Part 1 |
Save page Remove page | Previous | 1 of 2 | Next |
|
|
small (250x250 max)
medium (500x500 max)
Large
Extra Large
Full Size
Full Resolution
|
This page
All
|
ft~ United States ~~i~g 1 Department of ~ Agriculture Food and Nutrition Service Office of Analysis and Evaluation Food Stamp Research: Results from the Income Survey Development Program and the Promise of the Survey on Income and Program Participation r • 1 I ,. .. I I J : I l- PREFACE This report was prepared by Mathematica Policy Research under contract no. FNS 53-3198-5-20 from the U.S. Department of Agriculture, Food and Nutrition Service, Office of Analysis and Evaluation. This report represents a team effort in which a number of individuals made important contributions in addition to the authors. We gratefully acknowledge their assistance. We particularly want to acknowledge the sustained help and encouragement of Steven Cole of the Food and Nutrition Service. The authors would also like to thank the individuals from Mathematica Policy Research who helped with this report: John Czajka, Pat Doyle, Tom Fraker and Irene Lubitz contributed important advice and materials; and Lucia Wesley and the secretarial staff typed and formated the manuscript. March 1986 Sharon Long Harold Beebout Felicity Skidmore CONTENTS . I. II. Ill. IV. v. OVERVIEW • ••••••••••••••••••••••••••••••••••••••••• 1 A. Promise of SIPP............................... 1 B. Background. • • • • • • • • . • • • • • • • • • • • • • • • • • • • • • • • • • • 1 C. Objectives of this Monograph.................. 5 D. Organization of this Monograph................ 5 THE SIPP AND ISDP SURVEYS ••••••••••••••••••••••••• 6 A. The History of SIPP••••••••••••••••••••••••••• 6 B. The 1979 ISDP Research Panel.................. 7 C. The Structure and Contents of SIPP............ 10 ECONOMIC STATUS OF TH.E FOOD STAMP TARGET POPULATION. • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • 11 A. Asset Holdings of Food Stamp Participant and B. Nonparticipant Households ••••••••••••••••••• The Pattern and Adequacy of Multiple 11 Program Benefits............................ 14 c. Multiple Program Benefits During the Course of a Year................................... 20 THE DYNAMICS OF FOOD STAMP RECIPIENCY ••••••••••••• 24 MONTHLY PATTERNS OF INCOME RECEIPT •••••••••••••••• 30 VI. PROGRAM PARTICIPATION AND LABOR SUPPLY BEHAVIOR VII. VIII. OF FOOD STAMP HOUSEHOLDS•••••••••••••••••••••••• 35 A. Determinants of Participation in the B. Food Stamp Program •••••••••••••••••••••••••• The Effects of Food Stamp Participation On The Market Labor of Female Heads of Household ••• ISSUES IN THE USE OF PANEL DATA ••••••••••••••••••• 35 38 45 A. Analytical Issues............................. 45 B. Data Issues................................... 58 ISDP RESEARCH AND THE PROMISE OF SIPP ••••••••••••• 65 REFERENCES. • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • 6 6 LIST OF FIGURES APPENDIX FIGURE A- SURVEY WAVES-- SIPP THROUGH 1984 •••• A-1 , .. LIST OF TABLES APPENDIX TABLE A - SUMMARY OF THE METHODOLOGY OF MULTIVARIATE STUDIES ON THE DYNAMICS OF FOOD STAMP PROGRAM PARTICIPATION•••••••••••••••••••••••••••• A-4 APPENDIX TABLE B - SUMMARY OF THE METHODOLOGY OF MULTIVARIATE STUDIES ON THE DYNAMICS OF AFDC PROGRAM PARTICIPATION•••••••••••••••••••••••••••••••••• A-8 I. OVERVIEW A. PROMISE OF SIPP A major new data collection effort by the Census Rureau has enormous promise as the base for analyses of who the Food Stamp Program is serving and how effectively the program is achieving its purpose of raising the food purchasing power of low-income households. This data collection effort is the Survey of Income and Program Participation (SIPP) which, in the 1984 panel, is following a sample of more than 20,000 households, collecting detailed data on economic and household characteristics on a month-by-month basis over two and one-half years. The wealth of longitudinal behavioral data potentially available from SIPP for addressir~ questions related to eligibility and participation in food stamps and other programs is much richer than any previously available data. ~or the first time the actual behavior of households and individuals can be traced over time and potentially valuable new insights gained about the monthly patterns of household composition, income change, receipt of program benefits, and program turnover. Such new insights should in turn lead to improved targeting of program benefits, better estimates of program cost, and improved measures of the adequacy of program benefits in fulfilling program objectives. This ability to address a more comprehensive set of program questions using SIPP should provide a basis for better informed policy decisions. B. BACKGROUND Prior to SIPP, the analysis of the Food Stamp Program (FSP) as well as other income transfers and human resource programs was limited by the lack of income data. The maiority of policy research relied on the March Income Supplement to the Current Population Survey (CPS) and the Integrated Quality Control System data base (IQCS), although some studies used the smaller Michigan Panel Study of Income Dynamics (PSID). March CPS. The March CPS, also known as the Annual Demographic File, is an annual survey of a nationally representative sample of about 60,000 households and three times as many persons. The March survey collects a substantial amount of information for households, families, and individuals, including: o Employment status, occupation and industry, hours worked o Earnings, other income, program participation o Age, race, sex, marital status, ethnicity, education. However, the income data available from the CPS has several limitations since the primary focus of the survey is to obtain estimates of the size and characteristics of the labor force. 2 First, the income measures from the CPS are for the previous calendar year (i.e., the March 1985 CPS asked questions about income received in 1984). This can lead to severe recall problems for the survey respondents. Second, the survey asks about income received from groups of income sources, where the income groups are not necessarily the combinations needed in addressing policy issues. Further, there is little redundancy in the income questions, reducing the extent to which the accuracy and completeness of the data can be checked. Finally, the CPS is not truly a longitudinal file since households are not followed for long periods of time, rather, the CPS is a series of successive cross-sectional files. IQCS. The IQCS is an administrative file used to estimate the amount of food stamp and Aid to Families with Dependent Children (AFDC) benefits issued in error on a state by state basis. The data is obtained from reviews of a sample of 150 to 1,200 cases (i.e., program participant households) in each state over a six month period on a continuing basis. The full sample consists of approximately 45,000 cases. As a byproduct, the IQCS provides a source of detailed monthly data on the characteristics of participants in the FSP and AFDC, including: o Employment status, work registration o Earnings, other income, program participation o Assets; shelter, medical care and dependent care expenses o Age, race, sex, marital status, ethnicity. Although the IQCS provides all the detailed information needed to determine the household's eligibility for food stamps, the file is limited to food stamp and AFDC participant hous~holds. Consequently, issues dealing with the household's decision to participate in the FSP cannot be effectively addressed using the IQCS. Furthermore, since the IQCS is not a longitudinal file it is not possible to examine the dynamics of household participation in the FSP. PSID. The PSID, like the CPS, is an annual survey of a nationally representative sample. However, the PSID is a truly longitudinal data base with a primary focus on providing information on the income of families, particularly low-income families. Begun in 1968 with a sample of 5,000 families, the PSID provides annual information for families and individuals which includes: o Employment status, occupation and industry, hours worked .