Trends in the Consumption and Income of Poor Families*

Transcription

1 PRELIMINARY AND INCOMPLETE Trends in the Consumption and Income of Poor Families* Bruce D. Meyer University of Chicago, Northwestern University and NBER and James X. Sullivan University of Notre Dame August 1, 2004 ABSTRACT There has been a long-standing debate over how the well-being of the poor has changed over time. This debate has received renewed interest in light of notable increases in income inequality during the 1970s and 1980s and, more recently, dramatic changes in the tax and transfer programs that affect the poor. A substantial literature has examined trends in summary measures of income and consumption inequality such as variances and Gini coefficients. Other work has examined the relative quality of income and consumption data for those with few resources (Meyer and Sullivan, 2003). This paper investigates more fully changes over time for those at the bottom of the distributions of income and consumption. We examine several datasets to determine what trends are robust to the choice of survey. We find sharp differences between recent trends in measured income and consumption. We first document these sharp differences, focusing on families headed by a single mother. Income changes are often as much as twenty percent below consumption changes at very low percentiles, but are often as much as twenty percent above consumption changes at somewhat higher percentiles. Changes in food consumption also have a somewhat different pattern than those for overall consumption. We then explore several explanations for these differences, including changes in income underreporting, changes in prices of goods such as housing, and other reporting and measurement issues. We also explore what can be learned about well-being from looking at the components of consumption or from examining the consumption of non-market time. *We would like to thank seminar participants at the Fundacion Ramon Areces, Universitat Pampeu Fabra, the University of California, Berkeley, the University of Virginia, and the Institute for Research on Poverty at the University of Wisconsin, Madison for their helpful comments. We would also like to thank Vladimir Sokolov for excellent research assistance. Meyer: Harris School of Public Policy, University of Chicago, 1155 E. 60 th Street, Chicago, IL bdmeyer@uchicago.edu Sullivan: University of Notre Dame, Department of Economics and Econometrics, 447 Flanner Hall, Notre Dame, IN sullivan.197@nd.edu

2 I. Introduction There is a long-standing debate over how the well-being of the poor has changed over time in the U.S. This debate has intensified in light of notable increases in income inequality in the 1970s and 1980s and, more recently, dramatic changes in tax and transfer programs targeted at poor families. Much research has informed this debate. One strand of this literature examines income or consumption poverty rates. Consumption poverty rates are generally lower than income rates when the same dollar cutoff is used for both poverty measures, but there is controversy over the trend in recent years. 1 Another strand of this literature examines overall measures of inequality. This research, while not focused on the poor, has also generated a fair amount of disagreement; while some researchers argue that overall inequality has fallen, most argue that it has risen. 2 Lastly, there is a body of research on the effects of the recent welfare and tax reforms which, in large part, increased assistance to those with jobs, but cut assistance to those not working. This research has noted that reforms have sharply altered the patterns of program receipt, employment, and earnings. However, within this body of research there is little consensus on how the well-being of poor families has changed over the past decade. This paper investigates more fully changes in the well-being for those at the bottom of the income and consumption distributions. We first report the trends for these measures of wellbeing at all points in the distribution for a sample of single mothers in the 1990s. 3 We show that how one measures well-being is of critical importance by contrasting the very distinct patterns of income and consumption. Our analyses in this paper do not directly address why true income or consumption changes over time, but rather, we investigate why the trends for these measures of well-being are so strikingly different. After documenting the trends, we explore possible reasons for the differences. We then further examine changes in well-being by using data on consumption disaggregated by category as well as data on non-market time. In earlier work we analyzed the arguments for income or consumption measures of wellbeing of the poor. 4 We argued that conceptual issues almost always favored consumption over income and that measurement issues were more evenly split. However, for low-resource single mothers we argued that even the measurement issues favored consumption. In analyses of the association between other adverse outcomes and either low income or low consumption, we found that consumption tended to be a better predictor. In a second paper we examined trends in 1 See Cutler and Katz (1991), Slesnick (1993) and Triest (1998). 2 See Cutler and Katz (1991), Johnson and Smeeding (1998), Blundell and Preston (1998), Krueger and Perri (2003), Attanasio, Battistin and Ichimura (2004), and Autor, Katz, and Kearney (2004). 3 We focus on single mothers and their children because they constitute a large share of the poor, especially poor children, and because they can be selected based on demographic characteristics rather than being selected on having low values for the outcome of interest, such as low income or low consumption. See Section IV for a more detailed discussion of why we focus on single mothers. 4 See Meyer and Sullivan (2003).

