Price Indexes, Inequality, and the Measurement of World Poverty

TL;DR: In this article, the authors discuss the measurement of world poverty and inequality, with particular attention to the role of purchasing power parity (PPP) price indexes from the International Comparison Project.

Abstract: I discuss the measurement of world poverty and inequality, with particular attention to the role of purchasing power parity (PPP) price indexes from the International Comparison Project. Global inequality increased with the latest revision of the ICP, and this reduced the global poverty line relative to the US dollar. The recent large increase of nearly half a billion poor people came from an inappropriate updating of the global poverty line, not from the ICP revisions. Even so, PPP comparisons between widely different countries rest on weak theoretical and empirical foundations. I argue for wider use of self-reports from international monitoring surveys, and for a global poverty line that is truly denominated in US dollars.

Summary

I. Global poverty and global inequality

• Many countries have their own national poverty lines and poverty headcounts.
• Indeed, it is the 1993 revision that "established" sub-Saharan Africa as the region with the highest headcount ratio, a fact that has dominated subsequent discussions of world poverty; prior to revision, the measured prevalence of poverty in South Asia was substantially higher than in sub-Saharan Africa.
• The bottom two graphs use data from the Penn World Table (PWT), and compare version 5.6, which used PPPs referenced on 1985, with version 6.2, which uses PPPs referenced on 1993.

II. The construction of international purchasing power parity exchange rates

• In order to understand why the ICP has such a profound effect on their perceptions of global poverty and inequality, the authors need to understand something about how it works.
• In the 2005 ICP, the "gluing together" of the regions was accomplished using a third stage, the "ring," in which 18 strategically chosen countries, at least two per region, were asked to price a new, common detailed list of more than 1,100 items.
• Another problematic issue here is that smoked bonga (or indeed Kellogg's cornflakes, a more important example) may be available, but is rarely eaten in country c, so that it is only stocked in specialty shops at very high prices.
• These "prices," together with the regional aggregates of expenditures on each basic head expressed in regional numeraire currency, are used in a global multilateral aggregation to give regional PPPs, one index for each region, that allow the within-region PPPs to be scaled up to global PPPs.
• These weights are appropriate for national income accounting purposes, but do not reflect the consumption patterns of people who are poor by global standards.

III. Global poverty and the consumption patterns of the poor

• To turn the standard PPPs into poverty-weighted PPPs, or P4s, it is necessary to weight the multilateral indexes, not with the expenditure patterns from the national accounts, but with the expenditure patterns of people in each country who live around the global poverty line, patterns that can only come from household survey data.
• DD use India as base so that, for example, a figure of 1.692 for Nigeria shows that an Indian converting naira prices into rupees at the official exchange rate would find that consumption prices in Nigeria are 1.692 times consumption prices in India.
• The main message here is that the way in which the ICP prices and expenditures are aggregated, and the inconsistencies and differences in definition between national accounts and survey estimates of expenditure patterns are as, or more important in affecting the estimates of consumption PPPs than whether or not they are adapted to the expenditure weights of the global poor.

IV. Poverty lines and poverty counts

• Most of the increase in the world poverty count with the revision in the ICP can be attributed to two main factors only one of which, the treatment of housing rental, is directly attributable to the ICP itself.
• Both of the factors are of general interest.
• I deal with each issue in turn, starting with the least important.

A. Actual and imputed rent for housing

• Housing rental, including imputed rental for owner occupiers, is a difficult ("comparison resistant") area for the ICP, see Deaton and Heston (2010) .
• For the African and Asian regions, the 2005 ICP had to fall back on an imputation.
• Because the ICP is primarily focused on obtaining "volume" measures of GDP in international currency, it was decided to impute rental by assuming that, for countries in Asia and Africa, the volume of rental was a fixed proportion of GDP.
• What is neutral for quantities is not neutral for prices.
• For countries that make little or no allowance for owner-occupier rentals in their national accounts, this "imputed" parity will be very low.