-· · o Earnings, other income, program pat;:_ticipation o Assets o Age, race, sex, marital status, ethnicity, education. While the PSID is a longitudinal file covering a long time period, the sample of households is relatively small compared to the March CPS or IQCS. SIPP. While the CPS, IQCS, and, to a lesser extent, the PSID were the only data bases capable of supporting a large portion of policy analyses, as the policy questions asked became more detailed and sophisticated, major weaknesses in the use of these data for policy analysis became apparent: o The March CPS and the PSID measure economic data and household composition on an annual basis, whereas eligibility and benefit calculations for most federal programs are based on a monthly accounting period. o Asset holdings and liabilities are not measured in the CPS and are measured on an annual basis in the PSID. o Complicated program rules and insufficiently detailed data in the March CPS and the PSID makes the identification of households eligible for program assistance difficult. o The IQCS contains information for program participant households only. o The information on participation in other, non-food stamp, programs is limited in the March CPS, the PSID, and the IQCS. o It is known from comparison with program records that the CPS data underestimate transfer income, retirement and disability income, unemployment compensation, and property income. The inability of the March CPS, the PSID, and the IQCS to provide the income and household composition data needed in addressing the more complex policy questions led to increasing appreciation of the need for more detailed and more frequent information on assets, income flows, noncash transfer income, household composition, and participation in government programs for both program participant and nonparticipant families, households, and persons. Although the CPS could potentially have been expanded to fill this need, rather than contemplating the major changes that would have been necessary in the survey instrument and procedures of the CPS--for which collecting income information is in any case a secondary goal--the decision was made to design a major new Survey of Income and Program Participation. 3 4 The primary advantages of SIPP over previous data bases are: o The monthly reference period allows need and eligibility to be assessed on the same accounting period as is used in program administration. o The longitudinal data allows questions related to the dynamics of income receipt and program participation to be addressed. o The richness of detail of the data allows a variety of issues, such as participation in multiple assistance programs, to be examined. o The potential matching of SIPP with administrative data would allow a wider variety of administrative issues to be addressed. The promise of SIPP has already been demonstrated in a series of research studies on the Food Stamp Program based on the test survey for SIPP. The 1979 test panel, known as the Income Survey Development Program (ISDP) Research Panel, is very similar in content and structure to SIPP. The results from the set of food stamp studies based on the ISDP data, as presented in the body of this report, provide important new information on the economic status and behavior of FSP recipient households.~/ However, there are serious difficulties to be surmounted which were indicated by the ISDP experience and are becoming increasingly clear as SIPP starts to be used. Whereas before SIPP the call was for more and better data, it is now becoming clear that data collection efforts of this size and complexity present analytical, data preparation and data management problems that are orders of magnitude more difficult than any faced before. Discussion has turned from problems of inadequate data to problems of how best to use the complex data. The experience in using the ISDP data on the FSP research studies reported in this paper indicate ways that the problems of data preparation and data management as well as the conceptual problems of longitudinal data can be handled. A key ~/In addition to the research reported here, the ISDP data proved very useful for providing rapid information to FNS on very specific policy issues. For example, the ISDP was used to examine the adequacy of the earned income and child care deductions in the food stamp benefit formula. SIPP should be useful in this context as well. rationale for this monograph is the hope that by sharing these ISDP experiences prospective SIPP research will benefit from the wide professional critique. C. OBJECTIVES OF THIS MONOGRAPH There are two basic objectives of this monograph. The first is to report the findings from a number of studies of FSP households all of which were based on the ISDP test panel. These studies are important in that they represent the best current information on questions such as how different asset limits would impact FSP eligibility or how FSP program exit rates are associated with factors such as finding a job. However, they are preliminary as more definitive research can soon be conducted based on the larger samples, longer time period, and more tested procedures of SIPP as well as on the more refined analytical procedures that are emerging from the earlier efforts. The second objective is to provide a basis for learning from the ISDP experience. The research reported here can provide important insights into the complexities which arise in the use of longitudinal data files such as the ISDP and SIPP. Maximizing the benefits of the ISDP-based work is contingent upon the sharing of the approaches used, results obtained, and limitations of this research with both the policymaking and research communities. D. ORGANIZATION OF THIS MONOGRAPH This monograph is divided into two major sections. The first section provides an overview of the SIPP and ISDP surveys (Chapter II) and briefly summarizes the existing food stamp research which uses the ISDP test panel--Chapter III presents research on the economic status of the food stamp population, Chapter IV summarizes studies on the dynamics of food stamp participation, Chapters V summarizes research on the monthly patterns of income receipt, and Chapter VI present studies dealing with program participation and labor supply behavior of food stamp households. The second section of the monograph presents a discussion of the analytic and data issues which arise in research using panel data (Chapter VII). The final chapter summarizes the ISDP research and the promise of SIPP. ~ · ' 6 II. THE SIPP AND ISDP SURVEYS A. THE HISTORY OF SIPP A longitudinal survey of the scope of SIPP had never been designed or fielded on a nationally-representative basis. As a result, a major development effort--called the Income Survey Development Program (ISDP)--was authorized in 1975 by the Office of the Secretary of the U.S. Department of Health and Human Services (DHHS) to precede the