3 2 the consumption of single mothers at various percentiles of the distribution. 5 We relied on consumption based on a priori arguments, but did not compare income and consumption measures in detail. In this paper we show that income and consumption changes are sharply different. In other words, the measurement issues that we previously highlighted are crucial to understanding recent trends in the well-being of the poor. Two very recent research projects have examined roughly the same time period and demographic group that we study. Primus (2003) emphasizes recent decreases in reported income at low percentiles of the single mother population. In contrast, Blank and Schoeni (2003) downplay the declines at the very lowest percentiles, and instead emphasize substantial increases in reported income through the rest of the bottom half of the distribution of single mothers. We confirm these income patterns using Current Population Survey (CPS) and Consumer Expenditure (CE) Survey data. However, the reported income patterns differ very sharply from what the consumption data show. Using data from both the CE Survey and Panel Study of Income Dynamics (PSID), we find that consumption data show neither the sharp decline at low percentiles nor the large increases at the remaining percentiles in the bottom half of the distribution of single mothers. We suggest that it requires great confidence in the income data to believe the income trends given the large changes in both the sources of income and the reporting rates of the main income sources for single mothers during this period. Preliminary simulations suggest that under-reporting may play a substantial role in the difference between income and consumption changes in recent years. While sources of income change dramatically, changes in consumption shares are small but informative. We examine changes in food, housing, transportation, and child care expenditures in particular. There is some suggestion of higher rents in public or subsidized housing as a primary cause of increases in consumption at the very bottom, with transportation and food away from home accounting for increases as well. Child care expenditures appear too small, on average, to be important. These patterns differ greatly across income and consumption deciles. In this paper we do not attempt to separate out the causal effects of macroeconomic conditions and specific policies on income and consumption. We do however focus on the 1990s which were a time of sharp changes in policies and outcomes. Figure 1 reports changes in months of welfare receipt and weeks worked by income decile from the CPS between and Welfare receipt fell by more than 40 percent for all deciles besides the 2 nd and weeks worked rose sharply, more than doubling for those in the 3 rd decile of the income distribution. The remainder of the paper is organized as follows. Section II discusses the relative merits of income and consumption as measures of material well-being. Section III summarizes past work on income and consumption trends for poor single mothers in recent years along with some related research. In Section IV we describe our methods using the CPS, CE Survey and PSID data. Section V presents the trends for both income and consumption at different points in the distributions for single mother families. In Section VI we explore several possible explanations for differences that arise in the trends for income and consumption. In Section VII 5 See Meyer and Sullivan (2004).

4 we examine components of consumption and measures of non-market time to provide some evidence on the degree to which changes in total consumption reflect actual changes in wellbeing. In Section VIII, we conclude. 3 II. Income and Consumption as Measures of Well-Being In previous work we evaluate the merits of income and consumption as measures of wellbeing. 6 Conceptual arguments as to whether income or consumption is a better measure of the material well-being of the poor almost always favor consumption. For example, consumption captures permanent income, reflects the insurance value of government programs and credit markets, better accommodates illegal activity and price changes, and is more likely to reflect private and government transfers. Reporting arguments for income or consumption are more evenly split, with key arguments favoring income and other important arguments favoring consumption. Income data are easier to collect and therefore are often collected for larger samples. For most people, income is easier to report given administrative reporting and a small number of sources of income. However for analyses of families with few resources these arguments are less valid. These families tend to have many income sources. Income appears to be substantially under-reported, especially for categories of income important for those with few resources. Furthermore, the extent of under-reporting appears to have changed over time. Meyer and Sullivan (2003) provides evidence that for disadvantaged groups consumption is a better predictor of well-being than income. For example, we examine other measures of material hardship or adverse family outcomes for those with very low consumption or income. These problems are more severe for those with low consumption than for those with low income, indicating that consumption does a better job of capturing well-being for disadvantaged families. Meyer and Sullivan (2003) also provides evidence that commonly used household surveys have substantial under-reporting of key components of income. Weighted micro-data from these surveys, when compared to administrative aggregates, show that government transfers and other income components are severely under-reported and the degree of underreporting has changed over time. Comparisons of survey micro-data to administrative microdata for the same individuals also indicate severe under-reporting of government transfers in survey data. There is also some under-reporting of expenditures, but because expenditures often exceed income, we might be more concerned about over-reporting of consumption, of which there is little evidence. We also present strong evidence that income under-reporting and mis-measurement are particularly problematic for those with few resources such as low-educated single mothers. Expenditures for those near the bottom greatly exceed reported income. This result is evident in the percentiles of the expenditure and income distributions, and in comparisons of average expenditures and income among low-educated single mothers. These differences between expenditures and income cannot be explained with evidence of borrowing or drawing down 6 See Meyer and Sullivan (2003) for a more detailed discussion of these arguments.

