TL;DR: In case you came in late… progress is possible! The number of people living in extreme poverty – both as a percentage of global population and in terms of the absolute number of people – has fallen a lot. Long-term data from five key countries suggests that this happens gradually (rather than in big generational jumps). There are many pathways out of poverty (including, sometimes, people not really changing much about how they live), and no silver bullets.
One of humanity’s great recent achievements has been the reduction in the share of the global population living in extreme poverty, from an estimated 41% in 1981, to 36% in 1990, to 9% in 2015. Development economists have made great strides toward measuring the decline of poverty (Chen and Ravallion, 2010) and characterizing the lives of people who remain poor in spite of it (Banerjee and Duflo, 2007). But systematic descriptions of how it happened—of the anatomy of poverty decline—are rare. This paper aims to provide a description that is as comprehensive as the available data will allow.
By “how,” we mean in part what happened to the livelihoods of individuals and households as they moved out of (and often, as we will see, back into) extreme poverty. Did they find work in a factory? Start their own business? Move to a city? And so on. We will also emphasize that changes like these within the life of any one person, or cohort or generation, may be more or less important than changes across cohorts or generations, as the less poor young replace their poorer parents and grandparents in the population. So our initial concern will be to understand the demographic structure of poverty decline.
We focus on five countries—China, India, Indonesia, Mexico, and South Africa—that collectively accounted for 75% of global poverty decline during 1990-2015. We include these countries because they have available both (a) 3 or more waves of cross-sectional household surveys spanning 15 or more years, and (b) 3 or more waves of panel household surveys spanning 9 or more years, all spanning large portions of the national population and most fully representative of it. They also provide a degree of global representativeness in the sense that they span the major regions of the developing world, and—in the case of China, India, and Indonesia—include some of the most important contributors to global poverty decline.
…
The data have many imperfections, certainly relative to data from wealthier countries… But they are generally high quality by the standards pertinent to work on extreme poverty and thus, in our view, afford a valuable opportunity to better understand a crucial episode in human history. We study poverty in the now-conventional sense of living on less than $2.15 per day in 2017 PPP dollars (World Bank, 2023), using both consumption- and income-based measures as available.…
We use the repeated cross-sectional data to investigate how the pace of poverty decline within cohorts compares with that between cohorts. This exercise is in part motivated by recent results from Porzio et al. (2020) suggesting that cohort effects play an important role in structural change during the development process. We confirm that as poverty has fallen in aggregate, successive birth cohorts have entered adulthood at progressively lower poverty rates. However, we also find that the pace of this between-cohort process roughly matches the within-cohort pace of poverty decline. Poverty rates are similar across the age distribution at a point in time; aggregate poverty decline manifests in downward parallel shifts of a flat cross-sectional age profile. This fact holds true whether we use consumption or income, where available, as well as whether we focus on household heads, as is common practice in the economics literature, or take the whole population into account.
The downward march of flat cross-sectional age profiles has two noteworthy implications. First, changes in the age structure of the population—aging, for example—cannot account for poverty decline. With flat cross-sectional age profiles, the aggregate poverty rate is invariant to the age structure, at least in an accounting sense. Second, most poverty decline occurs over people’s lives, rather than through successive cohort replacement. Cohorts are interchangeable in their poverty rates at a point in time, so cohort replacement plays little role in the trajectory of poverty rates.…
We first document that progress out of poverty, while substantial, was not irreversible. Many households that were initially poor exited poverty, but many that were not poor at some point subsequently entered it. The share exiting poverty conditional on starting in it ranged from 45% in Rural India to as high as 67% in China. But the share entering poverty conditional on starting non- poor was also substantial, ranging from 13% in Indonesia to as high as 34% in Mexico.
…We then examine changes in the livelihoods of households that exited or entered poverty—the activities in which they engaged, and the income they obtained from them. Overall, no one pathway out of poverty predominates. For almost all countries and dimensions of activity, only a minority of the households that exited poverty did so while changing their primary means of earning a living. And households that entered poverty often changed their activities in similar ways. For example, the share of poverty-entering households that switched from agriculture to non-agriculture is 60% – 100% of the corresponding share of poverty-exiting households. One important exception, as we will shortly note, applies in the more advanced economies of Mexico and South Africa, where transitions to wage work played a predominant role.
