1 of 21

How did poverty fall?

v20250106. Based on Armentano, Niehaus & Vogl, 2025. “How Poverty Fell.”

1

2 of 21

Perception…

Question: Over the last 20 years, the proportion of the world population living in extreme poverty has…

Almost doubled?
Remained more or less the same?
Almost halved?

Source: Ipsos Public Affairs survey conducted in 12 countries in 2017 on behalf of Gapminder. Sample reweighted to balance demographics. See here for details.

3 of 21

…versus reality

From 1981 to 2019, the share of the world’s population living in extreme poverty (less than $2.15 in 2017 PPP USD per capita) fell dramatically, from 44% to 9%

Source: Armentano, Niehaus & Vogl, 2024. “How Poverty Fell.” Underlying data from World Bank World Development Indicators.

4 of 21

What happened?

Did kids start life better-off than their parents? Or did adults make progress during their lifetimes? Or both?

How did people make progress? Did they switch from farming to factory jobs? Start their own businesses? Move to cities? Did women start doing paid work?

These are not questions about what caused poverty to fall, but they help guide our thinking about what causal relationships to focus on.

5 of 21

Countries

Five countries—China, India, Indonesia, Mexico, and South Africa—have high-quality data of both types available.

Collectively they accounted for 70% of poverty decline post-1990.

Source: Armentano, Niehaus & Vogl, 2024. “How Poverty Fell.” Underlying data from World Bank World Development Indicators.

Explain that high-quality data on households in or near extreme poverty is quite rare, so that if we want to study them we are always going to be somewhat limited by data availability.

Here we are fortunate that there are five countries that all meet data standards we define in the paper and which collectively account for the majority of poverty decline in the post-1990 period.

(The specific criteria we require are that a country have 3 or more waves of cross-sectional household surveys spanning 15 or more years, and 3 or more waves of panel household surveys spanning 9 or more years, all spanning large portions of the national population if not fully representative of it. We make a slight exception to include China even though details of the sampling design for some of its surveys are not available, so that we cannot know with certainty how representative they are, because China is such an important country for poverty decline.)

You could point out that in fact three of these countries--China, India, and Indonesia--account for the bulk of the poverty decline. Mexico and South Africa are much smaller contributors, but also interesting because their economies are different (more industrialized).

6 of 21

Kinds of data & what you can do with them

7 of 21

Kinds of data & what you can do with them

8 of 21

Kinds of data & what you can do with them

In contrast, in a panel design we track the exact same households over time. We may limit ourselves to tracking only the households that we sampled in the first wave, or we may add additional young households in subsequent waves, as depicted here.

This design is much harder - we have to track these people no matter what, even if they have moved to a different part of the country! What that looks like in practice is a lot of hard work trying to track down people who have moved, asking their neighbors where they went, trying to get phone numbers to get in contact with them, etc.

But the payoff if we do all that hard work is that now we can see how things changed over time for specific households, including households that exited poverty and any that entered poverty.

(This is probably a good place to pause and take questions to clear up any confusion.)

9 of 21

Data cleaning & harmonization

Data excerpt from the Indonesian Family Life Survey

We observe household 2031400 in all rounds, but household 2230600 only in rounds 1, 3, and 5

The survey stopped collecting information on food transfers received after round 1

Household 2230600 consumed a substantial amount of self-produced food despite not having a “farmbusiness.”

Here you can talk through the kinds of issues that researchers typically deal with when working from survey data, even from the very highest-quality surveys. (The examples here are from the Indonesia Family Life Survey).

The first callout simply points out that we see one of these households in all 5 rounds of the survey, but see the other in only 3. So that will limit how much we can say about changes in poverty over time for the latter household.

The second callout highlights an example of a way that the survey changed over time. In the first round, households were asked about the value of food transfers they received from other households. In subsequent rounds they were not. This means that when we want to construct measures of income or consumption we have to worry about how to do so in ways that keep the different rounds of the survey as comparable as possible. And of course you can imagine that once we start trying to make comparisons across different surveys-- one from Indonesia, say, versus one from China - we have to deal with many more of these discrepancies.

The third callout highlights something that just seems weird: this second household is recorded as consuming a good deal of food that it produced itself in all three rounds (val_food_selfprod), but also reports not having a farm in all three rounds (hh_farmbusiness). Why is that? Is it because they weren’t selling any of the food they made, so didn’t consider their farm to be a “farmbusiness”? How should we classify them in terms of whether they are working primarily in agriculture or not? These are some of the judgment calls we have to make.