5 4 wealth; we show that these families rarely have substantial assets or debts. In fact, fewer than half of all single mothers with a high school degree or less have any financial assets. 7 We argue that the difference between income and consumption for these disadvantaged families is largely explained by the misreporting of income. Thus, our null model for these analyses (and the current paper) is that true income equals true expenditures. In other words, for the families on whom we focus those with limited resources saving or dissaving to compensate for fluctuations in income is minimal. III. Past Research on Trends in Income and Consumption for Single Mother Families A few recent papers examine income or consumption trends for single mother families. In Meyer and Sullivan (2004), we find that the level of total consumption for single mothers increases in real terms during the 1990s. Relative to comparison groups, there is some evidence that consumption for single mothers near the bottom of the consumption distribution also increases. In most cases, we report a statistically significant increase in relative total consumption for single mothers between and Our results also show that some of these gains in consumption for single mothers occur after, but these changes are smaller and in many cases not statistically significant. Nevertheless, across our different specifications we repeatedly reject the hypothesis that relative consumption falls by more than 5 percent, providing strong evidence that the material well-being of single mothers does not appreciably decline in response to changes in welfare and tax policy. However, we emphasize that increases in consumption may not necessarily imply an increase in well-being if these increases come at the cost of reduced leisure time. Also, we emphasize that because we do not examine consumption at the very bottom our results provided no evidence on how reforms affected low-educated single mothers below the 15 th percentile of the consumption distribution. Through controls for unemployment rates and the use of comparison groups we build the case for these changes being largely due to welfare reform and the EITC. Two other papers that examine income show that the trends for income during this period differ noticeably at different points in the income distribution. Income falls sharply for single mother families near the bottom of the income distribution, while income rises sharply for other single mother families. In the context of an evaluation of welfare reform, Primus (2003) uses CPS data to examine recent changes in income for single mothers, focusing on changes for the bottom ten percent of the distribution. He examines a measure of income that reflects the disposable resources available to the family by including in his income measure not only pre-tax money income but also the EITC and the cash value of non-cash transfers such as Food Stamps. We reproduce the key table from his work as Table 1. His Group 1" which includes single female-headed households without a cohabitating male shows substantial increases in income between 1993 and 1996 up to the 20 th percentile of income. However, between 1996 and 2000 there are declines, often substantial ones, for those in the bottom 10 percentiles. These data are 7 See Sullivan (2004) for more evidence on asset holdings for low-educated single mothers.

6 5 the basis for his observation that welfare reform has caused many individuals at the bottom to lose income. Blank and Schoeni (2003) also analyze CPS data for the 1990s. They compare pre-tax money income in the period to that in the 2000 period. They suggest that income at the very bottom may be reported with substantial error and they are wary of conclusions based on observed movements in the bottom few percentiles of the distribution. On the other hand, they emphasize changes for the remaining part of the bottom half of the distribution of single mothers, noting that strikingly, many poor families have increases in their income of around 30 percent. These patterns can be seen in Figure 2 which is a reproduction of Figure 1 from Blank and Schoeni (2003). These changes are also attributed to welfare reform. While these changes noted by Primus and Blank/Schoeni could be due to improved macroeconomic conditions and other causes, they coincide with sharp changes in policy. As a starting point for our analyses, in Figure 3 we report trends for a CPS sample similar to that used by Primus and Blank/Schoeni. These trends shows the key results in Primus and Blank/Schoeni, the decline in income at low percentiles and the large gains through much of the remaining bottom half of the distribution. The trend for money income for our sample of single mother families is very similar to the trend reported in Figure 2 for a sample of children without both parents present. Because many single mothers receive substantial shares of income from non-money income sources such as the EITC or Food Stamps, we also examine the trends for other income measures that better approximate disposal income, or the resources individuals have available for consumption. As shown in Figure 3, the trends for these other measures of income are noticeably different from those for money income. Adding in Food Stamps reduces the gains in income, while adding in income tax payments and credits does the reverse, bringing us back some of the way towards the original figure. The measures that better capture disposable income show declines in income for a wider range of percentiles at the bottom than is evident using the original income measure. These measures also show much lower increases in income through much of the remaining distribution. IV. Data and Methods This paper presents recent trends for income and consumption, emphasizing how the trends for these measures of well-being differ at the bottom of the distributions. The results that follow focus on single mother families. We restrict attention to this narrowly defined demographic group for several reasons. First, this restriction allows us to concentrate on families with children that have limited resources a group with an especially high social welfare weight. Single mother families are disproportionately represented among families in poverty. Second, selecting the sample based on demographic characteristics is preferable to restricting attention to families that report limited resources, because the latter approach will systematically bias comparisons of income and consumption by conditioning on the variables under study (Meyer and Sullivan, 2003). Third, tax and transfer programs targeted primarily at single mother families changed dramatically during the 1990s. This has sparked much debate about how these

