Is poverty about poverty or about distribution?
Two billion people climbed out of extreme poverty in a single generation. A household at the bottom of Detroit is richer than nearly all of them — and still gets counted as poor. Both numbers are honest. They answer different questions.
The absolute-poverty story (and its triumph)
“The poor are no less rational than anyone else — quite the contrary. Precisely because they have so little, we often find them putting much careful thought into their choices.”
— Abhijit Banerjee & Esther Duflo, Poor Economics, 2011
When Banerjee, Duflo, and Kremer won the 2019 Nobel, the committee was honoring a way of seeing poverty: not as an abstraction to argue about, but as a deficit you can measure, intervene on, and test. The deficit is concrete — calories, a roof, a dose of deworming medicine — and on that definition, the world has just done something extraordinary.
Start with the definition the World Bank uses to run the largest poverty-tracking exercise in history. Extreme poverty is living on less than \$2.15 a day, measured in 2017 purchasing-power-parity dollars — an absolute floor, the same number in Lagos, La Paz, and Lahore. The choice of floor is deliberate. Below it, the binding constraint is survival: enough calories to work, a roof, clean water, the basic health to stay alive. Above it, the constraints loosen. This is the absolute-deficit apparatus: poverty is a gap between what a household consumes and what it takes to function biologically, and the gap is the same gap whatever country you stand in.
The apparatus has a sharp empirical edge. Because the floor is fixed, you can count heads against it across decades, and the count tells a story almost no one expected. In 1990, roughly 36% of humanity lived below the extreme-poverty line. By 2019 it was under 9%. More than a billion people crossed the floor in thirty years — the largest measured improvement in material welfare in recorded history. Most of the lift came from a handful of places: China first and largest, then India, Vietnam, Indonesia, Bangladesh, growth-led economies integrating into world markets. The empirical episode that the absolute apparatus is standing on is told in full in Economic History Ch.17 (China's reform and the Asian century).
Why does the floor matter so much for policy? Because of what marginal welfare looks like near it. Write a household's welfare as utility over consumption, $u(c)$, with a subsistence threshold $\bar{c}$. Just above $\bar{c}$, an extra dollar buys survival itself — the marginal value of consumption is enormous. Well above $\bar{c}$, that same dollar buys a nicer phone, and the marginal value has collapsed. That asymmetry is the entire policy case for targeting: a dollar moved to a household near the floor does vastly more good than a dollar spread across everyone. Floor-lifting beats across-the-board.
Let welfare be $u(c)$ with $u'(c) > 0$ and $u''(c) < 0$, and let $\bar{c}$ be the subsistence floor. The subsistence structure is the limiting behavior near the floor:
$$\lim_{c \to \bar{c}^{+}} u'(c) = +\infty, \qquad \text{poverty deficit} = \max(0,\ \bar{c} - c)$$The poverty headcount counts households with $c < \bar{c}$; the poverty gap sums their deficits. Both are anchored to the fixed floor $\bar{c}$, not to anyone else's consumption.
Below the line, every dollar buys survival — food, shelter, the ability to keep working. Above the line, dollars start buying things that are nice but not necessary. That gap is why the apparatus says: don't spread help thinly across everyone, push the households closest to the floor across it. The whole policy follows from where the floor sits, and the floor sits at the edge of staying alive.
Two more pieces complete the apparatus. The first is Amartya Sen's capability approach, read here in its development register: poverty is not low income as such but the failure to reach a minimum set of functionings — being adequately nourished, being able to read, being mobile, being able to take part in the life of the community. In a low-income setting those functionings are anchored to subsistence: the capability you lack is the capability to stay alive and healthy. The second is the method that turned the apparatus operational. Rather than argue about poverty from theory, Banerjee, Duflo, Kremer and the J-PAL network ran randomized controlled trials — deworming, microcredit, conditional cash transfers, teaching at the right level — measuring which specific interventions actually moved consumption, health, and schooling. The 2019 Nobel was the discipline's recognition that this works. For the full apparatus, the chapters below open in place.
“The number of people living in extreme poverty has fallen from nearly 2 billion in 1990 to about 700 million today. That is one of the greatest achievements in human history.”
