Adjusted Poverty Measures and the Distribution of Title I Aid: Does Title I Really Make the Rich States Richer?

Federal and state governments in the United States make extensive use of student poverty rates as indicators of educational need and subsequent drivers of educational funding.¹ The federal government uses Census Bureau estimates of poverty in distributing Title I funds, for example. And thirty-four states distribute categorical funds or additional dollars using student weights that are based on measures of student poverty, such as the share of students eligible for free or reduced price lunches under the National School Lunch Program (Verstegen and Jordan 2009).

Unfortunately, the measures of student poverty currently used for policy making (and policy analysis) are geographically biased. Despite well-recognized differences in the cost of living from one state to another or from one city to another within a state,² the income thresholds currently used to determine the numbers of students living in households that are below the poverty level are not adjusted in any way for regional differences in the cost of living. That is, the current counts of students living in poverty, which are regularly used to determine how educational and other public dollars are distributed, do not reflect any adjustment for regional differences in the cost of living. For instance, the U.S. Census Bureau produces the Small Area Income and Poverty Estimates (SAIPE) for school districts using a set of income thresholds that increase with family size but are the same in rural Alabama as in New York City.³ Meanwhile, the share of students eligible for free or reduced price lunches, a proxy for student poverty that is commonly used in both research and policy, depends on the same set of nationwide income thresholds as are used to produce the SAIPE for school districts.⁴

The potential measurement problems arising from a lack of regional cost of living adjustments in the poverty estimates are obvious.⁵ Poverty rates are intended to measure the share of families living at or below a designated standard of living. A family with the poverty threshold level of income ($22, 113 for two parents and two children in 2010) would be more able to buy more goods and services—and therefore maintain a higher standard of living—in a low cost-of-living state like Iowa, however, than in a high cost-of-living state like California. As a result, the current regime, with its fixed poverty thresholds, tallies relatively too many children as living in poverty in states or regions with relatively low cost of living. The reverse is true in states or regions with relatively higher cost of living.

The implications of this measurement error for education policy are large, and frequently overlooked. Most of the discussion about the bias in the poverty numbers naturally focuses on the implications for anti-poverty programs like the Supplemental Nutrition Assistance Program (SNAP), Temporary Assistance for Needy Families (TANF), or Medicaid. Key education policies are also affected. The question raised in this study is: What would be the implications for the distribution of federal poverty-based education aid if we regionally adjusted the current poverty statistics reported by the U.S. Census Bureau to better reflect differences in the real (i.e., cost-adjusted) incidence of student poverty?

Title I is the largest educational grant program in the United States, and is specifically designed as a source of compensatory funding for low-income children. In 2011–12, the U.S. Department of Education appropriated $14.5 billion in federal aid through the Title I program (USDOE 2012).

There are four components to the Title I funding formula.⁶ The first two components, Basic grants and Concentration grants, allocate funding to each school district based on Census Bureau estimates of the number of children in poverty (i.e., the SAIPE) and the average per-pupil expenditure in the state.⁷ The higher the average per-pupil expenditure in the state, the larger is the Title I allocation. Each of these two components allocates a constant dollar amount per child in poverty within each state, but school districts must have at least 15 percent of the children in poverty to receive a Concentration grant. Basic and Concentration grants constituted roughly 54 percent of Title I funding in 2012 (USDOE 2012).

Targeted Assistance grants are the third component of the Title I formula. Targeted Assistance grants provide funding to school districts that increases as the share of children in poverty increases and as average per-pupil expenditure in the state increases. As a result, districts with a higher share of students in poverty receive larger Targeted Assistance grants per pupil than districts with a smaller share of children in poverty. As with the Basic and Concentration grant programs, the only determinants of school district aid are the state average expenditure per pupil and the estimated number of students in poverty. Targeted Assistance grants constituted approximately 23 percent of Title I funding in 2012.

