DAVID BRADY, Duke University
Abstract
Despite serious methodological problems, quantitative studies of poverty by U.S. sociologists predominantly rely on the official U.S. measure. After reviewing the shortcomings of the U.S. measure, this article examines several theoretical and methodological advances in poverty measurement. After synthesizing literature on poverty measurement, I argue that ideal measures of poverty should: (1) measure comparative historical variation effectively; (2) be relative rather than absolute; (3) conceptualize poverty as social exclusion; (4) assess the impact of taxes, transfers, and state benefits; and (5) integrate the depth of poverty and the inequality among the poor. Next, this article evaluates sociological studies published since 1990 for their consideration of these criteria. Due to sociology's neglect of these criteria, this article advocates for three alternative poverty indices: the interval measure, the ordinal measure, and the sum of ordinals measure. Finally, using the Luxembourg Income Study, I examine the empirical patterns with these three measures, across advanced capitalist democracies from 1967 to 1997. Estimates of these poverty indices are made available for future research.
In the first few pages of his classic The Truly Disadvantaged (1987), William Julius Wilson lamented the paucity of poverty scholarship by sociologists in the 1970s and early 1980s. Reacting, in part, to the climate surrounding Moynihan's (1965) study of the African American family and Wilson's earlier book, The Declining Significance of Race, many scholars avoided the study of
I appreciate helpful comments from two Social Forces reviewers as well as Art Alderson, Jason Beckfield, Clem Brooks, Katy Fallon, David Jesuit, Jane McLeod, Phil Morgan, Jason Schnittker, and especially Michael Wallace. This article was presented at the 2001 American Sociological Association meetings in Anaheim, California. It received the Aldi J. M. Hagenaars Luxembourg Income Study Memorial Award, and the Karl Schuessler Award for Graduate Research Excellence at Indiana University. Direct correspondence to David Brady, Department of Sociology, Duke
University, Box 90088, Durham, NC, 27708-0088. E-mail:brady@soc.duke.edu.
? The University of North Carolina Press Social Forces, March 2003, 81(3):715-752
716/ Social Forces 81:3, March 2003
poverty during this period. Fortunately, Wilson's 1987 book triggered an enthusiastic resurgence in the study of poverty, and sociological research subsequently proliferated. Since 1990, 53 empirical, quantitative studies published in prominent and relevant sociological journals have featured poverty as the dependent variable or as a key independent variable. Many other articles, and far more scholarly books, analyze poverty with a qualitative or theoretical approach. Sociologists have answered Wilson's call, and poverty research has experienced a vibrant reinvigoration.
Sociology's revitalization of poverty research has produced significant empirical findings, theoretical contributions, and policy applications. At the same time, several conventional methodological practices have become widely accepted. Though these conventions demonstrate effective scientific replication, this consensus has simultaneously obscured one very crucial methodological concern. On the whole, the measurement of poverty has not received the scrutiny it deserves. Most sociologists of poverty rely on estimates
of the official U.S. measure of poverty - presuming that such statistics are both valid and reliable. In a few cases, sociologists modestly augment the U.S. measure with slight alterations or by supplementing it with other indicators. However, most archival quantitative data sets supply researchers with dichotomous variables identifying respondents as below or above the official level. Typically, scholars use these simple dummy variables and hope that measurement issues are resolved. As a result, the vast majority of sociological studies of poverty use
a seriously problematic measure of poverty.
Despite these problems, sociologists' measurement of poverty would be
somewhat acceptable if no feasible alternatives existed. To the contrary, how-
ever, a wealth of scholarship focuses on this very issue. Several social scientists
have devoted their entire careers to devising innovative and useful techniques
for the measurement of poverty.' Though a few of these techniques are im-
practical and/or flawed, significant methodological and theoretical advances
have been made. Unfortunately, despite the relevance of this research and its
popularity across the social sciences, sociologists have not sufficiently integrated these advances.
Of course, these alternatives would not warrant sociologists' attention if different measures of poverty produced identical or even similar empirical conclusions and policy implications. This is, however, far from the case. Hagenaars (1991:134) explains, "Both the population of poor and the extent of their poverty appear to depend to a large extent on the definition chosen." Betson and Warlick (1998) convincingly demonstrate that the number, composition, and trends in U.S. poverty significantly depend on the particular measure chosen. In fact, simply ascertaining whether poverty has increased in the last thirty years elicits dramatically different answers with contrasting measures of poverty (see Jorgenson 1998; Triest 1998). Thus, poverty measurement decisions have very real, substantive, and policy consequences
Sociological Measurement of Poverty / 717
that potentially affect the scientific inferences of research (Hill & Michael 2001; Iceland et al. 2001).
Overall, this article attempts to provide a guide for sociologists while
facilitating a stronger connection between sociology and the theoretical and
methodological advances in the measurement of poverty. First, I revisit the
shortcomings of the official U.S. poverty measure. Second, I discuss several
emerging theoretical and methodological advances in the measurement of
poverty, and advocate five criteria for the measurement of poverty: (1) measure
comparative historical variation effectively; (2) be relative rather than absolute;
(3) conceptualize poverty as social exclusion; (4) assess the impact of taxes,
transfers, and state benefits; and, (5) integrate the depth of poverty and
inequality among the poor. Third, I evaluate sociological research since 1990
for its application of these criteria. Fourth, I advocate for three alternative
indices that resolve the problems of sociological measures of poverty while
adhering to these criteria: the interval measure, the ordinal measure, and the
sum of ordinals measure. Finally, with the Luxembourg Income Study (LIS),
this article empirically examines these indices across advanced capitalist democracies from 1967 to 1997.
Shortcomings of the Official U.S. Measure
In recent years, many scholars, journalists and policy makers have criticized
the official U.S. measure of poverty (e.g., Ruggles 1990). Betson and Warlick
(1998:351) emphasize that the U.S. measure "is commonly acknowledged to
be inadequate for measuring poverty." Wilson (1991:3) argues that the U.S.
measure "does not capture the real dimensions of hardship and deprivation,
it also does not reflect the changing depth or severity of poverty." The Family
Support Act of 1988 called for a scientific review of the U.S. measure. In 1995,
the National Research Council (NRC), and specifically the Panel on Poverty
and Family Assistance, published the results of this scientific review (Citro &
Michael 1995). In this report, the NRC panel, which included many of
America's most influential poverty researchers, broadly concluded that the U.S. measure is outdated and should not be retained.2
Mollie Orshansky (1965), of the Social Security Administration under President Johnson, constructed the measure in 1963. Johnson reportedly sought
a measure that was sufficiently conservative to render the eradication of pov- erty as an attainable goal of his War on Poverty (Betson & Warlick 1998; Katz
1989). Orshansky used family consumption data from 1955 (Wilson 1991) and what she called a "crude" calculus of family budgets.3 Though seemingly logi- cal, Wilson (1991:2) stresses that the U.S. measure "represents arbitrary income thresholds" which have little relevance to contemporary American society. The
718 / Social Forces 81:3, March 2003
dubious origins and significant elapsed time since the measure's inception, at a minimum, warrant regular review and updating. The NRC panel went one step further, arguing, "Our major conclusion is that the current measure needs to be revised" (Citro & Michael 1995:1). In turn, poverty analysts increasingly conclude that the antiquated official poverty line is no longer appropriate.4
Moreover, many analysts fault the U.S. measure's lack of reliability because it obscures differences in the extent of poverty among population groups and across geographic contexts and provides an inaccurate picture of trends over time (Citro & Michael 1995; Haveman 1987). Because the measure remains unchanged after thirty years, significant demographic, economic, and policy changes are ignored (Blank 1997). Specifically, the NRC noted the increased labor force participation of mothers, the related escalating need and expenses for child care and health insurance, differences in health status, and the inappropriateness of antiquated family size adjustments (Betson & Warlick
1998; Citro & Michael 1995). Lichter (1997) laments the unsophisticated equivalence scale (see below), which does not reliably measure poverty across family sizes and forms. Also, Ruggles (1990) explains that the relative share of family budgets devoted to different goods and services has changed.5 Foster (1998) adds that over time the U.S. measure has depreciated from its value in 1963 and become unreflective of what a family really needs to avoid poverty. Because of rising consumption and living standards, the NRC concluded that updating the poverty threshold solely with inflation is increasingly inadequate (Citro & Michael 1995). In short, the U.S. measure lacks reliability due in large part to the limited and weak means of adjusting the measure since its inception.
