Using the American community survey to create a National Academy of Sciences–style poverty measure: Work by the New York City Center for Economic Opportunity

Abstract

The need to improve the U.S. poverty measure has received renewed attention as state and local governments have initiated antipoverty efforts and wish to judge their effect. This paper describes the New York City Center for Economic Opportunity's implementation of the National Academy of Sciences' recommendations for measuring poverty. The center's decision to use the Census Bureau's American Community Survey as its principal data source created the project's central challenge; many of the items needed to construct the academy's measure of resources are not included in the survey and needed to be estimated through a variety of methods. The resultant measure creates a higher poverty rate and a demographic profile of the poor that is quite different from that generated by the official measure. The paper concludes with observations about these differences and how this new picture of poverty has begun to influence policymaking in New York City. © 2010 by the Association for Public Policy Analysis and Management.

This is an unusual Professional Practice in a couple of respects. For example, this Professional Practice has a guest editor, Kenneth Couch, from the University of Connecticut. Ken and I, plus some of the authors in this section, were participants in a conference on measuring poverty, social exclusion, and well-being that was held at the Organization for Economic Cooperation and Development, held in Paris, in March 2009. Also, unlike most Professional Practice publications, this one focuses on measurement of a key construct for policy analysts, public managers, and policy makers—poverty. So essential is this construct that I felt it was important to relay the European experiences and thinking on these issues and its relevance for the current poverty measurement in the U.S. I hope you enjoy this section as much as I did.

  Maureen A. Pirog, Editor-in-Chief, JPAM

The inadequacies of the official U.S. poverty measure are well known to social scientists. In 2006 they became vividly clear to New York City policymakers. Mayor Michael Bloomberg convened a Commission for Economic Opportunity and asked its members to develop new ideas for addressing poverty in the city. In the course of their work the commissioners became frustrated with how little the current poverty measure could tell them about either the degree of economic deprivation in the city or the effect of programs intended to alleviate it. In its report to the mayor, the commission urged that, in addition to new antipoverty programs, New York City should develop a better method to count the poor (Commission for Economic Opportunity, 2006). Mayor Bloomberg embraced the suggestion and poverty measurement has become part of the mission of the organization created to implement the commission's recommendations: the New York City Center for Economic Opportunity (CEO).

This paper summarizes CEO's initial effort to apply the National Academy of Sciences' (NAS) recommendations for measuring poverty using the U.S. Census Bureau's American Community Survey (Center for Economic Opportunity, 2008). The first section provides our rationale for adopting the NAS method. We then lay out the steps we took to construct the measure. The third section provides some of the results of our work, contrasting poverty rates derived from the official and NAS-recommended methodologies and offering some thoughts about what is driving the pattern of differences. We conclude with remarks about how New York City is making use of the alternative measure.

RATIONALE FOR CEO'S USE OF THE NATIONAL ACADEMY OF SCIENCES' ALTERNATIVE

CEO set out to create a poverty measure that is useful for policymaking. Because poverty measurement is a controversial topic, and there are a variety of ways to measure it, our choice was guided by two criteria. First, the alternative measure should be grounded in a substantial body of research and supported by experts in the field. The credibility of a “CEO poverty measure” would rest, in part, on the degree to which it is based on research by, and consensus among, expert analysts. Second, the new measure should be easily understood by the nonexpert public. This suggested that rather than a radical departure from the familiar, if flawed, official measure, a new approach should seek to maintain its structure (economic resources measured against a set of thresholds that are derived from expenditures on necessities) but improve its components.

A more useful measure of poverty, it is widely recognized, requires a more comprehensive definition of family resources than what is used in the official measure, pretax cash income. Although tax credits and near-cash, in-kind benefits have been a growing share of government antipoverty expenditures for decades, this support to low-income families is uncounted by the official poverty measure. Policymakers want to know how an effort to increase the take-up rate for food stamps or to provide a new tax credit to working parents would lower the poverty rate. They cannot get an answer from the official method.

CEO also believed that to better understand the adequacy of antipoverty policy, a family's resources needed to be measured against a more realistic set of poverty thresholds. The official poverty thresholds are uniform across the nation. The poverty line that defines who is poor in Manhattan is the same as that applied in rural Mississippi. The need to account for New York's relatively high cost of living is an obvious concern, where housing costs put a tight squeeze on family budgets.