10 of 21

What does it mean to be poor?

11 of 21

How do we measure who is poor?

In the U.S., the federal government defines poverty as having household pre-tax income below a threshold (e.g., in 2021 a family of four was considered poor if their income was below $21.37 per person per day)

In low-income countries, extreme poverty is usually defined as having a level of per-capita consumption below a threshold (e.g., $2.15 per person per day).

Question: why, do you think, do we often measure poverty differently in low-income countries?

The main issues to cover here are that (a) in low-income countries we generally do not have reliably data on income as we do in rich countries, where there is third-party reporting of wages, interest, capital gains, etc., so that we tend to think that we can more reliably measure consumption, and (b) people in low-income countries often have very volatile incomes (e.g. if they are farmers whose income depends on the weather, pests, ag prices, etc.) and try to smooth their consumption through various means -- so how much someone earns in a given year may not give us a very accurate picture of their actual standard of living.

For these reasons, development economists usually measure poverty in terms of people’s consumption, rather than their income.

For our purposes in this project, it will also be useful to look at consumption. But for some questions we will also need to look at income - if we want to know how people’s sources of incomes changed over time, for example, we can’t get that from consumption. So we are going to look at both.

12 of 21

Income versus consumption poverty

Source: Armentano, Niehaus & Vogl, 2024. “How Poverty Fell.”

13 of 21

Poverty rates over age, by survey round

Source: Armentano, Niehaus & Vogl, 2024. “How Poverty Fell.”

Now we’re going to look at the same data, but separated out by the age of the household head.

The way this graph works is that the y-axis is the poverty rate, the x-axis is the age of the household head, and each grey line represents a different survey round.

From survey round to survey round poverty rates tend to fall - that’s easier to see for China, India, and Indonesia, though it’s also true for Mexico and South Africa.

The other thing that’s interesting about the graphs is that any one of the grey lines looks pretty flat - it’s not the case that younger people tend to be poorer than older people, or vice versa. It seems more like people of all ages are making progress getting out of poverty in parallel.

It looks like maybe there is an exception for young people in Indonesia, but otherwise these all look pretty level. (If people are curious, that upward slope among people in their 20s in Indonesia is driven by selection into household headship - if you look at individuals rather than heads then even that portion of the graph becomes flat.)

14 of 21

Linking people in the same generation

Source: Armentano, Niehaus & Vogl, 2024. “How Poverty Fell.”

Now we’re going to look at the same data, but connect people who are part of the same cohort -- that is, people who were born in the same year -- so that you can see what is happening to them over time. That’s what the colored lines in this picture do.

This is a little harder to see for Mexico and South Africa, because the series are all on top of each other, and a bit easier to see for the other countries, so you can focus on those if you like.

What you can see here is that for every cohort, people are getting substantially less poor over the course of their lifetimes. And that’s one of the key takeaways. Remember, it could have been that people who start out poor largely stay poor, but then their kids are less likely to be poor because they are able to get access to better opportunities, education, etc. What we’re seeing here is that it’s true that younger generations are starting life out less poor than older ones, but that the older ones are also getting progressively less poor over the course of their lives.

One implication is that it will be useful to study those people and try to understand, what actually happened during their course of their lives as they moved out of poverty?

15 of 21

Interpretation

Kids did start their adult lives substantially less poor than their parents had…

…but adults continued to make substantial progress out over poverty over the course of their lives, at a similar rate…

…so that in aggregate the majority of poverty decline was accounted for by intra-generational rather than inter-generational progress.

16 of 21

Poverty dynamics

Question: which do you think better describes what happened as poverty fell?

Some people escaped poverty, but many stayed stuck in a “poverty trap”

Many people escaped poverty, but some also fell back into it

17 of 21

Many people exited, some fell back in

Source: Armentano, Niehaus & Vogl, 2024. “How Poverty Fell.”

Indonesia

The answer is (b), the main issue is that a lot of people who had escaped poverty fell back into it.

This pictures are one way of seeing that. We’re showing you all five countries, but you can focus if you like on Indonesia which we’ve made larger so its easier to see. The same basic things that are true of Indonesia are true of the other countries as well.

The way this graph -- works is that on the left hand side we’ve sorted households based on how poor they were at the start of the panel survey - the top group were living below the poverty line, the next group was living at between 1-2x the poverty line, the third group at 2x the poverty line or more.

And then the flows showed were people moved. So for example you can see that a lot of people who started out poor exited, to either the 1-2x or even the >2x category.