Similarly, many scholars (e.g., Jorgenson 1998; Slesnick 1993) argue that the U.S. measure lacks validity because it fails to capture the complex nature of poverty. Many increasingly burdensome family expenses (such as health and child care) are not encompassed in the U.S. measure. In addition, the measure ignores cash, near-income, and in-kind public assistance, as well as the taxes that effectively alter a family's disposable income (Betson & Warlick 1998; Citro
& Michael 1995; Lichter 1997). Neglecting these government benefits, the U.S. measure violates the transfer axiom (A. Sen 1976; see below), fails to grasp the financial reality of poor families, and significantly underestimates the extent of poverty in the U.S. (Ruggles 1990).
Because these validity problems have varied over time and place, reliability is also compromised. Taxes, such as the Social Security payroll tax, have varied across the U.S. and have increased enormously since the measure's inception (Betson & Warlick 1998; Lichter 1997). Moreover, future policy initiatives (e.g., the extension of Medicaid, and the Children's Health Insurance Plan) will not be captured by the measure (Citro & Michael 1995). In sum, the U.S. measure lacks both validity and reliability and warrants revision.
Sociological Measurement of Poverty / 719
Though the construction of a flawless measure is unlikely, the NRC and others suggest important revisions to improve poverty measurement. First, many scholars argue for a more significant temporal revision to the measure as living standards and consumption rise. Second, many analysts argue simply that the threshold should be raised to include those just above or near the poverty line. Third, a rising consensus argues for a reorientation toward a relative standard. While the U.S. measure purports to delineate a family's absolute level of minimum needs, the NRC recommended that the threshold be explicitly refocused on the relative consumption of contemporary U.S. families. With the NRC's proposed alternative measure, analysts have demonstrated significantly different historical trends in U.S. poverty (Betson & Warlick 1998; Triest 1998); much higher poverty rates (Hill & Michael 2001); smaller gaps between child and adult poverty rates (Iceland et al. 2001); and, importantly, different social consequences of child poverty (Hill & Michael 2001).6 Unfortunately, despite these scholars' efforts, the government has not implemented significant changes to its measure. While a lack of political will probably explains the inaction of the U.S. government, sociologists have no such justification. It is time that sociologists moved away from this flawed measure.
Measuring Comparative Historical Variation in Poverty
For decades, sociologists have sustained interest in the comparative patterns and historical trends in poverty and inequality (e.g., Alderson & Nielsen 1999; Brady & Wallace 2001; Casper, McLanahan & Garfinkel 1994; Firebaugh 2000). Unfortunately, however, the discipline's reliance on the U.S. measure has limited our contribution to the understanding of these dynamics. For research on poverty to advance, measures must be developed to gauge carefully the comparative historical variation in poverty.
Many analysts document that the paramount variations in poverty are cross- national (e.g., Atkinson 1998a). Hence, explaining these significant cross-country differences is essential to understanding poverty in contemporary societies (Cantillion 1997). In addition, a comparative perspective provides leverage in assessing the influence of causal factors such as economic change, public policy, and demographic shifts. Given these benefits, several methodological concerns emerge when comparing poverty across nations. Several analysts note that different measures of poverty produce small but important differences in the rank ordering of nations (Atkinson 1998a; Hagenaars 1991). Therefore, scholars must guard against noncomparability and measurement error when observing these sensitive cross-national differences (Atkinson 1990). Also, cultural differences in the definitions of family units often obscured some of the national differences in poverty. Until recently, the scarcity of high-quality data needed for such comparisons compounded these issues (Cantillion 1997). Hence, these
720/ Social Forces 81:3, March 2003
methodological and data concerns must be carefully addressed in order to advance the comparative analysis of poverty.
Though cross-national differences are probably larger, historical differences remain important to the study of poverty as well. For example, assessing the trends in U.S. poverty remains a highly controversial issue (Betson & Warlick 1998; Jorgenson 1998). Understanding historical trends in poverty is also important for separating the effects caused by sheer demographic shifts versus other causal factors such as social policy and long-run economic changes (Danziger & Weinberg 1994; Ruggles 1990). Though some research finds that poverty levels remain relatively stable over time within nations of the Organization for Economic Cooperation and Development (OECD), significant temporal variation does occur (Cantillion 1997). Fortunately, recent advances in data collection allow analysts to use more sophisticated measures to scrutinize over-time comparisons (Ravallion 1998).
Assessing comparative historical variation in poverty levels offers a great deal to the sociology of poverty. To maximize our contribution, two issues bear careful consideration. First, given the difficulty in comparing poverty across societies and across time, to make general inferences about causal processes, scholars need measures that grasp the same phenomena in each society. Second, given the diverse meanings, nature, and content of poverty in various societies, scholars need to broaden the very definition of poverty. To assess what are essentially culturally specific and historically contextualized phenomena, scholars need a broad conceptualization of poverty. Though seemingly contradictory, the next two sections argue that we need to embrace both concerns simultaneously.
Relative Versus Absolute Poverty Measures
For many years, a vigorous debate persisted over relative versus absolute definitions of poverty (see A. Sen 1979, 1983; Madden 2000). Relative and absolute definitions of poverty tap into fundamentally divergent notions of difference and deprivation (Shanahan & Tuma 1994). Also, absolute and relative standards produce different policy implications and accounts of the experience of poverty, and somewhat differ in the extent of poverty determined (Townsend 1980). Despite this historically contentious debate, poverty scholars increasingly conclude that in advanced capitalist democracies, a relative definition is more appropriate (Atkinson 1998a; Gordon 1972; Hagenaars 1991; Madden 2000; Ravallion 1998; A. Sen 1992). Relative measures usefully capture changes in necessities over time and place, which is particularly relevant to such nations. Scholars also conclude that a relative measure more effectively gauges comparative historical variation across comparable nations in a historical period. Alternatively, absolute measures of "basic needs" are most
Sociological Measurement of Poverty / 721
useful in developing countries vulnerable to famine and underdevelopment. By
reviewing the strengths and weaknesses of absolute and relative measures, this section establishes a relative standard as a criterion for an ideal measure of
poverty.
ABSOLUTE MEASURES OF POVERTY
Absolute measures involve a cross-nationally and historically constant and fixed threshold, which distinguishes poor from nonpoor. For example, except for inflationary adjustments, the U.S. measure is absolute over time, regions, and family types. Absolute measures assume that a certain material level purchases an essential bundle of goods necessary for well-being. For example, the World Bank defines poverty absolutely as living on less than one dollar per day. Thus, in developing countries, absolute measures can also be tied to absolute definitions of well-being, such as infant mortality, life expectancy, and caloric intake (see Y. Bradshaw et al. 1993). A. Sen (1992, 1999) argues that when studying developing countries, absolute measures should be retained. Nevertheless, absolute measures suffer from serious limitations that render them inappropriate for advanced capitalist democracies.
Importantly, scholars have grown skeptical about whether a fixed bundle of goods or absolute threshold of well-being can capture the complexity of poverty. This is exacerbated when an absolute standard is employed regardless of historical and national contexts (Atkinson 1998a; Smeeding, O'Higgins & Rainwater 1990).7 Smeeding, the director of LIS, and his colleagues (1993:246) avoid an absolute measure because it "conveys an unwarranted objectivity." Further, they argue that it has become the "widely held view among scholars working in this arena that a poverty standard cannot be established indepen- dently of the economic and social context within which needs arise and are defined" (247). Due to these reasons, poverty measurement debates have moved away from absolute measures. Though absolute measures effectively assess pov- erty in developing countries, relative measures are more appropriate in ad- vanced capitalist democracies. Relative poverty measures cannot capture the absolute deprivation of households but more accurately grasp the notion of relative deprivation.
RELATIVE MEASURES OF POVERTY
Relative measures generate specific poverty thresholds for each society in each time period from patterns in the income distribution. Typically, relative measures begin with a threshold of 50% of the median income. People below such a threshold are considered too far down on the queue of the scarce resource of income to be fully integrated into society (Shanahan & Tuma 1994). Hence, relative measures reflect the difference in living conditions between the
722 / Social Forces 81:3, March 2003
poor and the majority of society, rather than some abstract standard. Though relative measures emerge from the distribution within a particular society, "using a relative line does not amount to measuring inequality nor does it imply that poverty is by definition 'always with us."' (Foster 1998:337).
A relative measure's greatest theoretical virtue is that it is entirely grounded in national and historical context (Townsend 1980). Relative measures
advantageously measure deprivation according to a particular society's cultural
norms and customary, prevailing standards of necessities (A. Sen 1979). As
Adam Smith ([1776] 1937) argued, poverty is a lack of those necessities that
"the custom of the country renders it indecent for creditable people, even of
the lowest order, to be without" (quoted in Ruggles 1990:xv). Scholar and
activist Michael Harrington often argued that poverty should be gauged
according to the living standards of the mainstream of contemporary society.8
Thus, relative measures frame poverty as a social and, hence, sociological condition.9
Theory aside, relative measures are also superior because they provide leverage for policy analysis and sociology. In fact, several European poverty scholars argue for nation-specific relative measures over Europe-wide standards (see Atkinson 1998a). Even U.S. policy makers have long conceded that as a society's standard of living rises, more expensive consumption is forced on the poor to remain integrated into society (President's Commission on Income Maintenance Programs 1969). For example, arguably the most important trends in children's poverty entail relative deprivation, since massive relative deprivation is what disadvantages children in human development, human capital, and life chances.10 Relative measures are embedded in the social context and thus are far more valuable for sociological research on the causes and trends
in poverty.