fielding of SIPP. The ISDP program sponsored extensive research into ways in which the measurement, collection, and processing of income, transfer program, and wealth data could be improved. It also undertook a series of test panels to try out alternative data collection and processing methods, the last of which (the 1979 ISDP Research Panel) was sufficiently large to provide reliable national estimates of many individual and household characteristics. The intent was to fully exploit the experience of the 1979 ISDP Research Panel in order to produce an optimal strategy for fielding the first and later waves of SIPP. Subsequent events prevented this intent from being fully realized. The ISDP program was halted in 1982. Recognizing that the 1979 ISDP test panel was a valuable resource in itself, several government agencies, including the U.S. Department of Agriculture, Food and Nutrition Service (FNS), funded the work necessary to release public use versions of the cross-section files. In addition, the FNS funded the longitudinal linking of the data across waves, a data access system, and the FSP research discussed in this paper. In the midst of the efforts to render the data usable for methodological and public policy research, the Census Bureau obtained the funding needed to conduct the full SIPP. Faced with stringent deadlines for fielding the initial wave, the Census Bureau relied heavily on the ISDP in developing questionnaires, data collection strategies, and processing systems. As a result, SIPP is very similar to the 1979 ISDP test panel-sharing several of the characteristics that led to the complexities in the ISDP data processing five years ago. The ISDP experience and associated research has thus acquired unique importance as a way to learn not only about the promise but also the potential pitfalls of using the SIPP data. B. THE 1979 ISDP RESEARCH PANEL The 1979 ISDP Research Panel consists of a nationally representative samp~e of approximately 7,500 households.~/ The sample design oversampled both low and high income households in order to improve the precision of measurement at both ends of the income distribution. The individuals in the initial sample households were followed for approximately one and one-half years. The survey consisted of six three-month rounds (or waves) of interviewing using a technique called staggered interviewing. That is, the sample of households was divided into three groups of equal size (called rotation groups) and the survey was administered to the individuals in one group each month. The first rotation group received the Wave 1 interview during February 1979 with subsequent waves occurring every three months thereafter. Figure 11.1 illustrates the outcome of this survey technique. Note that the third rotation group, although surveyed every three months, was not administered the Wave 4 instrument during the fourth round of interviewing. Instead, they were administered the Wave 5 questionnaire with the Wave 6 survey following three months later. As a result, for one-third of the sample there are five rather than six waves of data. The survey instruments used in the 1979 ISDP Research Panel included five core modules and a series of supplemental modules which differed by wave. The five core modules were administered during each of the first five waves of interviewing and covered the following topics: o Household composition and characteristics, and person characteristics at the time of the interview o Recipiency of income from assistance programs in each of the three months prior to the interview o Earned income and employment status for each of the three months prior to the interview l/This is actually a subset of all households in the original design. The total sample initially included approximately 8,300 households from the area frame sample and 3,000 households from samples drawn from program records. The 7,500 households are those households from the area sample successfully interviewed in early 1979 and subsequently reinterviewed throughout 1979 and early 1980. 7 00 Reference lob nth 1978 Nov Dec 1979 Jen Feb Mar "pr May Jun Jul "ug Sep Oct Nov Dec 1980 Jen Feb Mer "pr Mey FIGlflE 11.1 SURVEY W"VES -- 1979 I SOP RESE ... RCH P"'£l l---------------------------------------------1979----------------------~-~~-~~~-~~~~------------l l---------------------198o-----------------l Feb Mllr "pr ""'~ Jun Jul "ug Sep Oct Nov Dec Jen Feb Mllr "pr Mil~ Jun ~ tJ = P11nel Rote"tlon Group 1 --> ,-----~ b • Penal Rote"tlon Q-oop 2 c • Panel Rotation Group 3 " b " b c II --- b c -------> " c " b -- " b c Ill ------5 core modules,---------- b c ------> " attitudes. c " b a b c I IV ----------------------------------~~~=:!::~:::~~--------------------------: ___ 1_______ : : ~------->1 c v ---------------------------------5e~~~~b~~e~;~~~~-------l__ b c , _____ ,C 5 core ooodu las, b d lsebll I ty, personal history (...,..rlage, jobs, educ., migration) ----------------------------------------- ----------------------attitudes, reasons tor---------------------------- nor>-pert lclpetlon In "FOe, Med lcald. Food Stamps, SSI • 5 core modu las non-lncCJIIeproducIng assets, 11 11 fecyc le" earnIngs, past- secondary educ., child cere, school breektest end school I unch. (2/3 sample) c a c a b " b b 5 core modules, net worth, pension cCN"erZ~ge. VI ------> 1-c----------- c a c " b " b b Annual lnc~e roundup, cepltel gains lind losses, ta>c:es, I n-kl nd I ncoooe (trlnge benetlts, services, Incl. WIC, Enar(jy "ss 1t). o Unearned income for each of the three months prior to the interview o Characteristics of children under age 16 at the time of the interview. Each of the supplemental modules was administered once during the survey, with Waves 2 through 5 containing at least one of the supplemental sets of questions. Figure I I.1 also summarizes the contents of each of the five waves. As implied above, the reference period for most of the data collected was the three calendar months preceding the interview month. Hence, as can be seen from Figure II.1, the calendar period covered for each wave varies by rotation group. Wave 2 was particularly important for FSP research. W"i th supplementary funding from FNS, in addition to the five core modules, as noted, extra questions were asked on Wave 2 on workrelated expenses, shelter costs, and asset holdings which allowed the identification of FSP eligibles as well as participants. This is the only source of nationally-representative integrated data on (1) income, (2) household composition, (3) work-related expenses relevant to allowable deductions against earned income, (4) asset information relevant to the resources test, and (5) shelter and other expenses allowable as deductions in computing FSP benefits. Such information does not exist for any other ISDP wave; current plans call for it to be collected in the course of SIPP. However, the SIPP data will be more difficult to use as it is scattered through different topical modules to be administered in different waves. The dispersion of the FSP eligibility information throughout the different topical modules of SIPP is the result of the competing needs of the various federal agencies and programs and the academic community. Given the limited time available within each topical module and across the set of modules, it is necessary to choose among questions which are intended to aid in program planning (e.g., questions concerning FSP eligibility) and questions which address important social issues (e.g., questions concerning retirement