📈 economics

The Ergodicity Gap: A Framework for Understanding When Population Evidence Describes Individual Lives

This paper introduces the Ergodicity Model of Well-being (EMW) to demonstrate that population-level averages often fail to represent individual life trajectories due to non-ergodic factors like compounding resources and absorbing states, thereby creating a systematic positive gap between aggregate evidence and personal experience.

Original authors: Shay Tsaban

Published 2026-08-31

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Shay Tsaban

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine trying to understand a single human life by looking at a photograph of a crowd. In the field of well-being research, this is exactly what happens every day. Governments, employers, and scientists routinely survey thousands of people at a single moment to measure happiness, life satisfaction, and mental health. They calculate an average score from this snapshot and assume it tells them what a typical person's life looks like over time. This approach relies on a hidden assumption: that the average of a group at one moment is the same as the average of one person's life across many years. For decades, this assumption has gone unchallenged, treated as a simple matter of statistics. But a new analysis suggests this assumption is fundamentally broken, not because the surveys are poorly designed, but because human lives do not work the way the math assumes they do.

The problem lies in a concept called ergodicity. In simple terms, a process is ergodic if the average experience of a group of people at one time is identical to the average experience of a single person over a long period. If you were watching a crowd of people walking through a park, and the group's average speed matched the speed of any single walker over an hour, the system would be ergodic. However, human lives are rarely like that. They are shaped by two powerful forces that break this symmetry. First, the resources that make life go well—like money, health, or social connections—often grow or shrink in a way that compounds. Just as interest builds on interest in a bank account, small advantages or disadvantages in life tend to multiply over time, causing people to drift further apart. Second, life contains "absorbing states," which are difficult conditions that, once entered, are hard to leave. These include severe depression, long-term unemployment, or chronic illness. Once a person falls into one of these states, they often stop participating in surveys or leave the workforce entirely, effectively disappearing from the data set.

A new framework called the Ergodicity Model of Well-being, developed by researcher Shay Tsaban, brings these ideas together to show why population averages systematically mislead us about individual lives. The model argues that when we look at a snapshot of a population, we are seeing a distorted picture. Because the people who fall into the hardest states are the ones most likely to drop out of the study, the remaining group looks happier and more stable than the reality of the population actually is. The average score calculated from the survivors is not just a little off; it is a systematically optimistic lie. It overstates what an individual can expect to live through because it silently excludes the very people whose lives are going the worst.

The research does not just point out a flaw; it maps out exactly how the error happens and in which direction. The author shows that the gap between the group average and the individual reality is not random noise. It has a specific shape and a predictable direction. The size of this gap depends on how much life's resources compound and how sharply the connection between those resources and reported happiness bends. But the direction of the error is fixed by the fact that people in the worst situations leave the sample. As time goes on, the group average gets further and further from the truth, not closer. The longer a study runs, the more it loses the people who need to be seen the most, making the final average look better than it really is.

This finding changes how we should interpret decades of well-being research. For example, the field has long debated whether people adapt to major life events like divorce or disability. The standard view, based on population surveys, suggests that most people bounce back to their previous level of happiness. However, this model suggests that the apparent adaptation is an illusion created by the data. The people who do not bounce back are the ones who stop answering the surveys or leave the workforce. The surveys only see the people who recovered, so the average looks like everyone recovered. In reality, a significant minority may be stuck in a state of unhappiness that the population average completely misses.

The implications extend to how governments and organizations make decisions. Institutions that use these population averages to guide policy are effectively optimizing for the wrong thing. If a company tries to improve employee well-being by raising the average score, they might succeed by making the already-happy employees slightly happier, while ignoring the few who are on the verge of burnout. Because the people on the verge of burnout are the ones most likely to quit and disappear from the data, the company's average score might go up even as the actual experience of the typical worker gets worse. The model predicts that by focusing on the group average, institutions are systematically under-investing in preventing the irreversible harms that drive people out of the system.

The paper offers a way to fix this, not by abandoning population data, but by adding two missing pieces of information. To truly understand a population's well-being, we need to know not just the average score of those still present, but also the rate at which people are falling into states they cannot leave, and how long they stay there. These are facts that can be found in administrative records of long-term sickness, hospitalization, or job loss, but they are rarely combined with survey data. By calculating these rates, we can correct the population average to get a realistic estimate of what a typical life actually looks like.

This work does not claim that population surveys are useless. They remain the best tool we have for many purposes. But it insists that we stop treating them as a direct window into individual destiny. The average of a crowd is not the same as the average of a life. When we confuse the two, we risk building policies that look good on paper but fail to protect the people who need help the most. The solution is to acknowledge the gap, measure the missing pieces, and accept that the story of a life is written in time, not in a single snapshot.

Technical Summary: The Ergodicity Gap in Subjective Well-Being Research

Problem Statement
The paper addresses a fundamental epistemological flaw in the human sciences, particularly in subjective well-being (SWB) research: the routine inference from population-level evidence (ensemble averages) to individual life trajectories (time averages). While economics has long debated whether a "representative agent" can describe an aggregate, this article investigates the converse: whether an aggregate describes any individual's life. The author argues that this inference is only valid under the condition of ergodicity, where the average of many realizations at a single moment equals the average of one realization over a long time. The paper posits that psychological and economic processes governing well-being are generally non-ergodic, rendering population statistics a systematically biased guide for individual expectations.