The remaining 23 percent of Title I funding comes from Education Finance Incentive grants (EFIG), which depend not only on the number of children in poverty and the state average per-pupil spending, but also on (1) a measure of state fiscal effort (the percentage of per-capita income spent on elementary and secondary education, relative to the national average) and (2) a measure of state school funding equity (a weighted coefficient of variation in district per-pupil expenditures, wherein children in poverty have a greater weight than other children.⁸ States with higher fiscal effort receive larger EFIG allocations than other states. Meanwhile, states with lower funding equity (i.e., those with a higher weighted coefficient of variation) receive smaller EFIG allocations than other states, and the EFIG allocations are distributed more progressively within. The equity factor does not take state efforts at compensatory education into account, so that a state where all of the districts had equal expenditures per pupil would be deemed more equitable (and receive more Title I funding) than a state where all of the low-poverty districts had equal expenditures per pupil, and all of the high-poverty districts had higher expenditures per pupil. More importantly, the equity factor does not take regional differences in the cost of education into account, so states that equalized nominal expenditures would be deemed equitable, and states that adjusted nominal expenditures to perfectly equalize the purchasing power of school districts would be deemed inequitable.

Perceived flaws in the design of the Title I funding formulas have drawn considerable attention lately (Carey and Roza 2008; Liu 2007, 2008; Miller 2009a; Miller and Brown 2010a, b). Critics believe that Title I funding is wrongly targeted because distribution of EFIG funding tends to favor wealthy states and larger urban districts.⁹ In other words, the critics argue that Title I funding makes rich states richer by allocating disproportionate funding to states that have greater fiscal capacity, and that Title I funding favors large districts in urban areas over comparably poor small districts in rural areas (Liu 2007, 2008; Miller 2009a; Miller and Brown 2010a, b).

In a frequently cited work, Liu (2008) examined regional disparities in the Title I funding formulas. His work focused on the relationship between Title I funding allocations, school district size, and state poverty rates. His analysis adjusted Title I funding for regional differences in the cost of education using the National Center for Education Statistics Comparable Wage Index (NCES-CWI), but did not similarly adjust the poverty rates for regional differences in the cost of living. He concluded that (1) “By allocating aid to states in proportion to state per-pupil expenditures, Title I reinforces vast spending inequalities between states to the detriment of poor children in high-poverty jurisdictions,” and (2) “small or mid-sized districts that serve half or more of all poor children in areas of high poverty receive less aid than larger districts with comparable poverty” (Liu 2008, p. 973). Liu's article and related papers laid the groundwork for numerous other policy briefs to follow (Miller 2009a, Miller and Brown 2010a, b).

In this study, we extend Liu's analysis to control not only for regional differences in the cost of education, but also for regional differences in the poverty thresholds. In the first part of our analysis, we construct alternative poverty measures based on regional differences in the prevailing wage for workers with the typical characteristics of the working poor. In the second part of our analysis, we use national data on all school districts from 2007–08, 2008–09, and 2009–10 to estimate the average Title I revenues per student in poverty after adjusting poverty rates for geographic differences in the cost of living and adjusting Title I revenues for geographic differences in the cost of education. We evaluate the distribution of cost-adjusted Title I funding per child in poverty by state and by locale within-state, and find that disparities do exist. Our analysis, however, turns conventional wisdom on its head. We find that when we fully adjust for regional differences, Title I funding patterns disproportionately favor rural school districts in low cost-of-living states. We conclude with policy recommendations for revising Title I funding formulas.

We follow a three-step strategy for constructing alternative poverty measures for all school districts in the contiguous lower forty-eight states. In the first step, we estimate the prevailing wage for individuals with the typical characteristics of the working poor in various labor markets across the United States. We use those wage levels as our best estimates of the income thresholds for poverty for each labor market. In the second step, we used those adjusted income thresholds to recalculate child poverty rates for each labor market area. The final step is to use the difference between current and adjusted poverty rates at the labor market level as a poverty adjustment factor to adjust school district level poverty rates for all districts located within each labor market. We discuss each step in turn.

Our approach to regional poverty adjustment uses hedonic wage analysis to adjust the poverty income thresholds for regional differences in the cost of living and the access to desirable local amenities (such as public services, good climate, low crime rates, quality schools, or access to shopping and medical facilities). Essentially, we presume that if the prevailing wage in Chicago for a worker with poverty-level characteristics is 10 percent above the national average, then the poverty income threshold in Chicago should also be 10 percent above the national average.