decisions or health). Although the 1979 ISDP permitted more comprehensive anrl extensive analyses of the FSP than had previously been possible, 1979 was not a typical year for the FSP. In that year, the structure of the FSP changed dramatically as the Food Stamp Act of 1977 was implemented. The provisions of that Act included a reduction of the net income eligibility standard, tighter restrictions on deductibles, an increase in the asset limit, and, most importantly, the elimination of the food stamp purchase requirement. Following the elimination of the purchase requirement and 9 the implementation of the other provisions of the Act, participation in the FSP increased rapidly, rising by about 3.4 million people over the following year. Given these changes, 1979 cannot be considered a representative year for the FSP. Furthermore, the more recent changes in the FSP due to the provisions of the Omnibus Reconciliation Act of 1981, the Food Stamp and Commodity Distribution Amendments of 1981, and the Food Stamp Act Amendments of 1980 and 1982, suggest that there is a need for FSP analyses using the more recent data of SIPP. C. THE STRUCTURE AND CONTENTS OF SIPP The structure of SIPP, as noted, bears a close resemblance to the structure of the 1979 ISDP test panel.2/ First, SIPP has a similar instrument design--which combines a set of core modules on levels of monthly economic well-being and changes in these levels over time with a set of topical modules which vary by wave. Some of these are fixed and assigned to particular waves; others are variable and designed in response to particular government agency program and policy analysis requirements. Second, SIPP has a similar wave, rotation group, and reference period structure, although spread of over a four-month rather than a three-month period. Thus, each SIPP wave has four rotation groups, which are interviewed (in rotation) every four months. For each interview month the reference period is the previous four months (that is, four months ago, three months ago, two months ago, and last month). The sample of households is larger and the period of time covered is longer for SIPP than for the 1979 ISDP test panel. Through 1984, about 20,000 households were interviewed under SIPP, 5,000 each month. In 1985, a second panel of households was introduced bringing the total sample size of SIPP up to about 35,000 households. Each individual in the original sample frame will be interviewed every four months for a period of two and one-half years. Appendix Figure A shows the content and reference periods of SIPP for 1984, 1985, and most of 1986. 2/This description draws heavily on Roger A. Herriot and Daniel Kasprzyk, The Survey of Income and Program Participation. SIPP Working Paper Series no. 8405. Washington, DC: U.S. Bureau of the Census, 1984. '. ~ III. ECONOMIC STATUS OF THE FOOD STAMP TARGET POPULATION Included in the ISDP based research on the economic status of food stamp research are studies 'Yhich address the asset holdings of low-income and food stamp households (Bickel and MacDonald, 1981), the adequacy of benefits from food stamps in conjunction with other income maintenance programs (MacDonald, 1983, 1984), and the pattern and adequacy of multiple program benefits over time (Doyle, 1985; MacDonald, 1985). In this chapter, the studies dealing with each of these topics are summarized in turn. A. ASSET HOLDINGS OF FOOD STAMP PARTICIPANT AND NONPARTICIPANT HOUSEHOLDS Eligibility for food stamps is restricted to households which, in addition to meeting other requirements, have assets below a statutory limit. Such rules are designed to limit participation in the FSP to households without holdings of assets which can be readily liquidated and are therefore potentially available for the purchase of food. In 1980, Congress set the limit at $1,500 for most households; households of two or more, at least one of whom is age 60 or older, are allowed up to $3,000 in assets.~/ Liquid assets, such as bank accounts, stocks, and bonds, are counted toward the limit. In addition, a portion of the value of some vehicles is counted. A homeowner's house and lot are not counted~ Certain other nonliquid or employment-related assets, such as personal effects and tools of a trade, are also excluded. Before the 1979 ISDP test panel there were no nationally representative data on assets available in enough detail to assess the dimensions of current asset holdings by low-income households. Wave 2 of the 1979 ISDP test panel provided the necessary data for such an assessment. The report by Bickel and MacDonald presents findings from this data base on the types and value of assets held by FSP participants and several categories of nonparticipants.l/ In addition, in order to assess the sensitivity of food stamp eligibility to the resource test, Bickel and MacDonald simulate the impact of altering the FSP asset limits. ~/Although the $1,500 limit was increased briefly to $1,750 under the Food Stamp Act of 1977, the $1,500 and $3,000 asset limits were first established in 1971. 2/This section summaries Gary Bickel and Maurice MacDonald, Assets of Low Income Households: New Findings on Food Stamp Participants and Nonparticipants. Washington, DC: Food and Nutrition Service, U.S. Department of Agriculture, 1981. 11 Data and Methodology Findings 12 The Bickel and MacDonald work is of particular interest for two reasons. First, Wave 2 of the 1979 ISDP test panel contains more detailed information on assets than any contemplated wave of SIPP. Second, in contrast to SIPP, the 1979 ISDP Wave 2 contains the integrated assets and expense information necessary for accurate determination of the pool of program-eligible units. Wave 2 of the 1979 ISDP test panel provided information on assets of all household members, including savings accounts checking accounts, cash on hand, certificates of deposit, stocks, bond, cars, trucks, other vehicles, homes, insurance policies, all types of income-producing properties (rental, commer~ial, industrial, farm), undeveloped land, and other assets such as trusts, estates, and mortgages. Certain types of assets were not included in the data collected. However, for most households, pension and retirement funds were the only potentially significant omission. Additionally, sufficient other information on households' incomes and characteristics was collected in order to distinguish between households eligible for food stamps, whether or not they actually participated, and households not eligible. The sample used for this study included about 7,200 households. The analysis focused primarily on the actual food stamp eligibility unit or "food stamp household" within each sample household. All households were classified first according to the FSP income eligibility test and then the resource test. They were further classified into participants and nonparticipants according to whether they had reported receiving food stamps in any of the previous three months. (Three months was used to provide a broader measure of participants, given the relatively rapid entry and exit rates for the food stamp caseload.) Food stamp participants and three nonparticipant groups were defined to permit comparisons of asset holdings. The FSP nonparticipant groups were: 1. FSP eligible nonparticipant households 2. Low-income households which were ineligible for food stamps on the basis of their assets 3. Households which were ineligible for food stamps on all other grounds. The limits on assets have a major impact on FSP eligibility. Over twelve million persons who otherwise would have been eligible for food stamps had assets high enough to exclude them from eligibility. Based on this finding, Bickel and MacDonald estimated the savings in potential benefits to be $2.