Work that Counts, and Work that Doesn't: Reconciling Experimental and Observational Labor Supply Elasticities
Listen
Radio episode about this paper
Transcript
Introduction to the show: ident: Economics. The day in economics: theory, mechanisms and the evidence beneath them.
Ruth: We're looking at the paper: "Work that Counts, and Work that Doesn't".
Warren: Here's the gist of the paper. Combining a real-effort emoji-counting experiment with platform records, self-reported expectations and recalls, and smartphone screen-time logs, this study compares how workers on Prolific respond to randomized wage changes in the experiment versus wage variation in their actual work.
Ruth: The results reveal that workers’ behavior in experiments is remarkably different from real life.
Warren: In the observational window, workers respond to expected wage changes just as they do in the experiment, but are much less responsive to unanticipated changes.
Ruth: However, once conditioned on recall error in self-reports, behavior aligns between the two windows.
Title and authors: Ruth: So we're looking at this paper called "Work that Counts, and Work that Doesn't: Reconciling Experimental and Observational Labor Supply Elasticities." It’s by Adamo and Cantarella. Essentially, they’re asking if the way people react to wage changes in a controlled experiment matches how they react in their actual jobs.
Warren: Yeah, they combine an emoji counting experiment with records from platforms like Prolific, plus self-reports and screen time data. They’re comparing how workers change their work when the wage changes randomly in the test versus when it actually changes at work.
Iris: The core finding they present is that workers act very differently in experiments compared to real life. In the observational window, they respond to expected wage changes just like in the experiment, but they are much less responsive to unexpected ones.
Ruth: That’s a big difference because it points directly at how expectations matter when we look at real labor supply decisions. They also find that if you condition on recall error in self-reports, the behavior starts to align between the two windows.
Warren: Recall error being a source of error for self-reported data, but also a sign of inattention, which they argue affects effort itself when unexpected wage changes go unnoticed. That’s an interesting link between measurement and actual behavior.
Iris: What really sticks with me is how the intertemporal elasticities look. In the observational window, they are remarkably muted compared to the experiment, actually smaller than what you'd expect from life-cycle models.
Ruth: And that’s a contradiction because life-cycle models predict that when workers perceive a wage change as transitory relative to their expected path, they work more today. But here, the observational window shows something different.
Warren: They also find that uncompensated elasticities are more stable between the two windows, and they are close to this intertemporal recall-adjusted figure, which suggests there aren't much income effects at play in this comparison.
Iris: That stability is driven by a process called attention to realized earnings. Workers who keep track of their earnings and can recall them accurately respond to wage changes in the observational window just like they do in the experiment.
Ruth: So, it seems that being able to track what you’ve actually earned is what bridges that gap between the two windows of data collection. But those who fail to track their earnings are much less responsive when things are unexpected.
Warren: That leads into how they reconcile the wide variation in elasticity estimates we see between observational and experimental studies, showing that once measurement error is handled, uncompensated elasticities are similar across both windows.
Iris: They also show this helps confirm the static gain-loss map by demonstrating under which conditions labor supply is shaped by deviations from expectations, finding that relative losses have about twice the effect of relative gains in the experiment.
Ruth: So it sounds like when things are visible in an experiment, loss aversion is really amplified compared to when we look at real work. But if you look at the reference week, losses and gains have similar effects there.
Warren: The paper suggests workers act neoclassical with respect to the wage they perceive, responding to expected wages with an elasticity close to one, and income effects play a negligible role overall.
Iris: The key distinction they make is that deviations from expectations matter only when workers actually perceive them, and when they do, relative losses weigh more than relative gains.
Ruth: And the difference between the two windows boils down to how much attention workers pay to their realized pay—the ones who track earnings respond strongly in both settings, while those who don't barely respond at all.
Warren: The implication for me is that this highlights how worker expectations and visibility of deviations are primary levers for influencing labor supply. Firms and platforms can gain leverage by shaping what workers expect or controlling how they observe their pay.