There are three reasons that we use differences in prevailing wage levels as our measure of regional differences in the cost of living rather than relying on a market basket approach as in Renwick (2009), Marks et al. (2010), or Short (2011). First, differences in the prevailing wage reflect not only differences in the price of food and shelter, but they also reflect any differences in important community characteristics, such as climate, crime rates, or public amenities (Roback 1982; Gyourko and Tracy 1989). As such, they provide a more complete measure of the income needed to maintain a reasonable standard of living in each community.

Second, market-basket approaches presume that all families choose the same bundle of goods and services in all locations. Using differences in the prevailing wage to measure regional differences in the cost of living allows for the possibility that families may choose a more modest dwelling in amenity-rich locations like San Francisco than they would choose in other parts of the country.

Finally, wage data are available for all parts of the country, making it possible to develop cost-adjusted poverty thresholds for labor markets throughout the forty-eight states under analysis.

We estimate the prevailing wage for the working poor using a hedonic wage analysis modeled after Taylor and Fowler's 2006 Comparable Wage Index (CWI). The Taylor-Fowler CWI measures the prevailing wage for college graduates in 800 U.S. labor markets. Our current analysis estimates the prevailing wage for workers who do not have a college degree. We take this approach because most of the population living below the poverty threshold does not have a college degree, and the geographic pattern of wages may be different for college graduates than for other workers.

Following Taylor and Fowler (2006), we used a maximum likelihood regression and data from the 2008, 2009, and 2010 American Community Survey (ACS) to generate estimates of the annual wage and salary income of individuals who have at most an associate's degree. The dependent variable was the log of annual wage and salary earnings. The independent variables were age, age squared, the amount of time worked, and a series of indicator variables for gender, race, educational attainment, occupation-by-year interactions, and industry-by-year interactions.¹⁰ In addition, the estimation includes an indicator variable for each labor market area and random effects by state.¹¹ Appendix table A.1 presents coefficient estimates and standard errors from the hedonic wage model.

As with the Taylor-Fowler CWI, we used the regression estimates to construct a Poverty-CWI.¹² The Poverty-CWI captures the differences in wage levels required to compensate workers for differences in the labor market specific attributes, such as the prices of goods, services, and other amenities (climate, crime rates, access to quality schools, medical facilities, parks, museums, etc.) associated with different regions within a state. The fact that individual wages not only contain information on the price levels of goods and services, but also capture the perceived amenities offered in different locations, distinguishes a comparable wage index from other cost indices based on a basket of goods (e.g., the familiar consumer price index or regional price parity rental prices). It is our use of labor market analysis that distinguishes this study from previous attempts (e.g., Renwick 2009, 2011) to adjust poverty rates for geographic differences in the “cost of living.”

In effect, the Poverty-CWI provides an estimate of the differences in the total perceived cost of living in different regions/labor markets as reflected in labor market outcomes. The index is centered at 1.00 representing the national average, so that deviations from this figure denote how much more or less in percentage terms it costs to compensate workers to live and/or work in different labor markets. For example, a value for a given labor market of 1.25 indicates that it costs approximately 25 percent more than the national average to hire a comparable worker in this location, whereas a value of 0.75 indicates that it would cost 25 percent less than the national average to hire a similar staff person. Figure 1 provides a map of the Poverty-CWI across all counties in the mainland United States, with darker counties representing higher cost areas and lighter ones representing lower cost areas.

We next calculated the cost-adjusted poverty thresholds for each labor market by multiplying the Census poverty thresholds for each family configuration by the corresponding Poverty-CWI value for each labor market. There are different poverty thresholds for different family configurations. To be precise, there are forty-eight family configurations differentiated by overall size of family and number of children under the age of eighteen, each with its own unadjusted poverty threshold set by the Census. Because there are forty-eight existing family configuration-specific thresholds, there are forty-eight cost-adjusted poverty thresholds in each of the 778 labor market areas.¹³

The second step uses the existing unadjusted (Census) poverty thresholds and the new cost-adjusted poverty thresholds to count the number of school-aged children living in families below these levels in each labor market, according to the 2008, 2009, and 2010 ACS (the same data as were used to estimate the Poverty-CWI). The total number of children living in poverty within each labor market can thus be summed to provide the cost-adjusted and unadjusted poverty counts in each labor market.

