Rural Poverty Transitions: A New Look at Movements in and out of Poverty José D. Pacas Research Scientist, Minnesota Population Center, University of Minnesota Elizabeth E. Davis Professor, Department of Applied Economics, University of Minnesota Preliminary Draft, March 14, 2018. Please do not cite or share, comments welcome. Email: pacas002@umn.edu; Corresponding author. Paper presented at the March 2018 Rural Poverty Conference: Fifty Years After The People Left Behind in Washington, DC.
Abstract This paper uses linked Current Population Survey (CPS) March Supplement data to examine short-term poverty transitions for families living in nonmetropolitan areas of the US over a twenty-year period from 1996 to 2016. Detailed information on family income sources and family composition allows decomposition of the Supplemental Poverty Measure (SPM) into factors corresponding to poverty exits and entries. While the SPM estimates a slightly higher poverty rate overall than the Official Poverty Measure (OPM), the direction of the nonmetro-metro poverty rate gap differs depending on which poverty definition is used. Based on the SPM, the nonmetro poverty rate is lower than the metro SPM in each year and ranges from a low of 11.7% in 1998 to 13.6% in the Great Recession, and a high of 14.0% in 2012. However, the likelihood of a family falling into poverty is non-zero at all income levels, and nonmetro families are slightly less likely to remain in poverty from one year to the next compared to metro families. Annual rates of entry into and exit from poverty are similar in metro and nonmetro areas. Changes in resources (as opposed to changes in family composition) are associated with most poverty transitions. Overall, changes in wages and salaries are more important for metro families, and while these are also important for nonmetro families, changes in Social Security, farm income and medical expenses play a larger role in poverty entries and exits for nonmetro than for metro families. Despite the short-term nature of the linked CPS panel data, the depth of information about income sources and large sample size offers new insight into the dynamics of poverty entry and exit for nonmetropolitan families over two decades.
1 Introduction In 2016, 40.6 million Americans lived in poverty, and while only 6.9 million (17%) of the poor lived in nonmetropolitan areas, the official poverty rate in nonmetropolitan areas exceeded that of metro areas (Semega, Fontenot and Kollar, 2017). According to the US Census Bureau, in 2016 the official poverty rate was 15.8% for individuals living outside of metropolitan areas, compared to 12.2% for those living within metro areas. The USDA Economic Research Service reports that the official rate of poverty in nonmetropolitan counties has consistently been higher than in metro counties since the 1960s (Economic Research Service, 2017). Official poverty statistics are estimated by the US Census Bureau based on data from the Annual Social and Economic Supplement of the Current Population Survey (CPS-ASEC) using methods developed initially in the 1960s. The Official Poverty Measure (OPM) has been criticized on a number of methodological grounds (Renwick and Fox, 2016), though it continues to be an influential measure of the nation s economic well-being. Given concerns about the OPM methodology, since 2011 the Census Bureau has published the Supplemental Poverty Measure (SPM) to provide an additional measure to assess poverty in the US. Among other modifications, the SPM takes into account a broader set of factors influencing families resources, including government programs, medical expenditures, and differences in the cost of living in different regions. These differences are important for studying poverty in nonmetropolitan areas, as the metro-nonmetro gap in poverty rates can reverse depending on the poverty definition used. In 2016, based on the SPM, the poverty rate outside metropolitan areas was lower than in metropolitan areas (12.8% versus 14.1%, see Table A3). This paper makes use of newly available historical estimates of the SPM extending back to 1996 to offer insights into understanding poverty in nonmetropolitan areas (see Wimer et al. (2017)). 1 The SPM, with its detailed categories of income sources and necessary expenses, provides a more nuanced assessment of families economic well-being than the OPM. Further, we use the CPS-ASEC to analyze poverty dynamics, because cross-sectional 1 These data have been developed by the Center on Poverty and Social Policy (CPSP) at the Columbia School of Social Work. 2
estimates of poverty rates provide little information about whether and how families move in and out of poverty. Despite the fact that the CPS-ASEC provides nationally representative data on income and is the data source for the official poverty statistics, this paper provides one of the few studies to use linked CPS data to examine poverty dynamics. Numerous studies have shown that changes in employment and family composition are frequently associated with poverty transitions. Cellini, McKernan and Ratcliffe (2008) summarize much of the US literature on poverty dynamics. Most studies find that employment and earnings changes are common, although household structure and composition changes are also associated with poverty entry and exit (e.g., Bane and Ellwood (1983); Cellini, McKernan and Ratcliffe (2008); McKernan and Ratcliffe (2005)). Brown and Hirschl (1995) note that controlling for determinants of poverty transitions such as household and contextual factors did not fully explain the higher poverty rate in rural versus core metro areas. Most of the studies done to date have used the Survey on Income and Program Participation (SIPP), Panel Study of Income Dynamics (PSID) or National Longitudinal Survey of Youth (NLSY) for longitudinal analysis of poverty transitions, and so have relatively small sample sizes for nonmetropolitan areas. Our results highlight the importance of understanding differences in poverty definitions and data sources. The SPM and the OPM estimate different patterns of poverty rates and churn rates over time in metro and nonmetro areas. The paper provides support for using the linked CPS data to study poverty, given the availability of the SPM and the ability to track (short-term) poverty transitions as well as decompose the resource changes that occur with poverty transitions. We find that the likelihood of a family falling into poverty is non-zero at all income levels and annual rates of entry into and exit from poverty are similar in metro and nonmetro areas. Nonmetro families are slightly less likely to remain in poverty from one year to the next compared to metro families. Overall, changes in wages and salaries are more important for metro families, and while these are also important for nonmetro families, changes in Social Security, farm income and medical expenses play a larger role in poverty entries and exits for nonmetro than for metro families. Despite the short-term nature of the 3