-9 billion in 1981 (19R1 dollars). Several other general patterns in the asset holdings of FSP participant households and others are noteworthy: o Nearly one-half (49 percent) of FSP participant households had no countable assets at all, while 91 percent had countable assets of $500 or less. o FSP participant households had far fewer assets than households not eligible for the program. When all assets except homes were considered, 37 percent of participant households had no assets, and 90 percent had less than $1,500. By comparison, 3 percent of FSP ineligible households had no assets, and only 21 percent had less than $1,500. o FSP participants tended to have few liquid assets. Fifty-one percent of the participant households had no liquid assets, 93 percent had $500 or less. There were no participant households in the sample with stocks or bonds. o FSP participant households had few assets in any other specific asset categories. Approximately one-half (51 percent) had no car, only 36 percent owned homes, 9 percent had life insurance policies, and 2 percent had rental property. No FSP participant in the sample reported any farm or business interests, undeveloped land, mortgages, royalties, estates, or trusts. Ry contrast, 87 percent of FSP households ineligible on the basis of either assets or income owned a car, 70 percent owned their homes, and 43 percent owned life insurance. o Among households eligible for food stamps, those who actually participated tended to have slightly fewer assets than those who did not. Thirty-seven percent of FSP participants, compared to 30 percent of FSP eligible nonparticipants, had no assets (excepting homes). o Households disqualified from food stamps solely on the basis of their assets were a relatively well-off group in terms of the types and value of assets they owned. Two-thirds (68 percent) of these households had total assets (not including homes) in excess of $5,000. Since the data for this report were collected in the Spring of 1979, both the income and asset limits in the FSP have changed. Thus, the figures in this report may not represent the current holdings of food stamps household. More recent data from SIPP is needed in order to obtain up-to-date information on the impact of the asset limit on FSP eligibility. 13 Data and Methodology 14 B. THE PATTERN AND ADEQUACY OF MULTIPLE PROGRAM BENEFITS The debate about whether and/or how to limit federal expenditures for income transfer programs has again focused attention on questions about the adequacy and equity of benefits available from the mix of social insurance, cash welfare, and in-kind programs that help maintain income security. With this debate comes disagreement about tradeoffs hetween the extent of income security provided and its costs. It has been argued that the gradual expansion and extension of income transfer programs has made multiple benefits available that provide more than is necessary for adequate support, and that it has reduced the incentive to work. Opponents reply that without multiple benefits there would be gaps in program coverage, and that the current system does not adequately provide for all.3/ Once again, this issue could not he properly examined before the 1979 ISDP test panel became available s:i:nce the ISDP provided, for the first time, concurrent information on monthly participation in a variety of income transfer programs, along with income information for the same period. The work by MacDonald on multiple program participation and the adequacy of benefits approached the issue by examining the relationship between income and poverty status4/ before transfers and the relationship including transfer income for recipients of all major income support programs. The paper includes comprehensive descriptions of the incidence of multiple ]/This section summarizes Maurice MacDonald, Multiple Benefits and Income Adequacy for Food Stamp Participant and Nonparticipant Households. Washington, DC: Mathematic Policy Research, February 1983. Prepared for the U.S. Department of Agriculture, Food and Nutrition Service. Summary results fro~ an extension of the analysis to include the impacts of nutrition and housing benefits are also included. See Maurice MacDonald, Multiple Benefits and Income Adequacy: Impacts of Nutrition and Housing Benefits. Madison, WI: Institute for Research on Poverty, August 1984. ~/The official Office of Management and Budget (OMB) 1979 annual poverty line was used as the household con ~nmption need standard in evaluating the adequacy of benefits. Three month incomes below one-quarter of that level were classified as inadequate to meet the households needs. (For more information on the official poverty threshold and for poverty statistics see U.S. Department of Commerce, Bureau of the Census, Current Population Reports, Series P-60.) benefit recipiency, the effects of benefits on poverty status, and the characteristics of recipient and nonrecipient households for different programs. It then focuses on the effectiveness of the FSP in alleviating poverty within the context of multiple benefits. The research reported here used Wave 2 of the 1979 ISDP test panel to examine the actual reported receipt of multiple benefits. It analyzes the incidence and effects of receiving benefits from one or more of the six most important cash and inkind transfer programs: Food Stamps (FSP); Public Assistance (PA) including Aid to Families with Dependent Children and other welfare programs; Unemployment Insurance (UI); Old Age, Survivors, and Disability Insurance (OASDI); Supplemental Security Income (SSI); and Medicaid. The combined effect of a somewhat larger list of transfer programs on the adequacy of benefits as measured by national poverty guidelines is also examined. In estimating the effects of transfers on benefit adequacy, transfers were added sequentially to market (pretransfer) income in the same order that these programs count income from other programs in determining benefit entitlements: social insurance (OASDI and UI) first, then cash welfare (PA and SSI), followed by food stamps (FSP).2_/ (Medicaid was omitted from the income count since it is seen as meeting special rather than normal income requirements.) The sample used for this study included about 7,200 households. Figure 11I.1 displays categories of households, to explain how the analysis of multiple benefit recipiency was organized. As can be seen, the primary focus is on households that were eligible for food stamps in Spring 1979. Sufficient information on income types, assets, and relevant expenditures was collected by ISDP to classify all households into food stamp eligible or ineligible groups • .2/There is a substantial literature on the evaluation of in-kind benefits (e.g., Smeeding, 1982, 1984; Manser, 1981) which proposes several different methods: market value, recipient or cash equivalent value, government cost, and poverty budget share. Government cost, since it includes administrative costs, would be likely to yield the highest valuation, while the poverty budget share would yield the lowest since it places a limit on the value of in-kind benefits. The approach used by MacDonald to value food stamp benefits was market value. 