Methodological Framework
The author synthesizes two distinct research traditions to construct the Ergodicity Model of Well-being (EMW):

Psychological Non-Ergodicity (Molenaar, 2004, 2007): Establishes that psychological processes are non-ergodic because they are neither stationary (properties change over time) nor homogeneous (processes differ across individuals). This invalidates the transfer of between-person results to within-person cases.
Economic Non-Ergodicity (Peters, 2019): Establishes that economic processes are non-ergodic due to compounding (multiplicative dynamics). In such systems, the time-average growth rate of an individual differs systematically from the ensemble average growth rate, which is driven by a shrinking minority of favorable trajectories.

The EMW framework is built on four tenets:

Ontological: Well-being is a property of trajectories (time averages), not static populations.
Dynamic: Resources generating well-being compound, causing the ensemble average to track a vanishingly small minority of trajectories.
Topological: Well-being trajectories contain absorbing states (e.g., severe depression, chronic unemployment, death) that are difficult or impossible to leave, making the state space decomposable.
Survivorship: Entering an absorbing state tends to remove a person from observation (attrition). This survivorship bias fixes the sign of the ergodicity gap as positive: population well-being overstates what an individual can expect to live through.

Key Contributions and Formal Derivations
The paper moves beyond qualitative critique by deriving closed-form expressions for the ergodicity gap under three elementary processes:

Multiplicative Dynamics (Compounding):
- Using geometric Brownian motion for resources, the paper demonstrates that the ensemble average grows at rate $\mu$ , while the typical individual trajectory grows at $\mu - \sigma^2/2$ .
- The gap in reported well-being depends on the curvature of the mapping function $f(x)$ from resources to reports.
- Result: If the mapping is less concave than logarithmic, the ensemble average of reports exceeds the typical report. If the mapping is exactly logarithmic (as often found for income and life evaluation), the gap vanishes. If more concave, the order reverses. Thus, compounding sets the magnitude of the gap but not necessarily its sign.
Permanent Absorption with Exit:
- A model where individuals enter an absorbing state with hazard $h$ and leave observation.
- Result: The observed ensemble average remains at the high well-being level ( $w_H$ ) because only unabsorbed individuals are seen. The expected time average for an individual declines as the observation window $T$ lengthens. The gap $\Delta(T)$ increases with both the hazard $h$ and the window length $T$ , approaching the full well-being decrement ( $w_H - w_L$ ) as $T \to \infty$ .
Recurrent Absorption with Partial Attrition:
- A model where absorption is reversible but a share $q$ of absorbed individuals are missing from observation.
- Result: The gap is strictly positive whenever $q > 0$ and the absorbed state has positive prevalence. The gap is independent of the observation window length in a stationary population but grows with the share of missing absorbed individuals.

Results and Calibration
The paper calibrates these models using parameters typical of major depression, long-term unemployment, and disabling illness.

Magnitude: The calculated gaps range from 0.1 to 1.0 scale points on an 11-point scale. These magnitudes are of the same order as the effect sizes the field treats as substantively important.
Direction: The survivorship mechanism ensures the gap is positive. Population figures systematically overstate the well-being an individual will experience over time.
Reinterpretation of Literature: The framework reinterprets four contested literatures:
- Hedonic Adaptation: Observed adaptation in panels may be an artifact of composition (absorbed individuals dropping out), masking a large minority who do not adapt.
- Income and Well-being: Cross-sectional slopes may not equal longitudinal slopes due to the non-ergodic nature of income dynamics.
- Organizational Interventions: Programs may raise the ensemble mean while failing to reduce the hazard of entering irreversible negative states, leaving the typical employee's time average unchanged or worse.
- Inequality: Cross-sectional dispersion does not capture the dispersion of life trajectories (e.g., everyone spending 20% of life in poverty vs. 20% of people spending 100% of life in poverty).

Significance and Claims
The paper claims that the ergodicity gap is not merely statistical noise or a limitation of measurement, but a theoretical object with a determinate structure.

Theoretical: It provides a formal justification for why population evidence licenses no inference about individual lives when processes are non-ergodic. It unifies the methodological concerns of psychology (non-stationarity) with the decision-theoretic concerns of economics (multiplicative growth).
Institutional: The paper argues that institutions optimizing population well-being (ensemble averages) face a perverse incentive: they are led to under-invest in preventing irreversible harm (absorption) because keeping people in the sample (even if suffering) lowers the measured average, whereas improving the lives of the already well raises it without changing the hazard.
Practical: The framework suggests that closing the gap requires collecting two currently missing quantities: the rate of entry into irreversible states and the duration of stay in those states. It proposes that policy and research should shift focus from the ensemble mean to the hazard of absorption and the time average of the typical trajectory.

The paper concludes modestly, noting that the framework does not replace population research but demands a correction for the systematic, predictable error introduced by non-ergodicity. It asserts that the condition required for valid inference from populations to lives has gone largely unexamined in the field dedicated to studying how lives go.

Technical Summary: The Ergodicity Gap in Subjective Well-Being Research

More like this