— the standard framing of the World Bank’s \$2.15/day line
Is a single dollar-a-day line the right way to count the poor?
One number, the same everywhere on earth, is what makes the great-decline story countable. It is also what the critics attack first. Pull on the line and the whole apparatus comes with it.
The apparatus and its honest critics
“We have to abandon the habit of reducing the poor to cartoon characters and take the time to really understand their lives, in all their complexity and richness.”
— Abhijit Banerjee & Esther Duflo, Poor Economics, 2011
Banerjee and Duflo are making the strongest case for the apparatus: stop arguing about poverty in the abstract and go find out, experiment by experiment, what actually moves a household across the floor. The absolute-deficit frame is what makes the experiments legible — you can measure consumption uplift, deworming take-up, default rates, test scores, because there is a fixed thing called the deficit to move. The intellectual ancestry is named in History of Economic Thought Ch.16 (Development economics): Lewis's surplus-labor model, Sen's capabilities, and then the experimental turn that made the apparatus operational. This is mainstream development economics' working tool, and it has a Nobel.
“Randomized controlled trials cannot automatically trump other evidence, they do not occupy any special place in some hierarchy of evidence, and the lack of explicit prior structure limits the ability to learn from them.”
— Angus Deaton, “Instruments, Randomization, and Learning about Development,” Journal of Economic Literature, 2010
Deaton, himself a Nobel laureate in poverty measurement, is not the relative-poverty critic — that voice arrives in Stage 2. His objection is from inside the absolute apparatus: a clean experiment in one village tells you what worked there, not why, and not whether it transfers. Lant Pritchett presses the same point harder — chasing well-measured small effects can crowd out the messy, hard-to-randomize question of how poor countries grow rich, which is what actually moved the billion. The critique softens the apparatus's claim to be the only right way to see low-income poverty without yet flipping to a different apparatus. That flip is the next stage's whole job.
Where this leaves us
On its own terms, the absolute-poverty apparatus has won. The 1990–2020 lift is the largest measured welfare gain of the modern era, China and India and Vietnam did most of the lifting, and the experimental method that operationalized the frame has been recognized at the discipline's highest level. The within-apparatus critics — Deaton, Pritchett — sharpen it rather than overturn it. But every claim in this stage rests on one assumption: that the binding constraint is consumption below a survival floor. The apparatus is built around $\bar{c}$. So here is the question it cannot answer on its own terms. Walk into a country where no one is below the floor — and people are still called poor. What are you measuring then?
Walk into Detroit, Cleveland, parts of Los Angeles, and ask who is in “poverty.” Every household is far above \$2.15 a day — richer than nearly everyone the great decline lifted. By the absolute apparatus the answer is: almost no one. By the apparatus we haven't reached yet, the answer is: roughly one in six. Same people, same data, opposite verdict. The difference isn't the threshold. It's the lens.
Flip the apparatus
“The distribution of wealth is too important an issue to be left to economists, sociologists, historians, and philosophers. It is of interest to everyone, and that is a good thing.”
— Thomas Piketty, Capital in the Twenty-First Century, 2014
Across the rich world the at-risk-of-poverty rate — the share of households below 60% of their country's median income — sits between 14% and 22%. In countries where extreme poverty by the \$2.15 line is statistically zero. The OECD does not think those people are imaginary. It thinks the absolute apparatus is looking at the wrong thing.
Now make the flip. Keep the same humans, the same income data, the same welfare surveys — and change what you anchor poverty to. Instead of a fixed survival floor $\bar{c}$, anchor it to the local distribution. The OECD's operational definition: a household is in poverty if its income falls below 60% of the national median. Peter Townsend, who built this tradition in his 1979 Poverty in the United Kingdom, put the principle in plain words — a person is in poverty when their resources exclude them from the customary activities and conditions of their society. The threshold is no longer about staying alive. It is about being able to participate.
This is the relative-poverty apparatus, and it changes three things at once. It changes the metric — distance from the median, or share of aggregate income held by the bottom decile, instead of distance from a global floor. It changes the welfare object — utility now carries a social-comparison term, $u_i(c_i, \tilde{c})$, where $\tilde{c}$ is the local median, not the subsistence line; the same dollar yields different welfare depending on what everyone around you has. And it changes the policy direction. If poverty is distance from the median, you reduce it by compressing the distribution — a universalist welfare state, progressive transfers, wealth taxes, a universal basic income — not by lifting a fixed floor that no one is below.