1'5 16 FIGURE II I. 1 CATEGORIES OF HOUSEHOLDS (IN MILLIONS) FOR MULTIPLE BENEFITS ANALYSIS, SPRING 1979 All Households ~~----------------- (80.6) ------------------ Food Stamp-Eligible (12.4) Food Stamp Ineligible (68.2) Food Stamp Participants Food Stamp Nonparticipants Poor (3.8) Not Poor (4.8) (64.4) Poor ( 3.3) SOURCE: I Not Poor (1.5) (7 .6) Poor (5.5) I Not Poor (2.1) MacDonald, M. Multiple Benefits and Income Adequacy for Food Stamp Participant and Nonparticipant Households. Washington, D.C.: Mathematica Policy Research, February 1983. Findings In Spring 1979, there were 28 million households who reported receiving benefits from one or more of the six major assistance programs studied. Less than one-quarter of these had benefits from two or more major programs. The other three-fourths received benefits from only one program. A closer look shows that about two-thirds of all households reporting benefits from any of the major programs received only OASDI. The principal findings of the analysis of the receipt of multiple benefits were: o Social insurance recipients (recipients of OASDI and UI) frequently do not have multiple benefits and are usually ineligible for welfare. Only 17 percent of OASDI recipient households and 34 percent of UI recipient households received benefits from other programs. o In contrast, multiple ben_efits were widespread among food stamp, Medicaid, and cash welfare recipients. In each of these programs, over 80 percent of recipients obtained benefits from at least one other program. About two-thirds of food stamp participant households had three or more of the six major benefits (most frequently, food stamps, Public Assistance, and Medicaid). o Many households below the poverty line received few or no benefits. Of pretransfer poor households who remained poor when all transfer income was counted, 38 percent did not receive any of the six major benefits, 32 percent received only one henefit, and 27 percent received food stamps only. Based on the analysis of the adequacy of the program benefits it was found that receipt of multiple program benefits did not lead to high income. The observed effects of multiple benefits on benefit adequacy include the following: o Of all pretransfer poor households, over one-half remained poor, 40 percent were removed from poverty by OASDI and/or UI, cash welfare and food stamps together removed another 5 percent. o Food stamps alone reduced poverty by 1 percent. If all eligible households had participated in the FSP, pretransfer poverty would have been reduced by 2 percent. o Among the pretransfer poor households who received food stamps, most (80 percent) remained poor after all transfers were counted. Of the 14 percent of food stamp households with money incomes below one-half of the poverty line, less than one-third still had incomes below that level after food stamp benefits were counted. , ., 18 Perhaps a more informative way of examining multiple benefit effects is to look at actual dollars. The dollar difference between each household's quarterly income and the quarterly equivalent of the poverty threshold for that household when added together for all poor households is defined as the poverty income gap. In general, all three major types of transfers (i.e., social insurance, cash welfare, and food stamps) had greater percentage effects in reducing the poverty gap than they did in reducing poverty counts. This supports the finding that most of the effect of transfers was to help those with incomes below onehalf of poverty. Among food stamp recipients, social insurance appeared less effective than did cash welfare in reducing the poverty income gap. In particular, it was found that: o Among all households nationwide, social insurance closed over 50 percent of the Spring 1979 $20 billion poverty income gap, cash welfare reduced it by 11 percent, and food stamps reduced it by another 4 percent. After all benefits had been considered, one-third of the original gap between pretransfer income and the poverty line remained. o Among FSP participant households, social insurance reduced the Spring 1979 gap between the pretransfer incomes of FSP participants in poverty and the poverty line by 23 percent, cash welfare reduced it about one-third, and food stamps closed about one-fifth. After all benefits had been considered, about one-quarter of the gap between incomes and the poverty line remained. o In contrast, for FSP eligible households who did not participate, over one-half the poverty income gap remained after cash benefits were added. Figure III.2 summarizes the results of the multiple benefit analysis by showing the Spring 1979 distribution of FSP eligible households relative to the official poverty line as they appeared (1) before government transfers, (2) after major cash transfers, and ( 3) after major cash transfers and food stamp transfers. Before government benefits were considered, 80 percent of FSP participants fell below national income poverty guidelines. This included 64 percent below one-half of the poverty line and an additional 16 percent between 50 and 100 percent of the poverty line. After all major transfer programs were considered, 64 percent of FSP participant still remained below income poverty guidelines, including 4 percent of FSP participants who remained below one-half of poverty. It is important to note once again that the FSP has changed considerably since Spring 1979. Cons~q.uently, more recent data from SIPP is needed to reflect the current adequacy of program benefits. Percent of 30 Households FIGURE III.2 EFFECT OF MULTIPLE BENEFITS ON INCOME AS A PERCENT OF THE POVERTY LINE, 1979 F'ood Stomp Porticlpant.s ~._----~r-----------T-----------~------ Pretransfer Income Percent of 3D Households With Welfare and Social Insurance With Welfare, Social Insurance, a!'ld Food Stamps F'ood Stamp EJI;ible Nonpartidpants ~----------,-----------------T---------- Pretransfer Income With Welfare and Social Insurance (but ~ Food Stamps) Household Income as a Percent of the Poverty Line ~over 130% ~ 100-129% §§ 50-99% ~Under 49% 19 20 A later supplementary analysis of multiple benefit receipt used Wave 4 of the 1979 ISDP test panel to include the School Lunch . Program and housing subsidies and Wave 6 to include the Special Supplemental Food Program for Women, Infants, and Children (WIC) (MacDonald, 1984). While this analysis is not strictly comparable to the main analysis primarily because using Waves 4 and 6 necessitated focusing on pretransfer poor households rather than FSP eligible household, this difference is not a major limitation on the analysis since the pretransfer poor define closely the subset of households who are eligible for most of the benefits of interest. The principal findings of this analysis were: o About 77 percent of both FSP and school lunch recipients reported receiving benefits from another program--compared with 88 percent for cash welfare. o When only cash benefits are counted, 11.2 percent of households were poor in late 1979 (the time of Wave 4). Adding the market value of food stamps, school lunches, and housing subsidies reduced the percent poor by 1.4 percentage points._§/ o The WIC data from Wave 6 indicate that approximately 1 percent of all households and 3 percent of all pretransfer poor households reported receiving WIC. Of the pretransfer poor WIC recipients, 38 percent reported receiving food stamps in Wave 4 and 62 percent reported both food stamps and school lunches. There were no households with school lunch only in Wave 4 who reported receiving WIC in Wave 6. C. MULTIPLE PROGRAM BENEFITS DURING THE COURSE OF A YEAR The studies presented in the previous section considered the receipt of benefits from multiple assistance programs at any point in in the interview wave. By using the five waves of the ISDP test panel, the patterns of transitions between multiple benefit categories over the year (Doyle, 1985) and the receipt ~/In estimating the value of benefits from food stamps, school lunch, and housing benefits, MacDonald ue ed the ·market value approach. Data and Methodology Findings of multiple benefits for the 12 months of calendar year 1979 (MacDonald, 1985) could be considered.?