Iris: They flag a limitation regarding the horizon over which reference points are formed, and whether income effects might become more important in other contexts outside of this specific setting.
Ruth: Overall, once measurement error is dealt with, the elasticities look positive and close to one at both margins, but the difference between windows is concentrated in those relative losses that workers react to strongly only when they are actively experimenting.
Warren: It’s a nuanced picture: workers are neoclassical regarding what they expect, but their response to deviations depends on whether they see them in real time or just in an experiment. That’s the main thrust of "Work that Counts, and Work that Doesn't."
The paper's summary: Ruth: So, this paper is trying to untangle why we see such big differences between how people behave in an experiment versus how they actually behave when they are working at home or on a platform like Prolific.
Warren: Right, it’s basically comparing the hard data from that controlled lab setting against real-world data from actual jobs to figure out where the confusion is coming from.
Iris: The main finding is that workers act very differently in those experiments compared to their day-to-day work. In the observation window, they respond to expected wage changes almost exactly like they do in the experiment, but they are way less sensitive when things change unexpectedly.
Ruth: That’s what’s really striking because it shows that just observing behavior doesn't tell you much about how people react when things are uncertain.
Warren: And the authors point out that this difference is tied to how much attention a worker pays to the money they actually earn while they are working.
Iris: They say that if you condition workers on recall error in their self-reports, then their behavior starts to look more similar between those two settings.
Ruth: That’s because recall error isn't just a mistake in reporting; it seems to be a symptom of not paying enough attention when things are unexpected.
Warren: So, when we look at the math, the study shows that the intertemporal elasticities are much smaller in that observational window compared to what life-cycle models predict.
Iris: And they find that these intertemporal elasticities are even smaller than the uncompensated ones, which is a bit weird because it contradicts some of those standard economic predictions.
Ruth: It suggests that in the real world, when things aren't happening right then and there, workers don't change their long-term plans as much as we thought.
Warren: But then they also find that uncompensated elasticities are pretty stable between the two windows, which is close to this recall-adjusted figure they found earlier.
Iris: That stability comes down to the attention workers give to their realized earnings; people who track their pay respond just as much in both settings, but those who don't barely react at all.
Ruth: It really boils down to visibility—whether a deviation from expectations is visible right now or if it’s just something that happened somewhere else.
Warren: The paper also shows how loss aversion plays a bigger role in the experiment than in real life, where relative losses have about twice the effect of relative gains when those deviations are visible.
Iris: And they conclude that workers seem neoclassical regarding what they expect, meaning if you ask them what they expect to earn, their response is pretty direct.
Ruth: But when those expectations are deviated from—when you see a change—it depends on whether it's a loss or a gain, and whether the deviation was visible in the first place.
Warren: The big implication here for platforms and companies is that they can have real power by shaping what workers expect or by controlling how much they notice about their earnings.
Iris: They flag that the study doesn't fully cover how long these reference points last, so we don't know if this effect holds up over a very long time horizon.
Ruth: It’s a lot to unpack—it sounds like the next step is figuring out how those reference points actually form and update during an ongoing task.
The paper's improvements: Tom: So, now we're looking at how they suggest fixing the study to make it even better. They aren't just stopping there with what they found; they’re suggesting how to make this kind of research more useful for understanding labor supply in general.
Ruth: They propose adding a way to measure attention workers pay to the money they actually earn, which would let them separate those who keep track from those who don't.
Warren: That makes sense for the quantitative side because it gives you a predictor to model how responsive labor supply will be in the observational window versus the experimental one.
Iris: And they suggest an AI capable of modeling reference dependence based on how visible those deviations are, predicting that reference dependence pops up when changes are visible during work but not in an experiment.
Ruth: That connects back to what we said earlier about visibility—it suggests we can anticipate labor supply responses based on whether a wage change is salient during the actual work or only in the test.
Warren: They also want to improve predictions of intertemporal elasticities by using recall error as a moderator, because they found that workers who don't track their earnings have much smaller intertemporal elasticity, and high recall error will show that trend even more.