CEO concluded that it should base its poverty measure on an alternative methodology that, at the request of Congress, had been developed by the National Academy of Sciences' (NAS) Panel on Poverty and Family Assistance in 1995 (Citro & Michael, 1995). The NAS panel recommended that the poverty thresholds reflect family needs for food, clothing, shelter, and utilities plus “a little more” for other necessities; that the threshold be updated annually by the change in median family expenditures on these items; and that the thresholds be adjusted to reflect differences in the cost of living across the U.S.

The NAS panel also recommended that in addition to cash income, the resource measure should account for the effect of tax liabilities and credits, along with the cash value of near-cash, in-kind benefits that families can use to obtain the necessities represented in the threshold. The panel further suggested that resources be adjusted to reflect necessary work-related expenses such as commuting costs and child care. Because what a family must spend to maintain the health of its members is unavailable for purchasing other necessities, the panel also proposed that medical out-of-pocket expenses be subtracted from income.

CREATING THE CEO POVERTY MEASURE

The NAS panel and subsequent research provided a straightforward method for calculating New York City–specific poverty thresholds and offered a conceptually consistent definition of resources. CEO's decision to base its measure on the Census Bureau's American Community Survey presented the key challenge in implementing the NAS methodology. Many of the items needed to construct a measure of resources consistent with the NAS' methodology are not included in the survey. As detailed below, these were estimated for each family through a variety of approaches including program rules, administrative data, and imputation techniques.11

Establishing the New York City Poverty Threshold

The NAS panel did not recommend a specific poverty line; instead, it suggested that the threshold fall between the 30th and 35th percentile of the distribution of the amounts that families spend on the items in the threshold. (These percentiles were equivalent to 78 percent and 83 percent, respectively, of the median level of spending on these goods at the time of the report.) The panel also offered an upper and lower bound for the “little bit more” that it recommended be included in the threshold, a multiplier ranging from 1.15 to 1.25 times the food, clothing, shelter, and utilities expenditure estimate.22 In its NAS-related alternative poverty measures research, the Census Bureau has used the mid-point of the percentage of the median (80.5 percent) and multiplier (1.2) for miscellaneous expenses (Dalaker, 2005; Short et al., 1999; Short, 2001). CEO continued that practice. For 2006, this methodology produces a U.S.-wide poverty threshold for a family composed of two adults and two children of $21,818.33

The academy argued that because living costs were not uniform across the United States, the poverty thresholds should be geographically adjusted. Since research indicates that the largest source of the disparity in inter-area living costs are differences in housing and utility costs, the panel recommended that only the part of the threshold that is made up of shelter and utilities expenditures be adjusted. It further suggested that the U.S. Department of Housing and Urban Development's Fair Market Rents could be used as the adjustment factor (Citro & Michael, 1995, pp. 182–201).

In its NAS-related research reports, the Census Bureau has used 44 percent as the share of the total threshold that represents shelter and utility expenditures. For 2006, this share equaled $9,600 for the NAS reference family of two adults and two children. CEO adjusted this amount to take account of the high cost of housing in New York City. We compared the New York metropolitan area fair market rent (FMR) for a two-bedroom apartment to the national average (weighted by population) for a similar apartment and found that New York City rents for such apartments were 1.45 times the national average. This factor was applied to the U.S.-wide shelter and utilities share of the threshold, generating a new shelter and utilities portion of the reference-family threshold of $13,920.44 When this is added to the non–shelter and utilities portion of the threshold (which remains unchanged), the total threshold for the reference family of two adults and two children comes to $26,138. This threshold is about 20 percent higher than the U.S.-wide NAS threshold and about 28 percent higher than the official poverty line.

Once a threshold for the reference family has been set, thresholds need to be calculated for families of other sizes and compositions. Our study used the three-parameter scale developed by David Betson. This equivalency scale is now used in the Census Bureau's experimental poverty measure reports and has gained wide acceptance among poverty researchers (Betson, 1996). Table 1 compares the CEO against the official poverty thresholds for a variety of families.

Table 1. Comparison of poverty thresholds, 2006.