/ Both the study by Doyle and the study by MacDonald used a linked longitudinal file created from the five waves of the 1979 ISDP test panel. This longitudinal file was created for the analysis of the dynamics of FSP participation and is discussed in the context of that work in the next chapter. By using the linked longitudinal file, the patterns of the receipt of benefits by households from multiple assistance programs over the full 12 months of 1979 could be examined. The five assistance programs considered in the studies were: Old Age, Survivors, and Disability Insurance (OASDI), Unemployment Insurance (UI); Supplemental Security Income (SSI); Aid to Families with Dependent Children (AFDC); and Food Stamps (FSP). Eleven program participation categories were defined. When only one of the five ma.ior programs was reported by the household for a month the benefit type was simply that benefit. When more than one benefit was reported, the household was classified by whether or not it received food stamps as well as by the largest benefit amount from the other cash programs. The residual group for each month consisted of households without benefits from any of the five programs analyzed plus recipients of one very infrequent benefit combination of two major benefit programs (AFDC and UI). Doyle combined several multiple assistance categories because of small sample sizes. On average, there were 3,174 households per month in the sample for the Doyle study and 3,205 for the MacDonald study. The study by Doyle examined the extent to which households switched between nine multiple benefit categories and the nature of those transitions. The study found that one-third of the sample of households reported some change in the multiple program category in which they participated during 1979 and onefifth reported three or more transitions during the year. Other findings of interest include: o Households receiving OASDI only were the most stable group, with only 18 percent reporting transitions over the year. UI L/This section summarizes MacDonald, Maurice Serial Multiple Benefits and Monthly Income Adequacy. Washington, D.C.: Mathematica Policy Research, March 1985. Prepared for the U.S. Department of Agriculture, Food and Nutrition Service and a memorandum from Pat Doyle of Mathematica Policy Research the U.S. Department of Agriculture, Food and Nutrition Service. 21 22 recipients were the least stable, with all recipient households reporting at least one transition. o Of those households receiving food stamps, the most stable households were those also receiving OASDI, with almost three-fourths reporting no transitions over the year. Households receiving food stamps only or food stamps and SSI were the least stable. Over three-fourths of both groups of households reported at least one transition. The final food stamp group, households receiving food stamps and AFDC, was somewhat more stable, with 40 percent of all households reporting no transition. o The most frequent transitions for households receiving food stamps only or food stamps in conjunction with some other program were to multiple program categories which included food stamps. For example, over 70 percent of the households receiving food stamps only which 'experienced a transition added participation in OASDI, SSI, or AFDC to their food stamp benefits. o Of those households not participating in any assistance program in January, over one-fourth had begun participating in at least one program by December. For 5 percent of the households, the program they entered was the FSP. o The number of months between transitions is short ( t to 3 months) for households with two or more transitions and approximately 90 percent of such households return to their original multiple benefit category. o For those households observed to have only one transition approximately 50 percent had no benefits in January but began receiving benefits during the course of the year, 25 percent were participating in one program for the full year and supplemented that program at some time during the yeEr, and 20 percent simply discontinued benefits at some point in the year. MacDonald examined average monthly participation rates and average length of participation for eleven multiple benefit categories and the effects of the different benefit combinations on the adequacy of the benefits. The principal findings of the MacDonald study were as follow: o Among those who received any benefits from the five transfer programs for the whole of 1979, those with only OASDI or only Ul were by far the most frequent beneficiaries. Households receiving only food stamps were third in frequency at 7 percent of all 1979 recipients. About 30 percent of those who received benefits from any program received them from two or more programs at some time during 1979, most of whom had benefits from only two programs. o The adequacy of different monthly benefit combinations was evaluated by comparing incomes with and without the cash benefits and the face value of any food stamps they received to a monthly poverty line determined by the net income limit for food stamp eligibility in July 1979.~/ By that standard, sample households would have been in poverty for 35 percent of the months if they had not received assistance. The benefits they actually received were sufficient to reduce the percentage of 1979 months spent in poverty to 25 percent. o Whatever their income without the monthly benefits, receiving any single one of the cash benefits was always more advantageous for the household than receiving food stamps alone. Yet monthly food stamp benefits were sufficient to move about 30 percent of those recipients who would have been below onehalf of the monthly poverty line above that standard. ~/The monthly net income limit for the FSP is less than the official annual poverty line divided by twelve. Data and Methodology 24 ; ' ' IV. THE DYNAMICS OF FOOD STAMP RECIPIENCY The analysis of program turnover is crucial, among other reasons, in order to shed light on the major issues of long-term dependence on public transfers--how long do households typically remain recipients? What are the turnover patterns of different types of households? What factors affect turnover? The ISDP 1979 test panel provided the first nationally representative data with which monthly program caseload turnover rates could be analyzed. Such a task required monthly income and related information necessary to determine eligibility, detailed household characteristics, and monthly observation of program participation status. There are two studies which make full use of the longitudinal nature of the ISDP in looking at turnover in the FSP. The first study (Carr, Doyle, and Lubitz, 1984) generated overall estimates of turnover rates in food stamp participation and compared them with estimates generated from earlier, less representative data. In addition, separate estimates of turnover for