7 6 policy changes have affected the well-being of these families. 8 This paper contributes to this debate in particular by arguing that how one measures well-being is critical when examining the effects of these policies. Fourth, single mothers have historically received a large fraction of their income from sources that are not reported well on surveys. For example, public transfers such as welfare and food stamps account for a large share of income for many single mothers. These transfers tend to be sharply under-reported in surveys. Moreover, there is evidence that this under-reporting has increased noticeably over time. There is also evidence from the early 1990s that single mothers receive a substantial fraction of their earnings from informal work (Edin and Lein, 1997). This off-the-books income is likewise often under-reported in surveys. As many single mothers left welfare and informal work and moved into the formal labor market, the under-reporting rates for income may have changed significantly. By looking at this group, we can more closely examine whether under-reported and unreported income sources play an important role in differences between income and consumption trends at the bottom. To examine the trends in income and consumption, we draw on data from three nationally representative surveys: the Interview Survey of the Consumer Expenditure (CE) Survey, the Current Population Survey (CPS), and the Panel Study of Income Dynamics (PSID). The CE Survey, which is conducted by the Bureau of Labor Statistics (BLS), is a rotating panel survey of approximately 7,500 families each quarter (5,000 prior to 1999). Each family is interviewed for up to five consecutive quarters. The CE Survey reports data on both income and expenditures for the same families, as well as detailed data on demographic characteristics and employment for each member of the family age 14 and over. 9 Expenditure data are reported at the family level only. For more information on the CE Survey see Meyer and Sullivan (2003) or Bureau of Labor Statistics (1997). The CPS is a nationally representative monthly survey of approximately 60,000 households. The CPS is the most commonly used source of nationally representative income data and the source of our official poverty data. We use the March CPS files that include the Annual Income Supplement data. In the March interview respondents are asked to provide detailed retrospective information including usual hours worked, weeks worked during the previous year, and income for the previous year from a variety of sources including earnings, asset income, monetary transfers, and Food Stamps. The March CPS does not include data on expenditures. The PSID is an annual longitudinal survey that has followed a random sample of families, their offspring, and coresidents since The survey provides detailed economic and demographic information on both the family and individual level for a sample of about 7,000 families each year. Although the PSID does not survey families about all expenditures, it does 8 See Blank (2002) and Grogger et al. (2004) for overviews. 9 The main unit of observation in the CE Survey is the consumer unit, which includes all related family members or two or more persons living together who use their income to make joint expenditure decisions. For a subset of individuals within a dwelling to be considered a separate consumer unit in the CE Survey, at least two of the three major expense categories housing, food, and other living expenses have to be made jointly.

8 7 collect data on food and housing expenditures at the family level. We define our sample of single mother families as those families headed by a single woman between the ages of 18 and 54 who lives with her own children only and who has at least one of her own children under the age of 18 living with her. 10 This excludes single mothers living with other related or unrelated adults unless the adult is a child of the female head. We use sample weights from each survey so that all results reported in the following section are representative of the U.S. population of primary families headed by single mothers. In this paper we present the trends for both income and consumption as well as components of consumption. We examine several different measures of income, including pretax money income as well as other measures of spendable resources. Our measure of disposable family income includes all money income including earnings, asset income, and public money transfers for all family members. From money income, we deduct income tax liabilities including state and federal income taxes, and add tax credits such as the EITC. In addition, we add the face value of Food Stamps received by all family members. This income measure more accurately reflects the resources available to the family for consumption than the gross money income measure currently used to calculate official U.S. poverty figures. The calculation of after-tax income in this study varies slightly across surveys. For the PSID, we calculate tax liabilities and credits using TAXSIM (Feenberg and Coutts 1993). In the CPS, state and federal income taxes, payroll taxes, and tax credits are imputed by the BLS using respondent income and family characteristics. Tax data in the CE Survey are based on reports from the respondent. Another reason income may differ across surveys is that the PSID imputes missing values for components of income, while the CE Survey does not impute missing values for income. For this reason, our samples from the CE Survey include only complete income reporters excluding those with missing data for primary sources of income. About 10 to 15 percent of CE Survey respondents are classified as incomplete income reporters. Our measures of family consumption in the CE Survey are derived from expenditure questions. Three major adjustments distinguish our measure of total consumption from the measure of total expenditures reported in the CE Survey. First, our consumption measure excludes spending on individuals or entities outside the family. For example, we exclude charitable contributions and spending on gifts to non-family members. Second, consumption does not include spending that is better interpreted as an investment such as spending on education and health care, and outlays for retirement including pensions and social security. Finally, reported expenditures on durables tend to be lumpy because the entire cost of new durable goods is included in current expenditures. To address concerns about this lumpy nature of expenditures on durables, we convert reported housing and vehicle spending to service flow equivalents for our measure of consumption. For a detailed description of how we calculate these service flows, see Meyer and Sullivan (2001). See Appendix 1 for more precise definitions of our various measures of income and consumption in each survey. We also adjust both income and consumption to express these measures on the same 10 A woman is classified as single if she reports her marital status as never married, divorced, separated or widowed. Own children include blood-related sons and daughters, step children and adopted children.