linked CPS panel data, the depth of information about income sources and large sample size offers new insight into the dynamics of poverty entry and exit for nonmetropolitan families over the past two decades. 2 Background and Prior Literature This study builds on two strands of research literature: studies of poverty dynamics in general, and second, rural poverty more specifically. First, prior research on poverty dynamics establishes the importance of changes in employment status and earnings associated with poverty entry and exit. Cellini, McKernan and Ratcliffe (2008) conclude, based on a thorough review of the poverty dynamics literature, that employment changes are the leading cause of poverty entries and exits. Changes in family composition can also lead to poverty transitions (McKernan and Ratcliffe, 2005). Most studies of poverty dynamics in the US have used longitudinal survey data from the PSID or the SIPP. Indeed, of the 21 studies reviewed by Cellini, McKernan and Ratcliffe (2008), all but four used the PSID or SIPP. The shortcomings of the OPM to define poverty and study poverty transitions are well known and discussed in these studies (e.g., McKernan and Ratcliffe (2005)). Numerous studies have examined differences in the incidence and persistence of poverty in rural and urban areas of the US. Weber et al. (2005) review studies that examine county-level factors and those analyzing individual-level factors to assess whether there exists something unique or different about poverty in rural places. The geographic concentration of persistent poverty is well established (Weber et al., 2005). Based on the OPM, poverty is higher in nonmetropolitan than metropolitan counties, and highest in remote rural counties (Fisher, 2007; Jolliffe, 2003; Partridge and Rickman, 2008). Explanations for rural-urban differences in poverty rates focus on differential economic opportunities and social and demographic characteristics of residents. Rural residents are more likely to be employed and have two adults in the household (Findeis et al., 2001). At the same time, rural areas tend to offer lower wage jobs, more part-time work, and fewer work supports such as formal child care 4
(Davis and Weber, 2001; Gibbs, 2001; Gibbs, Kusmin and Cromartie, 2004). A growing number of studies use the linked CPS-ASEC data to study poverty or related topics. Hokayem and Heggeness (2014) use two-year panels to analyze transitions of into and out of near poverty, based on 100 to 125% of the official poverty threshold. Feng (2013) estimates poverty transition rates using linked CPS-ASEC data to illustrate issues with matching methods and assumptions. Both Hokayem and Heggeness (2014) and Feng (2013) rely on the OPM, rather than the SPM, in contrast to Pacas (2017). Hardy, Smeeding and Ziliak (2018) come closer to the SPM framework by using the linked CPS ASEC data to analyze the influence of policy and economic factors on growth in joint participation in safety-net programs over a thirty year horizon. This study builds on the prior research to address four main objectives: 1) Compare the OPM and SPM poverty rate trends over time in metro and nonmetro areas; 2) Describe and compare poverty churn rates (entry and exit rates) in metro and nonmetro areas; 3) Describe changes in resource components of SPM for families entering and exiting poverty; and 4) Analyze changes in specific resource components of SPM for families entering and exiting poverty to determine the frequency and importance of these changes for metro and nonmetro families. In this paper, we focus on resource changes rather than on family composition changes. In effect, Pacas (2017) showed that the driving factor in poverty transitions was not so much the family composition change alone but rather the resource change that accompanies the family composition change. That is, holding income constant, family composition changes do not have a large effect on poverty rate changes. Our future work will focus more directly on how families experiencing particular events (eg. divorce, death, etc.) are more/less likely to fall in or out of poverty. 5
3 Data and Methods 3.1 Defining OPM and SPM The OPM and SPM generally differ in their treatment of family units, resources, and poverty thresholds. This section gives a general overview of the main differences between the two poverty measures. 2 A simple way to think of the SPM is as a series of extensions to the OPM. As explained by the US Census Bureau: Concerns about the adequacy of the official measure culminated in a congressional appropriation in 1990 for an independent scientific study of the concepts, measurement methods, and information needed for a poverty measure. In response, the National Academy of Sciences (NAS) established the Panel on Poverty and Family Assistance, which released its report, Measuring Poverty: A New Approach, in the spring of 1995. In March of 2010, an Interagency Technical Working Group on Developing a Supplemental Poverty Measure (ITWG) listed suggestions for a new measure that would supplement the current official measure of poverty (Renwick and Fox, 2016). Three main concepts define both poverty measures. First, there are a series of thresholds that define the minimum level of income needed for a family and these depend on family size and family composition. The OPM thresholds are based on the cost of a minimum yet adequate diet for a family of four (2 parents and 2 children) in the 1960s (see Table 1 for the 2016 poverty thresholds). The poverty thresholds are a function of the number of people in the family crossed with the number of related children under 18 which is, in essence, the number of dependents in the family. As the family size increases, the poverty threshold increases. And, within a given family size, as the number of children increases, the poverty threshold decreases. 2 A more thorough examination can be found in Fox et al. (2015) and Renwick and Fox (2016). 6