The relative-poverty indicator anchors to the distribution itself:
$$p_i = \mathbf{1}\!\left[\, c_i < 0.6 \cdot \mathrm{median}(c) \,\right]$$And social welfare is no longer a sum of independent utilities but carries inequality aversion — the Atkinson social welfare function, $W = \frac{1}{1-\varepsilon}\sum_i c_i^{\,1-\varepsilon}$, where the parameter $\varepsilon$ measures how much society dislikes dispersion. That function is derived in full in the inequality walkthrough; here it is enough to see that it penalizes spread, where the absolute apparatus penalized only the deficit below $\bar{c}$.
The absolute apparatus asks: can you eat? The relative apparatus asks: can you take part in a society where eating isn't the question? In a rich country the things you need to function as a citizen — housing near a decent school, a car, a phone, healthcare, a reliable internet connection — are defined by what everyone else has, not by a calorie count. Fall far enough below the median and you're locked out of all of it, even if you'd be wealthy on a global scale. That exclusion is what this apparatus calls poverty.
Notice who reappears: Sen, in a different register. The capability approach from Stage 1 is genuinely flexible — capabilities can be anchored to subsistence functionings (being nourished, being alive) or to socially-required functionings (being able to appear in public without shame, in Sen's own example). Same author, same apparatus, two anchorings. That dual-use is not a contradiction; it is the clearest sign that the choice doing the real work is not which threshold but which anchor. The formal welfare machinery — social welfare functions, the general-equilibrium welfare framework — lives in the A-book; don't re-derive the Atkinson function here, it has a home.
The empirical face of this apparatus is the post-2014 turn. Piketty's Capital, the Saez–Zucman work on top-wealth shares, the “elephant curve” of Lakner and Milanovic — all of it reads the same decades that the absolute apparatus read as triumph and finds a different story underneath: within-country distributions pulling apart even as global headcounts fell. The economic-history spine of that turn — the post-2008 distribution salience — runs through Economic History Ch.19 (The 2008 crisis and after), with the rich-world inequality rise that first drove relative deprivation back to prominence set in Ch.16 (Stagflation and the neoliberal turn).
The welfare-function machinery behind this apparatus — the Atkinson function, inequality aversion, the impossibility of costless redistribution — is built up at full depth in the sibling walkthrough Is inequality a problem economics can solve? (Stage 1). This walkthrough cites it rather than re-deriving it; the work here is the apparatus flip, not the apparatus internals.
“A universal basic income would not just be a check. It would be a tectonic shift in the relationship between the individual and the state, a way of valuing all the unpaid labor that holds society together.”
— Annie Lowrey, Give People Money, 2018
Is universal basic income a serious idea, or a category error?
Under one apparatus UBI is the obvious move. Under the other it's a waste — untargeted, capability-blind, money handed to people who aren't below any floor. The apparatus decides before the economics does.
Distribution as the question vs. measurement that manufactures the problem
“Poverty is not just a lack of money; it is not having the capability to realize one's full potential as a human being.”
— in the tradition of Amartya Sen and Peter Townsend; cf. A. B. Atkinson, Inequality: What Can Be Done?, 2015
This is the relative apparatus at full strength. Townsend's 1979 work and Atkinson's 2015 program both insist that in a developed society poverty is exclusion from a standard of living the society treats as normal — and that exclusion is fixed not by lifting a survival floor but by changing the distribution itself. The post-2008 revival of this tradition, with Piketty as its most-read node, is traced in History of Economic Thought Ch.17 (Modern pluralism): distribution returns to the center of the question, with Townsend and Atkinson named as the predecessors the post-2014 turn reactivated. Under this apparatus the rich-country poor are not a measurement artifact. They are the point.
“Most current poverty measurement ignores the great bulk of taxes and transfer payments. Once you count them, the great majority of what is called poverty in rich countries simply disappears.”