Family Composition	CEO	Official	CEO/Official
Source: U.S. Bureau of the Census and CEO calculations. a ^a Adult is non-elderly in official threshold.
One adult,a no child	$12,114	$10,488	1.155
Two adults,a no child	$17,081	$13,500	1.265
One adult, one child	$18,280	$13,895	1.316
One adult, two children	$21,702	$16,242	1.336
One adult, three children	$24,906	$20,516	1.214
Two adults, one child	$23,006	$16,227	1.418
Two adults, two children	$26,138	$20,444	1.279
Two adults, three children	$29,116	$24,059	1.210

Measuring Family Resources

Once calculated, the appropriate poverty lines are assigned to families and then compared against their resources to determine if the family members are poor. CEO used the New York City public use micro-data sample from the 2006 Census Bureau's American Community Survey (ACS) to represent the city's population and as the principal source of information for calculating family resources. The ACS is now the largest of the Census Bureau's annual demographic surveys, covering roughly 3 million addresses across the United States. The sample (over 25,000 households in New York City) is sufficiently large to analyze poverty across demographic groups and neighborhoods. Without this rich level of detail and the ability to track year-to-year changes, our measure would be far less useful for understanding poverty in New York City. The ACS also contains much information relevant to poverty status, such as living arrangements, school enrollment, educational attainment, race, citizenship, and employment, as well as income from a variety of sources, including earnings, social security, public assistance, and Supplemental Security Income, along with interest, dividends, and rental income.

Although the 2006 ACS provided data on cash income and the value of food stamp benefits, many of the other elements of a family's resources are not collected in the survey. These include taxes, participation in school-based nutritional assistance, receipt of housing assistance, commuting costs, childcare expenses, and medical out-of-pocket spending. The remainder of this section describes the methods used to develop estimates for these items.

Taxation

The CEO income tax model estimated net taxes (liabilities minus credits) for New York City tax filers. This was done in several steps: tax filing units (filers and their dependents) were created within each household, with the filing status of each unit determined by its composition (for example, a married couple filing jointly or a single filer). Then a simulated tax return for federal, New York State, and New York City income taxes was constructed for each tax filing unit. ACS income variables were used to estimate adjusted gross income. To compute taxable income, all filers were given a standard deduction and dependent exemptions were applied based on the number of dependents assigned to each filer in the creation of the tax unit.

Next, the model calculated income tax liabilities using the appropriate tax bracket and tax rate for the taxable incomes. Net income taxes were then estimated by applying the appropriate tax credits against the liabilities. The credits in the model were those that are most relevant to low-income taxpayers. These are listed in Table 2. Finally, payroll taxes were applied against earnings.

Table 2. Tax credits in the 2006 CEO model.

Federal	N.Y. State	N.Y. City
Child and dependent care	Household credit	Household credit
Elderly and disabled	Child tax credit	School tax relief
Child tax credit	Child and dependent care	EITC
EITC	EITC
Additional child tax	Real property tax credit
Education credit	College tuition

The difference between pretax and post-tax income is most apparent as taxpayers' incomes approach $20,000. Below $20,000 the refundable tax credits more than offset income tax liabilities and payroll taxes, and posttax income exceeds pretax income by an average of $346. But many tax credits phase out for filers with incomes between $20,000 and $40,000. Their after-tax income is, on average, $4,196 below their pretax income. The net effect of tax programs on income also varies by family composition, an issue explored in the third section.

Nutritional Assistance

A variety of public programs help low-income families meet their nutritional needs. Participation in these programs also frees a family's limited income for other uses. Although assistance usually takes the form of an in-kind benefit, the NAS recommended that the cash-equivalent value of this assistance be added to a family's income. CEO added the value of food stamp receipt and participation in the school lunch program to the resource measure.

Food stamps (available to families below 130 percent of the federal poverty guidelines) are used like cash to purchase food; a dollar in benefits is valued as a dollar in money income (Citro & Michael, 1995, p. 224). Information on how much households receive in food stamps was collected in the 2006 ACS. Following Census Bureau practice, CEO prorated the value of the food stamp benefit across the family units within the household. By this method, there were 1.3 million people in New York City who received an equivalent of $831 million in food stamp benefits. The median benefit per participating family was $1,853.

The national school lunch program offers free lunches to all school children whose family income is below 130 percent and reduced-price lunches to school children whose family income is between 130 and 185 percent of the federal poverty guidelines. To estimate the contribution this program makes to family income, we used ACS school enrollment and income variables to establish a child's eligibility. Data provided to CEO from the New York City Department of Education indicated that the participation rate among eligible children was effectively universal. We therefore treated each eligible child as a participant.

The study used the Census Bureau's 2006 dollar value for free and reduced-price school lunch–$2.505 per day for free lunches and $2.109 for reduced-price meals–to calculate the cash-equivalent value of participation in the program (Welniak, chief of the Income Statistics Branch of the Census Bureau, personal communication, May 14, 2008). The school lunch value was then multiplied by 175 school days. This established an annual value of $438 for those children who received free lunches and $369 for those who received reduced-price lunches if they attended school regularly. The value of the lunch subsidy was then assigned to each family based on number of eligible children. The median annual benefit for families with children receiving free or reduced-price lunches was $738.