On the other hand, the SPM estimates a threshold based on the 33rd percentile of expenditures on food, clothing, shelter and utilities calculated from the Consumer Expenditure Survey (CE) for the US. By doing so, the SPM thresholds more accurately capture the costs of living of the typical US household. As with the OPM, the thresholds are adjusted for family size and composition. Most importantly, these SPM thresholds are adjusted geographically while the OPM thresholds are the same throughout the US. The SPM thresholds are adjusted using 5-year estimates of median gross rents for different metropolitan statistical areas and non-metro areas for a total of 358 adjustment factors. 3 Second, families are defined to capture units of people who share resources. The OPM considers two distinct resource-sharing units: families and unrelated individuals. Treating families as a resource-sharing unit is uncontroversial. However, one of the criticisms of the OPM is that excluding unrelated individuals from families excludes people who are arguably sharing resources with the rest of the family. The most noticeable omission of the OPM family definition are unmarried partners (or cohabiting couples). The SPM adapts the OPM definition to account for different possible resource arrangements. Specifically, SPM families include unmarried partners and their relatives, co-residing unrelated children, and foster children. In essence, the SPM takes people considered to be unrelated individuals under the OPM and classifies them as a resource-sharing unit when appropriate. All other unrelated individuals in a household are considered as their own resource-sharing unit. For example, a renter living within someone s household would be considered an unrelated individual under both the OPM and SPM. 4 Third, resources for a family are then compared to the poverty threshold to define the poverty status of a given family. The definition of resources of a family also vary across OPM and SPM, though the OPM-defined resources are completely encompassed in the SPM. The OPM uses total cash income in the prior calendar year before taxes as its measure of 3 These thresholds are available at https://www.census.gov/hhes/povmeas/methodology/supplemental/ overview.html. 4 Arguably, even renters may be sharing resources, to an extent, with the rest of the household which is why many researchers choose to use households as the unit of analysis (McKernan and Ratcliffe, 2005). 7
resources. Table 2 presents the complete accounting of the cash income resources used in OPM and SPM resource definition. The total cash income in Table 2 are the full set of components used in OPM s definition of total resources. As is clear in Table 2, resources for the SPM attempt to measure a more complete picture of the resources that a family actually has at their disposal, taking the OPM resources as its starting point. The goal of the SPM resource measure is to monetize all the resources that a family has to spend on food, clothing, shelter and utilities. Therefore, SPM resources add non-cash public benefit recipiency to the total cash income of the OPM. The CPS collects information on the yearly receipt amount of the Supplemental Nutrition Assistance Program (SNAP), and the Low-Income Home Energy Assistance Program (LIHEAP). For Supplementary Nutrition Program for Women Infants and Children (WIC), National School Lunch Program and housing subsidies, the CPS asks about the receipt of the subsidy but imputes the amount. From these resources, the SPM subtracts resources that are not available for a family s consumption on food, clothing, shelter and utilities. The first, and perhaps the most obvious, are taxes. The Census Bureau imputes various different tax components using a model that simulates the tax liability of each family. Tax liabilities include federal and state income taxes, property taxes, Federal Insurance Contributions Act taxes (FICA) and federal retirement payroll deductions. To these liabilities, tax credits are subtracted out: the Earned Income Tax Credit (EITC), Child Tax Credit (CTC) and Additional Child Tax Credit (ACTC). In other words, these credits are added into a family s total resources. The second set of components that are subtracted from SPM total resources are necessary expenses. The SPM accounts for three in particular. The first set are work expenses. Using data from the SIPP, the SPM takes a national estimate of the reported expenses on commuting and work-related expenses (for example, uniform purchases). 5 Second, child care expenses are directly asked in the CPS and subtracted from the total resources. The third 5 Specifically, the SPM uses a value of 85 percent of median weekly expenses multiplied by the number of weeks respondents in the CPS reported working in the year. 8
set of expenses are child support paid which are also collected in the CPS. The last set of expenses that are subtracted from SPM total resources are medical expenditures. Medical expenditures have been acknowledged as a significant part of a family s budget and therefore were added as a question to the CPS. The CPS collects the amount a family pays for health insurance premiums and other medical expenditures not covered by insurance (i.e. prescription drugs and copayments). Medicare Part B premiums are subtracted for those reporting recipiency of Medicare. As Table 2 shows, there are nearly 40 components that constitute total resources under the SPM. 4 Data The data used in this paper come from two sources. First, the CPS-ASEC data come from IPUMS-CPS (Flood et al., 2017) for calendar years of 1996 through 2016. Note that the CPS-ASEC typically reports the year of survey implementation but asks about the previous calendar year. To simplify, we refer only to the calendar year. In effect, this means that data from 2016 come from the 2017 CPS-ASEC. The SPM is only available from the Census Bureau from 2009 