Iris: That ties the measurement error back into the actual economic behavior; if you mess up your report, it signals a process of inattention that mutes the response to unexpected wage changes.
Ruth: Plus, they want better models for loss aversion based on deviation visibility, predicting that relative losses with respect to expected pay have about twice the effect of relative gains when those deviations are visible in an experiment.
Warren: That would be useful because it gives us a clearer picture of how much workers care about a loss versus a gain depending on the context.
Iris: They also suggest implementing dynamic reference point updating during online tasks, which lets them simulate how workers adjust their expectations once the experiment actually starts running.
Ruth: It seems like they’re moving from just describing what happened to building an AI that can predict how workers will react in real-time based on these underlying behavioral mechanisms.
Warren: If this works, it means platforms and companies could really start leveraging knowledge about worker expectations to influence labor supply more effectively.
Iris: The limitation they mention is the horizon over which these reference points are formed, so we can't assume this works perfectly forever without a way to model that update process.
Conclusion: Tom: So we're wrapping up this look at "Work that Counts, and Work that Doesn't: Reconciling Experimental and Observational Labor Supply Elasticities." Basically, we saw how behavior differs between a controlled experiment and real work.
Ruth: The core message is that workers are neoclassical about what they expect to earn, but their reaction to deviations from those expectations depends entirely on whether those changes are visible in real time.
Warren: So even when things change unexpectedly, it’s not just about the size of the change; it’s about whether you can track your earnings while you're working.
Iris: I think what this means for econometric identification is that the variation driving the result isn't just a simple wage change, but how much attention is paid to that wage during a specific window.
Ruth: Exactly, and when we look at those numbers, the difference between windows was mostly concentrated in relative losses—workers react strongly to losing money in an experiment but not at all in real life.
Warren: That distinction about loss aversion being stronger when deviations are visible is something I’ll keep thinking about for my trading models.
Iris: And they did flag that the study doesn't fully cover the long-term reference points, which is a limitation we have to keep in mind for future work.
Ruth: So, while this paper gives us a solid framework on how expectations and visibility shape labor supply decisions, it opens up questions about how these reference points actually evolve over time.
Warren: Next up, we’re looking at something totally different—online Bayesian inference methods for Dirichlet process mixture of binary logit models. That sounds like a completely different kind of quantitative problem to tackle.
Mattia Adamo, Michele Cantarella
IMT School for Advanced Studies Lucca · Technical University of Denmark - DTU
econ.GN, q-fin.EC
Submitted: 2026-05-28
Updated: 2026-11-09
License: http://creativecommons.org/licenses/by/4.0/
Importance score: 92/100
The gist: Combining a real-effort emoji-counting experiment with platform records, self-reported expectations and recalls, and smartphone screen-time logs…
Key concepts
- Intertemporal Elasticities
- This measures how a worker's future labor supply changes in response to a wage change now. The study found these elasticities are much smaller in real work compared to the experiment, contradicting some economic models. This difference is linked to whether workers pay attention to their earnings.
- Recall Error
- This refers to mistakes made when self-reporting past earnings. Recall error acts as both a source of measurement error and a symptom of inattention. High recall errors are predicted to lead to muted intertemporal elasticity, especially in real work where wage changes might go unnoticed.
- Reference Dependence
- This describes how a worker's response to a wage change depends on their current reference point—what they expect or what they have recently seen. The paper suggests that this dependence is strongest when deviations from expectations are visible, meaning workers react differently based on whether the deviation is salient.
- Uncompensated Elasticities
- These elasticities measure the response to wage changes without considering income effects. The study found these are more stable between the experiment and real work, suggesting that income effects play a negligible role in determining labor supply responses across different settings.
