Why workers may respond less to pay changes they do not notice

Two ways to measure a response to pay

Economists describe workers’ responses to wages using labor supply elasticities: the percentage change in work associated with a 1% change in pay. An elasticity of one means that work changes by roughly the same percentage as the wage.

Mattia Adamo and Michele Cantarella distinguish responses to expected pay from responses to departures from that expectation. A higher expected wage can make working more attractive, but the prospect of greater income can also make additional leisure affordable. A temporary, unexpected increase instead makes working now more attractive relative to working later—provided workers recognize it as temporary.

Their study asks whether these responses look different inside an experiment and during ordinary platform work, even for the same workers.

Following workers across both settings

The researchers recruited 1,632 participants in Great Britain through Prolific, where people earn money by completing research studies. Of these, 1,561 completed the experimental task, and 1,377 completed it alongside a validation step checking their recalled activity against platform records. Individual analyses used smaller samples depending on available data and eligibility.

Participants reported their earnings, hours and completed tasks during their most recent working week, along with what they had expected. They also consulted Prolific records for logged earnings and task counts. Differences between recalled and recorded values supplied a measure of recall accuracy.

Working hours were not independently logged. To address errors in reported hours and calculated wages, the researchers used statistical instruments based partly on smartphone-recorded social media use and platform task statistics. Social media screen time was an indirect proxy, not a direct measurement of work. These corrections depend on assumptions about how the instruments relate to work and reporting errors.

The experiment involved three sequences of emoji grids. Participants selected matching symbols, earning a fixed reward for each successfully completed grid. Grids grew progressively larger, so earning the next reward required more time. Workers could leave a sequence whenever they chose.

Each person received a target reward based on their stated five-minute earnings expectation, plus higher and lower rewards in randomized order. The deviations were randomly set at 5%, 10%, 15% or 20%. Randomly displayed pay benchmarks also helped the researchers isolate variation in stated expectations. Live displays showed rewards, earnings and progress, making experimental pay readily visible.

A substantial gap—and a clue

In the experiment, the estimated elasticity of time spent working to unexpected pay deviations was about 1.14–1.15. Excluding time classified as idle reduced it to around 0.90. Responses to expected pay were smaller: approximately 0.79–0.85 for total time, depending on the statistical instrument.

During ordinary platform work, responses to unexpected deviations were much weaker. The main corrected estimates were approximately 0.29 using hourly wages and 0.40 using per-task wages. Responses to expected wages were larger, around 0.81 and 0.97 respectively.

Comparisons restricted to the same workers preserved this contrast, including after accounting for stable individual differences. But earnings recall changed the picture. In the main models evaluated at perfect earnings recall, the observational response to unexpected pay was approximately 0.89–1.04, compared with experimental estimates around 1.11–1.14. Those differences were not statistically significant. Larger recall errors were associated with weaker observational responses, but not significantly different experimental responses.

That pattern fits the authors’ attention explanation: workers may adjust their effort to the pay they perceive rather than every change in their actual earnings. However, statistical indistinguishability does not establish identical behavior. An alternative instrument produced larger remaining gaps, including one statistically significant difference.

Experimental responses were also asymmetric. Below-expectation pay produced roughly twice the response to above-expectation pay in several specifications, consistent with loss aversion. Here, that means a stronger reduction in work when pay falls below expectations—not working harder to recover a loss. Ordinary platform work did not show the same consistent loss-dominant pattern.

What remains uncertain

The study does not demonstrate that attention causes the gap. Recall accuracy was measured, not randomly assigned, and could reflect unobserved worker characteristics. Expectations for the reference week were elicited afterward, creating another potential source of error. Observational estimates lack the experiment’s strict causal identification.

The sample also comprised recently active platform workers using Apple or Samsung smartphones, not a representative cross-section of employees. Results from short, piece-rate tasks may not transfer to salaried jobs, higher earnings or longer periods.

This briefing draws on the supplied paper text of an arXiv preprint, version three, dated 9 November 2026. Its central contribution is evidence consistent with an attention mechanism, rather than proof of that mechanism: understanding wage incentives may require measuring what workers notice as well as what they earn.