onwards. We use the Historical SPM developed by Wimer et al. (2017) from 1996 through 2008. In order to construct the panels necessary for this analysis, linking keys are provided by the Census Bureau that longitudinally identify people and dwellings in the CPS. The procedure used here generally follows that outlined in Rivera Drew, Flood and Warren (2014) and Flood and Pacas (2016). Since the linking process is well-documented, the process is not covered in depth here. More important for this analysis is that there are two main issues with linking that affect the analysis of poverty transitions. First, the rotation pattern of the CPS allows for a given respondent to be found in at most two CPS-ASECs. Respondents for the CPS are selected at the dwelling-level in a given month; more specifically, it is the physical dwelling that is selected and not the set of people living at that dwelling. Once a dwelling is selected for participation, the entire household roster is interviewed (typically one 9
respondent responds for the entire household). The CPS is administered to this dwelling the next three months for a total of four interviews in a given year. After these four months, the dwelling is out of the CPS for eight months at which point the CPS is given to the dwelling another four times for the same calendar months. For example, a dwelling selected for the CPS in January of 2017 will be interviewed in February, March and April of 2017 and then January through April of 2018. Thus, a dwelling will only participate in the CPS-ASEC twice. In effect, this means that the CPS can be linked to create two-year panels, at most. 4.1 Metro v Nonmetro Because we use the CPS-ASEC, we are constrained to the geographic definitions made available which are limited to metropolitan-nonmetropolitan residence. Metropolitan areas are defined by the Federal Office of Management and Budget. Nonmetropolitan areas include counties that are outside the boundaries of metropolitan areas. A metropolitan area includes a central county with urbanized areas of at least 50,000 people along with nearby counties that are connected to the central county by a substantial amount of labor-force commuting (Economic Research Service, 2017). 4.2 Descriptive stats - Linked v non-linked sample There are over 4 million families in the CPS-ASEC from 1996-2016 but only about 1 million are in the linked sample with two observations. The linked sample is older on average, more likely to have no children, and more white. The linked sample is 41 years old on average, three quarters of whom are non-hispanic white, and 90% are US citizens (Table 3) while the non-linked sample is 37 years old on average, two-thirds are non-hispanic white, and 88% are US citizens (Table A1). While attrition is a concern, the people who drop out after the first year are very similar to the full sample as can be seen by comparing the descriptive statistics in Appendix Tables A1 and A2. All differences are less than one percentage point (the largest is 0.26 percentage points in absolute value). 10
4.3 Linked sample We discuss the characteristics of the linked sample in some detail because these characteristics are important context for understanding the observed patterns in poverty transitions and resource changes. We see some important differences in the socio-demographic characteristics of individuals and family structure between metro and nonmetro. From Table 3, we can see that the sample is slightly older on average in the nonmetro, 43 versus 41 years, in the metro areas. More families have no children in nonmetro areas (59 vs. 56%), and a smaller share are two-parent families with children (32 vs 35%). The nonmetro sample contains a higher percentage of white, non-hispanic persons. Notably, the share of Hispanics in the nonmetro linked sample is only 4% compared to 12% in metro areas. The difference in shares of African-Americans is smaller (by about 3.5% points). Nearly all of the nonmetro participants are US citizens (98%) compared to 89% of the metro participants. Differences in the composition of the nonmetro and metro characteristics may be related to differences in poverty transitions. Table 4 presents key characteristics of four groups separately for those in metro and nonmetro areas. The four groups include those who were not poor in either year ( never poor ), those who were poor in both years ( always poor ), those who entered poverty and those who exited. Those who were poor in both years were younger on average than any of the other groups, and the average age is higher in nonmetro than metro areas for each subgroup. Family structure varied across the poverty groups as well. Those who were not poor in either year were much less likely to be single adults with no children or single parents and more likely to be two-parent households. In contrast, only 16% of families poor in both years are two-parent households. While the never poor families have equal proportions of male and female references, females are disproportionately represented in families who were poor in one or both years. As expected based on crosssectional data, those who were not poor in either year are disproportionately white and US citizens. African-American and Hispanic families are overrepresented among families experiencing poverty, especially in the metro areas. SPM total net resources can be divided into those components that add to a family s 11