But what you also see is that a fair number of people who were initially in one of the above-poverty groups fell back into poverty. That pattern of “churn” is very consistent in our data.

And that’s important because it changes how you think about the problem of what we can do to accelerate poverty reduction. If you thought the main issue was that people stay trapped in poverty, you’d focus on how to get them out of those traps. But if the main issue is that a lot of people make some progress and then fall back into it, then you’d focus on what we can do to protect people from falling back in.

For example, a typical setback that a household might hit might be when the primary breadwinner gets sick. They lose that income for a while. Maybe they have to pay for medical treatment, and maybe to afford that they have to sell off some assets, some livestock, some land, etc. And maybe even then it isn’t enough and the sick person dies - then the household has been hit with a really big setback, on top of which they will need to pay for a funeral and burial.

18 of 21

What changed?

Question: which group do you think accounted for a majority of poverty decline:

People who changed sectors (e.g. farming to non-farm)
People who migrated
People who switched occupations (wage- versus self-employment)
People in households where women entered or exited the workforce

Next we’re going to move to studying all those people in the flow diagram that did move out of poverty.

I want you think about what might have changed during the course of their lives that might explain it.

For each of these options, raise your hand if you think that the group it names accounted for a majority of poverty exits.

Note that more than one of these statements could be true, so you can raise your hand for as many of them as you think are true in the data.

(Go through one by one.)

The answer actually turns out to be that none of these are true. In all of these cases, there were some people who exited poverty - some who exited poverty while changing sector, while migrating, while switching occupations, etc. But in no case did a majority of those who exited poverty make the corresponding change. So actually there is a sense in which many people exited poverty while continuing to largely do what they had been doing.

19 of 21

Example: (not) leaving the farm

Source: Armentano, Niehaus & Vogl, 2024. “How Poverty Fell.”

Here is an example. This graph focuses on whether people worked primarily in agriculture or in a non-agriculture sector. Traditionally, economists see moving out of agriculture and into manufacturing or services as a big part of the process of development.

The way the graph works is that the width of each bar tells you the share of the population in each group. So for example the width of the dark orange bar tells you the share of the population that started in ag and ended in ag.

And the height of each bar tells you the next exit rate for that group. So for example, for China, we see that the rate at which households exited poverty was actually highest for those that stayed in the ag sector.

In fact what we see is that in all countries, people who transitioned out of agriculture (the light orange bars) accounted for less than half of total poverty reduction.

That’s true even though in two countries - India and Indonesia - they had the highest rates of net poverty exit. The issue was that there just weren’t that many households that switched. So one way to read that would be that switching out of agriculture was a good strategy in those places, but there are barriers to doing so. But notice that even in those countries the net exit rate was almost as high amount households that stayed in ag as among those that switched out of it.

Notice also that in Mexico and S Africa these are middle-income countries and you see most of the workforce is already in non-ag.

20 of 21

Outside help

Source: Armentano, Niehaus & Vogl, 2024. “How Poverty Fell.”

Another thing we can do is look at the sources of households income, splitting them into those that exited poverty (the solid bars here) and those that entered it (the hollow bars).

Here I want to draw your attention in particular to the last bar in each series, the share of income people received from transfers. Transfers means both transfers from government, e.g. anti-poverty programs, as well as potentially from friends and family if they help you out during hard times.

If you thought a major reason people made progress out of poverty was because these kinds of transfers became more generous, because governments or aid agencies started giving people more, then you would expect to see the share of income from transfers rise a lot among households that exited.

In fact you see that it rose a little in some cases, fell a little in some cases, but was never really a big change with one exception which is Mexico, where it fell a lot for households that exited poverty.

In contrast, you can also see that the transfer share always rose more for households that entered poverty than for those that exited it.

So what this looks like is less like big new programs lifting people out of poverty, and more likely safety nets doing what they are supposed to do - when people hit a setback so that they fall into poverty, this is at least partially offset by getting more generous help from others. So in that sense transfers may have helped keep more people from falling back into poverty. But they aren’t what’s driving all of those exits from poverty in the first place.

21 of 21

People living in extreme poverty have made enormous progress getting out of it in recent decades

Younger generations did begin life less poor than older ones, but older generations also made progress out of poverty at the same rate

The main reason progress was not even faster was not that many people stayed “stuck” in poverty, but rather that some fell back into it

People most often made progress out of poverty “in situ,” without changing the kind of work they did, where they lived, etc. or getting meaningfully more “outside” help.