Capability Deprivation and Income Poverty in the United States, 1994 and 2004: Measurement Outcomes and Demographic Profiles

Some of these constructs attempt to capture multiple dimensions of poverty or deprivation in an integrated framework. Wagle’s (2002, 2005, 2008a, b) treatment, for example, incorporates capability, the topic of interest here, as one of the different dimensions of poverty together with economic well-being and social inclusion. ‘Relative deprivation’ is another construct applied to capture individual frustration or unsatisfying emotional experience using economic as well as non-economic indicators (Coleman 1990; Rankin and Quane 2000; Stewart 2006; Walker and Smith 2002).

While the term ‘capability poverty’ has been included in the lexicon of international poverty research, ‘capability deprivation’ is used here to differentiate from ‘income poverty’ even though both concepts indicate ‘deficiency’ or ‘shortfall’ of something tangible or intangible.

As a multidimensional construct, capability would have to be jointly determined by one’s status on its various dimensions. These dimensions can also be complex and unobservable, measurable only by using multiple indicators of their own. In this paper, however, I use ‘dimensions’ and ‘indicators’ interchangeably.

As will be clear shortly, one of the alternatives is to measure deprivation status directly without ascertaining the cardinal capability scores. In this case, the issue of aggregation can be avoided altogether. The issue of weighting indicators still needs to be dealt with, however, as some indicators may be less important than others.

Incorporating value judgments, for example, the UNDP (2006) uses equal weights in aggregating various macro indicators to determine the Human Poverty Index where as Wagle (2005, 2008a) uses an empirical procedure (factor analysis) to aggregate various micro indicators of capability.

The widely cited disagreement between Sen (1985b) and Townsend (1985) is quintessential of the tension between using absolute and relative criteria, with the former arguing for absolute and the latter for relative criteria.

The degree of knowledge, for example, changes consistent with or even beyond what its formal indicators, such as educational attainment, can capture. A specific example is educational attainment which has significantly improved in the past few decades in the US, making college education, once considered beyond a norm, more typical today. Any capability deprivation threshold focusing on educational attainment would have to be relative to the given time and context.

They also apply some subjective deprivation criteria developed from a survey and find that the measurement outcomes may be rather different between different approaches.

In this case, the degree of membership would be computed as \( \frac{{(D_{j}^{**} - D_{j} )}}{{(D_{j}^{**} - D_{j}^{*} )}} \)for each indicator, with the weighted average of all indicators determining the degree of membership to the capability poor.

Depending on one’s motivation, measurement with more elaborate degrees of deprivation can be useful in some situations. In the policymaking world, however, one typically needs to be categorized either as deprived or as not-deprived, with being deprived qualifying the person for some policy supports.

This is to note that I include only one common factor or dimension in the expression with the expectation that capability will turn out to be the only factor to significantly load on the indicators. In actuality, however, this can be thought of as a matrix of common factors as determined by the indicators used.

This partly reflects the actual degree of capability since one with the overall capability score below the reference score would be regarded as deprived even if he or she scores much better on one or some of the indicators.

I expect data from these years to approximate the economic trend for the 1990 and 2000s. Admittedly, the economic and labor market changes of the 1990s reached the pinnacle by 1998–1999, with the cycles of the 2000s still in effect. Because these years witnessed the expanding and shrinking modes of the economy, data from these waves capture important changes in the economic and non-economic behavior of households.

Correlations between them are around 0.86 as reported in Table A2 (in the “Appendix”).

The occupational prestige scores refer to the ‘social standing’ of the different occupations as assessed by the respondents. The 1980 occupational prestige scores used here were developed from the GSS data of 1989 using the 1980 Census occupational classifications (NORC 2006).

This is also to note that my attempt to use the ‘treated with respect’ variable measuring the actual level of respect, which the respondents enjoyed in their work life, was unsuccessful since this question was not included in the 1994 wave of the Survey.

Five percent of the respondents had missing values on prestige where as up to 52% (33% in 1994 and 52% in 2004) of the respondents had missing values on health condition.

This is justified since all of the important socio-demographic variables have complete data. The variables used in making such predictions include age, gender, nativity, race, marital status, religion, household size, number of adults, number of children, number of earners, education, income, region, dwelling type, and occupation. While not all variables would turn out to be significant in each case, I use a consistent set of predictors as it would not lead to any more or less biased predictions. Because the original set of indicators being imputed contains discrete values, I recode the imputed values to gain consistency.

