US Multidimensional Poverty by Race, Ethnicity and Motherhood: Evidence from Pennsylvania Census Data

Abstract

The combined influence of gender and race has been a defining feature of poverty in the USA, especially for single mothers. Recent applications of capability-based multidimensional poverty (MP) measurement to US data have examined race and gender, but little attention has been given to the intersection of the two. We address this gap in the literature on multidimensional poverty by employing household-level US Census data for the years 2006–2010 that are from Pennsylvania, a state with key income poverty indicators close to the mid-poverty values for all fifty states. We employ a dual cut-off procedure to present MP measures by levels of population sub-groups. The poverty ranking by single motherhood shows Hispanics are the most deprived, Whites as the least deprived, and African–Americans coming in between. Our findings suggest that the provision of child care facilities can prove effective for poverty reduction; and the improvement of language skills is likely to be critical for Hispanics.

Appendices

Appendix 1: Aggregate Level Decomposition

From our modified sample given in Appendix Table 9.8 (bottom row), we obtained an Income Poverty Headcount (YPH) of approximately 0.094, while the Multidimensional Poverty Headcount (MPH) according to our indicators is 0.168 along with a Poverty Intensity (A) of 0.453. Thus, we obtain our Multidimensional Poverty Index M ₀ as the multiple of MPH and A, namely, 0.076.

Table 9.8 Decomposition, ethnic subgroups

• 1. Decomposition by Racial and Ethnic Sub-groups

To examine the distribution of deprivation among ethnicities, we decompose the index by ethnic sub-groups. Appendix Table 9.8 shows that the M ₀ has the same poverty ranking as the YPH. Hispanics are the most deprived of the sub-groups, Whites are the least deprived of the sub-groups, and the African–American sub-group is in the middle. Similar to the general index, the M ₀ index for each sub-group is smaller than its YPH. Statistically, we observe that the M ₀ indices of African–American and Hispanic sub-groups are three and three and one-half times greater, respectively, than the M ₀ index of the White.

• 2. Breakdown of the Dimensions

We decompose the aggregate M ₀ as indicated by Eq. (9.4). For the full sample (which has the M ₀ index as 0.076), the most significant dimension is the dimension of work status (i.e., the employment status of husband and wife) with a non-deprived household defined as having at least one member employed full time. It has an index of 0.12, which contributes 41% of the total deprivation. The second one is the dimension of education, which captures 30% of the deprivation with an index of 0.09. The other two dimensions, dimension of income deprivation and dimension of living standard, with their indices from 0.04 to 0.05, have less impact than the work and educational dimensions (Table 9.9; Fig. 9.2).

Table 9.9 General dimension breakdown

Fig. 9.2

Breakdown, general

In the remainder of this Appendix, as in the main text, we discuss the results of applying our Quadruple Decomposition Model to the above sample.

• 3. Decomposition by Household Type Sub-groups

Examining Appendix Table 9.10 for household type decompositions, we obtain an index for the married couple household sub-group and male-headed household sub-group of 0.05 and 0.08, respectively. By contrast, the index for the female-headed household sub-group (0.15) has a notably higher ranking than the M ₀ index of the married couple household sub-group and twice as large as the index of the male-headed household sub-group. Female-headed households are 22% of the sample; they contribute 42% to the deprived population, whereas the married couple households are 64% of the population, yet their contribution to the deprived population is just under 43%; the male-headed households’ percentage contribution remains unchanged. Therefore, the female-headed household sub-group is the most-deprived sub-group in this decomposition.

Table 9.10 Decomposition, African–American sub-group

• 4. Summary: Aggregate Level Decompositions

From the two decompositions above, we observe that the most-deprived sub-groups are the following: Hispanics , African–Americans , and female-headed households.

Appendix 2: Robustness Test

An alternative approach to MP is fuzzy set theory. Its application draws strength from the idea that it is futile to attempt exact measurement of poverty since the concept is inherently ambiguous. The fuzzy set approach takes into account the vagueness of the distinction between the poor and non-poor, allowing partial membership in a poverty set based on a membership function, F*π, where F is the degree of membership in the [0, 1] interval, with π = 1 for those definitely poor and with π = 0 for those definitely non-poor. For those who are poor to some degree, 0 < π < 1 (Cerioli and Zani 1990). The in-between set then allows for different definitions of poverty groups. ¹⁶ A simplified version of this approach is to focus only on the lower end of the poverty scale and to examine changes in the measurement of poverty resulting from degrees of high chronic poverty (Wagle 2009).

