The Easterlin Paradox

The three ways to be right

Easterlin's position: the paradox is real and has strengthened. His central claim has narrowed and sharpened since 1974. It is now specifically about the long-run time series: over periods long enough to average out business cycles — he uses ten years and more — there is no significant relationship between the growth rate of income and the growth rate of happiness, in either rich or poor countries. The 2010 PNAS paper with Laura Angelescu McVey, Malgorzata Switek, Onnicha Sawangfa and Jacqueline Zweig assembled series for 37 countries, including 17 developing and 11 transition economies, and reported exactly that. Short-run correlations exist, he argues, because recessions make people miserable and recoveries relieve them — which tells you about fluctuations, not about growth.

Stevenson and Wolfers's position: the paradox was a data artefact. Their 2008 Brookings paper argued that earlier work had too few countries, too short a period, and measures too coarse to detect a relationship that is real but logarithmic. With the Gallup World Poll they found: a strong cross-country gradient in log income; a within-country gradient of similar magnitude; no evidence of satiation at high incomes; and, in the time-series data available, positive associations between growth and rising wellbeing in most countries. They were explicit that this undermines the relative-income story, because a purely relative account predicts no cross-country gradient at all.

The reconciling position. Several researchers — including Deaton, and Easterlin himself in places — accept that the cross-sectional and cross-national gradients are real and logarithmic, while holding that something specific about national growth over time prevents the gains from materialising. Candidates: rising aspirations moving in lockstep with income; growth accruing disproportionately to the top so that the median household's log income barely moves; and social costs of growth — insecurity, longer hours, weaker communities, greater inequality — offsetting the material gains.

That third candidate is worth taking seriously because it makes the paradox a finding about how countries have grown rather than about the limits of money. It is also much harder to test.

The methodological fault lines

Cardinality of the scale. Comparing "the average ladder score rose 0.2 points" across countries or decades assumes the scale means the same thing to everyone, always. Analyses that treat the responses as merely ordinal, and that allow for different people using the scale differently, can reverse conclusions. This is a live technical literature and it is not a side issue: the entire dispute is conducted in units whose interpersonal meaning is assumed rather than established.

Ceiling effects. Reported happiness is bounded; income is not. If a large share of a rich country's population is already at or near the top of the scale, growth cannot show up. This mechanically produces something that looks like the paradox without any adaptation or comparison at work.

Which measure. Life evaluation, positive affect, negative affect and domain-specific satisfaction behave differently against income. Studies that disagree frequently turn out to be using different dependent variables. Reading any claim in this area requires knowing which question was asked.

Identification. Almost none of this evidence is causal. Income and happiness are correlated for many reasons — healthier people earn more, happier people are promoted, and both respond to third factors. The small number of quasi-experimental studies, including lottery-winner studies and cash-transfer trials, generally find positive but modest effects of income on wellbeing, which is consistent with a range of positions in the main debate.

Inequality. The distributional point deserves separate mention. In the United States, real median household income grew far more slowly than GDP per capita from the 1970s onward. If wellbeing tracks the log income of the typical household rather than the mean, then a large part of the "flat happiness despite growth" observation is explained without any appeal to adaptation or status. Whether this rescues the material account or vindicates the social account of the paradox depends on what you thought the paradox was claiming.

What both sides agree on

It is worth being explicit, because the disagreement is narrower than the literature's tone suggests. Both camps accept that:

• Income and wellbeing are positively related in cross-sections, robustly.
• The relationship is approximately linear in log income, so absolute gains matter less and less.
• Very large income differences correspond to real and substantial wellbeing differences — nobody disputes that the average Norwegian reports a better life than the average Burundian.
• Relative comparison and adaptation are real psychological phenomena with measurable effects.

The dispute is about the magnitude of the time-series relationship in the long run, and about how much of the cross-sectional gradient survives when everybody moves together.

1949. James Duesenberry published Income, Saving and the Theory of Consumer Behavior, introducing the relative income hypothesis: consumption and satisfaction depend on income relative to others and to one's own past. The idea that Easterlin later found in the happiness data was already on the table.

1965. Hadley Cantril published The Pattern of Human Concerns, with the self-anchoring ladder scale and cross-national survey data. This is the data Easterlin used.

1974. Richard Easterlin published "Does Economic Growth Improve the Human Lot? Some Empirical Evidence" in Nations and Households in Economic Growth: Essays in Honor of Moses Abramovitz, edited by Paul David and Melvin Reder (Academic Press), pages 89–125. The paper is careful and hedged — Easterlin was explicit about the limitations of the data — and it invented a field.

1970s–1990s. The finding was mostly ignored by economists and taken up by psychologists. Philip Brickman and Donald Campbell's 1971 "hedonic treadmill" and their 1978 lottery-winners-and-accident-victims study supplied the psychological mechanism. Ed Diener built the measurement literature. Economics stayed away, largely because the profession was uncomfortable with self-reported data.

1995. Easterlin published "Will Raising the Incomes of All Increase the Happiness of All?" in the Journal of Economic Behavior and Organization, restating and extending the case. Around the same time Robert Frank, Andrew Oswald and Richard Layard began bringing the material into mainstream economics.