9 8 scale across observations with different family sizes by dividing each outcome by the appropriate poverty threshold for each family. 11 We experimented with other methods of adjusting for family size, but given the types of analyses that we perform and the narrow demographic group on which we focus the results are not sensitive to the precise equivalence scale used. 12 Differences in reference periods for income and consumption in the CE Survey and PSID may affect comparisons of these outcomes. The reference period is longer for income (previous twelve months) than for consumption (previous three months). 13 Thus, our trends for income show the percentage change in an annual measure, while the trends for consumption show the percentage change for a quarterly measure. This is not of particular concern for comparing the means of these outcomes, however, because annual averages must have less variance than annualized measures over a shorter period, our expenditure and consumption measures are overdispersed relative to those for annual measures. To simplify the analysis of the trends for income and consumption, we group the data into two separate periods: 1993 to and 1997 to The first period begins after the end of the recession in the early 1990s, and ends prior to the passage of major welfare reform legislation in The second period starts after PRWORA was implemented in most states. While these policy changes do make this period particularly interesting to examine, we do not address the identification of the causal effects of these reforms in this paper. In the CE Survey the reference period for expenditures is the previous three months and for income it is the previous twelve months. Thus, our first period includes data from the second quarter of 1993 survey through the first quarter of 1996, and the second period includes data from the second quarter of 1997 survey through the first quarter of 2001 survey. Stacking the quarterly surveys for single mothers yields 3,075 family-quarter observations in the first period and 4,470 family-quarter observations in the second period. In the PSID the reference period for 11 This follows the approach used by Blank and Schoeni (2003). We use the official poverty line as defined annually by the Census Bureau. These thresholds vary by the size of the family and the age of its members. This poverty-line adjustment, converts our measures of income and consumption to constant dollars, because the poverty thresholds are adjusted annually for inflation using the CPI-U. 12 For example, we also adjust these measures using a scale for the number of adults and children in the family. We use a scale factor equal to s/(mean of s), where s= 1/(number of adults + number of children*0.7) 0.7. This is a fairly standard equivalence scale that follows National Research Council (). 13 Respondents in the CE Survey generally report income only in the second and fifth interviews. Income reported at the second interview is carried over to the third and fourth interviews unless a member over 13 is new to the CU, or a member of the CU that was not working at the time of the second interview is working in a subsequent interview. In these cases new values for family income are reported. 14 Our analysis is not sensitive to the precise specification of these two periods.

10 food expenditures is the previous month, and for income it is the previous calendar year. Thus, for food consumption trends in the PSID we include data from the 1993 through waves in the first period (N=2,049), and data from the 1997 and 1999 waves in the second period (N=817). 15 For income, we include data from the 1994 through 1996 waves of the PSID in the first period (N=2,154), and data from the 1999 and 2001 waves in the second period (N=827). 9 V. Trends in Income and Consumption Using CE Survey data, we can directly compare the patterns for income and consumption for the same sample of single mothers in the and 2000 periods. Figure 4 reports our key results for income and consumption measures. Several striking patterns are evident that show the sharp differences between income and consumption measures. The CE Survey income data show patterns that mirror those in the CPS income data. Total consumption, however, shows neither the pronounced drop for the bottom ten percent nor the sharp increase over the 10 th - 50 th percentiles that one sees with income. Total consumption does not increase as much for the bottom ten percent as for higher percentiles. Finally, food consumption shows small declines at low percentiles that disappear by about the 30 th percentile. The trends reported in Figure 4 are expressed in real terms using the CPI index for all items. One nice feature of consumption is that one can account for changes in the relative prices of various components. Thus, we also calculate the trends for total consumption and food consumption, adjusting major components such as food, housing, transportation, and apparel, using the respective CPI index for each component. This alternative approach had a negligible effect on the trends for total consumption and food consumption. We also examined income and consumption trends for single mothers for the years from 1984 through 1993, although we do not report these trends here. Throughout this earlier period consumption changes very little at all points in the distribution. The top half of the income distribution also shows little change for these earlier periods. At lower percentiles, however, income noticeably rises more than consumption. Table 2 reports some of the statistics behind Figure 4 including the 5 th percentile and the decile cutoffs for consumption and income in and We report these calculations for families of different sizes in different columns with consumption in the top panel and income in the bottom panel. One should keep in mind that these figures are for annual consumption and income for the entire family. As we emphasized in earlier work, the consumption numbers at low percentiles, even though they are shockingly low to the authors, are much larger than the corresponding income numbers. For example, the 5 th percentile of consumption for families with one child in the 2000 period is 2.35 times the 5 th percentile of income for the same families. It is also noteworthy that there is little or no tendency for the consumption numbers to rise with family size. In Figure 5 we report the same income and consumption measures for all other families excluding single mothers. This figure shows trends that are quite different from those for single 15 After 1997, the PSID is only administered bi-annually.