Adjusting for Housing Status

A credible method for measuring poverty in New York City must account for its high housing costs. As described above, the shelter and utilities proportion of the U.S.-wide poverty threshold was adjusted to reflect those costs. This adjustment provides a more realistic poverty line, but it fails to recognize that many New Yorkers can obtain housing of adequate quality at a lower price than that implied by the threshold. Public housing residents, recipients of tenant-based vouchers, some residents of rent-controlled apartments, or homeowners who have paid off their mortgages, for example, do not need to pay as much for housing as other New Yorkers. CEO captured the effect of these different circumstances by adding to family income the difference between what a family would need to spend in order to meet its housing and utility needs at market rates (represented by the appropriate shelter threshold) and its actual shelter expenditures. The median adjustment for families living in public housing was $3,830. For families living in privately owned housing, but receiving a rent subsidy (such as Section 8), the median adjustment was $5,328.

CEO employed a survey that is conducted every three years by the Census Bureau, the New York City Housing and Vacancy Survey (HVS), to identify housing status and calculate spending on shelter and utilities. This survey draws a sample of more than 15,000 households and collects detailed information on rents paid, subsidies received, the presence or absence of rent controls or stabilization, and a host of other housing-related information. We imputed this information into the ACS sample by using a hot-deck procedure to match households in the ACS to households in the HVS based on a set of common characteristics including neighborhood, whether the household was renting or owned its shelter, household size, household income, the race and ethnicity of the household head, and whether the house-hold head was elderly.

Work-Related Expenses: Transportation and Child Care

Social welfare policy in the U.S. promotes employment through a variety of incentives, most notably the Earned Income Tax Credit. But work often entails unavoidable expenses such as transportation and child care. These costs enter the NAS resource measure as a reduction in the income that would otherwise be available to a family to meet its needs.

Commuting Costs. CEO estimated the cost of transportation by using information on usual weekly work hours, weeks worked, and the journey to work collected by the ACS along with transportation agency data on the costs of commuting in New York City. We assumed an eight-hour workday and calculated the number of workdays per week for each earner based on his or her weekly hours. The number of work-related trips was capped at 14 per week. The cost per trip was estimated by New York City–specific information such as subway and bus fares. The annual transportation cost per worker was calculated by multiplying cost per trip, by number of trips per week, by number of weeks worked as reported in the ACS. Finally, the expenses per worker were summed across the family.

With this method, the median annual transportation cost for a full-time, year-round worker who traveled via subway or bus came to $940. If this same worker drove to work alone, the annual cost would be $1,824.50. There was no transportation cost assigned to those who walked, biked, or worked at home. The median cost per family was $940, an estimate that reflects the prevalence of single-earner families and the widespread use of buses and subways by the New York City workforce.

Child Care Expenses. Child care is an unavoidable expense for many working parents. Since the ACS does not include data on child care, CEO imputed data on weekly out-of-pocket child care expenditures available from the Census Bureau's 2001 and 2004 Survey of Income and Program Participation (SIPP) to the ACS sample. We employed a two-stage, regression-based approach to estimate weekly child care expenditures for working families with children (Citro & Michael, 1995, p. 255; Iceland & Ribar, 2001). First, the likelihood that a working family was paying for child care was estimated from the urban respondents in the SIPP sample, based on characteristics such as the number and ages of family members, race, ethnicity, education levels and hours worked by adults in the family, proportion of total family income earned by female family members, and participation in welfare programs such as Temporary Assistance to Needy Families (TANF). These characteristics were then used to estimate what each family already identified as paying for care was likely to spend.55 The predicted weekly child care expenditures for each family were multiplied by the lowest number of weeks in the past 12 months that any parent in the family had been working to arrive at an annual child care expenditure figure.

Our model estimated that among families with at least one parent working and one or more children less than 12 years of age, 45.1 percent pay for child care. Their median annual expenditures were $5,702. The likelihood that a working family is paying for child care and how much they pay for it varies with its income. Only about 20 percent of the families in the lowest deciles of the income distribution have any child care costs, and these families are estimated to have median annual costs in the $2,000 to $3,000 range.