With respect to sector, transitions out of agriculture played a limited role. They did not account for the largest share, let alone the majority, of transitions out of poverty in any country. In fact, an accounting decomposition credits transitions into agriculture with a meaningful share of poverty reduction in most countries. More generally, in every country, households that stayed in the same sector contributed more to poverty decline than households that changed sector.…
Rural-to-urban migration, in the three countries (Mexico, Indonesia and South Africa) where the panel surveys tracked migrants, similarly accounted for a limited share of poverty decline. These migrants did exit poverty at relatively high rates in Mexico and South Africa, but in Indonesia they did worse than those that did not move, and in all countries they were rare. Consequently, only between 0% (Mexico) and 11% (Indonesia) of households that exited poverty did so while moving from a rural to an urban location. Notably, however, rural-to-rural migrants accounted for a third of all net poverty decline in Indonesia.…
With respect to occupational choice, patterns are noticeably different in the less-developed economies (China, India and Indonesia) relative to the more-developed ones (Mexico and South Africa). In the former group, those who stayed or became self-employed contributed the most to poverty reduction: reminiscent of, for example, the occupational hierarchy induced by credit constraints in Banerjee and Newman (1993). In the latter group, on the other hand, transitions within and into wage work account for the bulk of poverty decline.
…
Finally, the panel data shed light on the role of redistribution. Redistribution could in principle contribute to poverty decline in two distinct ways: indirectly, by stimulating growth in income from other sources (as for instance Best et al. (2026) find in Brazil), or directly if it becomes more generous over time. In the data, we see that transfers (both public and private) played a largely“protective” role: they increased as a share of income for households that entered poverty, and decreased (or increased only slightly) for those that exited it, but did not become meaningfully more generous over time. India is a notable exception in this regard, however: in India, the gap between pre- and post-transfer poverty rates widened meaningfully over time, as public transfer generosity increased.14 One can thus see in the micro-data India’s exceptionalism as a country that chose generous redistribution—which Kuznets (1955) and others have associated with later stages of development—relatively early (Muralidharan, 2024).
…
… it may be fruitful to consider the relationship between the descriptive facts here and causal work, and in particular which causal work to prioritize in the future.We think of this issue as follows. If households were always able to do what is in their best interest, then neither descriptive nor causal work would be useful for improving households’ lives. But households living in or near poverty are often unable to make the decisions that would be best for them because they lack crucial information, cannot access well-functioning capital markets, or face various other hurdles and constraints. Viewed through this economic lens, descriptive results yield implications of (at least) three sorts.
First, observing people making meaningful progress while adopting a given approach rejects the extreme view that it is simply not possible. While such views are perhaps less common today, the history of thought about development contains some well-known examples. Analysts worried that certain economic pathways—subsistence agriculture, for example (Lewis, 1954; Rostow, 1991), or specialization in primary products (Prebisch, 1950)—were effectively dead ends that could not lead to further meaningful development. If nothing else, results documenting widespread progress out of poverty within sectors suggest that experimental work seeking to accelerate that progress need not be a fool’s errand.
Second, observing rapid progress by people who adopted a given strategy even while few others did likewise suggests that (financial) market failures bind.* The very high poverty exit rate of the few South African households that moved from rural to urban locations, for example, could be read in this way. Other explanations may also apply: heterogeneity (in returns, preferences, or unobserved costs) could yield this pattern in the absence of frictions. But observing it shifts our posteriors towards the existence of meaningful frictions—frictions that experimental work could then seek to relax in the specific contexts where observational returns are high.
Third, observing people making progress without making any visible change—as we often do here—suggests a few possibilities. We may simply be looking at the wrong decisions. We have emphasized decisions that are widely measured and that development economics has, in our reading, tended to emphasize, but others may be relevant. In a related vein, people may have escaped poverty thanks to a multitude of small changes, not one big one. For example, the key may be not whether one leaves agriculture for manufacturing, but rather the composite benefits of changes to crop mix, input use, planting technique, and so on. This reading calls to mind Banerjee and Duflo’s (2011) concluding argument that “small changes can have big effects,” proposing a research agenda devoted to finding many small but reliable improvements. And general equilibrium effects may have been important.
Vincent J. Armentano, Paul Niehaus, and Tom Vogl – How Poverty Fell
*Market failures bind: Here’s Claude on what that means:
If moving to cities is so effective at reducing poverty, why aren’t more people doing it? The most natural economic answer is that something is stopping them — a binding constraint. Perhaps they can’t afford the upfront costs of moving, can’t access credit to finance the transition, lack information about urban job prospects, or face housing market barriers
See also:
Little By Little 1: Raymond Briggs on 1940s Britain
Little By Little 2: Li Kunwa on indoor plumbing in 1980s China
Little By Little 3: Scarcity and Subsistence in rural Suffolk in the 1900s
Little By Little 4: Dolly Parton on the scar on her toes (rural poverty in 1950s Tennessee)
Not long ago; or, Little by little (5): Tyler Cowen on progress in living standards since 1900
Not long ago; or, Little by little (6): New York Tenement, 1889
Not long ago; or, Little by little (7): Fred Dibnah, Steeplejack – health and safety edition
Not long ago; or, Little by little (8): Singapore market, early 1970s
Not long ago; or, Little by little (9): Car ownership in China 1999-2019
Not long ago; or, Little by little (10): Switzerland’s iodine revolution
Not long ago; or, Little by little (11): Hot feet on hot days