Although a growing consensus of poverty analysts prefers a relative measure,
the debate between absolute and relative measures persists. Advocates of absolute measures concentrate on basic needs, because if those are unmet -
in terms of physiological subsistence and safety - poverty is truly present. Despite the persuasiveness of this claim, advocates of relative measures respond that the concept of "need" is actually relative itself -reflecting contextual norms of what is a "need" (Harrington 1981:188). Ruggles (1990) argues that consumption patterns have changed so dramatically over the past 40 to 50 years that defining the basic needs of American families is quite elusive. Other scholars go even further in problematizing the concept of basic needs. Ravallion (1998:21) notes that perceptions of "well-being" are contingent on the reference group's circumstances and argues, "There is an inherent subjectivity and social specificity to any notion of 'basic needs."' Hagenaars (1991:141) stresses that even nutritionists cannot agree about levels of calories needed for various ages, sexes, occupations and living conditions, and contends "the resulting estimates are not as absolute and objective as they are claimed to be."
Sociological Measurement of Poverty / 723
While most scholars agree that a desperate absolute level of deprivation does exist under which families are definitely poor, discerning an appropriate standard above that level remains ambiguous. Because such a minimal standard has limited utility in advanced capitalist democracies, most scholars argue that basic needs standards are less useful. Townsend (1980:300) argues, "any rigorous conceptualization of the social determination of need dissolves the idea of 'absolute' need." Overall, relative measures emerge as superior.
While most scholars now agree on the use of a relative measure in operationalizing poverty, much theorizing and debate continues about the conceptualization of poverty itself. Arguably, the most promising theoretical direction for the analysis of poverty is the literature on social exclusion. Though debates on social exclusion are prominent in Europe, the concept has not been fully integrated into the sociology of poverty in the U.S.
Conceptualizing Poverty as Social Exclusion
Recently, poverty analysts have grown dissatisfied with narrow theoretical conceptualizations and measures of poverty. In fact, narrow perceptions of poverty may fundamentally underestimate the extent and severity of poverty (Townsend 1980). European scholars have advanced the concept of social exclusion as an attempt to broaden the conceptualization of poverty and to facilitate measurement innovations (Cantillion 1997; Ormerod 1998; Paugam
1998; Procacci 1998; Wacquant 1995). Conceptualizing poverty as social exclusion can provide a novel and beneficial direction for the U.S. sociology of poverty. Potentially, the concept of social exclusion will suggest new sets of interesting sociological questions and provide different theoretical interpretations of old findings. Therefore, conceptualizing poverty as social exclusion is a criterion for an ideal measure of poverty.
Social exclusion is polysemic, having multiple meanings in different contexts
and for different purposes (Silver 1994, 1995), but we can identify important common elements. Social exclusion is the antithesis of the Durkheimian
concept of solidarity and connotes marginalization and irrelevance. Theorists characterize social exclusion to entail "the multi-dimensional character of
disadvantage and exclusion in modern market economies" (Cantillion 1997:30); multiple deprivation or "cumulative misery" (Schuyt & Tan 1988:14); those "who suffer from an accumulation of disadvantage which cannot be reached by macro-policies" (Dahrendorf 1990:151); and those difficult to reach with social policy (Engbersen 1991). The notion of social exclusion echoes Harrington's (1981:11) classic concern that "the poor are losing their links with
the greater world." In addition, social exclusion is consistent with Wilson's (1991) concept of social dislocation, which he describes as limited differential opportunities for economic resources, political privileges, organizationa
724/ Social Forces 81:3, March 2003
influence, and cultural experiences (see also Rankin & Quane 2000). Social exclusion can be understood as "people being prevented from participation in the normal activities of the society in which they live or being incapable of functioning" (Atkinson 1998a:27). In sum, social exclusion means incomplete citizenship and unequal access to the status, benefits, and experiences of typical citizens in society (Gore 1995).
Though social exclusion has multiple meanings, the concept can also be reduced to one central notion. If an individual is socially excluded, that person has a limited capability to effectively participate in society. Capability refers to the ability to function effectively in society and have the freedom to participate fully and equally with the mainstream. Capability offers a promising link between poverty and social exclusion, as social exclusion defines the lack of the basic capabilities that make one poor (A. Sen 1999)." A. Sen (1992) has formulated his arguments about inequality and poverty around people's substantive freedom of choice to achieve valuable functionings and well-being. A functioning member of society must have basic freedoms (or capabilities) to participate in society's main institutions (Barry 1973, 1998). Thus, the concepts of social exclusion and capability present an engaging, broadening direction for analysts of social inequality. To date, however, the connection to poverty measurement has not been fully articulated. I argue that social exclusion, and hence capability, facilitate the reconceptualization of poverty in two main ways.
First, these concepts explicitly and implicitly necessitate a relative measure
of poverty. Explicitly, policy debates for relative poverty measures have been
influenced by the concept of social exclusion (Barry 1998; Gore 1995). The
notion of social exclusion has been deployed in the European debate about the
community- and society-specific nature of poverty. In 1984, when the European
Commission constructed measures of poverty, the Council of Ministers overtly
linked their measures to social exclusion by defining poverty as "persons whose
resources are so limited to exclude them from the minimum acceptable way
of life in the Member State in which they live" (Atkinson 1998a:2). Silver
(1994) explains that the European Union's statistical service, Eurostat, and the
Luxembourg Income Study use relative measures of poverty due to a theoretical
interest in relative deprivation and social exclusion. Atkinson (1998b) adds that
using an absolute deprivation standard of poverty - with its emphasis on economic circumstances in isolation from others - has no relevance to social
exclusion. Last, the empirical reality of "new" poverty within Europe explicitly corresponds to what French scholars label "exclusion," with its relative deprivation, insecurity, and displacement (Silver 1994).
Implicitly, relative measures are theoretically consistent with social exclusion. Social exclusion theorists have also been influenced by Rawls's difference principle. Atkinson (1987) explains that a Rawlsian theory of poverty
is concerned with the "least fortunate group in society" and this group is the
Sociological Measurement of Poverty / 725
socially excluded who could be defined as poor. Rawls (1971) even suggested that this group could be defined as those with less than half of the median income and wealth, and noted that this could form a meaningful poverty standard. Finally, a theoretical interest in social exclusion necessitates an appreciation of its cultural and historical context. In turn, cross-nationally operationalizing poverty or social exclusion is quite difficult and requires a relational measure that is grounded in social context (Silver 1994).
Second, social exclusion and the economic market are strongly connected. Because poverty is primarily an economic phenomenon and social exclusion
is multifaceted and complex, the two may seem incompatible. It may even seem inappropriate to treat social exclusion as a market phenomenon rather than as a cultural, institutional, and social concept. However, the economic market is one of several main mechanisms triggering social exclusion. In postindustrial welfare states, a low level of economic resources is a principal precursor to social exclusion (Barry 1998). In effect, an interest in social exclusion demands an interest in economic inequality; as Barry (1998:22) puts it, "A government professing itself concerned with social exclusion but indifferent to inequality is, to put it charitably, suffering from a certain amount of confusion." The essence of social exclusion thus involves the dual marginalization by society's institutions and, especially the market (Gore 1995; Rodgers 1995).12
The link between social exclusion and the market, and hence poverty, is made even more clear by returning to capability. A. Sen (1992:110) asserts that "poverty is not a matter of low well-being, but of the inability to pursue well- being precisely because of the lack of economic means." Even with the nuanced concept of capability, a basic level of economic means is essential for escaping poverty. While this basic level of capability almost implies an absolute definition
of poverty, this need not be the case. Though poverty is absolute in terms of capabilities, it is plausible, and even appropriate, that poverty is relative in terms of economic resources (A. Sen 1983; Ravallion 1998). Therefore, capability can be an absolute concept entailing basic levels of social functioning, while being measured as a relative economic standing.13 Thus capability and social exclusion, as attributes of poverty, emerge from a relative measure of poverty.