resources (such as wages) and those that are considered expenses and (usually) decrease resources available for consumption, such as medical expenditures. Taxes net of credits are included in the expenses section because we do not currently have separate estimates of tax credits. If tax credits exceed taxes paid, the net will be a positive contribution to the family s resources, but will be shown as a negative expense. Similarly, business or farm losses are negative contributors to family resources. We compare the shares of resources and average dollar amounts for metro and nonmetro families. As seen in Table 5, income from wages and salaries are the most important component of family resources, accounting for about 63% of cash and noncash resources on average for both metro and nonmetro families in their first year in the panel. In the second time period, the share of income from wages falls, with a larger decline for nonmetro families, for whom wages are only 52% of total resources. Social security income is the second largest component, comprising 17% of resources for metro families and 23% for nonmetro families (in both time periods). Other retirement income and government subsidies contribute about 6% of families resources. Other sources of income such as rent, dividends, interest and other sources of assistance contribute similar small amounts on average for families in both metro and nonmetro areas. The patterns for farm and business income differ, however, for nonmetro families. For metro families, business income averages about 2-3% in the two periods, and farm income is negligible. For nonmetro families, the reported business income share in the first time period is negative (i.e., business losses reduce total resources on average) but the share increases to 1.5% in the second income. Taxes and work expenses are higher in dollar terms for metro than nonmetro families, because these estimates reflect the higher average wages in metro areas. The mean shares of each resource component for the CPS sample as a whole differ from those of the linked CPS sample in predictable ways. The linked sample is older, on average, so it is not surprising that the share of income from wages is lower, and the share from Social Security and retirement is higher. The shares of other resource components are similar across the samples (Table not presented). 12
5 Results 5.1 Overall poverty rate The SPM estimates a slightly higher overall poverty rate in each year between 1996 and 2016 compared to the OPM (Figure 1). The gap ranges from 0.3 to 1.96 percentage points, with the differences between 2003 and 2008 somewhat larger than earlier or later years. The SPM and OPM measures of poverty also differ for families in metropolitan and nonmetropolitan areas (Figure 2). In fact, the metro-nonmetro poverty rate gap reverses sign depending on whether one uses the SPM or OPM. Using the SPM, the poverty rate is higher for metro families than nonmetro, while the reverse is true using OPM. Possible explanations and implications of this difference are discussed below. 5.2 Poverty churn rates To begin, we present the overall poverty entry and exit rates as well as those remaining in poverty for the full linked-sample. Figure 5 shows that the rate of people staying in poverty across two years is persistently lower than the rate of people entering and exiting poverty. Focusing in on the metro-nonmetro differences, we plot the annual rates of poverty entry and poverty exit as well as the percentage remaining poor separately in Figures 4-6 and combined in Figure 7. Recall that the linked CPS captures transitions based on comparison of annual resources to the relevant threshold in two succeeding years. Figure 4 shows that the annual flow rate into poverty based on the SPM is similar for metro and nonmetro families, rising from about 6% in 1996 and peaking above 8% in 2013. The poverty exit rate also ranges from about 6% of the population in 1996 and increases over time to close to 8% (Figure 5). Similar patterns are seen in metro and nonmetro areas, although the nonmetro rate is less precisely estimated. About 5% of the U.S. population is poor in both years (Figure 6). However, the most noticeable difference is the lower rate for nonmetro families those remaining in poverty. The metro-nonmetro gap in the percentage who are poor in both years has increased over the 20 years from 0.4 to 1.6. With about 80% of all families not 13
in poverty in both years, the metro-nonmetro gap in the overall poverty rate results from a slightly higher level of poverty churn for families in metro areas. 5.3 Risk of entering poverty by income level Figure 8 illustrates an important finding with regards to poverty entries. For this analysis, each family is ranked by resources in the first time period, and aligned along the horizontal axis such that the poorest families are near the vertical axis and the richest families at the right margin. The lines show the percent of families at each percentile of the t1 resource distribution who enter poverty in the second year of the panel. The t1 metro and nonmetro poverty lines are indicated at 12.3% (nonmetro) and 13.1% (metro). Not surprisingly, families with resources less than the poverty line are likely to remain in poverty. Those who are near the poverty line are more likely to enter poverty than those with more resources in t1. The rate of remaining in poverty is lower for nonmetro than metro families, and the likelihood of falling into poverty for those near the poverty line is also lower for nonmetro families. While the nonmetro estimates have more noise, it appears that the likelihood of falling into poverty converges for nonmetro and metro families with t1 resources about about the 30th percentile. 5.4 Decomposing poverty transitions into resource components When families enter poverty in the second year of the panel, their total resources nearly always decrease, and similarly, those exiting poverty nearly always experience an increase in resources. Table 6 presents the average level of each resource component in t1 and the average change between time periods separately for nonmetro and metro families who enter poverty. Both nonmetro and metro families who enter poverty in t2 suffer a loss of 76-80% of their t1 resources level. Income from wages and salaries falls by nearly three-quarters, and farm income falls in half. The losses are similar (in percentage terms) between nonmetro and metro families. In t2, families have, on average, higher medical expenditures, with nonmetro families experiencing an average of 64% increase compared to 48% increase for 14