11 10 mothers. Income for families in the bottom decile of the rest of the population does not drop sharply. Nor does income rise between the 10 th - 50 th percentiles. Also, unlike the trends for single mothers, the trends in Figure 5 are very similar for different measures of well-being. At every percentile below the 90 th changes in after-tax income plus Food Stamps are not more than 5 percentage points greater than changes in consumption. We should emphasize that each line in Figures 3 through 5 sort the observations by the material well-being measure in question. Thus, for example, a family at the 10 th percentile of income is not necessarily the same family at the 10 th percentile of consumption. To see what has happened to the consumption levels of those in the low deciles of the income distribution, in Figure 6 we plot average consumption for each decile of the income distribution. For comparison, we also include the plot for changes in income by percentile from Figure 4. Figure 6 shows that consumption rises quite a bit for those with low income over our period. Consumption rises by over ten percent for families below the 18 th percentile of income, despite the fact that reported income falls for these same families. This result further suggests that the consumption-income difference is not due to some peculiar sorting of individuals over time, but one of different measures. The Panel Study of Income Dynamics (PSID) is another survey that provides consumption information for single mothers. Figure 7 reports the income and consumption measures of Figure 4 that are also available in the PSID. Several patterns are evident. PSID income shows increases over most percentiles. The increase is not confined to those between the 10 th and 50 th percentiles. 16 PSID food consumption is basically flat, with small declines for most of the percentiles below the 10 th, while at higher percentiles PSID food consumption shows small increases. These results show that the consumption changes are not an anomaly due to some aspect of the CE Survey. Food consumption in the CE Survey and PSID show quite similar patterns, but ones very different from the CPS and CE Survey income patterns, which are in turn similar to each other. VI. Possible Explanations for the Differences Between Consumption and Income We have clearly established that income and consumption have followed different trends in recent years. We see the same qualitative patterns in two income and two consumption datasets. While it is a tall order, we would like to examine if we can reconcile the differences between consumption and income. We have assembled strong evidence in prior work that transfers are under-reported and substantial evidence that earnings are under-reported. We have also shown that transfer underreporting has sharply increased in recent years and that the incentives to under-report income have sharply fallen so that now the incentives may be to shade reporting toward increased work. If the increased transfer under-reporting is disproportionately among those with low reported income and if increased earnings reporting is concentrated among those with the largest 16 We should note that updated data similar to that in Gouskova and Schoeni (2002) indicate that there seems to be a regime shift in the relationship between PSID and CPS income at low percentiles starting in 1997 with the PSID income becoming much higher.

12 11 increases in reported earnings, then it would seem that changes in reporting have the potential to explain the differences between income and consumption. At a less ambitious level, we are skeptical that income provides a good way to assess changes in well-being given that reported income shares attributable to transfers and earnings have changed, combined with reporting rates for some income components being well below one. Changes in reporting rates are likely to only bias income measures further. The top panel of Table 3 reports mean income from various sources by income decile for and The bottom panel of this table reports the share of total income from these sources. We can see that shares for income sources dramatically changes. The share of income from AFDC/TANF and Food Stamps in the bottom decile falls from 0.63 to In the second decile it falls from 0.65 to The share of income from earnings rises from 0.20 to 0.40 in the second decile and from 0.24 to 0.53 in the third decile. Table 4 shows AFDC/TANF under-reporting rates taken from Bavier (2002, with updates from the author). In percent of months of cash welfare were not reported, but by 2000 the share not reported had doubled to 38 percent. Despite the doubling of the under-reporting rate, the number of months not reported actually fell between 1993 and 2000, as one can see in Table 4. Thus, it is possible that declining true receipt could reverse the effect of a lower reporting rate. Since most of under-reporting is due to a failure to report receipt at all, 17 not reporting welfare could make income of non reporting recipients appear extremely low. We have performed some preliminary simulations of the effects of under-reporting of transfers. Specifically, we scale up the fraction of recipients in a given consumption decile of the CE Survey proportionately using the adjustment factors from Table 4. We use consumption deciles rather income deciles because the latter would systematically bias our results by conditioning on those with low reported income values. We only adjust for under-reporting of months of welfare received, not accounting for under-reporting of amounts conditional on reporting welfare receipt. We use the same approach both before and after welfare reform. We then examine how changes for the simulated income distribution compare to the changes in reported income. A full description of the simulation methodology can be found in Appendix 2. The effects of accounting for under-reporting in this way can be seen in Figure 8. Here we report changes in income by consumption decile as reported in the CE Survey as well as income adjusted for under-reporting. The figure shows that adjusting for AFDC/TANF and Food Stamp under-reporting in a conservative way fills in a bit of the trough in income and cuts off a bit of the peak, but most of the contours remain. A likely problem with this adjustment algorithm is that it does not specifically impute welfare income for those with high consumption, but low reported income. It is this group who we think are the likely nonreporters, but we do not have evidence, at least not yet, to support this contention. Such evidence we suspect would have to come from a new validation study. The question remains as to whether increased reporting of earnings with unchanged true income could explain the increases in reported income above the first decile of income. Suppose the reporting rate for earnings is equal to the national CPS average in Roemer (2000) of 17 See Moore, Marquis and Bogen (1996).