Terminology
Summary
Combining a real-effort emoji-counting experiment with platform records, self-reported expectations and recalls, and smartphone screen-time logs, this study compares how workers on Prolific respond to randomized wage changes in the experiment versus wage variation in their actual work. p. 0 Work that Counts, and Work that Doesn’t:
Labor Supply in Experiments and Real Work∗
Mattia Adamo1 and Michele Cantarella† 2,1
1
IMT School for Advanced Studies Lucca
2Technical University of Denmark - DTU
Abstract
Do workers behave the same in experiments as they do in real life? Combining a realeffort emoji-counting experiment with platform records, self-reported expectations and
recalls, and smartphone screen-time logs, we compare how the same workers on Prolific
respond to randomized wage changes in the experiment and to wage variation in their actual work. Combining a realeffort emoji-counting experiment with platform records, self-reported expectations and
recalls, and smartphone screen-time logs, we compare how the same workers on Prolific
respond to randomized wage changes in the experiment and to wage variation in their actual work. Our results reveal that workers’ behavior in experiments is remarkably different
from real life. Our results reveal that workers’ behavior in experiments is remarkably different
from real life. In the observational window, in fact, workers respond to expected wage
changes just as they do in the experiment, but are much less responsive to unanticipated
changes. In the observational window, in fact, workers respond to expected wage
changes just as they do in the experiment, but are much less responsive to unanticipated
changes. However, once we condition on recall error in self-reports, we find that behavior
in the two windows aligns. However, once we condition on recall error in self-reports, we find that behavior
in the two windows aligns. Recall error is then both a source of error for self-reported
data but also the result of a process of inattention that affects effort itself, as unexpected
wage changes often pass unnoticed.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
The study finds that workers’ behavior in experiments is very different from that in their actual work. p. 0 Combining a realeffort emoji-counting experiment with platform records, self-reported expectations and
recalls, and smartphone screen-time logs, we compare how the same workers on Prolific
respond to randomized wage changes in the experiment and to wage variation in their actual work. Our results reveal that workers’ behavior in experiments is remarkably different
from real life. We find that workers’ behavior in experiments is very different from that in their actual work. Intertemporal elasticities, in fact, are remarkably muted in the observational window when
compared to the experiment, to the point that are smaller than uncompensated elasticities,
contradicting the predictions of life-cycle models. Intertemporal elasticities, in fact, are remarkably muted in the observational window when
compared to the experiment, to the point that are smaller than uncompensated elasticities,
contradicting the predictions of life-cycle models. Uncompensated elasticities are more stable
between the two windows but are close to the intertemporal recall-adjusted figure, indicating
an absence of income effects. We find that this result is driven by a process of attention to realized earnings. Workers
who keep track of their earnings and are able to recall them accurately respond to wage changes
in the observational window just as they do in the experiment, while those who fail to track
their earnings are much less responsive.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
The study contributes by helping reconcile wide variation in wage elasticity estimates between observational and experimental studies. p. 4 estimated elasticities may reflect differences in how salient deviations from expectations are,
as unexpected wage changes often pass unnoticed. Contribution to the Literature Our results first contribute to the literature by helping
reconcile the wide variation in wage elasticity estimates, especially between observational and
experimental studies. , 2009; Gabaix, 2019), and small optimization frictions can
generate large differences between observed and structural labor supply elasticities (Chetty,
2012). We show that once measurement error is addressed, uncompensated elasticities are
similar across windows, and the remaining differences can be attributed to a recall channel, as
workers who do not keep track of their earnings do not perceive deviations from their expected
pay. We show that once measurement error is addressed, uncompensated elasticities are
similar across windows, and the remaining differences can be attributed to a recall channel, as
workers who do not keep track of their earnings do not perceive deviations from their expected
pay. Our findings also corroborate the static gain-loss map by showing under which conditions
labor supply is shaped by deviations from expectations. Our results
reconcile this literature by showing that reference dependence emerges when deviations from