Taxes, Transfers, and State Benefits
One of the most persuasive critiques of the U.S. measure of poverty is that it neglects taxes, financial transfers, and in-kind benefits. Of course, taxes and transfers make a significant impact on a family's finances. In fact, the deteriorating value of transfers is the main reason for the worsening of child poverty in recent decades in the U.S. (Lichter 1997). Further, taxes on poor families in the U.S. have steadily risen, making their financial standing actually
726 / Social Forces 81:3, March 2003
weaker than that of prior equivalent families. This neglect of taxes and transfers in configuring income violates A. Sen's (1976:219) Transfer axiom, "Given other things, a pure transfer of income from a person below the poverty line to anyone who is richer must increase the poverty measure." Ignoring these financial costs and benefits when measuring poverty is a fundamental theoretical and empirical problem (Ravallion 1998). Thus, ideal poverty measures must incorporate taxes and transfers and measure poverty before and after taxes and transfers (Danziger, Haveman & Plotnik 1981).
Though taxes and transfers most often are financial, the contribution of in- kind and near-cash benefits like housing assistance and food-stamps is essential (Townsend 1980). Poverty analysts often ignore state benefits when assessing family income because problems in measurement, valuation, and imputation of near- and non-cash income to individual households are quite formidable. This is troubling since these benefits affect the distribution of well-being between households; and ignoring these benefits yields misleading inferences about the relative well-being of various types of households (Smeeding et al. 1993).
Under the assumption that the most comprehensive definition of income is optimal for assessing familial welfare, the LIS analysts have made significant strides in incorporating taxes and transfers into their measures of income. Smeeding and his colleagues (1993) have assessed the value of, and imputed near-cash income to a variety of benefits. These benefits provide significant resources for families; and cumulatively, they have significant equalizing consequences by raising living standards and reducing poverty. While national differences exist in the nature and extensiveness of these benefits, their importance to the income distribution is universal.
Importantly, these benefits accrue from both the government and the pri- vate sector. Though government benefits typically provide larger consequences for the overall income distribution, private benefits are nontrivial. To be most effective, poverty should be examined both as it is generated in the private sector, and as it is mediated by the state.14 To the extent possible, private ben- efits should be considered as part of market income, and state benefits should be considered as part of ultimate state-mediated income. Both types of income - before and after taxes and transfers - are important to understanding the complex nature of poverty.
The Ordinalist Revolution
Following Sen's (1976) pioneering work, "The Ordinalist Revolution" (Hagenaars 1991) fundamentally redirected debates on poverty measurement. While only minimally impacting public policy (Osberg & Xu 2000), consideration of Sen's contribution is essential for any serious evaluation of
Sociological Measurement of Poverty / 727
TABLE 1: Alternative Poverty Measures Emerging from the Ordinalist Revolution
Symbol Definition Advantages Disadvantages
Headcount H Percentage of population
below 50% of median income
Income Gap I Difference between
Simple, dichotomous measure of the
percentage of the
population socially excluded
Continuous variable
of the average depth of poverty among the poor
Simple, parsimonious measure combining quantity and depth of poverty
Weightsmeasure so the deeply poor have more impact than
the barely poor
Provides clear, interpretable graphical representation
Ignores the depth of poverty among the poor
Ignores the quantity of poor people
Does not weight index with
income distribution
among the poor
Mayadd
unimportant information or
unneeded complexity
Less precise information
about distribution
of the poor than HI or O
Interval
Ordinal
population's median income and mean
income of poor
with H, standardized
by population's median income
HI Product of H and I
0 HI-(1+CV), where CV is
coefficient of variation
Sum of H for 60%, 50%, 40%, 30%, 20%, 10%, and 5% of median income
Sum ofOrdinals SO
poverty measurement. The contribution can best be explained by considering a series of measures that build on one another. Table 1 displays each measure that is relevant to this explanation. All of the measures discussed can be defined relatively.
First, poverty is measured commonly with a Headcount (denoted by H), the percentage of the population that is below a certain threshold of income. H is a simple dichotomous measure of poverty, offering an either/or account of who
is denied the basic minimum rights of citizenship or social inclusion (Atkinson
1998a). Despite its useful simplicity, H has received mounting criticism (Atkinson 1987). Sen (1976:219) calls H "crude" because it ignores the income
728 / SocialForces 81:3, March 2003
distribution of the poor and contains no information on the depth of poverty. A. Sen (1976:219) articulated this basic criticism of H as the "Monotonicity Axiom: Given other things, a reduction in income of a person below the line must increase the poverty measure."'5 Though it still has utility for describing the proportion of the population that is socially excluded, H is generally considered an imperfect measure of poverty (Myles & Picot 2000).
To address these concerns, we can estimate the depth of poverty of the poor. Conventionally, depth is measured as the poor's average difference from either the median of income or the threshold of poverty (Kakwani 1993). This average deprivation, the Income Gap (denoted by I), is normally standardized by the median income or threshold of poverty to render it comparable across populations. By considering I, rather than simply H, scholars more realistically capture the continuous quality of poverty. In reality, poverty is not a discrete condition that is immediately acquired or shed by crossing any particular income line (Watts 1968). Rather, poverty is an interval variable, as the desperately poor with zero income are worse off than the poor just below the poverty threshold.
Still though, I is imperfect as well. While H offers information on the percentage of the population that is poor, I details the depth of poverty of this sub-population. H is insensitive to the depth of poverty, while I is insensitive to the quantity of poor (A. Sen 1976). As a result, scholars have created new poverty measures by simply taking the product of H and I, that is HI (Atkinson
1987). Because it treats poverty as continuous, unlike the dichotomous H, I label HI the Interval Measure. Both H and I are equally important components because neither individually tells the whole story about poverty intensity (Osberg and Xu 2000).16
At this point, A. Sen (1976) offered his key contribution. He imposed Axiom
R, that the poverty gap (I) should be weighted to correspond to the rank order
in the interpersonal welfare ordering of the poor (see also Shorrocks 1995). A.
Sen argued that HI should be weighted such that the income gaps of the
poorest of the poor had more influence. In effect, HI should add a weight for
the income inequality among the poor. The Interval Measure was augmented
to form the Ordinal Measure of poverty (denoted by O), with the following formula:
O = H * I * (1 + CV),
where CV is the inequality among the poor.17
While more mathematically complicated versions of this formula exist
(e.g., Kakwani 1993; P. K. Sen 1986), several scholars have demonstrated that
O can be reduced in this way (Myles & Picot 2000; Osberg & Xu 2000).
Additionally, with this formula, O is easily decomposed into three parts that
can be analyzed separately to understand their specific changes and relative influence.
Sociological Measurement of Poverty / 729
As a final alternative, I have created the Sum of Ordinals Measure of poverty
(SO). SO is simply the sum of head counts for seven different thresholds, and thus builds on relational distribution measures of inequality (Handcock & Morris 1998). Specifically, I calculated H for 60, 50, 40, 30, 20, 10 and 5% of the median income, and summed the values. The SO measure mimics the properties of HI, and can be easily converted to something similar to O by weighting the lower thresholds (5, 10, and 20%) more greatly.
Overall, A. Sen's work provoked a fundamental rethinking of poverty measurement. In addition to the Headcount (H) or Income Gap (I), scholars can now use three more sophisticated measures of poverty: the Interval (HI), Ordinal (0) and Sum of Ordinals (SO). Each offers fruitful direction for sociological research on poverty.18 As Table 1 displays, each measure has advantages and disadvantages, which upon consideration assist measure selection. If an analyst seeks a simple, parsimonious measure that incorporates both the quantity and depth of poverty, HI is preferred. By contrast, if one decides that the deeply poor should disproportionately affect the index, O should be used. Unlike HI, O weights the index with the inequality among the poor and reflects the judgment that the deeply poor are more important than those near the threshold. Of course, many analysts may not agree with this judgment. Additionally, some evidence exists that the variation in O not captured by HI is empirically unimportant (Myles & Picot 2000; Osberg & Xu 2000). Hence, O often adds unneeded complexity that may obscure national comparisons of poverty (Atkinson 1987; Hagenaars 1991). Therefore if one seeks a sufficient yet parsimonious measure and prefers to avoid the complexity and assumptions of O, HI may be the best alternative. Finally, if one seeks to graphically represent descriptive analyses of the poor, SO offers advantages. Unlike HI or O, SO provides a clear, interpretable display of patterns in the distribution of the poor (see Figure 1 below). Of course, SO has the disadvantage of being a less precise measure than HI or O. In sum, analysts gain much by considering these multiple measures of poverty. Depending on a scholar's theoretical interests, each measure carries certain advantages and disadvantages. Generally, though, an ideal measure of poverty should integrate the depth of poverty and the inequality of the poor.
To summarize, this discussion provides five criteria for ideal measures of poverty. Each of these criteria emerges from an existing theoretical literature that has fully established their relevance to poverty measurement. Table 2 displays these criteria in summary form.