13 percent that he gives for wage and salary income and that the reporting rate for AFDC/TANF and Food Stamps is 63 percent as calculated in Bavier (2002), averaging over A reduction in welfare income compensated by an equal increase in earnings implied by the increase in earnings in Table 3, would imply 1, 10, 13, 15 and 12 percent increases in reported income in the 1 st through 5 th deciles, in order. This increase would just occur because wage and salary income is reported at a higher rate than welfare income. Thus, a shift from under-reported welfare income to earnings that are not under-reported could lead to a substantial part of the increase in reported earnings in deciles 2 through 5, even if true income has not changed. These illustrative calculations are based on just a shift in income from under-reported welfare to formal work that is reported fully. Another possibility is a shift from unreported informal work to reported formal work. The incentives for this type of shift during the welfare reform period were pronounced. The implicit tax rate on reported earnings under AFDC was often 67 percent or higher, while under the EITC a 32 or 40 percent subsidy rate is not uncommon. As the welfare rolls declined and the EITC was expanded, these changed incentives may have led a higher share of true earnings to be reported. VII. What We Can Learn from the Components of Consumption Unlike income, consumption data can be disaggregated into its components in order to explore more deeply the relationship between changes in total consumption and changes in wellbeing. For example, by looking at the components of consumption we can discern, in part, whether changes in total consumption reflect changes in the relative prices of different components. Looking at components of consumption can also shed light on the degree to which total consumption changes are the result of increased work expenses. Table 5 reports shares for various components of consumption for both periods. Many of the changes in these shares seem consistent with the trend towards increased work for single mothers during this time. Food at home falls while transportation increases for every decile of the consumption distribution. Over this same period, food away from home and child expenses also increase, but these components account for only a small fraction of total consumption, so increased spending for these components will not have a very important effect on changes in total consumption. Using data on these consumption shares we can decompose the change in total consumption into its major components in order to determine which components have had the largest impact on changes in the total. Figure 9 shows the overall change in consumption for each decile, as well as the contribution to the overall change from various components of consumption. We see that food fell at the bottom decile, but aggregate consumption did not fall because housing went up sharply. Overall, housing pulls total consumption up sharply in the bottom two deciles, while increases in transportation expenditures account for much of the increase in total consumption for deciles two through four, and to a lesser extent for higher deciles. One important fact that should be taken from this decomposition is that an analysis of changes in total consumption alone may lead to misleading conclusions about changes in wellbeing. For example, in the second decile consumption increases by 9.3 percent, but this increase is driven by higher housing and transportation spending. Moreover, increased spending on these components masks a drop in food consumption. There is some evidence that single mothers

14 13 substituted food away from home for food at home (see Figure 10), but because food at home is such a large fraction of total food spending for this group, total food consumption falls for percentiles below the 50 th. For these percentiles food falls despite the fact that total consumption increases. 18 Does the increase in housing consumption, particularly in the bottom two deciles, reflect an improvement in well-being for these families? Much of this increase in housing spending reflects higher rental payments, which went up by 17.6 percent in the bottom decile of consumption and by 28.5 percent in the second decile (Figure 11). 19 Rents for single mothers that receive public or subsidized housing assistance increase by even more in the second decile of the consumption distribution. Moreover, rent as a share of total consumption increased by 32 percent in the second consumption decile for renters who receive housing assistance. This suggests that these families may have qualified for smaller rent subsidies as their earnings increased, resulting in increased out-of-pocket rental expenses, but not necessarily improved housing conditions. Due to a lack of data on housing quality, it is difficult to discern how much of this increase in rents reflects improved living quarters. Looking at data on the number of rooms in the housing unit, we found no evidence that the increase in housing spending was the result of larger living quarters. The last remaining major component of consumption is non-market time. Table 7 reports rates of welfare receipt, average weeks worked during the year, and average hours worked during the year, by income decile. Figure 8 reports the same statistics but by consumption decile. It is well known that welfare and tax reform have been associated with an increase in work. These tables make clear where in the income and consumption distributions the increases in work and consequent losses in leisure and other non-market uses of time have occurred. Nonmarket time falls sharply for single mothers, especially those in the low deciles of the distributions of income and consumption. The decline in non-market time is somewhat more concentrated in the bottom two deciles of consumption than in the bottom two deciles of income. Hours worked more than double for the 2 nd through 4 th deciles of the income distribution, while they almost triple for those in the bottom decile of consumption. The increases in work are almost exclusively confined to the bottom half of the income distribution and the bottom seven deciles of the consumption distribution. VIII. Conclusions Earlier work by the authors suggested that income and consumption might yield very 18 Rates of food pantry use and gifts of food from others reported in Winship and Jencks (2004) do not suggest that there has been increased hardship among the poor. One should keep in mind though, that this evidence is only conclusive if the supply of assistance has not shifted or has shifted inward. 19 Less than 3.5 percent of all single mothers in the bottom consumption decile own a home in either period. For the second decile home ownership increase over time from 3.6 percent in the first period to 7.4 percent in the second period.