2005. Layard's Happiness: Lessons from a New Science took the argument to a general audience and to policymakers, arguing directly for taxation and policy reform on the basis of the relative income effect.

2008. Betsey Stevenson and Justin Wolfers published "Economic Growth and Subjective Well-Being: Reassessing the Easterlin Paradox" in the Brookings Papers on Economic Activity. Using the Gallup World Poll and reconstructed national series, they argued the paradox does not survive better data. This is the paper that made the question contested rather than settled.

2010. Two important papers. Easterlin and colleagues published "The happiness–income paradox revisited" in PNAS, drawing the long-run/short-run distinction and reporting no long-run relationship across 37 countries. Kahneman and Deaton published "High income improves evaluation of life but not emotional well-being" in PNAS, splitting the outcome measure and reporting the $75,000 plateau for emotional wellbeing.

2012. Easterlin, Robson Morgan, Switek and Fei Wang published "China's life satisfaction, 1990–2010" in PNAS, using the country with the largest income growth in history as the strongest available test.

2021. Matthew Killingsworth published "Experienced well-being rises with income, even above $75,000 per year" in PNAS, using experience sampling from 33,391 people and finding no plateau.

2023. Kahneman, Killingsworth and Mellers published "Income and emotional well-being: A conflict resolved" in PNAS, reporting the results of a pre-registered adversarial collaboration. Kahneman died in 2024; the paper is among the last of his career and is widely cited as a model for how scientific disputes should be conducted.

It made subjective wellbeing a legitimate economic variable. Before Easterlin, economics inferred welfare from behaviour and treated what people said about their own lives as unusable. There is now a large empirical literature, a dedicated journal, national statistical programmes measuring wellbeing in the UK, France, New Zealand, Bhutan and elsewhere, and an OECD standard for how to ask the question. Whether or not Easterlin's specific claim survives, the field would not exist without the paper.

It supplied the strongest empirical case against GDP as a welfare target. The argument that growth is not the same thing as improvement had been made on theoretical grounds for decades. Easterlin gave it data. The Stiglitz–Sen–Fitoussi Commission (2009), the OECD Better Life Index, the UK's national wellbeing measures and the World Happiness Report all descend from that move.

It sharpened the distinction between a mean and a distribution. Much of the force of the modern critique is that "national income rose" and "the typical household's income rose" came apart in several rich countries after 1970. Working out which of those wellbeing tracks is now a central question in the economics of inequality, and it started as a puzzle about happiness surveys.

It is a case study in how measurement disputes are settled. The 2023 adversarial collaboration is the concrete payoff. Two researchers with published opposing findings agreed in advance on a joint analysis with a neutral arbiter, and discovered that the disagreement was located somewhere neither had specified — in a subgroup. That method is being adopted in other contested areas precisely because of how well it worked here.

National wellbeing statistics. The UK Office for National Statistics has asked four wellbeing questions of a large sample since 2011; New Zealand runs a wellbeing budget; the OECD publishes guidelines on measurement. Whether these measures should influence policy, and how, is a direct descendant of the argument on this page.

Cash transfers and basic income. Trials of unconditional cash transfers in Kenya, the United States and Finland measure subjective wellbeing as a primary outcome. The results — generally positive effects on wellbeing, often larger for the poorest recipients, sometimes fading over time — are informative for the paradox because they are experimental. Adaptation shows up in some of these studies as an effect that decays, which is exactly Easterlin's mechanism observed directly.

The satiation question in policy. Whether there is a level above which extra income stops improving lives has direct implications for progressive taxation, for how development aid is targeted, and for the design of poverty thresholds. The 2023 finding that the answer differs by subgroup complicates all of these in a useful way: it suggests that at high incomes, spending on mental health may dominate spending that raises incomes further.

Social media and reference groups. If wellbeing depends on comparison, then technologies that widen and intensify the comparison set should reduce it. This prediction is testable and is being tested; the evidence on social media and adolescent wellbeing is contested in its own right, but the theoretical connection runs straight back to Easterlin's relative income mechanism.

Growth policy after the paradox. The practical question the paradox poses is not "should countries stop growing" but "what kind of growth". If insecurity, long hours and inequality offset material gains, then the composition of growth matters as much as its rate. That framing is now common in policy discussion, and it is a direct inheritance from a 1974 book chapter.

The specification at issue

Let $W_{it}$ be mean reported wellbeing in country $i$ at time $t$, and $Y_{it}$ income per head. The three stylised facts correspond to three different regressions, and conflating them is the source of most confusion.

Within-country cross-section. For individuals $j$ in country $i$ at time $t$:

$$W_{ijt} = \alpha_{it} + \beta_{\text{micro}} \ln y_{ijt} + \varepsilon_{ijt}$$

Estimates of $\beta_{\text{micro}}$ are consistently positive and reasonably stable across countries.