expectations are visible. In the experiment, where pay is displayed throughout the task, relative losses have about twice the effect of relative gains, consistent with loss aversion. In the experiment, where pay is displayed throughout the task, relative losses have about twice the effect of relative gains, consistent with loss aversion. In the
reference week, workers do not respond to relative losses any more strongly than to relative
gains.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
The study suggests that workers appear neoclassical with respect to the wage they perceive, responding to expected wages with an elasticity close to one, and income effects play a negligible role. p. 5 Putting uncertainty and expectations together, empirical evidence suggests that labor supply responses are in fact closer
to neoclassical predictions when wages are known (Fisher, 2025), when individuals put less
value on earnings (Exley and Terry, 2019), and when individuals are not loss-averse (Fehr and
Goette, 2007). Focusing on income effects, Andersen et al. Our results
reconcile this literature by showing that reference dependence emerges when deviations from
expectations are visible. In the experiment, where pay is displayed throughout the task, relative losses have about twice the effect of relative gains, consistent with loss aversion. We find that this result is driven by a process of attention to realized earnings. Workers
who keep track of their earnings and are able to recall them accurately respond to wage changes
in the observational window just as they do in the experiment, while those who fail to track
their earnings are much less responsive. This finding suggests that recall error plays a dual
role: first, as a source of error in self-reported data, and second, as a result of a process of
inattention that affects effort itself when workers fail to track wage variations. It follows that
expectations, and the attention workers pay to unexpected changes, are the primary drivers
of behavior, more so than very small income effects. Both life-cycle and reference-dependent models assume that workers keep track
of all realizations, be they expected or unexpected, but evidence on limited attention suggests
that this might not always be the case (Chetty, 2012; Gabaix, 2019; Bordalo et al. Expectation levels and their deviations are also hard to measure, and most studies do not
elicit them, but proxy them through the experimental design (e.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
The key implication is that workers’ expectations, and how visible deviations from them are, are among the primary margins through which labor supply can be influenced. p. 4 We show that once measurement error is addressed, uncompensated elasticities are
similar across windows, and the remaining differences can be attributed to a recall channel, as
workers who do not keep track of their earnings do not perceive deviations from their expected
pay. Our findings also corroborate the static gain-loss map by showing under which conditions
labor supply is shaped by deviations from expectations. If
within-worker elasticities are similar across environments, differences in macro elasticities may
reflect not preference heterogeneity but differences in how closely workers track their earnings. Our findings therefore also inform the design and evaluation of incentives within firms and organizations (Lazear, 2000), where compensation schemes may operate by making pay changes
more or less visible, and the assessment of market power in monopsonistic settings (Manning,
2020), where firms’ ability to set wages, fees, and informational environments influences labor
supply through what workers perceive as well as through prices. , 2004) expectations also act as
reference points, so that the same wage change can have different effects depending on whether
it is perceived as a gain or a loss. In applied research, however, these predictions often fail,
as the distinction between these two elasticities blurs and estimates vary widely from study
to study.Where the paper says it
Where the paper says it
Where the paper says it
The results suggest that once measurement error is addressed, elasticities are unambiguously positive and close to unity at both margins. p. 4 , 2009; Gabaix, 2019), and small optimization frictions can
generate large differences between observed and structural labor supply elasticities (Chetty,
2012). We show that once measurement error is addressed, uncompensated elasticities are
similar across windows, and the remaining differences can be attributed to a recall channel, as
workers who do not keep track of their earnings do not perceive deviations from their expected
pay. We find that workers’ behavior in experiments is very different from that in their actual work. Intertemporal elasticities, in fact, are remarkably muted in the observational window when
compared to the experiment, to the point that are smaller than uncompensated elasticities,
contradicting the predictions of life-cycle models. The literature operates a fundamental distinction between labor supply
responses to permanent and temporary wage changes, approximated by uncompensated and
intertemporal elasticities, respectively. While the uncompensated (or Marshallian) elasticity
reflects the impact of a permanent wage variation on labor supply affecting future income,
the intertemporal (or Frischian) elasticity reveals how much workers are willing to alter the