Medical Needs and Expenditures

The NAS recognized that medical out-of-pocket expenses (MOOP) can be a major factor in determining whether families have enough resources to meet their other needs. MOOP expenditures are not included in the American Community Survey, so they were estimated using the Agency for Healthcare Research and Quality's (AHRQ) Medical Expenditure Panel Survey (MEPS) for 2005. The MEPS includes data on out-of-pocket payments for insurance premiums and medical services.66

Because the distribution of medical spending is highly skewed, we used a hot-deck procedure to estimate MOOP expenditures by families in the ACS sample. Families in the MEPS were grouped into cells defined by demographic factors associated with levels of medical expenditures such as age of the family head, family size, income, employment status, education, and race.77 The 25th, 50th, and 75th percentiles of annual MOOP expenditures were calculated for the families within each respective cell. This information was then imputed into the ACS sample. Each family was grouped into the appropriate cell and randomly assigned one of the three levels of annual expenditures calculated for the corresponding cell in the MEPS.

Annual MOOP expenditures (unsurprisingly) rise with family size. They also depend on the age of the household head and family income. Median expenditures for low-income families are roughly $1,600 for those with an elderly householder, but half that for families that are not headed by someone 65 or older. Median spending by upper-income families with an elderly householder is close to $4,000, but less than $2,000 for families with a household head younger than 65 years old.

APPLYING THE POVERTY MEASURE

CEO's application of the NAS recommendations yields a New York City poverty rate of 23 percent in 2006.88 This is considerably higher than the corresponding official rate of 18 percent for that year. Although this may be an attention-getting difference, without further information it is not very useful for understanding poverty or assessing the adequacy of antipoverty programs. The questions we address in this section are: Why is this new poverty rate higher? For whom is it higher? And what accounts for the differences in poverty rates across demographic groups?

The Effect of Alternative Definitions of Income on the Poverty Rate

The different poverty rates generated by the CEO and official approaches result from both differences in where the poverty line falls and what gets counted as income. The impact of using the CEO thresholds on the New York City poverty rate is obvious: Holding all else equal, higher thresholds create a higher poverty rate. The more comprehensive resource measure could move the poverty rate in either direction. Expanding the definition of resources to include tax credits and the value of near-cash benefits will lower the poverty rate. But the CEO method also reduces the available resources by subtracting work-related and MOOP expenditures.

The effects of these additions and subtractions to the income counted in measuring poverty are shown in Table 3. It reports poverty rates using a progressively more inclusive set of income definitions that are compared against the CEO poverty thresholds. The first and most limited definition of income is pretax cash, the resource concept used in the official measure. This yields a poverty rate of 23.9 percent. Next, income is adjusted for taxation. The tax system increases income for families vulnerable to poverty and thus lowers the poverty rate, but only by a modest 0.7 percentage points, to 23.2 percent. The next adjustment to income adds the cash value of the food stamp and school lunch nutritional assistance programs. These benefits lower the poverty rate by 1.3 percentage points to 21.8 percent.99 Adding the value of the housing adjustment to resources reduces the poverty rate to 18.6 percent. The three additions nearly offset the effect of the considerably higher CEO poverty thresholds, returning the poverty rate to within 0.6 percentage points of the 18 percent rate we find using the official methodology.

Table 3. Alternative poverty rates, 2006.

Income Concept	Rate	Changea
Source: Tabulated from the CEO augmented American Community Survey, 2006. a ^a Percentage point difference from prior line.
1. Pretax cash	23.9
2. After-tax cash	23.2	−0.7
3. After-tax cash plus nutritional subsidies	21.8	−1.3
4. After-tax cash plus nutritional subsidies plus housing adjustment	18.6	−3.3
5. After-tax cash plus nutritional subsidies plus housing adjustment minus work expenses	20.4	1.8
6. After-tax cash plus nutritional subsidies plus housing adjustment minus work expenses minus MOOP	23.0	2.6

However, the resource measure must also account for work-related and medical spending. Subtracting the cost of commuting and childcare expenses from income brings the poverty rate to 20.4 percent. The final adjustment, for MOOP expenses, results in a poverty rate of 23 percent.

Differences in Poverty Rates Across Demographic Groups

Poverty rates are higher under the CEO measure than the official method across a wide variety of demographic characteristics. But the increases are not always uniform. Moreover, there are exceptions to the general pattern: children and people living in single-parent families. Because they generate some of the most informative patterns in our study, we report comparisons between official and CEO poverty rates by age and living arrangements.