metro families. Somewhat offsetting the decline in resources, families increase their receipt of government subsidies (SNAP, LIHEAP and housing subsidies) and reduce taxes paid. Families transitioning out of poverty in experience nearly symmetric changes in resource components, with large increases in wages and salaries driving most of the poverty exits. On average wages and salaries increased more than 200% for both metro and nonmetro families who moved out of poverty (see Table 7). Business income also increases by threefold or more, and farm income expands rapidly as well for nonmetro families, but remains a small share (1% or less on average) of families total resources. Social security grows for both metro and nonmetro families moving out of poverty but remains a larger share of resources for nonmetro families (37 vs. 28% of resources). Families see some decline in government subsidies in t2 as their other resources increase, but these are relatively small except for the decline in housing subsidies. 5.5 Measuring the importance of resource component changes for poverty transitions While certain resource component changes are large, with many possible resource changes the question remains how to determine which components are the most relevant for poverty transitions. This section aims to address two questions, first, for what percent of poverty transitions (entries or exits) does the expected change occur in a particular resource, and second, for what percent of transitions is the expected change pivotal in that the poverty transition would not have occurred without the resource change. When resources change in the expected direction (e.g., wages fall when families enter poverty while medical expenses increase), these changes may or may not be large enough by themselves to push the family above or below the relevant poverty threshold. We define a pivotal change in a resource component to be one that is large enough to result in a poverty transition, holding other resources constant. The frequency of pivotal changes provides a measure of the importance of changes in that resource component for poverty transitions. This section explains the simple method for formalizing the relative importance of resource changes to poverty transitions 15
presented in Pacas (2017). Consider the following example. A family of four people has total SPM resources of $60,000 in t 1 where $80,000 come from wages/salaries and medical expenses are $20,000. In t 2 wages/salaries drop by $55,000 to $25,000 while medical expenses drop by $20,000 to $5,000. The total SPM resources in t 2 are $20,000. This pushes the family into poverty since the poverty threshold is about $24,000. In this example, the change in wages/salaries and the change in medical expenses are expected. Neither one of these resource changes are pivotal, however, since the change of a single resource alone does not push that family into poverty. Now, consider the same family with the $60,000 in t 1. This time, the change in wages/salaries drops by $60,000 to $20,000 and medical expenses still drop to $5,000. Total SPM resources are $15,000. In this case, the change in wages/salaries is pivotal since the family would fall into poverty regardless of the change in medical expenses (i.e. $20,000 in wages/salaries already puts the family into poverty). The change in medical expenses is still expected because it moves in the expected direction. This example uses only two resources whereas total SPM resources consist of six major resource categories. To formalize this analysis consider families that are not poor in t 1 but enter poverty in t 2. By definition, the following is true: T otalresources t1 >Threshold t1 ) Not poor in t 1, T otalresources t2 <Threshold t2 ) Poor in t 2. With some algebraic manipulation, the following must be true: T otalresources t2 T otalresources t1 <Threshold t2 Threshold t1. For all practical purposes, the right-hand side is zero. Pacas (2017) established that, in practice, the changes in thresholds alone (i.e. family composition changes alone) have a negligible impact on poverty transitions in comparison to the change in total resources. Thus, to simplify analysis, it is defensible to study the following: TotalResources t2 TotalResources t1 < 0. 16
By definition, total SPM resources in any time period are defined as: TotalResources t = W orkincome t + OtherIncome t + GovSubs t MedicalExp t + Taxes t + NecExpenses t, where WorkIncome is the total family cash income from wages/salaries, OtherIncome is all other cash income, GovSubs is the total non-cash benefits, M edicalexp is the total family medical out-of-pocket expenditures, T axes are total taxes paid, and NecExpenses are a family s necessary work and child care expenses. The change in total SPM resources between two time periods is simply: TotalResources = TotalResources t2 TotalResources t1 = (W orkincome t2 W orkincome t1 )+(OtherIncome t2 OtherIncome t1 )+(GovSubs t2 GovSubs t1 ) (MedicalExp t2 MedicalExp t1 )+(Taxes t2 Taxes t1 )+(NecExpenses t2 NecExpenses t1 ) This can then be decomposed into specific SPM resource components and, for any given component, the equation can be arranged to isolate a particular change in one resource component. For example, income from wages/salaries can be described as follows: < (W orkincome t2 W orkincome t1 ) (MedicalExp t2 MedicalExp t1 )+(Taxes t2 Taxes t1 )+(NecExpenses t2 NecExpenses t1 ) (OtherIncome t2 OtherIncome t1 )+(GovSubs t2 GovSubs t1 ) This relationship states that for families who enter poverty, the loss in wages/salaries must be less than the change of expenses of a family net of their other income. Using this relationship, one could establish a counterfactual framework by setting all other changes in resources to zero. That is, if none of the other components changed, would this family still be in poverty? But, this approach does not take advantage of the actual changes in resources experienced by a family. Allowing resources to change, it is possible to define pivotal and expected changes. A pivotal change is defined as a change in resources 17