The Poverty Adjustment Factor (PAF) is calculated for each labor market by taking the ratio of the cost-adjusted to unadjusted counts of students living in families in poverty. A PAF value greater than 1.00 indicates that the current unadjusted (Census) poverty rate for the given labor market underestimates the true incidence of poverty, whereas values less than 1.00 show the current unadjusted poverty measure overstates the true amount of poverty. For example, a PAF value of 1.15 indicates that the cost-adjusted poverty rate (the true relative poverty for a given labor market) is 15 percent higher than the unadjusted poverty rate, and a PAF of 0.85 indicates that the cost-adjusted poverty rate is 15 percent lower than the unadjusted poverty rate.

Table 1 summarizes the average PAFs across labor markets within regions (Regional Educational Laboratories) established for research purposes by the Institute of Education Sciences.¹⁴ For the analyses herein as elaborated in the following section, we merge our poverty adjustment factors to school district level data, where school districts are clustered within labor markets. Ultimately, our intent is to discern the distribution of Title I resources to local education agencies, with respect to the estimated poverty rates in those local education areas and their location. The summaries in table 2 are based on local edu- cation area enrollment-weighted averages, using district-level enrollment data over the three-year period from 2008 to 2010.

On average, school districts within states belonging to the Northwest Regional Educational Laboratory have a poverty adjustment factor of 1.03. That is, the cost-adjusted poverty in districts in these states is, on average, three percent higher than the originally stated poverty. Districts in states in the Mid-Atlantic and Northeast regions experience the largest average cost adjustments to their poverty rates (12 and 13 percent, respectively) with the Western states close behind at 11 percent. Districts in the Southeast, Southwest, and Central (Plains/Mountain) have downward average cost adjustments to their poverty rates all on the order of 7 to 8 percent.

The expected pattern also holds for differences across locale within region. Whereas districts in northeastern metropolitan areas receive an average upward cost adjustment to their poverty estimates of 16 percent, districts in rural areas in those states receive an average downward adjustment of their poverty rates of 11 percent. The largest downward adjustments to poverty are found in rural labor markets in the Central and Southwestern states.

Nationally, the PAFs indicate that measured rates of child poverty are 5 percent too low in metropolitan areas, and 20 percent too high in rural communities, on average. As such, the PAFs clearly show that the geographic bias embedded in the SAIPE or the free and reduced price lunch statistics is large and economically meaningful.

The final step of this part of the analysis applies the calculated PAF to the unadjusted poverty rates for each school district in the labor market.¹⁵ This approach assumes that the relative (proportional) adjustment in the counts and proportions of students living in families below the poverty threshold are constant across districts within a given labor market. This assumption is necessary to apply the adjustments of poverty rates for jurisdictions such as districts, which are more granular than that at which the CWI and PAF are calculated (i.e., the labor market). Table 2 summarizes the state average poverty and adjusted poverty rates based on enrollment weighted calculations using district level data. The findings show that in states with high adjustment factors—states requiring higher nominal income to achieve comparable real poverty thresholds—average district-level poverty rates are adjusted upward by 2 to 3 percentage points. New Jersey, for example, goes from an average district poverty rate of 12 percent to an average district poverty rate of 14 percent. By contrast, states in regions where lower nominal incomes to achieve comparable real poverty thresholds experience reductions in estimated poverty of similar magnitude.

These seemingly subtle overall shifts in poverty lead to significant reshuffling of the rank order of states in terms of poverty. For example, New York State ranks eighteenth in unadjusted poverty, but it ranks fifth in cost-adjusted poverty. California, which ranks seventeenth in unadjusted poverty, ranks sixth after the cost adjustments are applied. Indeed, all of the states where the enrollment-weighted average PAF is greater than 1.00 experience an upward cost adjustment in both their poverty rates and their poverty ranks. Most states where the average PAF is less than 1.00 experience a downward adjustment in their poverty ranks. A few states, however, such as Georgia and Texas, experience a downward cost adjustment to their poverty rates but an upward adjustment in their poverty ranks.

