What Do We Want Out of Poverty Measures?

Back in 1906, Helen Bosanquet, an economist and social reformer in the UK wrote:

“The doctrine of the ‘Poverty Line’ shows itself peculiarly elusive to examination… like the ray of light thrown by the moon across the sea, it shifts its position to meet the eye of the observer wherever he happens to place it.” Bosanquet nonetheless acknowledged that poverty lines were “gratefully accepted by a public anxious for exact and not too difficult science.”

That’s a fair description of the problems with and uses of poverty lines a century later. Much like calculations of gross national income, they’re both far more complex and far more arbitrary than they look. And yet they can be powerful metrics in public discourse and policymaking. So, it is surely worth trying to make them better. Three recent papers make the point that we need higher standards and more gradation in our views of global poverty than the World Bank’s extreme poverty line (now set at $3.00) allows, and suggest solutions. But I think they also demonstrate no single metric will achieve all of what we want from a poverty indicator.

Olivier Sterck proposes we should replace dollar a day with its reciprocal: how many days to consume a dollar? It is particularly nice in that unlike a dollars-per-person measure, it fits with inequality aversion and the declining utility of income. While there is a one-unit move from two to one days per dollar (or from 50 ¢/day to $1/day) there is only a half unit move from one to a half days per dollar (from $1 to $2). Something similar is used by the nonprofit Coefficient Giving (formerly known as Open Philanthropy) in its benefit cost analyses, and it is simply a good way to help thinking about where a dollar will likely do the most good to improve global welfare.

Lant Pritchett and Martina Viarengo propose a global upper-bound poverty line of $21.50. That number speaks powerfully to how much the extreme poverty line is not a measure of adequate income. One way of getting to the $21.50 number they present is by asking what is the income across countries at which you can be 90 percent confident of a household having electricity, access to improved sanitation used by the household alone, and an improved source of drinking water, and where no children have died under five or is malnourished, and where all children have completed primary schooling? It is a bare-bones definition and yet about ten times the extreme line.

Aart Kraay and colleagues (including Sterck) propose a decomposable distribution sensitive welfare index which combines features of the last two papers in that it is pretty much equivalent to “how many days to get to a global upper bound poverty line.” And depth of poverty below a high poverty line does seem a nice solution to the problem that no one thinks passing $1 or $3 a day is a measure of global development success, but many still agree focusing on the very poorest should be the most urgent priority.

The papers do lead to a question though: what are these measures for, exactly? Will they appeal to an audience “anxious for exact and not too difficult science?” Are they useful for policy? And there I’m left in some doubt, in part based on the experience with the dollar-a-day line.

The advantage of average number of days to earn a dollar is you don’t need a line. There are many problems with lines—not least, and as Pritchett and Viarengo note, nature doesn’t usually jump. The problem is people like a line. The extreme poverty line is a case in point. Look at Google’s corpus of books as tracked in its Ngram Viewer: since 1990, “$1 a day” (the original line set by the World Bank) and “global poverty” rose in lockstep in frequency of mentions into the 2000s. Global poverty meant $1 a day, and global poverty only became a popular thing to write about after the introduction of $1 a day.

And the line really is—at least apparently—simple. A dollar a day to live on, easily burying all the stuff about purchasing power parity and different survey approaches leading to different outcomes, and inflation, and family size, and income versus consumption, and a lot else somewhere under the hood. Compare the decomposable distribution sensitive welfare index: “Our measures are based on a simple intuitive “average factor by which incomes must increase to reach 𝑧” interpretation,” say the authors. It is a sign of their considerable smarts that they assume that such a measure will in fact be simple and intuitive to all, even if it is certainly more straightforward to interpret than the Watts Index and a very useful addition to poverty metrics.

And this isn’t just about the discourse: the US poverty line is still used in determining access to a lot of government programs for example, and it is hardly the only country where that is the case. Because nature doesn’t jump, sliding graduation usually makes a lot of sense (see the US Earned Income Tax Credit), but even then, practically, you need a starting and/or ending point for the graduation process. We can’t get away from lines.

That said, in fact, the global poverty line hasn’t been used in policy very much. For example, national poverty lines can help target welfare payment distribution. But once you have allowed for GNI per capita, there is no relationship between country poverty rates and foreign assistance receipts per capita. Institutions like the World Bank use national income thresholds—GNI per capita—to decide which countries get concessional finance, not household income thresholds. I think they’re right about that. For what it is worth, my preferred easily memorable, utterly arbitrary line as a tool to focus grant assistance is the low-income country cutoff of about $1,000 per capita GNI.

And that is a problem with higher lines in the global policy discussion: we shouldn’t be satisfied with a world where anyone is living under the current Danish poverty line. If it isn’t good enough for a Dane, it isn’t good enough for all. But that is a fairly useless tool for focusing the pittance we distribute every year in overseas development assistance, worth about 0.2 percent of developing country incomes. You need to use it where it will make the most difference, in the poorest countries home to the people who have the greatest factor by which their incomes must increase to reach the Danish poverty line. Those are also the countries where the trade preferences should be most generous, the World Trade Organization exceptions most expansive, the Basel banking rules most liberally applied, and so on. If China, where about 90 percent of the population live under the Danish poverty line, counts as a target country for this kind of special treatment, any such policy proposal is simply dead.

When measuring poverty, you can add dimensions, consider relative or absolute poverty, set high or low lines, use monthly or weekly recall in your surveys. The choices are almost endless and likely there isn’t one right answer to many if not most of them. So, when creating a poverty measure, it is probably helpful to ask “what is it for?” before you begin. There can be a strong rhetorical value in measures that help make the arguments that a high quality of life is associated with a lot more money than indicated by a low poverty line, and that poverty is not a binary condition. These papers demonstrate that. But policy practicalities will still demand lines, and sometimes lines that are relatively low.

A final note: the papers point out an important and welcome truth about progress. It wouldn’t be reasonable for a poverty basket in Helen Bosanquet’s day to have as a component of the calculation for your poverty line that you were 90 percent confident of a household having electricity and no children dying under five if you surpassed it. Because however privileged in however many domains you were, and wherever you lived, that wasn’t going to happen a century ago. But today we need a more ambitious poverty line and poverty measures in considerable part because we’ve made progress—even the extreme poverty line is getting more generous. Long may that continue.