that is large enough to push a family into poverty even if other resources increase. Then it must be true for a change to be pivotal that a family would not enter poverty without the change in this resource. Then, setting (WorkIncome t1 would not enter poverty if the following holds: 0 > WorkIncome t1 )=0,afamily (MedicalExp t2 MedicalExp t1 )+(Taxes t2 Taxes t1 )+(NecExpenses t2 NecExpenses t1 ) (OtherIncome t2 OtherIncome t1 )+(GovSubs t2 GovSubs t1 ). In the case where this relationship holds true, the change in the work income is pivotal for exiting poverty, once all other components have changed. Using this final equation, the percent of families who enter poverty for whom the change in work income is pivotal to enter poverty can be calculated. First, define poor pivotal =1 if the equation above holds. Then, the percent is calculated as: P ivotalw orkincome = P poorpivotal P Poor to NonPoor. The expected condition can be constructed in a similar fashion. Since we know that the change in total resources must be negative in order for a family to fall into poverty, then the expected condition occurs when work income decreases regardless of the change in the other resources. The expected condition for families that enter poverty occurs when: (W orkincome t2 W orkincome t1 ) < 0. Intuitively, this percentage captures the number of families that enter poverty where the change in resources happen in the expected direction. That is, positive resources (e.g. work income) decrease for families entering poverty while negative resource (e.g. medical expenses) increase. Thus the expected conditions capture all the families that experience the expected change in resources while the pivotal condition captures the subset of these families where the change in one resource is enough to push the family into poverty. For families exiting poverty, the algebraic exercise is analogous to that just shown but where T otalresources t2 T otalresources t1 > 0. That is, all families that exit poverty experience a positive change in total SPM resources overall. 18
Looking first at poverty entries, over half of the metro families experience the expected change in wages (a decline) compared to only 44% of nonmetro families (see Table 9). And while one third of the metro families have a pivotal change in earnings, less than one quarter of nonmetro families do. Thus, while changes in wages and salaries are frequently important for both metro and nonmetro families, their role in poverty entries is more substantial for metro familes. Declines in business income occur in about 10% of poverty entries, but 4% or fewer entries have pivotal changes in this component. For nonmetro families, about 4% of poverty entries experience declines in farm income and about half of those (2% of all nonmetro poverty entries) experience pivotal declines (that is, large enough to result in the poverty entry holding all other resource components at their t1 levels). Increases in medical expenditures are common among poverty entrants (over half of all entries) but the percentage that are pivotal is small (6% in nonmetro and 4% in metro). Thus, medical expenditures by themselves typically are not large enough to result in a change in poverty status, but occur frequently with other resource changes that in combination result in a poverty entry. Government subsidies decrease for about one third of families entering poverty, but 4% or fewer experience these as pivotal changes. While net taxes paid and necessary work expenses increase with some poverty entries, these rarely are pivotal. The importance of specific resource changes for poverty exits is shown in Table 9. As with poverty entries, changes in wage income are very common for families exiting poverty. Wages increase in half of all metro families exiting poverty while only 42% of nonmetro families who exit poverty experience an increase in wages. Similarly, more metro families experience pivotal changes in wages (nearly one third, compared to less than one quarter of nonmetro families). Business and farm income increase for about 9% of metro families exiting poverty compared to 12% of nonmetro families. Fewer than half of these increases are pivotal, however, with increases in farm income pivotal for only 1.7% of nonmetro families exiting poverty. More than half of poverty exits are associated with increases in retirement income or Social Security, yet relatively few of these increases are large enough by themselves to result in a poverty exit (about 9-12% of poverty exits). Medical expenses decrease as well 19
for more than half of the families exiting poverty, but these by themselves are infrequently pivotal. Government subsidies increase for about half the families exiting poverty but in only 4% of cases are the increases large enough to be pivotal for the poverty exit. 6 Discussion and Conclusion The CPS-ASEC is an invaluable resource for studying the economic circumstances of American families given its extensive information on income and family structure. Leveraging this resource to study longitudinal events is somewhat limited, given the sampling design. However, through the use of two-year panels, one can examine transitions in and out of poverty (as well as other transitions in employment, program participation or family structure, for example). Compared to more traditional longitudinal surveys such as PSID, SIPP or NLSY, the sample sizes are large, with a long time horizon for historical comparisons. Perhaps more important for poverty research, is the available categorization of families as poor or not poor based on both the SPM and OPM. Neither measure of poverty is perfect, and neither measures consumption poverty or material hardship directly. The SPM provides a measure of the resources that the family obtains from public programs, both cash and in-kind, as well as making adjustments for cost of living in the area. These features suggest that using the SPM would be preferred, but at the same time, understanding which resource components are imputed, and how, as well as understanding the geographic adjustments, is necessary for assessing the observed metro - nonmetro patterns. One of the objectives of this research is to assess the linked Census data and SPM for analyzing rural poverty. The choice of SPM versus OPM clearly matters given that the metrononmetro gap in poverty reverses direction depending on the poverty measure. Based on the SPM, the poverty rate outside metropolitan areas is lower than inside metropolitan areas, in contrast to the OPM which measures a higher poverty rate for nonmetro than metro areas. The contrasting result demonstrates the importance of the geographical adjustment for cost of living differences, although these adjustments are not without controversy (Renwick, 2011; 20