Previous studies critiquing the distribution of Title I funding across states have relied on a single year of data (see Carey and Roza 2008; Liu 2008). We rely on the most recent three-year panel (2008 to 2010) of local public school district fiscal data from the U.S. Census Bureau's fiscal survey of local governments (F-33).¹⁶ Using panel data ensures that our analysis is not distorted by one-time anomalies in the annual data.

Our analysis focuses specifically on the distribution of Federal Compensatory Aid to local public school districts. We evaluate the distribution of (1) Federal Title I Revenue per Enrolled Pupil, (2) Federal Title I Revenue per Pupil in Poverty, (3) Federal Title I Revenue per Pupil in Cost-Adjusted Poverty, and (4) Cost-Adjusted Federal Title I Revenue per Pupil Adjusted for Geographic Differences in the Cost of Education (NCES-CWI) using Cost-Adjusted Poverty. In order to present a relatively simplified summary of Title I revenues, we evaluate those revenues by state and region. Prior research suggests that rich Northeastern states, such as New Jersey, New York, or Massachusetts, make out particularly well in Title I funding per pupil in need, whereas Southeastern states appear to be at a particular disadvantage.

Within each region, we also evaluate the distribution of Title I funding by locale, specifically by metropolitan, micropolitan, and rural or other areas within states and regions.¹⁷ Prior research suggests the highest per-pupil allocations of Title I aid occur in districts in metropolitan areas, with lower allocations in micropolitan, and especially rural, districts. But again, these disparities might be moderated after accounting both for the lower of the Title I dollar in metropolitan areas and for the higher poverty rates after adjusting poverty-income thresholds.

Table 3 summarizes the average Title I expenditures per pupil in poverty, cost-adjusted for regional variation in the value of the Title I dollar per pupil in unadjusted poverty (as in Liu 2008), and with adjustments to both poverty and purchasing power. Viewing the unadjusted data on Title I funding per child in poverty, it would appear that the highest rates of Title I allocation are in Northeastern metropolitan area school districts (at $1,867 per child in poverty). By contrast, the lowest Title I allocations are in Appalachian micropolitan districts and Western state rural districts, at less than $1,220 per pupil. In other words, the unadjusted pattern of Title I expenditures is largely consistent with previous research findings.

When adjusting for differences in purchasing power (as in Liu 2008) the Title I funding landscape shifts. We find that the highest Title I allocations are in micropolitan districts in Central states and rural districts in Central, Western, and Northwestern states, and the lowest Title I allocations per child in poverty are in metropolitan areas in Western states.

Adjusting for differences in poverty thresholds but not for differences in school district purchasing power yields the third set of columns in table 3. Again, we find that Title I allocations are highest in micropolitan districts in Central states, and lowest in Western metropolitan areas, on average.

Taking the final step and adjusting not only the purchasing power but also the poverty rates, we find that the pattern uncovered by the purchasing power adjustments is amplified. The highest, by far, Title I allotments per child in poverty are in micropolitan and rural districts in Central states and the lowest Title I allotments are in districts in metropolitan areas in Western states. On average, Title I allotments in metropolitan areas in the Western states are less than half of the Title I allotment in rural areas of the Central, Mid-Atlantic, Northeast, Northwest, or Southwest regions. On average, within each region, districts in metropolitan areas have significantly lower cost-adjusted Title I funding per adjusted-poverty child than do micropolitan or rural districts. These findings run in stark contrast with those popularized in policy reports intended to influence Title I re-authorization.

Table 4 drills down to examine Title I allocations in six large, diverse states. Again, the pattern is clear. Unadjusted, Title I revenues per poverty pupil appear to favor metropolitan districts over rural ones in five of the six states. (Texas is the lone exception.) In stark contrast, fully adjusting the allocations reveals that current Title I policy actually strongly favors rural and micropolitan districts in all six states.