— the argument of Phil Gramm, Robert Ekelund & John Early, The Myth of American Inequality, 2022
The steelmanned pushback is not “distribution doesn't matter.” It is that the relative apparatus, applied carelessly, double-counts. Gramm, Ekelund, and Early argue that standard rich-country poverty statistics measure pre-transfer income and so book as “poor” households whose post-tax, post-benefit, in-kind consumption is far higher — the redistribution the relative apparatus prescribes is, on their accounting, already in the data and uncounted. The deeper claim is that a relative line can never reach zero by construction, so “poverty” defined relatively is a permanent statistic no policy can retire. The absolute-deficit advocate says: measure consumption, count the transfers, and much of the rich-country poverty crisis is a measurement convention, not a material fact.
Where this leaves us
The apparatus determines the answer. Under the absolute frame, poverty in the United States is approximately zero by global standards; under the relative frame, the US poverty rate is comparable to most of its OECD peers, around one in six. Both are honest accountings of the same households and the same income data. And the policy direction follows the apparatus, not the facts: floor-lifting interventions if you adopt the absolute frame, distribution-compressing interventions if you adopt the relative one. So the question in this walkthrough's title is not a riddle with a hidden answer. It is the choice — the methodological decision that determines which policy direction is even coherent before any data is consulted. Which raises the obvious next problem: the discipline uses both apparatuses. When does each one apply?
Two apparatuses, two policy directions, one discipline that deploys both. So the question becomes: in which contexts? And the hardest case of all — is the largest success story economics can point to, the billion lifted since 1990, read the same way under both lenses? Stage 3 is where the verdicts diverge most sharply, and where we have to face the question underneath all of it: are “rich-country poverty” and “\$2.15-a-day poverty” even the same thing?
Two apparatuses for two contexts (and the unresolved middle)
“The global middle class — mostly in Asia — gained enormously, the global top 1% gained enormously, and the people who lost out were the lower middle classes of the rich world. The same chart is a triumph and a crisis, depending on where you stand.”
— on the “elephant curve”; Branko Milanovic, Global Inequality, 2016
The elephant curve is the single most-argued-about chart in distribution economics — and it is the same data, read two ways. The body of the elephant is the billion the absolute apparatus celebrates. The collapsed trunk is the rich-world stagnation the relative apparatus indicts. The curve isn't ambiguous because the data is murky. It's two pictures because it answers two questions.
No new apparatus in this stage. The work is to lay the two we have against each other on the cases where they disagree, and then to face what the disagreement means. Start with the three boundary cases.
Rich-country poverty. Read through the absolute apparatus, a US household at the 5th percentile is rich — many times above the \$2.15 line, above the \$6.85 line, materially better off than almost everyone the great decline lifted. “Poverty” there looks like a measurement artifact. Read through the relative apparatus, US relative-poverty rates (around 17%) sit right alongside the OECD average, and the apparatus and the data agree the phenomenon is real. The verdict: in rich-country settings the relative apparatus is the appropriate tool, because the binding constraint genuinely is social participation, not survival. Lifting a floor no one is below is the wrong intervention shape.
Low-income-country poverty. Here the absolute apparatus is the right tool, and the relative one is a secondary diagnostic. The 1990–2020 lift is the discipline's largest empirical success; targeted transfers and capability provision demonstrably work; the RCT evidence validates the frame where survival is the binding constraint. The relative reading still has something true to say — within-country inequality rose in China, India, and Vietnam over exactly the period their headcounts fell — but in a setting where people are dying of want, distribution is the second question, not the first. The economic-history record of that lift sits in Economic History Ch.17 (China's reform and the Asian century).
Middle-income boundary cases. Brazil, South Africa, Mexico, Turkey. Both apparatuses bind, both generate predictions, and the literature uses both — often in the same paper. The verdict here is the honest one: the boundary is empirical, not categorical. There is no clean line in the development path where survival stops being the question and participation starts; the discipline applies whichever apparatus fits the binding constraint, and at the boundary both do.
Which leaves the question this whole walkthrough has been circling. Are “rich-country poverty” and “\$2.15-a-day poverty” the same phenomenon at all? The honest answer is: probably not. They are different binding-constraint regimes — one where the constraint is biological survival, one where it is social participation — and the word “poverty,” doing double duty across both, hides that the underlying mechanism is genuinely different. We are not measuring one thing badly in two places. We are measuring two things and naming them the same. The reframe's whole payoff is that this choice, normally invisible, is now in the open.