B. Setting the global poverty line

• Since the first dollar-a-day poverty counts in 1990, the line has been set by taking the national poverty lines of a group of the poorest countries in the world, converting them to international dollars using PPPs from the ICP, and taking a simple average.
• Consider first India, which has a low poverty line relative to its living standards, and suppose that India has recently moved across the line from left to right.
• There is a further drop to R18.05 once the authors exclude three basic headings that are not covered by household surveys, FISIM (financial intermediation services indirectly measured, the profits of banks and insurance companies), prostitution, and actual and implicit rental for housing.
• Instead, Deaton and Dupriez (2009, Table 13 ) calculate "star" PPPs comparing the US with each of the 62 countries they use, with each country's currency first converted into international rupees using the P4s.

V. Quality and inequality around the world

• (The negative relationship would be even steeper using the old estimates of per capita GDP, according to which the poorest countries were richer than shown.).
• The figure also shows that the largest revisions are in Africa and Asia; the average ratios of new to old are 1.42 for Africa and 1.33 for Asia-Pacific.
• Since Africa and Asia were linked to the other regions using regional PPPs-one for each region-calculated from the ring using Diewert's method-these regional PPPs are a natural point of investigation for investigating the increase in inequality.
• Without the upward revisions to the African and Asia-Pacific PPPs, there would have been no increase in measured inequality.

A. Country coverage, imputations, and inequality

• The 2005 ICP has price data for 146 countries, many of which were imputed in earlier rounds, either by comparison with countries at similar level of development, or by updating old data, or some combination of the two.
• (Other variables are also included, see Changqing Sun and Eric Swanson (2009) and Deaton and Heston (2010) for more details.).
• These imputations target the logarithm of PPP, or equivalently the logarithm of per capita GDP in international dollars, and deliver unbiased estimates in ideal conditions.
• The imputed values will have less cross-country variations than the unobserved true values.
• Over successive rounds of the ICP, as fewer PPPs are imputed, and more actually measured, measured inequality will rise.

B. Comparisons between rich and poor countries in the ring

• The most important question for international comparisons of prices is how to make useful comparisons across widely different countries, countries in different regions of the world, at different levels of per capita GDP, and with different patterns of consumption and of relative prices.
• The most influential basic headings are mostly those that would be expected.
• When tastes are identical but not homothetic, the superlative index approximates the cost of living for a level of living intermediate between the base and the comparison which, when the authors are comparing Cameroon or Zambia, with Britain or Japan, is not obviously helpful in the sense that it over-weights the cost of living in Cameroon and Zambia.

C. Tectonic regional PPPs

• Not surprisingly, some of the same goods in Table 7 reappear in Table 8 , because their prices are high throughout Africa, but several of the basic headings not included in the ring play a large part in shaping the regional PPPs.
• That pharmaceuticals and motor cars are more or less offset by housing in Africa hardly builds confidence in the numbers, though it is hard to see any evidence in Table 8 that would support the idea that prices in Africa are systematically overstated.
• If the authors make a rough correction for all of these issues, raising Asia/Pacific and Africa by 5.9 percent, then China by 20 percent (urban bias plus single bias from regional calculation), and India by 15 percent (same reasons as China), the Gini coefficient falls from 0.543 to 0.524.
• What is disconcerting is not the inability to find a cause or group of causes for the increase in measured inequality, but rather the sense that all of the intercontinental comparisons are fragile, and can easily be disturbed by factors that the authors do not know how to handle.

VI. Why don't we just ask people?

• I now return to the main theme, the question of poverty measurement.
• Poverty counts can also be very sensitive to survey design.
• The survey based counts cannot satisfy another demand, which is for annual monitoring of world poverty.
• So there is something to be said for directly asking people around the world how their lives are going, whether they have enough, or whether they are in financial difficulty, and in cases where there are reliable income data, turning those reports into poverty lines.
• In Deaton (2008) , I used the 2006 data to show that there was a close, and approximately linear, relationship between the national average value of the ladder and the logarithm of per capita GDP in international dollars, and Betsy Stevenson and Justin Wolfers (2008) subsequently showed that the relationship also holds within countries using the WP income data.

VII. Conclusion

• Particularly in my discussion of international inequality, I emphasized the difficulties of making comparisons of real income between countries, especially countries with very different relative prices and economic structures.
• If the Indo-centric approach is not acceptable, the world poverty line could remain, as now, an average of poverty lines from poor countries although, once again, those poverty lines need not and should not be changed over time; this is what was done until the current revision.
• The "surveys" column replicates this calculation with survey expenditure weights in place of national accounts weights, and the "poverty weights" uses a poverty weighted PPP with expenditure weights around a global poverty line that is the poverty-weighted average of the international rupee value of 50 national poverty lines.