U.S. Sociology and The Measurement of Poverty
Given the influence of these methodological advances across the social sciences and internationally, it seems plausible that sociologists would incorporate them as well. Certainly, as the sociology of poverty grows, these techniques potentially offer leverage for research on poverty. But have sociologists used these advances
730 / Social Forces 81:3, March 2003
TABLE 2: The Criteria for Ideal Measures of Poverty
1. Measure comparative and historical variation effectively. 2. Be relative rather than absolute.
3. Conceptualize poverty as social exclusion.
4. Assess the impact of taxes, transfers, and state benefits.
5. Integrate the depth of poverty and the inequality among the poor.
in their work? Specifically, what are the patterns in the sociological literature regarding the measurement of poverty? I answer these questions with a content analysis of major U.S. sociology journals. As Table 3 displays, I analyzed every quantitative sociological study published from 1990 through 2001 in seven relevant journals: American Journal of Sociology, American Sociological Review, Demography, Research in Social Stratification and Mobility, Social Forces, Social Problems, and Social Science Research. In sum, I examined 53 articles that featured poverty as a dependent or key independent variable.
The vast majority (69.8 %) of sociological studies use the official U.S. measure. This is surprising given that several important critiques of this measure were published prior to or in 1990 (for example, Ruggles 1990), and since Wilson's 1990 American Sociological Association Presidential address (published as Wilson 1991) criticized the U.S. measure and called on sociologists to conceptualize poverty as social dislocation. Unfortunately, sociologists continue to use the measure despite its problems and prominent calls for change. Particularly problematic, some studies attempt to analyze historical variation in poverty with the U.S. measure, despite its temporal unreliability (e.g., Eggers & Massey 1992). Six studies (11.3%) made modest alterations to the U.S. measure, including measuring the percentage of the population below 125% of the official threshold. Five studies (9.4%) included a broader scale where the U.S. measure was one of several indicators (e.g., South & Crowder 1999). Also, a few studies examined the U.S. measure in combination with other measures of poverty (e.g., Eggebeen & Lichter 1991). Though these last few studies provide some improvement to the flawed U.S. measure, they are few and far between. Plausibly, it might be better to change our measurement strategies altogether rather than attempting to modify this flawed indicator.
While my approach departs from their strategies, some sociologists have produced innovative alternatives to the U.S. measure. Twelve studies (22.6%) examine recipiency of welfare programs (such as AFDC) as a proxy for poverty (e.g., Harris 1993; McLeod & Shanahan 1993). Four studies (7.6%) examine severe poverty, including families with less than half of the U.S. measure. A slightly larger percentage (15.1%) assessed long-term poverty (e.g., Devine, Plunkett & Wright 1992; Quillian 1999). Finally, eight studies (15.1%) avoid criticisms of poverty measures by simply measuring income, as an ordinal or interval variable, and studying low-income families (e.g., Brooks-Gunn et al.
Sociological Measurement of Poverty / 731
TABLE 3: Content Analysis of Empirical, Quantitative Sociological Research and the Measurement of Poverty between 1990 and 2001
Frequency (Percentage of Total)
Total studies 53 U.S.officialmeasure37(69.81%)
Augmentation of U.S. measure (e.g., 150%) 6 (11.32%) Broader scale, including U.S. measure 5 (9.43%) Recipients of assistance (e.g., AFDC) 12 (22.64%) Deep or severe poverty (e.g., 40% U.S. threshold) 4 (7.55%)
Persistent or long-term poverty
(e.g., length of time under U.S. measure) 8 (15.09%)
Low income 8 (15.09%) Relative measure (e.g., 50% of median income) 5 (9.43%) Examine Comparative/International Variation in Poverty 4 (7.55%) Include post-tax or post-transfer measure 5 (9.43%) Measure inequality among the poor 0 (0.00%)
Note: Table includes all studies in American Journal of Sociology,
Demography, Research in Social Stratification and Mobility, Socia
Social Science Research in which poverty was the dependent var variable.
1993; Duncan et al. 1998; Huff-Corzine et al. 1991; Li Wu 1996). Compared to most sociological studies, embrace issues of relative deprivation, social exclusion would argue that these studies do not quite sufficient in the measurement of poverty.
A smaller number of articles incorporate some o theoretical advances. Five studies (9.4%) use a relative surprisingly, three of the four papers that examine c variation in poverty also use a relative measure (e Kenworthy 1999). The sheer paucity of sociologi comparative international variation is unfortunat predominant case of study for U.S. sociologists of poverty unique and potentially anomalous position in the glob 10% of sociological studies include a posttax or posttran (e.g., Butler 1996; Duncan & Rodgers 1991). It is equall sociologists do not examine posttax and posttransfer prime criticism of the validity of the U.S. measure. sociological study uses a measure that incorporates ine The complete neglect of A. Sen's Ordinal Revolution sociological research.
732 / Social Forces 81:3, March 2003
Overall, there is a fair amount of diversity in the sociological measurement of poverty, with several scholars incorporating innovative and useful techniques. Further, several studies acknowledge important advances in poverty measurement and make note of the many conceptual and methodological limitations of the U.S. measure (e.g., Eggebeen & Lichter 1991). However, on the whole, sociological research has not sufficiently integrated the theoretical and methodological advances in poverty measurement. While it is unlikely that this content analysis is fully exhaustive of all sociological research on poverty, the coverage of seven journals and twelve years provides an illustration of the dominant patterns in the sociology of poverty. By and large, U.S. sociologists too often rely on the problematic U.S. measure.19 Despite the very important contributions of sociological research, our collective potential for scholarly and policy debates is probably limited by these measurement deficiencies.
Data and Methods
Selecting poverty measures has consequences for social policy priorities and
theoretical conclusions (Atkinson 1998a; Hagenaars 1991; Haveman 1987). My
efforts center around establishing criteria and developing measures that ar
accessible, statistically defensible, and methodologically feasible (Atkinson 1987
Citro & Michael 1995). To further evaluate these measures, I conducted a se-
ries of analyses with the LIS. The LIS provides cross-nationally and historically
comparable individual-level data sets. Cumulatively, LIS provides almost stan- dardized data - what the LIS staff call "Lissifed" data with similar variables
across data sets, similar samples, and equalizing weights, which all allow for population estimates (Cantillion 1997). Of course, the LIS data is not perfect and has important methodological limitations (see the LIS Web page: www.lisproject.org). However, the advantages outweigh the disadvantages, and the LIS has great capability to advance the comparative historical scholarship of inequality and poverty. To get estimates of poverty, I first conducted analy-
ses with 74 different data sets to compute a poverty statistic in a given country
in a given year. This analysis generated data on 18 countries with between one and seven time points each, resulting in an unbalanced sample of 72-73 cases.20
The LIS contains data on income, but not wealth or consumption. However, for theoretical (see Hagenaars 1991) and practical (see A. Sen 1992) purposes, income may be preferable to wealth for this type of analysis. People have reasonably good records of income because of taxation; however, less high- quality data exists on wealth, and people simply do not have good records of consumption (Ruggles 1990).21 Following the practices of other LIS analysts (e.g., Smeeding et al. 1993), two measures of income: market-generated and state-mediated income. Market-generated income includes all sources of
Sociological Measurement of Poverty / 733
TABLE 4: Descriptive Statistics for Alternative Poverty Measures Based on Luxembourg Income Study Data, 1967-1997
Coef. of Mean Std. Dev. Variation
Market generated (N = 72) Headcount (H) 31.743 4.387 .138 Poverty income gap (I) .879 .048 .054
Interval poverty (HI) 28.050 4.867 .174 Inequality among poor (CV) 1.360 .544 .400 Ordinal poverty (0) 67.748 2.735 .365 Sum of ordinals (SO) 181.469 35.890 .198
State mediated (N = 73)
Headcount (H) 9.711 3.846 .396
Poverty income gap (I) .681 .065 .095 Interval poverty (HI) 6.616 2.674 .404 Inequality among poor (CV) .478 .202 .422 Ordinal poverty (0) 9.818 4.133 .421 Sum of ordinals (SO) 38.958 13.961 .358
income prior to government taxes and transfe includes all sources of income after taxes and
LIS assigns value to near-cash benefits to pr measure of income.23
Following convention in poverty research (At and among LIS analysts, I standardize the incom family size with an "equivalence scale." Whi different scales (see Triest 1998), the LIS staff e of over 30 different scales and concluded that o approximates all others (see Buhmann, Rainw
1988). Accordingly, I use this equivalence sca Organization for Economic Cooperation and De 1998a). The OECD scale weights the head of the adults as .5, and children as .3, owing to th household's resources that heads, other adults,
Empirical Analyses
The following empirical analysis explores the con measurement decisions. By demonstrating that result in different empirical patterns, this sophisticated measures of poverty provide new information on poverty (Ruggles 1990).