amount of time worked today when only today’s wage varies. This finding suggests that recall error plays a dual
role: first, as a source of error in self-reported data, and second, as a result of a process of
inattention that affects effort itself when workers fail to track wage variations. It follows that
expectations, and the attention workers pay to unexpected changes, are the primary drivers
of behavior, more so than very small income effects. In the experiment, where pay is displayed throughout the task, relative losses have about twice the effect of relative gains, consistent with loss aversion. In the
reference week, workers do not respond to relative losses any more strongly than to relative
gains.Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
Where the paper says it
The study concludes that workers (flexible-hours workers at least) appear to be neoclassical with respect to the wage they perceive and reference dependent in its response to deviations from it. p. 2 Both life-cycle and reference-dependent models assume that workers keep track
of all realizations, be they expected or unexpected, but evidence on limited attention suggests
that this might not always be the case (Chetty, 2012; Gabaix, 2019; Bordalo et al. Expectation levels and their deviations are also hard to measure, and most studies do not
elicit them, but proxy them through the experimental design (e. Our results
reconcile this literature by showing that reference dependence emerges when deviations from
expectations are visible. In the experiment, where pay is displayed throughout the task, relative losses have about twice the effect of relative gains, consistent with loss aversion. Our findings therefore also inform the design and evaluation of incentives within firms and organizations (Lazear, 2000), where compensation schemes may operate by making pay changes
more or less visible, and the assessment of market power in monopsonistic settings (Manning,
2020), where firms’ ability to set wages, fees, and informational environments influences labor
supply through what workers perceive as well as through prices. The remainder of the paper is structured as follows.Where the paper says it
Where the paper says it
Where the paper says it
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
-
Incorporate a mechanism to measure attention to realized pay as a key predictor of behavioral response. The improved system can distinguish between workers who
keep track of their earnings
and those who do not, allowing it to predict how responsive labor supply will be in the observational window versus the experimental window. -
Develop an AI capable of modeling reference dependence based on deviation visibility. The system should be able to predict that
reference dependence emerges when deviations from expectations are visible,
suggesting it can better anticipate labor supply responses based on whether a wage change is salient during work or only in an experiment. -
Improve prediction of intertemporal elasticities by incorporating recall error as a moderator. The AI can use the finding that
workers who do not keep track of their earnings are much less responsive
to predict that workers with higher recall errors will show muted intertemporal elasticity, especially in the observational window where deviations are less salient. -
Enhance prediction of loss aversion based on deviation visibility. The system should be able to model how responses vary by sign of deviation, specifically predicting that
relative losses with respect to the expected pay have about twice the effect of relative gains
when deviations are visible in a setting like an experiment. -
Implement a mechanism for dynamic reference point updating during online tasks. The improved system can simulate how workers
update their reference point once the experiment starts,
allowing it to predict thatintertemporal elasticities are small in the first sequence
and increase as workers gain exposure to earlier sequences.
Abstract
Do workers behave the same in experiments as they do in real life? Combining a real-effort emoji-counting experiment with platform records, self-reported expectations and recalls, and smartphone screen-time logs, we compare how the same workers on Prolific respond to wage changes in the experiment and in their actual work. Our results reveal that workers' behavior in experiments is remarkably different from their behavior in real life. In the observational window, in fact, workers respond to expected wage changes just as they do in the experiment, but are much less responsive to unanticipated changes. However, once we condition on recall error in self-reports, we find that behavior in the two windows aligns. Recall error is then both a source of error in self-reported data and the result of a process of inattention that affects effort itself, as workers do not respond to unexpected wage changes that pass unnoticed.
Related papers
- Huge If True: Learning from the Size of the Surprise
- The Rebundling of Freelance Work in the AI Age
- Dynamic investment and coordination in teams: Theory and experimental evidence
- Does Frontier AI Strengthen Cyber Defense? Evidence from Project Glasswing
- Generative AI at Work
- What is Work For? Labor, Politics, and Meaning in the Age of AI