Professional Documents
Culture Documents
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp
.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.
American Economic Association is collaborating with JSTOR to digitize, preserve and extend access to The
American Economic Review.
http://www.jstor.org
By JAMES E. FOSTER*
Should poverty be measured using an "ab- choosing poverty lines and in aggregating the
solute" or a "relative" approach?This age- data into an overall index of poverty. A gen-
old question in poverty measurement is once eral taxonomy is presented, and the question
again on the agenda, due to the ambitiouspro- of robust comparisons is addressedwithin this
posals of PatriciaRuggles (1990) and the Na- general framework.Special attentionis paid to
tional Research Council of the National distinguishing between (i) the general concept
Academy of Sciences (Constance Citro and underlying the poverty line and (ii) the partic-
Robert Michael, 1995) to alter the way U.S. ular cutoff chosen. The paper concludes with
poverty is measured. Their wide-ranging sug- a discussion of "hybrid" poverty lines and the
gestions include a new "hybrid" approachto associated parameterthatis likely to play a key
setting the poverty threshold that, unlike the role in futurediscussions: the income elasticity
current absolute method, is sensitive to of the poverty line.
changes in the general living standard,but less
sensitive than a purely relative approach.The I. Elements
proposals also recommend using aggregate in-
dexes of poverty beyond the usual "head- Poverty measurement is based on a com-
counts," such as well-known "gap" measures parison of resources to needs. A person or
and indicators of the distributionof resources family is identified as poor if its resources fall
among the poor. Importantrelative notions of short of the poverty threshold. The data on
poverty enter at this "aggregation" step as families are then aggregatedto obtain an over-
well. The effects of the various recommenda- all view of poverty.
tions on the trend and cross-sectional profiles There are many ways of defining resources,
of poverty are actively being explored (see constructing thresholds, and aggregating the
e.g., David Betson and JenniferWarlick, 1997; resulting data (see e.g., Ruggles, 1990; Martin
Thesia Gamer et al., 1997; David Johnson et Ravallion, 1994; Citro and Michael, 1995).
al., 1997). At the same time it may prove use- Virtually all partition the population into
ful to consider some of the conceptual mea- groups of families (or resource-sharingunits)
surement issues arising from the proposals. with similar characteristics, and I follow this
This is the direction taken in the present study. approachhere. Let 0 denote the raw data, con-
This paper evaluates the multiple notions of taining information on resources received by
relative and absolute poverty that arise in families, their demographic and other charac-
teristics, and perhaps other data (e.g., con-
sumption distributions) needed to construct
t Discussants: David S. Johnson, U.S. Bureau of Labor poverty thresholds. Let m be the number of
Statistics; Patricia Ruggles, U.S. Department of Health distinct groups, with nk nfk()- being the
and Human Services; BarbaraWolfe, University of Wis- number of families in group k. Once a specific
consin; ChristopherJencks, HarvardUniversity. definition of family resources has been fixed,
* Department of Economics, Vanderbilt University,
this yields a distribution of resources among
Nashville, TN 37235. I thank the discussant, David S. the families in group k, denoted by the nk_
Johnson, for his insightful commentary.Financial support
from the John D. and CatherineT. MacArthurFoundation dimensional vector x k = Xk ( ) . The poverty
throughthe Network on Inequalityand Poverty in Broader threshold for families in group k is denoted by
Perspective is gratefully acknowledged. the number Zk = zk(0); a family is identified
335
Michael O'Higgins and Stephen Jenkins adjusted across demographic groups. One ap-
(1990).2 proach is to apply repeatedly the procedurefor
Using a relative line does not amount to setting poverty lines to each group separately
measuring inequality (although theorem 6 in and thereby arrive at m independent thresh-
Foster and Anthony F. Shorrocks [1988a] pro- olds. However, as noted by Ruggles ( 1990 Ch.
vides one important link) nor does it imply 4), this can lead to odd (rnonmonotonic)be-
that poverty is by definition "always with us" havior of the poverty line as famiily size
(see Anthony Atkinson, 1975 p. 189). And changes. An alternativeapproach sets the line
while many studies regardabsolute lines as be- in one reference group and then derives the
ing especially low and relative lines as being remaining thresholds using an "equivalence
high, this is not necessarily the case. If living scale" to account for the differing needs of
standardsare rising and thresholds are pegged different-sized families. The typical scale pro-
at Za = Zr in some initial period, then Za < Zr vides the rate at which a dollar for one group
for all subsequent periods, but Za > Zr for all translatesinto dollars for another. So if group
previous periods, as emphasized by Citro and 1 is the reference, and Sk is the conversion rate
Michael (1995 p. 132). In any isolated period, from group 1 to group k, then Zk = 5k l be-
it is not possible to tell whether a given thresh- comes the poverty line for group k.
