Poverty — By the Numbers?

Global economic policymakers like to give grandiose speeches about helping "the poor." But how many people are really "poor"? Like all attempts at economic measurement, that question is difficult to answer. In a conversation with Prakash Loungani, Angus Deaton — a Princeton University professor who has studied Indian poverty statistics intensively — explores the difficulties.

Loungani: The World Bank's estimate that 1.2 billion people live on less than a dollar a day is cited everywhere. How reliable is this estimate?

Deaton: There's surely a very large margin of error in the estimate. Even small changes in the design of a survey can often have dramatic impacts on the poverty estimates.

For instance, you could lower the estimate of the number of poor in India by 175 million just by shortening the "recall period."

Tell me how that would work.

To measure poverty, you have to survey people. The surveyor asks them to recall their expenditures — how much they spend on food, on clothing and so forth.

Evidently, the surveyor has to chose: Over what period do you ask people to recall their expenditures? Over the past week? Over the past month? That's the recall period.

What happens when you extend the recall period?

Choosing a one-week recall period generally yields higher expenditures — and therefore lower rates of poverty. Conversely, if you choose a one-month recall period, people tend to forget some of the things they have purchased.

What about India's experience?

India has long used a 30-day recall period. In recent years, the statistical authorities in India experimented to see what difference the recall period makes to the estimate of the number of poor.

They found that shifting to a one week recall period would essentially halve the number of poor in India. That must be the most successful poverty-reduction program in the world!

Beyond the data problems, isn't it true that we have seen more poverty reduction in recent decades?

I have found that the number of people living in poverty has declined at a steady rate over the last 20 to 30 years. But at the same time, there is no evidence of a pick-up in the rate of decline since the reforms of the 1990s.

I estimate India's poverty rate at 28% in 2000. Scholars at the Delhi School of Economics — working independently and using methods quite different from mine — have reached similar conclusions.

Why are India's poverty statistics a matter of global concern?

The answer is simple. India accounts for about a third of the world's poor. So coming up with a more reliable estimate of India's poor goes a long way towards getting a better estimate of the world's poverty rate.

Why did you initially suspect that there was a problem with the Indian poverty statistics?

According to India's national income accounts, the country has had robust economic growth over the last decade.

However, when the national survey statistics — which are the source of the poverty estimates — came out, they showed that average consumption has essentially been flat over the last decade. These two stories about what has happened in India cannot both be right.

What are the types of statistical problems that might cause the usual poverty statistics to be misleading?

In order to compare poverty rates across countries — that is, to make the kind of $1 a day numbers that are cited everywhere — you have to use purchasing power parity (PPP) exchange rates.

Revisions to those PPP exchange rates — which are common — can wreak havoc with the poverty estimates.

The World Bank has been caught in this trap. In the 1997 World Development Report, before the Asian financial crisis, Thailand is shown as having a poverty rate of only 0.1% of the population.

How seriously was that number treated?

That figure was attributed at the time by then chief economist Joe Stiglitz to the positive force of the Asian economic miracle. In fact, any one familiar with Thailand would not have believed that the poverty rate was that low.

Essentially, Thailand's poverty rate was less a demonstration of the Asian miracle than of the dangers of inappropriate exchange rate conversion.

It's a bit disconcerting when the World Bank's dream of a world free of poverty can be realized simply by misusing exchange rate data.

Other measurement problems involve comparing poverty rates in regions within the same country — particularly between the urban and rural areas.

Countries often have good data for urban centers, but not for the countryside. The countryside, however, is often where most of the poor live. This can turn into a big problem.

Does this matter in practice?

I think the unavailability of good price indices for rural areas is in part responsible for the very conflicting views of what impact the Asian crisis had on the poor, say, in Indonesia.

If the poverty data are so error-ridden, why don't we rely on other — directly measured — socio-economic indicators?

We do. We have plenty of statistics on life expectancy, infant mortality, literacy — these are all things that people look at in order to supplement the poverty numbers.

Amartya Sen — recipient of the 1998 Nobel Prize in economics — has been the intellectual force behind this broader look at deprivation.

The United Nations Development Program has come up with a Human Development Indicator that aggregates all this information in a certain way.

Should we then just ignore the poverty numbers altogether?

No, that's clearly going too far. We do have a notion of poverty — like we have a notion of being cold or being hot. People can generally identify who in their community is poor.

But it's one thing to have a rough notion of poverty in your community — and quite another to come up with an estimate of the number of poor in the whole developing world.

What is the problem then?

I object to the pretense that at the end of this series of decisions, we can draw a very sharp cut-off, namely a numerically-defined poverty line.

Doing so would encourage a rather "Mr. Micawber"-ish view of things. Remember Charles Dickens’ famous character from David Cooperfield?

It was Mr. Micawber who claimed that "Annual income twenty pounds, annual expenditure nineteen pounds six, result happiness. Annual income twenty pounds, annual expenditure twenty ought and six, result misery."

In other words, happiness is on one side of the numeric line and misery on the other.

This is simply misleading. We should admit that the poverty numbers have large margins of error, but keep working to improve them.

What institutional changes are needed to get some quality control on the poverty numbers?

Helping countries resolve statistical issues is something that the World Bank and the International Monetary Fund should do a lot more of.

That may be difficult for the IMF — since some still call on it to leave the "poverty business" altogether.

I'm in favor of the IMF staying in the poverty business, within limits. I was persuaded by First Deputy Managing Director Stanley Fischer's remarks on why poverty is central to the Fund's mission.

He said that the IMF cannot use the "Von Braun defense" — "I just put the rockets up and it's someone else's business where they fall" — to keep out of poverty.

What are some areas that the IMF could focus on?

The IMF also had a long-standing interest in accurate price indices — because of the need to get accurate measures of real monetary aggregates, real exchange rates and the like.

The IMF these days actually issues guidelines on how to provide macro data and assess its quality. I think that should be extended to poverty data.

Note: This interview was adapted from the IMF Survey (Vol. 31, No. 13, July 8 2002).