734/ SocialForces 81:3, March 2003
Table 4 displays the descriptive statistics for the poverty measures and their related components for 72-73 observations. Besides the mean and standard deviation, I also list the coefficient of variation (CV). For all poverty measures and components, I report market-generated and state-mediated poverty.
Consistent with past research and the European Union's threshold of 50% of the mean income (Atkinson 1998a), the headcount (H) is computed as 50% of the median income. Unsurprisingly, market-generated H is considerably higher than state-mediated H. Owing to the vast comparative-historical variation in the generosity of welfare states, the CV for H is also much larger for state-mediated poverty, indicating greater variation in H after considering taxes and transfers. The poverty income gap (I) is computed by subtracting the average poor household's income from the median, and standardizing over the median. Like H, state-mediated I is smaller than market-generated I, and again the state-mediated CV of I is larger than the market-generated CV of I for MG. Because H and I are smaller for state-mediated poverty, it is clear that the welfare state reduces the number of poor and the depth of poverty. In turn, state-mediated Interval Poverty (HI) is much smaller than market-generated Interval Poverty (HI), since HI is simply the product of H and I separately. As with the two components, the CV of state-mediated HI is much larger than the CV of market generated HI. While state-mediated HI poverty is significantly smaller than market-generated HI, greater variation exists across countries and years, in state-mediated HI than market-generated HI.
To calculate the Ordinal Measure of poverty (O), I estimated the inequality among the poor. After censoring the sample above 50% of the median income, I calculated the CV of income for the sub-sample of the population that was poor. Thus, CV in this case refers to the dispersion of income among the poor. Since many people have no market income, market-generated CV of income is much larger than state-mediated CV of income. To calculate O, I synthesized CV of income for poor households with HI. Consistent with the components, market-generated O is much greater than state-mediated 0, and the CV of state-mediated O is slightly larger than market-generated 0. Whether utilizing H, I, HI, or 0, state-mediated poverty is more dispersed yet smaller than market-generated poverty. As with the others, market-generated SO is much larger than state-mediated SO, despite more dispersion in state-mediated SO relative to market-generated SO.
Table 5 displays statistics on market-generated poverty and the comparative ranking of 16 nations circa 1995 (ranked such that 1 has the most poor, and 16 has the least poor). Nations like Belgium, Denmark, and Italy consistently have more poverty, while countries like Finland, Switzerland, and Canada have markets that generate less poverty. Importantly, the ranking of countries substantially changes depending on the measure. For example, though Sweden has the third most poverty according to the simple headcount measure (H), it
Sociological Measurement of Poverty / 735
TABLE 5: Patterns in Market-Generated Poverty with Different Measures across 16 Western Nations, circa 1995
Poverty Interval Inequality Ordinal Sum of Headcount Income Poverty among Poverty Ordinals
County Year (H) Gap (I) (HI) Poor (CV) (O) (SO)
Australia 1994 34.3 (8) .919 (3) 31.508 (8) 1.505 (5) 78.934 (5) 205.4 (7)
Belgium 1997 40.1 (1) .947 (1) 37.984 (1) 2.167 (1) 120.296 (1) 257.3 (1)
Canada 1994 29.9(13) .842(14) 25.174(14) 1.025(12) 50.973(14) 157.7 (14)
Denmark 1995 38.9 (2) .903 (7) 35.109 (2) 1.420 (6) 84.954 (3) 230.3 (2)
Finland 1995 30.1 (14) .811 (16) 24.405 (15) .847(16) 45.070 (15) 148.7 (15)
France 1994 36.2 (6) .895 (8) 32.404 (5) 1.301 (8) 74.577 (6) 212.6 (5)
Germany 1994 36.2 (6) .917 (4) 33.192 (4) 1.660 (2) 88.277 (2) 220.5 (3)
Italy 1995 36.8 (4) .919 (2) 33.820 (3) 1.390 (7) 80.823 (4) 217.2 (4) Luxembourg 1994 29.1 (15) .916 (5) 26.661 (12) 1.637 (3) 70.303 (9) 178.4 (12) Netherlands 1994 33.9 (10) .879 (10) 29.805 (9) 1.163(10) 64.458 (10) 190.4 (9)
Norway 1995 34.3 (8) .848(12) 29.081(10) .981(13) 57.603 (12) 181.9 (11) Spain 1990 31.3(11) .908 (6) 28.435(11) 1.583 (4) 73.459 (7) 188.2 (10) Sweden 1995 37.9 (3) .851 (11) 32.246 (6) .949(14) 62.851 (11) 199.3 (8) Switzerland 1992 24.2 (16) .839 (15) 20.315 (16) .918(15) 38.959 (16) 123.9 (16) U.K. 1995 36.3 (5) .882 (9) 32.024 (7) 1.248 (9) 71.989 (8) 209.4 (6) United States 1994 31.1 (12) .846 (13) 26.302 (13) 1.053(11) 53.995 (13) 165.6 (13)
Index correlation
with H (N = 72) .671 .974 .416 .682 .933 Rank correlation
with H (N = 16) .542 .929 .347 .711 .884
Note: The numbers in parentheses are relative rankin poverty with a particular measure and 16 being the l
drops to having the sixth most poverty up and using the interval measure (HI). Further most poverty according to the Ordinal M inequality among the poor (CV). In anoth fifteenth most poverty with H, but slides to with HI, and the ninth most poverty with O
Though some consistencies exist, the rank poverty is sensitive to the measure emp correlation coefficients between H and the o
of cases. While H is highly correlated with HI a with I, the CV, and O. Even though high correl the limited dispersion in Income Gap (CV o only modestly correlated with I and CV, it i O if they are interested in the depth or in
736 / Social Forces 81:3, March 2003
TABLE 6: Patterns in State-Mediated Poverty with Different Measures across 16 Western Nations, circa 1995
Poverty Interval Inequality Ordinal Sumof Headcount Income Poverty among Poverty Ordinals
Year (H) Gap (I) (HI) Poor (CV) (O) (SO)
Australia 1994 13.7 (2) .745 (5) 10.210 (2) .612 (5) 16.454 (2) 63.1 (2) Belgium 1997 7.7 (14) .652 (11) 5.020 (14) .467 (7) 7.365 (13) 30.3(13) Canada 1994 11.1 (4) .650 (12) 7.214 (6) .327 (15) 9.572 (9) 39.6 (7) Denmark 1995 9.1 (9) .776 (3) 7.063 (8) .820 (2) 12.853 (5) 45.5 (5) Finland 1995 4.7 (15) .638 (14) 3.000 (15) .811 (3) 5.432 (15) 18.6(15) France 1994 8.5 (11) .623 (15) 5.300 (12) .438(11) 7.619 (12) 30.4(12) Germany 1994 7.8 (13) .658 (10) 5.132 (13) .383 (13) 7.100 (14) 30.0(14) Italy 1995 13.3 (3) .695 (6) 9.240 (3) .451 (8) 13.405 (4) 52.1 (4) Luxembourg 1994 3.5 (16) .593 (16) 2.075 (16) .230 (16) 2.553 (16) 16.1(16) Netherlands 1994 8.7 (10) .882 (1) 7.672 (5) .647 (4) 12.638 (5) 39.2 (9) Norway 1995 8.5 (11) .669 (9) 5.688 (11) .447 (10) 8.229 (10) 34.7(11) Spain 1990 9.4 (7) .640 (13) 6.017 (10) .345 (14) 8.092 (11) 35.3(10) Sweden 1995 9.3 (8) .748 (4) 6.956 (9) .540 (6) 10.713 (7) 39.5 (8) Switzerland 1992 10.3 (6) .787 (2) 8.110 (4) .911 (1) 15.499 (3) 52.8 (3) U.K. 1995 10.6 (5) .674 (8) 7.143 (7) .449 (9) 10.348 (8) 42.1 (6) United States 1994 18.2 (1) .680 (7) 12.368 (1) .418(12) 17.538 (1) 67.5 (1)
Index correlation
with H (N= 73) .018 .980 -.082 .889 .954 Rank correlation
with H (N= 16) .479 .927 .037 .844 .927
Note: The numbers in parentheses are relative ranki poverty with a particular measure and 16 being the
Table 5 reveals weaker but similar pattern rankings of the 16 nations with H, and th
Table 6 displays state-mediated poverty Again, the rank ordering of nations is equa Consistent with previous research, countr the most poverty, and Finland and Luxe important departures from this pattern ap headcount (H) poverty, the high income g significantly elevate Denmark and the Ne or Ordinal (0) poverty. Also, Canada and S HI or O than with the simple H. It is com like the U.S., Australia, and Canada have m and much less market-generated poverty, w the Netherlands, Sweden, and Denmark exh et al. 1993; Korpi 8 Palme 1998). However,
Sociological Measurement of Poverty / 737
FIGURE 1: Comparison of State Mediated Poverty With Various Thresholds for Four Western Nations
20
| Italy 1986 10 El DSwitzerland 1992
BU.K. 1974 [ U.K. 1995
60% of 50% of 40% of 30% of 20% of 10% of 5% of Median Median Median Median Median Median Median
sophisticated measures of poverty, this simple pa complex.