old z is relative or absolute, nor is the distinc- This sort of equivalence scale is relative in
tion particularly important, since the same that the transformationfrom group to group is
numerical cutoff, however originally derived, multiplicative, and consequently group pov-
must lead to the same level of poverty. erty lines are proportionateto each other. An-
The key distinction between absolute and other possibility raised by Charles Blackorby
relative thresholds is not seen in the specific and David Donaldson ('1994) is for variations
values obtained at a given date, but in how the in family configurationto have an constantab-
values change as the distribution changes. solute effect so that, for example, adding an-
Thus, there is an important distinction to be other child is seen as an additionalfixed (real)
made between the general concept underlying cost to the family, independent of the size of
the poverty threshold, and the specific cutoff the base threshold.Relative equivalence scales
selected. For comparisons involving extended preserve the ratios of group poverty lines as
periods of time, or very different standardsof the base threshold changes; an absolute equiv-
living, the former is likely to be the more im- alence scale preserves the absolute differ-
portant issue (see also Ruggles, 1990 Ch. 3), ences. The two forms are indistinguishablefor
while the latter choice (of cutoff) is largely a single observation or if the reference thresh-
arbitrary(see Fuchs, 1969; Atkinson, 1975, old remains unchanged (as with an absolute
1987; Foster and Shorrocks, 1988b). This in- poverty line).
evitable arbitrarinesscasts doubt on the mean-
ing of the cardinal poverty levels obtained at C. Population
specific cutoffs and leads to a considerationof
the robustness of results to changes in the cut- The aggregation stage uses three notions of
off, a topic I will returnto below. absolute and relative poverty in constructing
poverty indexes. First, a relative or per capita
B. Equivalence Scale poverty index is independentof the population
size in the sense that "replicating" the popu-
A second entry point for relativities in pov- lation leaves the poverty value unaffected: for
erty measurement is where poverty lines are example, P(x, x; z) = P(x; z). Such a mea-
sure is based purely on the relative frequencies
of incomes in the income distribution.In con-
2 There are importantmeasurement issues in selecting trast, an absolute index is one whose value
the standardof living. Should it be the mean, the median, rises in proportion to the number of replica-
or some other representative income? Should it be from
the entire population or some reference group? Should it tions: for example, P(x, x; z) = 2P(x; z).
be for all expenditures or a significant subset? (For ref- The head-count ratio qln is relative in this
erences, see Citro and Michael [1995].) sense while the head-count q is absolute. An
Most results of this type are presented in a malized equivalent incomes can be compared
one-group framework with absolute thresh- using first-degree stochastic dominance over
olds; but in fact, the tools have far greaterap- the range (0, c ), while the tests for the nor-
plicability. As an illustration, suppose that the malized gap index and the Foster et al. (1984)
base threshold z1 and equivalence scale sk are index use second- and third-degree stochastic
relative, the index P is based on a notion of dominance, respectively. Atkinson's (1987)
absolute deprivation (hence decomposable) results go beyond these results to consider
but otherwise relative, and the only question variations in poverty indexes and indicate, for
is the specific cutoff a to be used in setting the example, that if there is an unambiguouscom-
relative poverty line. Suppose that for a spe- parison for H (and hence first-degree stochas-
cific value of a, say, a - 50 percent, the re- tic dominance), then virtually any acceptable
source distribution (x, ... , xm) has greater index P will agree with this conclusion. This
poverty than (y', ... , yi). When can one be illustrates the power of the head-count ratio in
sure that this will remain true for an entire this context.
range of a values, say (0, ii ) where ii > 50
percent? Let r be the standardof living under- IV. HybridMeasurement
lying the relative poverty line, and let rx and
ry denote the respective standardsin the dis- Many of the categories in my taxonomy al-
tributions. Constructa new "equivalent" dis- low for an intermediateposition to be chosen
tribution xik for demographic group k by in place of a pure relative or absolute ap-
dividing family resources by the equivalence proach. One particularlyinterestingexample is
scale sk, and then replicating by family size in the "hybrid" poverty threshold that is central
k, so that xk has one equivalent resource level to the proposal in Citro and Michael (1995),
for each person in group k. It is not difficult which is based on what lnight be termed a
to show that for P satisfying the above prop- "partial" standardof living: rpis the median
erties, the poverty level of the original distri- expenditure on certain basic goods. The
bution (x1 ..., xm) at the group-specific threshold z = arp has the same structureas a
thresholds z = s kz is simply purely relative cutoff (and in fact the robust-
ness result applies equally well to it). How-
Px 1
Pi,..., m; .