When the population is categorized by age, striking differences emerge between the poverty rates generated by the official and CEO methods. Under the official method, children had, by far, the highest poverty rate in New York. With the CEO method, the elderly become the city's poorest age group. As Table 4 indicates, the proportion of children who are living below the CEO poverty line is similar to the official method estimate (a difference of 0.6 percentage points). The CEO poverty rate for the elderly, by contrast, is 13.9 percentage points higher than the official rate and equals 32 percent.

Table 4. Poverty rates by age, 2006.

	CEO	Official	Percentage Point Difference
Source: Tabulated from the CEO augmented American Community Survey, 2006.
Age Group
Under 18	26.6	27.2	−0.6
18 to 64	20.0	14.5	5.5
65 & Over	32.0	18.1	13.9
Children (under 18), by Presence of Parent
Two parents	17.2	16.5	0.7
One parent	41.6	44.4	−2.8

Another interesting difference emerges when children are distinguished between those living in one- or two-parent families. The CEO poverty rate for children living with only one parent is 2.8 percentage points below the official rate. The CEO poverty rate for children in two-parent families is, by contrast, marginally higher (by 0.7 percentage point) than the official poverty rate. While this pattern of change narrows the difference in the poverty rate between these two groups of children, they remain enormous: The CEO poverty rate is 17.2 percent for children living with two parents, compared with 41.6 percent for children in single-parent families.

These patterns are echoed when individuals are grouped by the kind of family they are living in. In Table 5, living arrangements include: husband-wife or unmarried partner, single-headed families, and unrelated individuals. Within those categories, we separate families that do or do not include children under 18. The CEO poverty rates for persons living without children are considerably and consistently higher than are rates based on the official methodology, by 7.3 percentage points for those in husband-wife families, by 10.0 percentage points for persons with a single family head, and by 11.8 percentage points for unrelated individuals. By contrast, the differences between the CEO and official poverty rates are small for persons living in families with children. The CEO rate is 1.9 percentage points higher for those in husband-wife families and 0.9 percentage point lower for people in single-headed families than the official rate.

Table 5. Poverty rates for persons by family type, 2006.

Family Type	CEO	Official	Percentage Point Difference
Source: Tabulated from the CEO augmented American Community Survey, 2006. a ^a In the CEO measure unmarried partners are treated as spouses.
Husband-Wife/Unmarried Partnera
All	14.8	10.9	3.9
No children	13.3	6.0	7.3
With children	15.8	13.9	1.9
Single Family Head
All	30.9	27.8	3.1
No children	21.4	11.4	10.0
With children	36.5	37.3	−0.9
Unrelated Individuals
All	34.9	23.1	11.8

Resources and Between-Group Differences

The comparisons between the CEO and official poverty rates combine two effects: a different threshold and a more inclusive definition of income. The next two tables compare the six income concepts employed in Table 3 against the CEO poverty threshold. Thus they focus exclusively on the effects of the expanded resource definition on poverty rates across demographic groups.

Table 6 details how the progressively expanding income concepts affect poverty by age group and indicate which resource adjustments drive the different direction the poverty rate takes for children as opposed to the elderly. Under the first and most restrictive definition, pretax cash, children have a higher poverty rate than the elderly (33.9 percent compared to 27.5 percent). But as income resources are added, the poverty rate for children drops sharply, while the poverty rates for working-age and elderly adults decline more modestly.1010 Under the fourth income concept (after taxes, nutrition assistance, and the housing adjustment), the poverty rate for children is down to 21.2 percent, while it stands at 16.4 percent for 18- through 64-year-olds and 25.3 percent for the elderly.

Table 6. Alternative poverty rates by age group, 2006.

Income Concept	Under 18	18 to 64	65 and Over
Source: Tabulated from the CEO augmented American Community Survey, 2006.
1. Pretax cash	33.9	19.5	27.5
2. After-tax cash	31.2	19.5	27.0
3. After-tax cash plus nutritional subsidies	28.5	18.6	26.2
4. After-tax cash plus nutritional subsidies plus housing adjustment	21.2	16.4	25.3
5. After-tax cash plus nutritional subsidies plus housing adjustment minus work expenses	24.1	18.1	25.6
6. After-tax cash plus nutritional subsidies plus housing adjustment minus work expenses minus MOOP	26.6	20.0	32.0
Percentage point difference between income concept 6 and 1	−7.4	0.5	4.5

Then resources are removed from income. Subtracting work-related expenses lifts the poverty rate for children by 2.9 percentage points, compared to a 1.7 percentage point rise for working-age adults. There is virtually no change for the elderly.1111 The adjustment for MOOP expenditures drives the poverty rate for the elderly up by 6.4 percentage points, while the respective increases in poverty for children and working-age adults are only 2.4 percentage points and 1.9 percentage points. The combined effect of all these adjustments lowers the poverty rate for children by 7.4 percentage points, leaves the poverty rate for working-age adults essentially unchanged (a 0.5 percentage point rise), and increases the poverty rate for the elderly by 4.5 percentage points.