For this purpose, I use a specific threshold on each indictor. I use the high school level of education or 12 years of schooling as the cutoff point for educational indicators where as other cutoff points include ‘fair’ or ‘poor’ response categories in case of health condition and a score of 30 or lower in case of occupational prestige. These thresholds may look arbitrary given the sensitive nature of their usage. At the same time, however, these are highly conservative and intuitively appealing thresholds. The criterion of less than 12 years of schooling or less than high school equivalent education, for example, is fairly accurate to use as deprivation cutoff point given the new labor market situations in the US. This is consistent with the Bourguignon and Chakravarty’s (2003) application of the fourth grade of education in rural Brazil and, particularly, with Alkire and Foster’s (2007) application of the high school and equivalent education in the US. The self-evaluated health condition that is short of ‘excellent’ or ‘good’ disallows one to effectively function in the labor market or community and therefore can be considered as a basis for poor health. My operationalization of health condition is similar to that of Stewart (2006) who looked at health as a unidimensional indicator of deprivation. Establishing a convincing threshold for occupational prestige, however, is problematic since the scores do not have practically sensible values. With the possible range of 17 and 86, the established cutoff of 30 represents the scores associated with select sales workers, private household cooks, industrial machinery repairers, glaziers, food batch-makers, select machine operators, truck drivers, and construction trade helpers. Immediately below these are such occupations as garbage collectors, taxi drivers, mining occupations, carpenter apprentices, ground-keepers and gardeners, and elevator operators. While these occupations may provide large economic payoffs, their prestige levels are relatively low that the occupants are likely to deprive of the self-respect they need to effectively function in society. From the relative standpoint too, this threshold suggests less than 21% as deprived in 1994 and less than 18% as deprived in 2004. See Table A3 (in the “Appendix”) for the measurement outcomes following these thresholds on each indicator.

The second common factor secures a considerably smaller Eigenvalue, an indication of a small commonality to form this factor. See Table A4 (in the “Appendix”) for results.

The income poverty estimates are based on the official poverty thresholds for 1993 and 2003, the years to which data on household income refer. While the before tax incomes reported in the GSS data are not highly precise compared to those from other surveys, such as the Current Population Survey and the Survey of Income and Program Participation, they are helpful to identify the magnitude of poverty on the income space given that the poverty thresholds are specific to different household sizes especially at the lower end of the distribution. These thresholds are adapted from the Census Bureau (2008). Comparatively, these poverty estimates appear to be somewhat lower than the 15% official poverty estimates for 1994 but are very similar to the 12% estimate for 2004 (Census Bureau 1995, 2005).

This finding is not quite clear from Table 3, partly because the final measurement outcomes are not increasingly different.

As with any Logit model, the purpose is to determine the probability of being poor or deprived given the different values of the explanatory variables. The generic version of the model takes the following form:

\( \Pr (Y = 1\,X_{1} ,\,X_{2} ,\, \ldots ,\,X_{k} ) = F(\beta_{0} + \beta_{1} X_{1} + \beta_{2} X_{2} + \cdots + \beta_{k} X_{k} ) = \frac{1}{{1 + e^{{ - (\beta_{0} + \beta_{1} X_{1} + \beta_{2} X_{2} + \cdots + \beta_{k} X_{k} )}} }} \)

Where Y is the poverty or deprivation status and X ₁, X ₂, …, X _k are the various demographic characteristics. Rather than getting to the actual probability, however, the model outputs provide the effect of each variable on changing the log of the odds of being poor or deprived, which moves in the same direction as the probability of being poor or deprived.

The pooled versions of the model by combining data from 1994 and 2004 suggested a consistent role of the time factor. The results of this estimation are not reported here to save space and are available from the author.

Countries with comparable human development indices including Norway, Iceland, Australia, and Ireland made over 0.03 improvement compared to 0.01 of the US (UNDP 1996, 2006).

Sen’s (2000, 2006) more recent treatise on the subject has further expanded the purview of the approach from the inner strength and resources to the inner and outer resources including relational qualities of life. Partly, this has occurred in response to a charge that capability approach fails to accommodate the issues of social exclusion, which emanate not from the individual failures but from the failures of the larger society or the state. As Ballet et al. (2007) argue, furthermore, the agency aspect of well-being, which necessitates acting out of responsibility for the well-being of not only oneself but also of others, expands collective capability and freedom. The individual capability is therefore intricately entwined with the collective or relational capability further complicating the issue.

It is difficult to fully disentangle the non-demographic factors that may have contributed to these positive changes, as one’s performance on the indicators used to measure capability are developed over time and thus are hard to improve in the short term. But some of the credit may need to go to the booming economy that enabled many to gain education, move up in the career ladder, and maintain good health. The overall direction that the society is taking with wider recognition of the role of education in the labor market as well as elsewhere may have also played a positive role.