Furthermore, as table 5 illustrates, we find no evidence that Title I allocations are systematically lower for small districts than for large ones. If anything, our evidence suggests that once we fully adjust the allocations for regional differences there is either a negative correlation or no correlation between school district size and the Title I allocations per pupil.

Our analysis demonstrates that it is feasible to estimate, with publicly available data, cost-adjusted poverty measures for all school districts in the nation. It also demonstrates that a failure to make adjustments for regional differences in the cost of living leads to inaccurate measures of the percentage of students who are really living below the poverty threshold. We show that adjusting the poverty thresholds to account for differences in the cost of living can have large effects on our perceptions of relative poverty. Absent adjustment, the child poverty rates in California and New York are the seventeenth and eighteenth highest in the nation, respectively, whereas after adjustment they are sixth and fifth.

Our analysis also casts considerable doubt on conventional wisdom suggesting that Title I over-subsidizes districts in rich states and larger districts in metro areas. We find that—if anything—Title I fails to adequately support economically disadvantaged students in metropolitan areas.

As such, our analysis suggests that the Title I formula components that have been heavily criticized, in particular the role of state average per-pupil expenditures in determining allotments, are not leading to demonstrably inequitable outcomes of the sort found in prior work. Although it seems illogical on its face to provide states with poverty-based funding according to their own level of spending, there is no doubt that Title I aid should account for regional differences in the cost of education, and little doubt that current poverty measures fail to accurately reflect the geographic distribution of student need. A fair Title I funding formula that incorporated appropriate adjustments for geographic differences in the cost of education and in the poverty thresholds might still appear to favor “richer” states and school districts in large urban areas. A strong and positive correlation between fiscal capacity and Title I aid is not sufficient evidence that the Title I formulas are flawed.

There is other, more persuasive evidence that the existing Title I formulas are flawed. Our analysis demonstrates that the inaccuracy embedded in existing measures of student poverty is large and economically meaningful. Because Title I relies on biased measures of student need and fails to adjust for regional differences in the cost of education, it strongly favors school districts in low cost-of-living areas at the expense of school districts in high cost-of-living areas. Given that low-income minority students disproportionately attend school districts in high cost-of-living areas, this pattern is particularly disquieting.

Fortunately, it should be relatively straightforward to resolve the flaws in the Title I funding formulas. First, the baseline measure of student poverty should be changed to incorporate geographic differences in the cost of living. We favor adjusting the poverty thresholds using labor market analysis because that approach provides a more complete picture of regional differences in the cost of living. Nevertheless, other strategies such as the market-basket adjustments proposed by Meyer and Sullivan (2012) or Renwick (2009) also have merit. Second, Title I funding allocations should be adjusted for uncontrollable differences in the cost of education rather than state average expenditure levels, so that Title I funds have the same purchasing power in every district. The Taylor-Fowler CWI could be easily updated to use for such adjustments. Finally, we recommend that the formula used to distribute EFIG funding be revised to incorporate a more sophisticated measure of equity. The existing measure (a weighted coefficient of variation) measures equality, not equity, and therefore penalizes states that equalize school district purchasing power in the face of regional differences in the cost of education.

Our analysis focuses specifically on Title I funding but our basic conclusions also apply to compensatory aid programs within states. We find that geographically adjusting the poverty thresholds for differences in the cost of living leads to substantial changes in poverty rates within states, not just between states. As such, our analysis suggests that the compensatory education components of state aid formulas, which, like Title I, are based on geographically unadjusted measures of student need, may be over-targeting resources to rural districts and under-targeting resources to urban districts. The guidelines we suggest applying to the distribution and equity evaluation of Title I funding, therefore, also apply to state school finance formulas.

An early version of this work was supported by Regional Educational Laboratory Midwest, with funds from the Institute of Education Sciences (IES), U.S. Department of Education, under contract number ED-06-CO-0019. The content does not necessarily reflect the position or policy of IES or the Department of Education, nor does mention or visual representation of trade names, commercial products, or organizations imply endorsement by the federal government.