The comparison (in Table 5) of CEO and official poverty rates by family type indicated that the two methods produced very similar poverty rates for people living in families with children. CEO poverty rates were considerably higher than the official rates, however, for persons who were not members of families with children. Table 7 illustrates how the expanded definition of income shapes that pattern. It groups people into families with or without children less than 18 years of age. Among families with children, we report separate estimates for one- and two-parent families.1212 A more inclusive definition of income lowers poverty rates for persons living in families with children. By contrast, the relatively small effect of tax credits and in-kind benefits on the poverty rate for people in families without children is more than offset by their work-related and medical out-of-pocket expenditures.

Table 7. Alternative poverty rates by family type, 2006.

	With Children		Without Children
Income Concept	Two-Parent	Single-Parent
Source: Tabulated from the CEO augmented American Community Survey, 2006.
1. Pretax cash	20.0	45.7	13.7
2. After-tax cash	18.3	42.2	14.4
3. After-tax cash plus nutritional subsidies	17.1	38.1	14.0
4. After-tax cash plus nutritional subsidies	11.9	29.6	12.3 plus housing adjustment
5. After-tax cash plus nutritional subsidies plus housing adjustment minus work expenses	13.8	33.6	13.4
6. After-tax cash plus nutritional subsidies plus housing adjustment minus work expenses minus MOOP	15.8	36.5	16.1
Percentage point difference between income concept 6 and 1	−4.2	−9.2	2.5

The poverty rate for persons in families without children rises from 13.7 percent to 16.1 percent between the first and last income measures. Tax programs do not lower the poverty rate for members of these families (it edges up to 14.4 percent).1313 Nutritional assistance programs have a negligible effect, in part because these families do not include children who could be receiving free or reduced-price school lunches. The housing adjustment does have some impact (reducing the poverty rate to 12.3 percent). Including work-related expenses increase this group's poverty rate by a fairly modest 1.1 percentage points; they have no child care costs. Adding MOOP expenses raises their poverty rate to 16.1 percent.

A sharply different pattern emerges for people living in families with children; poverty rates decline markedly as the income concept becomes more inclusive. In addition, the cumulative declines in the poverty rate are much larger for single-parent families than for two-parent families. For single-parent family members, the poverty rate falls from 45.7 percent to 29.6 percent after taxes, nutritional assistance, and the housing adjustment are added to income. The poverty rate for two-parent families declines from 20.0 percent to 11.9 percent once the corresponding adjustments are made to their income.

When work-related expenses are deducted from income, the 4 percentage point rise in the poverty rate for single-parent families is considerably larger than the 1.8 percentage point change for two-parent families. One possible explanation for this is that single-parent families lack a second parent who can provide childcare. Subtracting MOOP spending from resources raises poverty rates for both one- and two-parent family members by roughly similar amounts, 2.8 percentage points and 2.0 percentage points, respectively. The total effect of all the adjustments to resources is a drop of 4.2 percentage points in the poverty rate for people living in two-parent families, and the poverty rate for members of single-parent families tumbles by 9.2 percentage points.

Some Observations

We see three important themes in these comparisons. The first is that in a policy environment in which support to poor families increasingly takes the form of tax credits and in-kind benefits, a poverty measure that recognizes these as income will provide a far different profile of poverty than one that only counts pretax cash. Despite a higher threshold, the poverty rate for children living in single-parent families, for example, is lower under the CEO than the official method. This is consistent with, and reflects, public policy's focus on families headed by single mothers.

It also appears that even for single-parent families the effect of tax credits and nutritional assistance on the poverty rate is modest. We believe that this is due, in part, to the manner in which these programs phase out and the coincidence between their phase-out ranges and the CEO poverty thresholds. The food stamp program is a case in point. A family loses its eligibility for the program if its income exceeds 130 percent of the federal poverty guidelines. In 2006 this equaled $26,000 for a family of four. The corresponding CEO poverty threshold is $26,318. The result, in this instance and in others, is that benefit levels are falling off or reaching zero just as families are approaching the point where more resources would move them from just below to above the poverty line.