Figures

Table 6: Alternative global poverty counts and poverty lines

Table 1: Poverty headcount ratios

Table 7: Most influential basic headings in pairwise Törnqvist indexes: ring African countries versus UK

Figure 1: Gini coefficients for population weighted national incomes

Table 10: Global and regional tracking of living standards, 2006 to 2009

Figure 5: Fractions of people that they are dissatisfied with their standard of living

Table 2: Root mean square distance of prices of consumption over 62 countries

Figure 4: Prices of consumption relative to GDP per capita, 18 ring countries

Table 9: Correlations between log GDP per capita and well-being measures from the Gallup World Poll

Table 4: Ratio of parity for actual and imputed rental to parity for household individual consumption, selected countries 2005

Table 3: Price of consumption in countries with largest numbers of poor people, 2005

Figure 3: Ratios of old to new PPPs for GDP as a function of per capita GDP

Table 5: Global poverty lines in world rupees per person per day

Table 8: Most influential basic headings in pairwise Törnqvist indexes: Africa and Asia versus OECD

Figure 2: The construction of the global poverty line

Full text

Price indexes, inequality, and the measurement of world poverty
Angus Deaton, Princeton University
January 10
th
, 2010
ABSTRACT
I discuss the statistical basis for measures of world poverty and inequality, with particular
attention to the role of purchasing power parity price indexes from the International
Comparison Project. Global inequality increased sharply with the latest revision of the ICP,
but the recent large increase in global poverty came from an inappropriate updating of the
global line, not from the ICP revision. ICP comparisons between widely different countries
rest on weak empirical and theoretical foundations. I argue for greater use of self-reports in
international monitoring surveys, and for a global poverty line that is denominated in US
dollars.
Presidential Address, American Economic Association, Atlanta, January 2010. I am grateful
to Olivier Dupriez, Alan Heston and Sam Schulhofer-Wohl who have collaborated with me
on related work, to them and to Erwin Diewert, Yuri Dikhanov, D. S. Prasada-Rao, and Fred
Vogel who have over the years taught me about international price comparisons, as well as
the Development Economics Data Group at the World Bank for supplying me with data and
for their patience and help with my questions. I also thank Tony Atkinson, Tim Besley,
Branko Milanovic, François Bourguignon, Anne Case, Olivier Dupriez, Bill Easterly, Alan
Heston and Martin Ravallion for comments on partial drafts. The views expressed here are my
own.

1
This lecture is about counting the number of poor people in the world. If we ask people
whether they consider themselves to be poor, or how much money someone would need to get
by in their community, they appear to have little difficulty in replying. Beyond the local level,
many countries, including the United States, regularly publish national counts of the number
of people in poverty, and while these numbers and the procedures for calculating them are
contested and debated, the estimates typically carry sufficient legitimacy to support policy,
see Connie Citro and Robert Michael (1995) for the history in the US, and Rebecca Blank and
Mark Greenburg (2008) for a recent proposal for reform. But once we try to calculate the
number of poor people in the world, matters are more complicated. The World Bank’s global
poverty count, which started with Montek Ahluwalia, Nicholas Carter and Hollis Chenery
(1979), and which became the dollar-a-day count in the World Development Report 1990, was
incorporated into international policymaking and discussion via the Millennium Development
Goals (MDGs)—the first of which is to “halve, between 1990 and 2015, the proportion of
people whose income is less than $1 a day.” This global count has an apparent transparency
that helps account for its rhetorical success: it is simply the number of people in the world
who live on less than a dollar a day, with the relevant dollar adjusted for international
differences in prices. But this transparency and simplicity is more apparent than real. There
are many complexities beneath the surface, and these are the subject matter of this paper,
particularly the calculation of the global line and the adjustment for international differences
in prices using purchasing power parity exchange rates. Price adjustment is also important for
other purposes, such as the construction of databases such as the Penn World Table, which
underpins almost all of economists’ empirical understanding of the process of economic

