 1,000 students at the University of Oslo. They asked participants to choose between jobs in two ways: hypothetical bundles of pay and amenities, and real job advertisements drawn from the local labor market. For some real ads the survey showed the advertised pay and for others it did not. The researchers then compared the participants’ salary beliefs with actual starting salaries from matched administrative data.
The survey used a dynamic design called a Bayesian Adaptive Choice Experiment to learn each person’s tradeoffs efficiently. Respondents saw a total of tens of thousands of job choices: 38,920 hypothetical job alternatives, 19,460 real job alternatives paired with belief questions, and 19,460 real job alternatives where pay was experimentally shown. The design let the authors measure stated preferences for amenities, elicit individual pay beliefs for specific ads, and test how exposure to explicit pay information on one set of ads affected beliefs about other, related ads.
From the hypothetical choices the study recovers sizable willingness to pay — that is, the share of pay people say they would give up for better amenities. On average respondents were willing to give up 23.3% of pay for a permanent contract, 15.1% for the option to work from home, and 8% for flexible hours. They required extra pay of 17.1% to accept shift work and 14.7% for each additional hour of daily commuting. Yet beliefs about actual salaries were biased. On average respondents under‑predicted starting pay by about 18%. They also thought higher‑amenity jobs paid more than they actually do. For example, respondents expected permanent jobs to pay about 8.5% more than temporary ones, while the administrative data show the true gap is only about 2.2%.
Showing pay information did move beliefs, but only modestly. Exposure to explicit pay on a related set of ads raised mean pay beliefs by about 4% and reduced the dispersion of beliefs by about 15%. In other words, the intervention closed roughly one‑fifth of the average under‑prediction and made beliefs more concentrated. Crucially, however, the intervention did not break the strong positive link between perceived pay and advertised amenities. After seeing pay information, people still inferred that better‑amenity jobs were better paid, and the implied tradeoffs in the stated choice tasks did not change significantly.
These results matter because they help explain a persistent puzzle in labor economics. Experiments and surveys often find that workers value amenities a lot. But observational studies using real wages often find little evidence that bad amenities are compensated by higher pay. The paper shows that biased pay beliefs and the tendency to use amenities as a signal of pay can make workers’ perceived tradeoffs differ from the tradeoffs that would hold under full information. Important caveats are that the experiment used students at one Norwegian university, so findings may not generalize to all workers or labor markets. The authors also note that eliciting beliefs is methodologically challenging, and that the pay information in the experiment was shown on related but not identical ads, so the way information spreads in real job searches could differ.