15 14 different pictures of the material well-being of the poor, and argued for the use of consumption when analyzing material well-being. In this work, we find that income and consumption give disparate summaries of the changes in single mothers circumstances in recent years. Income data suggest a sub-group that has experienced extreme deprivation, while most have done very well. On the other hand, consumption data suggest that the material circumstances of the poor have changed little. We argue that the consumption data should be preferred, but our conclusion is tentative at this point. A further analysis of consumption suggests a complicated picture of changes for lowresource single mothers. Some of the increases in consumption are potentially the result of the increased market work by single mothers during the 1990s. Over this period, housing expenditures rise, especially for those in public or subsidized housing, as do expenditures on food away from home and transportation. Child care expenditures rise somewhat, but are a very small share of most household budgets. Non-market time falls sharply for single mothers, especially those in the low deciles of the distributions of income and consumption. The decline in non-market time is somewhat more concentrated in the bottom two deciles of consumption than in the bottom two deciles of income.

16 15 References Atkinson, Anthony "Comparing Poverty Rates Internationally: Lessons from Recent Studies in Developed Countries." World Bank Economic Review 5(1): Attanasio, Orazio, Eric Battistin, and Hidehiko Ichimura (2004): Autor, David, Lawrence F. Katz and Melissa S. Kearney (2004): Inequality in the 1990s: Revising the Revisionists, Working Paper, MIT. Bavier, Richard An Early Look at the Effects of Welfare Reform. Unpublished manuscript, Office of Management and Budget, March. Blank, Rebecca and Robert Schoeni, Changes in the Distribution of Child Well-Being Over the 1990s, American Economic Review: Papers and Proceedings, 2003, 93(2): Blundell, Richard and Ian Preston Consumption Inequality and Income Uncertainty Quarterly Journal of Economics 113(2): Blundell, Richard, Luigi Pistaferri and Ian Preston Partial Insurance, Information and Consumption Dynamics. Working Paper 02/16, Institute for Fiscal Studies. University College London. Bollinger, Christopher R. and Martin H. David "Modeling Food Stamp Program Participation in the Presence of Reporting Errors." Journal of the American Statistical Association 92(3): Bollinger, Christopher R. and Martin H. David Estimation with Response Error and Nonresponse: Food-Stamp Participation in the SIPP. Journal of Business and Economic Statistics 19(2): Bound, John and Alan B. Krueger The Extent of Measurement Error in Longitudinal Earnings Data: Do Two Wrongs Make a Right? Journal of Labor Economics 9(1): Bound, John, Charles Brown, Greg J. Duncan, and Willard L. Rodgers Evidence on the Validity of Cross-sectional and Longitudinal Labor Market Data. Journal of Labor Economics 12(3): Bound, John, Charles Brown, and Nancy Mathiowetz Measurement Error in Survey Data. In Handbook of Econometrics, Volume 5, eds. James J. Heckman and Edward Leamer, Amsterdam: Elsevier. Branch, E. Raphael The Consumer Expenditure Survey: a Comparative Analysis Monthly Labor Review 117(12): Bureau of Labor Statistics Consumer Expenditure Survey, , Bulletin 955, U.S. Government Printing Office, November. Bureau of Labor Statistics BLS Handbook of Methods. Card, David and John DiNardo (2002): Skill Biased Technological Change and Rising Wage Inequality: Some Problems and Puzzles, NBER Working Paper No. w8769, February Clampet-Lundquist, Susan, Kathryn Edin, Andrew S. London, Ellen Scott, and Vicki Hunter Making a Way Out of No Way How Mothers Meet Basic Family Needs While Moving from Welfare to Work. MDRC.