Between-country cross-section. For a single period $t$:

$$W_{it} = \alpha + \beta_{\text{macro}} \ln Y_{it} + u_{it}$$

Stevenson and Wolfers's central empirical claim is that $\beta_{\text{macro}} \approx \beta_{\text{micro}}$, which is what you would expect if income operates absolutely, and not what a purely relative-income model predicts. Under a pure relative model, $\beta_{\text{macro}} = 0$ while $\beta_{\text{micro}} > 0$.

Within-country time series. With country fixed effects:

$$W_{it} = \alpha_i + \beta_{\text{time}}\ln Y_{it} + \gamma_t + u_{it}$$

Easterlin's claim is that $\beta_{\text{time}} \approx 0$ over long horizons. This is where the disagreement lives, and it is the hardest of the three to estimate, because within-country variation in $\ln Y$ over the available sample period is small relative to measurement noise.

The relative income model in its simplest form makes the distinction explicit. Suppose

$$W_{ijt} = \theta \ln y_{ijt} - \phi \ln \bar{y}{it} + \varepsilon{ijt}$$

where $\bar y_{it}$ is the reference income. Then $\beta_{\text{micro}} = \theta$ but $\beta_{\text{time}} = \theta - \phi$. The Easterlin paradox is the hypothesis that $\phi \approx \theta$: reference income tracks own income one-for-one in the long run, cancelling the gain. Stevenson and Wolfers's position is that $\phi$ is well below $\theta$.

Stated this way, the paradox is a single-parameter dispute, and its empirical content is entirely about how completely the reference point adapts.

Why the log matters so much

Every serious specification uses $\ln Y$, not $Y$. This is not a technicality; it is doing a great deal of the rhetorical work on both sides.

Under a log specification, wellbeing gains are proportional to growth rates, not to absolute increments. A country growing at 2% per year gains $\beta \cdot \ln(1.02) \approx 0.0198\beta$ per year. Over forty years that is $\beta \cdot \ln(2.2) \approx 0.79\beta$. With a within-country micro slope of roughly 0.3 ladder points per log point — a common estimate — that predicts a rise of about 0.24 points on a 0–10 scale over forty years.

That is small. It is within the noise of most national time series, and it is comparable to the drift you would get from changes in survey mode or question wording. The relationship Stevenson and Wolfers defend predicts an effect that Easterlin's data would struggle to detect. Both camps can therefore be consistent with the same numbers, and this is the most important thing to understand about why the dispute has not closed. It is a power problem as much as a substantive one.

Ceiling effects and scale use

Let $W^{}$ be latent wellbeing and $W$ the reported value on a bounded scale $[0, C]$ with thresholds. If the mapping is $W = \min(C, \max(0, g(W^{})))$ for some increasing $g$, then as the population distribution of $W^{*}$ shifts right, the observed mean rises less than proportionally and eventually not at all. This is enough to produce a flat time series from a genuinely rising latent variable.

The 2023 adversarial collaboration is an instance. Kahneman and Deaton's emotional wellbeing measure was constructed from dichotomous items — did you experience happiness yesterday, yes or no — which imposes a hard ceiling at "yes". People already answering yes cannot answer more yes. When Killingsworth's continuous experience-sampling measure was used, the ceiling disappeared and so did the plateau, except in the subgroup for which the flattening was substantive rather than mechanical.

Distinguishing a real plateau from a measurement ceiling requires either an unbounded measure or an explicit model of scale use. The literature that treats wellbeing responses as ordinal and estimates the thresholds — including work by Bond and Lang, and by Kaiser and Oswald — has shown that for many datasets, the sign of the estimated group difference can be reversed by an admissible choice of underlying scale. That is a serious result, and it applies to some comparisons in this literature and not others.

The subgroup structure found in 2023

The pre-registered joint analysis fitted the relationship between log household income and experienced wellbeing separately across the wellbeing distribution — that is, by quantile of happiness rather than only at the mean. The headline findings:

Happiness quantile	Slope of wellbeing on log income above ~$100k
Bottom ~15–20%	approximately flat
Middle	positive, similar to below $100k
Top ~30%	positive and steeper than below $100k

Averaging a flat segment with a steepening one produces the apparent plateau in the mean, which is how two competent analyses of overlapping questions reached opposite conclusions.

The substantive interpretation both authors endorsed: for people whose unhappiness has a cause money cannot address, more money does not help. For everyone else it continues to. This is a statement about heterogeneity, not about satiation, and it does not by itself bear on whether a whole country gets happier as it grows — though it does suggest that national averages will move sluggishly whenever a sizeable unhappy minority is included in them.

What would settle the original question

The honest answer is that the required evidence is close to unobtainable. Establishing whether long-run national income growth raises wellbeing needs: a consistent measurement instrument held fixed for many decades; enough countries with enough variation in growth rates; a credible source of exogenous variation in national income; and a way to separate the effect of growth from the effects of everything correlated with it. No study has all four, and the first alone rules out most of the existing series.

What can be done, and is being done, is to accumulate long consistent series going forward — the Gallup World Poll has now run since 2005, national statistical offices since around 2011 — and to test the mechanism directly through experiments on individuals. That is why the field's centre of gravity has shifted from national time series to cash-transfer trials and experience sampling. It is a retreat from the original question toward a tractable one, and it should be understood as such.