2
growth, as well as for the measurement of global income inequality, which is another of my
main concerns here.
One goal of this paper is to understand why almost half a billion people were moved into
poverty at the time of the revision of the purchasing power parity exchange rates in the 2005
round of the International Comparison Project (ICP). This same revision also increased global
income inequality, widening the apparent distance between poor and rich countries. I shall
argue that the increase in poverty had little to do with the ICP, and much to do with an
inappropriate increase in the global poverty line. The causes of the increase in inequality are
harder to pinpoint, but my investigations lead to skepticism about our ability to make precise
comparisons of living standards between widely different countries, for example between
poor countries in Africa and rich countries in the OECD.
In spite of the attention that they receive, global poverty and inequality measures are
arguably of limited interest. Within nations, the procedures for calculating poverty are
routinely debated by the public, the press, legislators, academics, and expert committees, and
this democratic discussion legitimizes the use of the counts in support of programs of
transfers and redistribution. Between nations where there is no supranational authority,
poverty counts have no direct redistributive role, and there is little democratic debate by
citizens, with discussion largely left to international organizations such as the United Nations
and the World Bank, and to non-governmental organizations that focus on international
poverty. These organizations regularly use the global counts as arguments for foreign aid and
for their own activities, and the data have often been effective in mobilizing giving for
poverty alleviation. They may also influence the global strategy of the World Bank,
emphasizing some regions or countries as the expense of others. It is less clear that the counts

3
have any direct relevance for those included in them, given that national policymaking and the
country operations of the World Bank depend on local, not global poverty measures. Global
poverty and global inequality measures have a central place in a cosmopolitan vision of the
world, in which international organizations such as the UN and the World Bank are somehow
supposed to fulfill the redistributive role of the missing global government, see for example
Thomas Pogge (2002) or Peter Singer (2002). For those who do not accept the cosmopolitan
vision as morally compelling or descriptively accurate, such measures are less relevant, John
Rawls (1999), Thomas Nagel (2005), Leif Wenar (2006).
The paper is organized as follows. Section I explains how the dollar a day poverty
numbers are calculated and how they depend on purchasing power parity exchange rates. It
shows that poverty measures are sensitive to the PPPs used in their construction, and
establishes the basic facts and puzzles to be addressed, particularly the increases in poverty
and inequality associated with the latest ICP revisions. Sections II and III are more technical
and can be skipped without losing the thread of the main argument. Section II discusses the
components of the construction of PPPs that are particularly important for measuring world
poverty and inequality, while Section III discusses how PPP indexes need to be reweighted
for use in poverty measurement. It argues, by reference to my related work with Olivier
Dupriez (2009), that the reweighting, although a clear conceptual improvement, matters less
than might be thought. Section IV returns to the main argument and is concerned with the
definition of the global poverty line, and I discuss ways of constructing the line based on the
international price indexes and the national poverty lines of poor countries. As is always the
case with poverty lines, how the poverty line is updated with respect to new information
deserves as much or more attention than how its original value is set. I argue that the updating

4
procedure in current use is incorrect, that it can result in reductions in national poverty
causing increases in global poverty, and that this explains why the global poverty counts
increased so much in the latest revision. Paradoxically, one of the main reasons that India (and
the rest of the world) became poorer was because India had grown too rich. I argue for a
definition of the line, and an updating method, that is substantially different from those
currently in use, and that preserves a better continuity with previous estimates.
Section V, which is again somewhat more technical, is an enquiry into the international
price indexes themselves, with a focus on the factors that affect global inequality. It starts
from the question of how to price comparable goods in different countries and whether the
ever more precise specification of goods by the ICP has had the effect of making poor
countries poorer relative to rich countries, widening our estimates of international inequality,
and causing the global poverty line to increase in dollar terms at a rate that is markedly slower
than the rate of inflation in the US. More generally, I discuss the special difficulties of making
comparisons between countries whose relative prices and patterns of consumption are very
different, for example between Japan and Kenya, or Britain and Cameroon. Such comparisons
are required if we are to make multilateral price index numbers for the world as a whole, and I
use data from the 2005 ICP to investigate their credibility. My analysis shows that these
comparisons rest on weak theoretical foundations and are fragile in practice.
Section VI returns to the main argument and looks briefly at an alternative monitoring
system based on asking people about their lives are going; I use data from the Gallup World
Poll which collects an annual sample of all the people of the world. Section VII concludes,
and speculates on the global system of income and poverty statistics as a whole. I argue that
we should be less ambitious and more skeptical in using the international data, particularly