Renwick et al., 2014; Renwick, Figueroa and Aten, 2017). Using the SPM to examine poverty transitions allows for a more detailed look at the accompanying changes in resource components for metro and nonmetro families. Overall the differences are perhaps smaller than expected, but this may reflect the inadequacy of the nonmetropolitan definition to identify families living in rural areas. In addition, estimates of certain resource components may not fully reflect metro-nonmetro differences. For example, work expenses are calculated using a figure based off of a national average of work expenses from the SIPP and applied uniformly across metro and nonmetro areas. While it is unlikely that work expenses are largely driving differences across OPM and SPM, this example highlights the sort of issues that require more attention for metro-nonmetro analyses. The results of this study, using linked CPS and SPM, generally confirm findings from other studies with regards to the importance of earnings and employment changes in explaining poverty transitions. With the addition of near-cash government subsidies and accounting for necessary expenses, the overall poverty rate and churn differ using the SPM and OPM. Social Security and retirement income stand out as particularly important resource components in nonmetro areas. However, more families may be economically vulnerable in metro than nonmetro areas, as evidenced by the higher likelihood of falling into poverty in the next year, particularly for those nonpoor who are close to the poverty line. The SPM allows for a closer look at the changes in specific resource components, the sum of which determine whether the family transitions into or out of poverty. Asecondobjectiveofthisresearchwastodocumenthowpovertyandpovertytransitions differ between rural and urban areas. The overall poverty rate, based on the unlinked sample and SPM, suggests that the percentage of the population who are poor is lower in nonmetro than metro areas. The churn rates of entry and exit into poverty are similar. The likelihood of remaining poor is lower for nonmetro families, however. Changes in resource components are similar, which is not surprising given the importance of wage income for most families. Overall, the patterns of resource changes were similar. As noted earlier, the lack of notable differences may reflect the use of nonmetropolitan categorization, given its availability in the 21
CPS, rather than rural. Additional research on predictors of poverty entries and exits may reveal starker differences between metro and nonmetro families. 6.1 Limitations While the use of the linked CPS-ASEC data and the newly available SPM estimates offer new insights into poverty transitions, we recognize a number of limitations of the study. In the two-year CPS-ASEC panels, a household is observed (at most) twice so these data cannot be used to study long-term poverty or multiple transitions in and out of poverty. Issues of attrition using the linked CPS data have been well studied, and may introduce unknown biases to the analysis. Unfortunately, the definition of nonmetropolitan is not ideal for studying rural populations, as about half of all people living in rural areas live in metropolitan counties (Economic Research Service, 2017). The analysis is descriptive, not causal, and the sequence of events cannot be determined by looking only at annual differences in changes in resource components. Nonetheless, given the difficulties of studying poverty transitions in general, and rural poverty in particular, the SPM estimates available in the linked CPS-ASEC data provide an in-depth look at patterns of (short-term) poverty dynamics over the past twenty years. This research provides a foundation for further research into the determinants and consequences of rural poverty. 6.2 Conclusion The causes and consequences of poverty differ across geographic regions because of differential access to jobs and other sources of income as well as differences in the cost of living. Understanding what drives poverty trends and transitions in a wealthy nation such as the US requires reliable and valid data. Using linked CPS-ASEC data, this study examines differences in metro and nonmetro poverty transitions between 1996 and 2016. Based on the SPM, the poverty rate outside metropolitan areas is lower than inside metropolitan areas, in contrast to the OPM which measures a higher poverty rate for nonmetro than metro areas. Despite the difference in overall poverty rate, the transition rates into and out of poverty 22
are similar for metro and nonmetro families. One difference we find between metro and nonmetro is that the risk of falling or remaining in poverty is lower for nonmetro families that are near the poverty line. At higher levels of income, the risk of falling into poverty are similar across metro and nonmetro families. Using a framework to establish pivotal changes in resource components, we quantify which resources are most important in explaining poverty entries and exits. The importance of specific resource components or necessary expenses are similar for the two groups, although changes in Social Security, farm income and medical expenses play a larger role in poverty entries and exits for nonmetro than for metro families. Nevertheless, overall the metrononmetro differences are relatively small. This lack of differences may reflect the inadequacy of the nonmetropolitan definition to identify rural areas, or may indicate that the economic, social and policy factors causing poverty are similar (on average) in metro and nonmetro areas. This initial work describing poverty transitions and resource changes using the linked CPS-ASEC data sets the stage for future work to analyze poverty in rural America. The linked CPS-ASEC data has been underutilized to study rural poverty despite the availability of SPM and large sample size. The recent release of SPM estimates extending back before 2010 provides new opportunities for analysis of anti-poverty programs and cyclical poverty trends. Specific steps we aim to take include looking at how particular life events (such as job loss, retirement, death, divorce, etc.) are associated with poverty entries and exits. Moreover, we look to study whether there have been changes in the relative importance of certain resource components in pre- and post-great Recession periods. Digging deep into the economic circumstances of families with the detailed CPS data will help inform policy to reduce the chances of falling into poverty or staying poor. 23