A third message, also related to our resource measure, concerns the poverty rate for the elderly. Much of the support low-income seniors currently receive takes the form of cash, either through Social Security or the Supplemental Security Income program. The positive effect of noncash assistance for this group is small and their health care costs are high. Given the perception that progress against senior poverty in the 20th century was a place where New Deal and Great Society programs clearly had their intended effect, our finding of a 32 percent poverty rate is unsettling.

USING THE NEW MEASURE

CEO's initial report on poverty in New York City was widely noted, and it immediately raised questions as to how the new measure would affect city policy. The answer is that the new measure is stimulating new thinking but change cannot (and perhaps should not) be dramatic or rapid. Much of what New York, or any city, does to support low-income families is to administer programs that are subject to federal and state statute or regulation. CEO's poverty measure cannot affect federal or state funding formulas, eligibility requirements for means-tested programs, or their benefit levels. There are no plans to use the CEO thresholds or its definition of family resources in any programs where New York City might have that discretion.

CEO's poverty measure is a social indicator; its value lies in the extent to which it tells us something new about populations in need. Where the CEO measure is beginning to influence local policy is in the area of program innovation. Mayor Bloomberg established the Center for Economic Opportunity to initiate and evaluate new programs, and the center has responded to its measure with plans to expand the populations it targets. The mayor's commission had recommended that innovation focus on families with young children, youth (persons 16 through 24 years of age), and the working poor. Our findings have prompted the center to expand its focus to the elderly.

The center is now working with New York City's Human Resources Administration and Department for the Aging to find opportunities to fashion new programs or build on existing ones that can reduce senior poverty. One initiative that is emerging from this work is an employment program targeted to older New Yorkers who have most, but not all, of the 40 quarters of earnings they need to qualify for Social Security benefits and eligibility for Medicare. This appears to be a particular problem for elderly immigrants who may have contributed to their families' well-being by providing child care or earnings from informal work. We expect that future poverty measurement work will continue to cast poverty in a new and more informative light and that, over time, the measure will become integral to the strategic planning of the many city agencies whose work addresses the needs of low-income New Yorkers.

• 1

  To avoid cumbersome language we use “family” to denote the unit of analysis in our study. Family includes one-person units if the person is an unrelated individual. Unmarried partners are treated as spouses.

• 2

  Miscellaneous necessities cover items such as some non–work related travel, household supplies, and personal care products.

• 3

  The NAS thresholds are calculated from the Bureau of Labor Statistics' Consumer Expenditure Survey. A description of this survey is available at http://www.bls.gov/cex/home.htm. The U.S.-wide threshold (labeled FCSU-CE) is posted at http://www.census.gov/hhes/www/povmeas/altmeas06/nas_experimentalthresholds.xls. Note that this threshold does not include principal payments by homeowners as an expenditure.

• 4

  This approach is a deviation from that taken in the Census Bureau's experimental poverty measures reports, where regional adjustments are carried out by grouping all households within each state into one metropolitan and one non-metropolitan area. This method would have put New York City in the same housing market as far lower housing cost areas such as Albany, Buffalo, and Syracuse. Our approach provides a more New York City–specific measure.

• 5

  The measure is out-of-pocket expenditures and thus reflects the effect of child care subsidies on spending by low-income families in the sample.

• 6

  These data, along with much help in using and interpreting them, were provided to us by Jessica Banthin and her staff at AHRQ.

• 7

  Although insurance status was not available in the 2006 ACS, the MEPS data indicate that insurance status was highly correlated with the family characteristics that the cells were based on.

• 8

  Throughout this section we refer to the “CEO measure” of poverty. As the second section detailed, the methods used to create the measure and estimate poverty rates from it represent CEO's attempt to apply the National Academy of Sciences' recommendations.

• 9

  Differences are taken from un-rounded numbers.

• 10

  Readers should bear in mind that although the classification of people in this table is by their individual-level characteristics, their poverty rate is being determined by their family circumstances; the child poverty rate, for example, reflects the income and expenses of the family they live in.

• 11

  This pattern reflects the higher work-related expenses that families with children would be expected to incur because of child care costs and the very low levels of work-related expenses for the elderly population; relatively few of them are in families or are living with people who are in the workforce.

• 12

  Unrelated individuals are not represented in the table.

• 13

  The federal, state, and city Earned Income Tax Credits are much more generous for families with children than they are for childless families, and, of course, childless families cannot make use of credits that are designed to offset the cost of raising children.