,mz) ever, median expenditures on basic goods do
not rise as fast as, say, median total expendi-
or the poverty in the equivalent distribution tures, and it is this empirical fact that gives z
given group l's poverty line. If one further its hybrid nature.
normalizes incomes by the standardof living, One could also imagine thresholds that are
then the poverty level is given by hybrid by construction, in that they depend di-
rectly on an absolute and a relative standard.
P(xI rx, ... , ?m/rx; a) . For example, consider a weighted geometric
averageof a relativethresholdZr = ar and an
Consequently, the judgment that (x', ..., xm) absolute threshold Za, namely, z =. zzI -p
has greaterpovertythan(y', ..., ym) is in fact where 0 < p < 1. This form of hybrid line has
robust in a if the property that a 1-percent increase in the
living standard r always leads to a p-percent
P(x/rx, ..,xmlr,,; a) increase in the poverty line. In other words, p
is the elasticity of the poverty line with respect
>P(y1Ir , a)
ym/ry; to the living standard,or what Gordon Fisher
( 1995 ) has termed the income elasticity of the
foralla E (0, ). poverty line. In general, p = (dzldr)(r/z) has
This last condition is in a form that allows a naturalinterpretationas a measure of the ex-
the application of results in Foster and tent to which a given threshold z is relative,
Shorrocks (1988b) and Atkinson (1987). So, with p = 0 correspondingto an absolute pov-
for example, the test for the head-count ratio erty line and p = 1 a fully relative one. The
H checks whether the two distributionsof nor- possibility of using a hybrid standardchanges
garner enough support to displace the current Dominance." Social Choice and Welfare,
standard. 1988b, 5(2/3), pp. 179-98.
s . 'Subgroup Consistent Poverty Mea-
REFERENCES sures." Econometrica, May 1991, 59(3),
pp. 687-709.
Atkinson, Anthony B. The economics of in- Fuchs, Victor. "Comment on Measuring the
equality. Oxford: Oxford University Press, Size of the Low-Income Population," in
1975. Lee Soltow, ed., Six papers on the size dis-
"On the Measurement of Poverty."
.0 tribution of wealth and income. New York:
Econometrica, July 1987, 55(4), pp. 749- National Bureau of Economic Research,
64. 1969, pp. 198-202.
._ "Measuring Poverty and Differences Garner, Thesia I.; Paulin, Geoffrey; Shipp,
in Family Composition." Economica, Feb- Stephanie; Short, Kathleenand Nelson,Chuck.
ruary 1992, 59(233), pp. 1-16. "Experimental Poverty Measurement for
Betson,DavidM. and Warlick,JenniferL. "Al- the 1990's." Mimeo, Bureau of Labor Sta-
ternative Historical Trends in Poverty." tistics, Washington, DC, 1997.
Mimeo, University of Notre Dame, 1997. Johnson, David; Shipp, Stephanieand Garner,
Blackorby, Charles and Donaldson, David. "Eth- Thesia. "Developing Poverty Thresholds
ical Indices for the Measurement of Pov- Using Expenditure Data." Unpublished
erty." Econometrica, May 1980, 48(4), manuscriptpresented at the Joint Statistical
pp. 1053-60. Meetings, Anaheim, CA, August 1997.
O'Higgins, Michael and Jenkins, Stephen. "Pov- Sen, Amartya K. "Poverty: An Ordinal Ap-
erty in the EC: Estimates for 1975, 1980, proach to Measurement." Econometrica,
and 1985," in Rudolph Teekens and March 1976, 44(2), pp. 219-31.
Bernard M. S. van Praag, eds., Analysing _-. "Description as Choice." Oxford
poverty in the European Community:Policy Economic Papers, November 1980, 32 (3),
issues, research options, and data sources. pp. 353-69.
Luxembourg: Office of Official Publi- _-. "Poor, Relatively Speaking." Oxford
cations of the European Communities, Economic Papers, July 1983, 35(2), pp.
1990, pp. 187-212. 153-69.
Ravallion,Martin.Poverty comparisons. Chur, Zheng, Buhong. "Can a Poverty Index Be
Switzerland:Harwood, 1994. Both Relative and Absolute?" Econo-
Ruggles,Patricia.Drawing the line. Washing- metrica, November 1994, 62(6), pp.
ton, DC: Urban Institute Press, 1990. 1453-58.