“It is the most powerful single image to come out of the study of global inequality — and it has been read as proof of globalization's success and proof of its failure, by people looking at exactly the same line.”
— on the elephant curve; Branko Milanovic, Global Inequality, 2016
Is the elephant curve a triumph or a crisis?
The same chart is the great-decline victory lap and the rich-world-stagnation indictment. People treat that as a debate about the data. It isn't. It's a debate about which question the chart is answering.
Different apparatus by development stage vs. one apparatus everywhere
“First you need to get people above the level where survival is the binding constraint. Distribution is the question that becomes urgent once that is done — and not before.”
— the development-economist's case; cf. Lant Pritchett and the Banerjee–Duflo program
This is the apparatus-by-context position argued at its strongest. The rich-country / low-country split is not a hedge or a political dodge; it tracks a real difference in the binding constraint. Where people are below subsistence, the absolute apparatus is correct and the relative one is premature — you cannot redistribute your way out of an aggregate shortfall of calories. Where survival is secured, the relative apparatus is correct and the absolute one is empty. The discipline has effectively resolved the choice context-by-context precisely because the contexts are genuinely different.
“The rich-world / poor-world split is a convenient story that lets the profession look away from the within-country distributions that are deteriorating fastest in exactly the places that are growing.”
— the distributional rejoinder; in the tradition of Piketty, Saez & Milanovic
The rejoinder is genuinely live. Distribution, this side argues, is not a luxury question that switches on at some income threshold — it bears on welfare at every development stage, and treating it as the rich world's concern is a political choice dressed as a methodological one. The evidence: within-country inequality rose sharply in the very economies whose headcounts were falling, and the dominant absolute apparatus had no slot to record it, so it went under-attended where it was worsening fastest. On this view the apparatus split doesn't track reality; it tracks whose poverty the discipline has decided to make visible.
Where this leaves us
Three things, in order. First: the two apparatuses are appropriate to different empirical settings, and the discipline has effectively resolved the choice context-by-context — absolute where survival is the binding constraint, relative where participation is, both at the middle-income boundary where both bind. That is a position, not a punt: it commits to which apparatus when, and to why the rich-country / low-country split is real rather than a hedge. Second: the “poverty problem” framing has been hegemonic in rich-country policy discourse where the “distribution problem” framing would be more accurate — UBI, wealth taxes, the post-2014 Piketty turn are all relative-framing-native, and arguing about them in absolute-framing language quietly confuses the policy. Third, and least resolved: whether “rich-country poverty” and “\$2.15-a-day poverty” are the same phenomenon at all — honest answer, probably not. The deepest payoff of the reframe is not a settled verdict on that third question. It is that the question is now visible, where standard exposition treats the choice of apparatus as a downstream technicality and hides exactly the decision that determines everything downstream of it. For the global-versus-within-country distinction at deeper policy depth, the sibling walkthrough Is inequality a problem economics can solve? (Stage 5) takes it further.
Where this leaves us
We started with two numbers that both have to be true: a billion people lifted across a survival floor since 1990, and a household at the bottom of Detroit — richer than nearly all of them — still counted as poor. The walkthrough's claim is that these are not in tension, because they are produced by two different apparatuses answering two different questions. The absolute apparatus measures distance from a fixed survival floor and prescribes lifting people across it; it is the right tool in low-income settings, and on its own terms it has won the largest welfare victory in modern history. The relative apparatus measures distance from the local median and prescribes compressing the distribution; it is the right tool in rich-country settings, where the binding constraint is participation, not survival. The disagreement between them is not about thresholds. It is about which apparatus, and that choice decides the policy direction before any data is read.
So the next time a poverty argument is put in front of you, ask the prior question before the policy question: which apparatus is this running on? If someone celebrates the great decline and someone else indicts rich-world stagnation, they are very likely both right and talking past each other, because they are measuring different things and calling both “poverty.” The deepest of the three verdicts is the one that stays open: rich-country poverty and dollar-a-day poverty are probably not the same phenomenon, and the discipline names them the same anyway. Seeing that — seeing the choice of lens that standard exposition keeps invisible — is what lets you tell which fight you are actually in.