Also, Table 6 reports the correlation coefficients between H and other measures, with the entire sample, and the correlation in ranking between H and other measures, with the subsample of 16 nations. For state-mediated poverty, H is highly correlated with HI and SO, is less but strongly correlated with 0, and is not correlated with I and CV. Though the measures follow similar patterns, the stark noncorrelation between H and I and CV suggests that if a scholar is interested in the depth and inequality among the poor, it is wise to utilize HI, O, or SO. With the rank ordering of nations, the correlation between H and other measures is less strong but similar.
Some readers may contend that these differences in ranking are small and that given the high correlations among measures, the simple H is still an adequate measure of poverty. To further consider this matter, I display in Figure 1 four cases with similar state-mediated H poverty. Italy at 10.6 in 1986, Switzerland at 10.3 in 1992, and the United Kingdom at 10.7 in 1974 and at 10.6 in 1995 have practically indistinguishable rates of poverty. However, a closer examination of one higher and five lower thresholds - to construct SO and thus include similar information to HI and O - reveals much greater variation. Switzerland has much deeper and more poverty than Italy and the United Kingdom, while poverty increased considerably between 1974 and 1995 in the United Kingdom. Further, although according to H, the United Kingdom
738/ Social Forces 81:3, March 2003
TABLE 7: Trends in U.S. Poverty with Official U.S. Measure and Interval, Ordinal and Sum of Ordinals Measures, 1974-1997
State- Market- State- State- Mediated
Generated Mediated Mediated Sum of
Interval Interval Ordinal Ordinals Offical U.S. (MGI) (SMI) (SMO) (SM SO)
1974 11.2 (6) 23.756 (6) 11.665 (5) 16.821 (5) 63.8 (5) Percentage change +4.464 +.960 -1.312 -2.848 -1.724 1979 11.7 (5) 23.984 (5) 11.512 (6) 16.342 (6) 62.7 (6) Percentage change +19.658 +2.769 +9.147 +6.902 +6.858 1986 14.0 (3) 24.800 (4) 12.565(1) 17.470 (2) 67.0 (2) Percentage change +1.429 +2.677 -3.191 -2.536 -1.642 1991 14.2 (2) 25.464 (2) 12.164 (3) 17.027 (4) 65.9 (4) Percentage change +2.113 +3.291 +1.677 +3.001 +2.428
1994 14.5(1) 26.302(1) 12.368 (2) 17.538(1) 67.5 (1)
Percentage change -8.276 -3.608 -3.064 -2.623 -2.074
1997 13.3 (4) 25.353 (3) 11.989 (4) 17.078 (3) 66.1 (3)
Percentage change, +18.750 +6.723 +2.778 +1.528 +3.605 1974-97
Note: The numbers in parentheses are relative rankings, with 1 being the greatest amount of poverty with a particular measure and 6 being the least. "Percentage change" is the rate of change between the previous and following years, defined as: ((Xt2- Xtl)/Xtl) * 100.
in 1974 would have the most poor and Switzerland would have the least, these countries reverse with a more sophisticated measure (that is, the United Kingdom's SO is 33.3 and Switzerland's SO is 52.8). Such vivid cases illustrate the potentially important role that HI, O, or SO would serve in capturing the complexity and depth of poverty ignored by H. If scholars are interested in the poverty rankings of countries, it is essential that more rigorous measures be used and that multiple measures are evaluated (Atkinson 1987; Cantillion 1997).
While cross-national variation in poverty is greater, historical trends are
important as well. The U.S. provides an informative case to explore historical
trends in poverty (see Table 7). With the official U.S. measure, poverty increased 18.8% between 1974 and 1997. This trend included small increases
between the years 1974 and 1979, 1986 and 1991, and 1991 and 1994. This trend also included a dramatic increase between 1979 and 1986, and a large decrease between 1994 and 1997. However, with more sophisticated measures, the trends in U.S. poverty are more complex. Because the U.S. measure does not fully include taxes and transfers, a comparison with market-generated HI is useful. With this measure, poverty increased a more modest 6.7% over the
Sociological Measurement of Poverty / 739
period. Also, while the years with the most and least poverty remain the same,
the third and fourth worst years reverse with the market-generated HI measure.
In short, a relative measure incorporating the income gap has different trends than the U.S. measure.
As discussed above, the U.S. measure is flawed because it ignores taxes and transfers, so a comparison with the state-mediated HI measure is also valuable. Within the U.S., poverty increased an even more modest amount of 2.8% with state-mediated HI over the period. In addition, contrary to the U.S. measure, with SM HI, poverty actually declined between 1974 and 1979 and between
1986 and 1991. Owing to differences in the depth of poverty and taxes and transfers, factors unobserved with the U.S. measure, state-mediated HI exhibited
very different trends. If interested in the inequality among the poor (CV), one
should also consider the state-mediated 0. With this measure, poverty
increased only a very small 1.5% between 1974 and 1997. Again, with state-
mediated O, poverty actually declined between 1974 and 1979 and between
1986 and 1991. As an alternative, one can examine the state-mediated Sum of
Ordinals (SO) and find similar patterns to state-mediated HI and state-
mediated O, with some shuffling of which years had more poverty. Clearly, the
perceived dramatic increase in U.S. poverty observed by the U.S. measure is
an artifact of measurement error. More realistically, the increase was much
smaller, somewhere between 1.5 and 6.7%. Replicating the decline in poverty between 1974 and 1979 and 1986 and 1991 with three measures establishes
these trends as actual, and vividly demonstrates the problems with the U.S measure. Different historical patterns emerge with different measures, and false understanding emerges with the U.S. measure.
As one final example of the problems of the U.S. measure, we can compare the quantity of households that would be misclassified with the U.S. measure As mentioned earlier, studies have demonstrated sizable differences between
the rates of poverty with the U.S. measure and the NRC's alternatives (Hill & Michael 2001; Uchitelle 1999). With the measures proposed in this article important differences materialize as well. As Table 7 indicates, 13.3% of the U.S. population was officially classified as poor in 1997. With the market- generated H, 30.1% of the population would be poor. This rate, which is more than twice as high, provides the closest comparison since the U.S. measure doe not consider taxes and transfers fully. Considering the impact of taxes and transfers on household income, a state-mediated H provides a more realistic picture of familial income. The state-mediated H for the U.S. in 1997 was 17. 6%; hence the official U.S. measure inappropriately classified 4.3% of U.S households as not poor. With a more theoretically and methodologically defensible measure, the U.S. rate of poverty in 1997 would have been 4.3% higher. Overall, the official U.S. measure provides inaccurate trends over tim and clearly underestimates the extent of poverty in the U.S.