We examine two alternative methods to check for the robustness of our M ₀ results when determining the poor vs. non-poor boundary. One method is to adopt a different value for the poverty threshold k in the dual cut-off method. The other is to allow for fuzziness or vagueness in the poverty boundary by the fuzzy set methodology discussed above. We employ both approaches for robustness to the ranking of high chronic poverty population groups. However, since the main purpose of our fuzzy set measures is to check for poverty rank robustness rather than obtaining a full set of fuzzy poverty calculations, we rely on simple fuzzy set measures of MP for two different types of boundaries separating the poor from the non-poor. ¹⁷ The first measure is based on counting a household as poor if deprived in at least two of the four capability dimensions. We call this the Fuzzy Set Lower Half (FSLH) measure of standard of living. Given our limited purpose in employment of the FSLH, we set the threshold for the FSLH at 0.375. This threshold value results in an aggregate FSLH population in poverty similar to that obtained by the M ₀ index. To check for robustness in vagueness in the poverty boundary, we also calculate and rank another fuzzy poverty measure based on counting a household as poor if it is deprived in all four dimensions of capability. We call this the Fuzzy Set Extreme Poverty (FSEP). Then, we compare poverty by all three measures to find out if our ranking by poverty populations changes.

We rely on the larger full sample in order to obtain a more reliable robustness test of the MPI method employed in this study. We use the dual cut-off method by changing the threshold k = 3 to values lower and higher than 3 so now k equals 2, or 3, or 4. We apply this procedure to the ranking of poverty status by the race of the household head. Second, we employ fuzzy set measures of MP based on the FSLH and FSEP. The results by the dual cut-off M ₀ method are displayed in Appendix Fig. 9.3 and are reported in Appendix Table 9.11. Appendix Fig. 9.3 shows that for each value of k, for instance k = 2, the M ₀ value for Hispanics is the highest and the value for Whites is the lowest with African–Americans in between. Note, however, that Appendix Table 9.11 also shows that the gap between White poverty and non-White poverty becomes more pronounced as we adopt k values closer to extreme poverty (third column).

Fig. 9.3

Robustness of Female-headed households with child. (a) % MPI: percentage of population deprived by MPI standard. (b) % FSLH: percentage of population deprived using. (c) % FSEP: percentage of population deprived using fuzzy set extreme poverty standard

Table 9.11 M ₀ according to different k

Appendix Table 9.12 and Graph IId focus on the main household populations by gender, race or ethnicity , and the presence of children that are identified above as high chronic poverty groups in Pennsylvania; comparing M ₀ with FSLH and FSEP values. Once again, allowing for vagueness in separating the poor and non-poor by FSLH and FSEP produces the same ranking as M ₀. That is, Hispanics have the highest poverty ranking and Whites the lowest and African–American are in between; and once again the FSEP gap between the White and non-White poor is more pronounced compared to the FSLH.

Table 9.12 Robustness of female-headed household with child using fuzzy set theory

We conclude that our capability-based measures of poverty appear to be robust to the cut-off point employed for M ₀ calculations and are robust if we allow for a substantial degree of fuzziness in the boundary between the poor and the non-poor.

Appendix 3: Poverty Profiles: Income-Based and Multidimensional

In this Appendix, we present findings from the estimation of probit models with the goals of predicting the probability of being in poverty conditional on the variables revealed in the previous section to be the important determinants of the MP index and identifying and comparing the poverty profile for Pennsylvania by the M ₀ index and Head-Count Income poverty. The dependent variable indicator is zero for non-poverty status and 1 for poverty status (if k > 3 in column 3, and if income is below the poverty line in column 2). The analysis is exclusively in terms of the combined household categories defined by race or ethnicity , gender, and the presence of children. With three racial/ethnic groups in this study, there are nine demographic categories that depend on the gender and marital status of the heads of households with and without children. The White non-married male-headed households act as the excluded base category. The results appear in Appendix Table 9.13, the second column for income-based poverty and the third for M ₀ poverty.

Table 9.13 Probit estimates by income-based poverty and by multidimensional poverty for probability of falling into poverty conditional on key features of the household

The poverty profiles suggested by the two approaches have a great deal in common, and both are highly consistent with the M ₀ analysis and results reported above. First, all categories in both models, with the exception of White male-headed households with children and White female-headed households without children in the third column, affect the poverty profile in the expected positive direction. Second, note that within each racial/ethnic category, the coefficient estimates for female-headed households with children are the largest in size and are the most significant. This suggests single motherhood is the most important feature that increases the chances of being poor in terms of income or M ₀. Still, there are equally significant racial/ethnic differences within the female-headed households with children in terms of the increased chance of being poor. This brings us to the third aspect: the estimated coefficient size for the female-headed households with children can be ranked as the largest for Hispanic , the next largest for African–Americans , and the smallest for Whites, with their statistical significance levels following the same ranking. In general, however, there is a considerable degree of agreement between the earlier M ₀ analysis and the probit results here; the latter confirm and reinforce the main results of the above M ₀ calculations.

The probit analysis for the likelihood of being poor, based exclusively on the intersection of race /ethnicity , gender, and marital status of heads of households, indicates that the demographic variables identified above are effective conditional variables for obtaining probability estimates of household poverty status in Pennsylvania.