740 / Social Forces 81:3, March 2003
TABLE 8: Results of Market-Generated and State-Mediated Interval Poverty in Western Countries, 1969-1997
Market- State-
Market- State- Generated Mediated Market- State- Generated Mediated Sum of Sum of Generated Mediated
Country Year Interval Interval Ordinals Ordinals Ordinal Ordinal
Australia 1981 25.285 7.844 Australia 1985 27.660 8.006 Australia 1989 26.538 8.084 Austria 1987 4.085 Austria 1995 7.833
165.3 47.0 181.0 48.0 172.2 47.6
21.8
47.3 227.3 20.7 232.8 23.4 250.6 27.5 159.8 84.8 142.4 66.2 130.7 52.5 134.1 42.5 145.4 43.7 154.9 44.2 206.6 43.7 219.8 35.5 230.1 42.8 106.0 23.7 99.9 25.7 178.9 35.5 80.8 35.3 186.9 32.8 218.7 44.0 172.4 35.7 198.9 31.0 219.0 25.7 181.3 23.4 221.0 24.4 198.3 24.9
58.171 11.002 63.778 11.108 59.437 11.216
4.945 12.498 146.120 4.807
130.562 5.506 154.035 6.578 55.945 23.728 46.160 17.800 41.132 13.299 41.548 10.241 45.949 10.771 50.152 12.260 73.487 11.072 78.970 9.334 84.414 11.400 29.810 5.126 27.839 5.921 66.425 11.967 23.252 8.269 69.212 7.967 98.449 14.257 57.176 9.751 66.308 7.997 107.797 5.460 56.538 4.800 92.360 5.222 78.668 5.655
Belgium
Belgium
Belgium Canada
Canada Canada Canada Canada Canada Denmark Denmark Denmark Finland Finland France France France France
Germany Germany Germany Germany Germany GermanyI
1985 33.020 3.350 1988 33.886 3.568 1992 36.189 3.973 1971 25.090 15.101 1975 22.737 12.165 1981 21.007 9.474 1987 21.818 7.662 1991 23.547 7.816 1997 24.643 7.860 1987 32.106 6.919 1992 34.051 5.738 1997 35.016 6.641 1987 18.236 4.006 1991 17.152 4.317 1979 27.455 6.146 1981 13.272 5.972 1984 28.603 5.199 1989 32.191 7.288 1973 27.040 6.215 1978 31.287 5.517 1981 32.095 4.225 1983 28.975 3.925 1984 32.486 4.267 1989 29.473 4.126
Implications for Future Research
These advances in the measurement of poverty pot sociological research on poverty. Plausibly, sociologis these advances for a number of reasons. Myles and Pi advances have not received wider circulation in par technical, mathematical quality of the literature. Furthe presume that these measures are not replicable or co data sets or contexts. Many analysts probably avoid t
Sociological Measurement of Poverty / 741 TABLE 8: Results of Market-Generated and State-Mediated Interval
Poverty in Western Countries, 1969-1997 (Cont'd)
Market- State-
Market- State- Generated Mediated Market- State- Generated Mediated Sum of Sum of Generated Mediated
Country Year Interval Interval Ordinals Ordinals Ordinal Ordinal
Ireland 1987 32.357 5.938
Italy 1986 29.590 6.564
Italy 1991 27.955 6.049
Luxembourg 1985 28.228 3.310
Luxembourg 1991 25.204 2.167 Netherlands 1983 31.454 5.877
Netherlands 1987 31.608 5.108
Netherlands 1991 28.119 5.149
Norway 1979 28.811 4.237 Norway 1986 26.338 5.258 Norway 1991 26.185 4.540
Spain 1980 26.240 7.893 Sweden 1967 30.681 7.740 Sweden 1975 29.675 5.016 Sweden 1981 31.335 4.069 Sweden 1987 31.525 6.846 Sweden 1992 34.540 6.214 Switzerland 1982 21.447 7.127 U.K. 1969 20.265 4.152 U.K. 1974 21.595 6.221 U.K. 1979 29.295 5.024 U.K. 1986 32.819 5.085 U.K. 1991 30.612 8.207
U.S. 1969 21.030
U.S. 1974 23.756 11.665 U.S. 1979 23.984 11.512 U.S. 1986 24.800 12.565 U.S. 1991 25.464 12.164 U.S. 1997 25.353 11.989
208.8 36.5 198.1 35.9 187.1 35.9 193.9 20.9 165.1 17.3
209.8 38.7 212.5 33.3 180.9 32.4 184.7 31.2 169.4 32.7 164.8 30.0 173.9 44.5 208.6 47.2 191.7 33.6 196.8 23.9 198.9 38.1 217.9 34.6 130.6 40.9 132.2 26.9 130.1 33.3 194.2 32.6 212.9 33.2 198.2 46.7 131.1
148.5 63.8 149.8 62.7 155.4 67.0 159.2 65.9 158.8 66.1
80.231 9.147 79.378 8.331 68.276 8.059
103.096 4.384 61.149 2.566 87.371 12.806 89.523 10.479 60.685 8.588 64.275 6.479 56.061 7.154 51.270 6.758 68.528 10.537 90.429 12.004 67.419 7.591 63.954 6.030 63.460 10.278 71.462 9.124 39.402 9.788 44.802 5.026 37.738 7.389 71.776 6.645 77.435 8.512 68.816 10.763 40.893
48.626 16.821 47.979 16.342 50.146 17.470 50.916 17.027 51.522 17.078
Note: Cases were excluded from this table that were included in Tables 5 a
avoid arguing first principles and justifying what reviewe controversial. Ultimately, I presume that the official U.S. legitimacy for sociological readers and convenience for soc None of these concerns, however, need prohibit sociologists and using more sophisticated measures of poverty.
In pursuit of this goal, I advocate for three alternative measu These three measures, the Interval (HI), Ordinal (0) and (SO), emerge directly from the recent literature on poverty
742 / Social Forces 81:3, March 2003
meet the aforementioned criteria. Unlike much of the previous literature, my discussion has been intentionally less technical in order to disseminate these advances to a wider audience. As mentioned earlier, each of the measures
carries certain advantages depending on one's research interests. To adhere to A. Sen's Axiom R, analysts should use 0; for graphical representations, analysts should use SO; and, for a simple, parsimonious adherence to the five criteria, analysts should use HI. Utilizing these measures, sociologists of poverty can proceed in a number of directions.
First, for additional analyses of the amount of societal poverty, I have
supplied further data on different poverty measures. Table 8 displays estimates
on the remaining cases available with the LIS as of May 2002. I have calculated
the market-generated HI, state-mediated HI, market-generated SO, state-
mediated SO, market-generated 0 and state-mediated 0 measures for 74 total observations. This data set includes 18 OECD nations and covers the historical
period 1967 to 1997. Hopefully, with the publication of these results, other scholars will utilize these estimates for analyses of the causes and consequences of comparative historical variation in poverty. Second, other individual-level data sets can easily produce these measures of poverty. Analysts can easily calculate these measures with basic descriptive statistics on the entire sample and the sub-sample of households that are below 50% of the median income.24 Future analysts should replace the U.S. measure with a relative measure that incorporates the depth of poverty.25 Of course, though not all data sets will have the detailed information on income that the LIS provides, analysts can estimate relative measures of poverty with as much information as possible. Overall, analysts will find that constructing and using these more sophisticated measures of poverty is quite simple. Because these measures are easy to incorporate, sociologists should use them in future research.
Conclusion
This article seeks to advance the sociological measurement of poverty. Unfortunately, much of sociology still relies on the U.S. measure despite serious methodological problems. These methodological limitations are crucial and probably limit sociology's collective contribution to the understanding of the causes and effects of poverty. As an alternative, I argue that scholars should cultivate measures of poverty that meet five criteria. First, scholars should use measures of poverty that effectively gauge comparative historical variation. Second, analysts should measure poverty as relative rather than absolute. Third, poverty should be conceptualized as social exclusion. Fourth, poverty indices should measure the depth and inequality among the poor. Finally, analysts should incorporate taxes, transfers, and state benefits when calculating household resources. These criteria reflect theoretical and methodological
Sociological Measurement of Poverty / 743
developments that will be useful for sociology. For sociological research on poverty to advance, it is essential that scholars embrace and incorporate these developments.
This article also provides an empirical analysis of patterns with three alternative measures: the Interval, Ordinal, and Sum of Ordinals. The empirical analysis with the LIS data demonstrates that the amount and the cross-national ranking of poverty fluctuate with the particular measure used. Further, simple headcount measures produce limited and potentially less accurate information about comparative and historical variation in poverty. In an analysis of poverty in the United States since the early 1970s, the U.S. measure clearly provides inaccurate results about trends. With state-mediated poverty, the Interval, Ordinal, and Sum of Ordinals measures display important declines in poverty that are not captured with the U.S. measure. Further, the U.S. measure misdiagnoses the increase in poverty since the early 1970s. To understand the causal mechanisms driving poverty, it would be valuable for scholars to consider multiple measures and utilize one of these more sophisticated measures.
If sociologists seek to make scientific inferences and inform public policy, it is imperative that new measures of poverty be developed and integrated into the discipline. At present, the contribution of U.S. sociology is probably limited
by the reliance on the U.S. measure and, in the few comparative studies, on a simple headcount measure. While the sociology of poverty has grown considerably over the past few decades, the discipline remains unfortunately out of step with advances in poverty measurement. Further, in the 1990s, the sociology of poverty cultivated increasingly sophisticated statistical analyses of poverty, yet left the more fundamental issue of measurement largely neglected. Plausibly, the sociology of poverty would benefit more by first scrutinizing the basic and primary methodological concern of measurement before proceeding with increasingly sophisticated statistical analyses of the causes and consequences of poverty.
Altogether, these criteria and alternative measures of poverty create a po- tentially fruitful new direction for research in sociology. With this redirection,
it is possible that sociology could experience a second reinvigoration of research on poverty. This article seeks to facilitate this second reinvigoration and to encourage sociology to benefit from more sophisticated measures of poverty that are grounded in theoretical and methodological advances.