important subgroups within the lowincome population were calculated and changes in eligibility status were estimated over time. The wide variation found by Carr et al. in food stamp turnover rates by type of household makes it important to identify what kinds of events cause households to enter and leave the food stamp caseload. The second turnover study (Lubitz and Carr, 1985) pursued this question by estimating the impact of particular events on the probabilities of entrance to and exit from the FSP. The ISDP data made such an analysis possible because it provided the month-to-month linked information needed to identify changes in household circumstances and to associate those changes with program entry and exit. The data base from which estimates of turnover in the FSP were derived was a linked longitudinal household file develoyed from the first five waves of the 1979 ISDP test panel~ Sociodemographic variables such as household composition, ethnicity, and education were included as well as indicators of food stamp receipt and all components of the eligibility determination process. For the purpose of the turnover studies, longitudinal household units were constructed according to the status of the principal person(s) in the unit and complete income data from all five ~/This section is primarily a summary of Timothy J. Carr, Pat Doyle, and Irene Smith Lubitz, Turnover in Food Stamp Participation: A Preliminary Analysis. Washington, DC: Mathematica Policy Research, July 1984. Prepared for the u.s. Department of Agriculture, Food and Nutrition Service. ISDP waves were used. Principal persons are generally the "reference person" in the terminology of the survey, and his or her spouse, if any. (They may be thought of as the heads of the household.) As a result of this approach, the composition and characteristics of units were allowed to change over time. For example, it is possible to observe an ongoing unit which with the passage of time gains or loses a household head or other member. For the purposes of the present analyses, the universe of households consisted of all units headed by a primary sample member. A primary sample member is an individual included in the initial sampling frame, and therefore ·followed throughout the survey to the extent possible. A household could be formed during any month and dissolved during any month. The only sample households excluded from the studies were units formed after the initial interview which were headed by individuals who were not present in Wave 1~ About 7,000 sample household units were retained in the analysis. The analysis in the study by Carr et al. was carried out in two phases~ First, a descriptive analysis provided an overview of the general level of turnover in food stamp eligibility and participation during 1979, the manner in which turnover varied over the course of the year, and the manner in which turnover varied across socioeconomic groups. Several indicators of turnover levels in program eligibility and participation were used. The most important of these were the entry rate and exit rate. For program participation, these are defined as the proportion of all households who did not receive food stamps in one month who were receiving food stamps in the next month and, similarly, the proportion of all households who did receive food stamps in one month who were not receiving them in the next month. Other measures of participation turnover used were the proportion of households that continuously received food stamps, the number of spells of food stamp participation during the sample period, and the average duration of food stamp participation. Analogous turnover measures were defined for program eligibility. Since such descriptive analysis cannot identify the independent effects of individual factors, the second phase of the analysis estimated a multivariate statistical model of program participation and eligibility. The probabilities of entering and exiting from the FSP (or to and from eligibility for the program) were estimated (using event history analysisl/) as functions of household characteristics expecterl to affect eligibility and participation. 2/see Tuma (1982) for a discussion of event history analysis. 25 Findings ; • . 26 There was substantial movement of households into and out of the FSP; the number of households who received benefits from the program over the course of a year was over 70 percent greater than the number who received benefits in any given month. Furthermore, there were significant variations in observed turnover across socioeconomic groups. Specific findings of interest include the following: o Of all households who received food stamps in a given month, 7.3 percent left the program within the next month. o Given that a household did not receive food stamps in a given month, there was an 0.53 percent probability that it would enter the program in the next month. o Of the households who were present in the sample for the full calendar year and reported receiving food stamps at any time, about one-third received food stamps for the entire year. o The probability that a household was a food stamp recipient household at least once in the course of the year was nearly twice (1.74) the probability that it participated in the program in a given month. o There were systematic variations in entry and exit rates across households. The lowest monthly exit probabilities were for households who received AFDC and/or other types of welfare, nonwhite households, households containing an elderly or disabled person, households in which no person was employed, households whose head has relatively little formal education, and households headed by a single person. o The highest probabilities of entrance into the program were for households who received AFDC, households headed by a single person with children, nonwhite households, large households, households in which no person was currently employed, households whose head had little formal education, and households in which an elderly or disabled person was present. The multivariate results tend to support the findings of the descriptive analysis. In particular, after controlling for other factors, entry rates were higher and exit rates lower for households with a nonwhite household head, no earner present, a single household head, elderly or disabled membe-rs, and AFDC recipients. The principal findings of the descriptive analysis of program eligibility are as follows: The Role of Changes in Household Circumstances '.~ o There appeared to be substantial turnover in food stamp eligibility. The probability that an eligible household became ineligible each month was about 17 percent, and the probability that a previously ineligible household became eligible was 6.3 percent. Both these probabilities are substantially higher than the corresponding probabilities of participation change. o The types of households with the highest propensity to become eligible for food stamps were those who received AFDC and other types of welfare, households headed by a single person with children, nonwhite households, households whose head had relatively little formal education, households in which no person was working, and households containing elderly or disabled persons. o The types of households with the lowest propensity to become ineligible were those who received AFDC and other types of welfare, those with a head over the age of 65, those with a single head, those with a single member, those in which no one was employed, and those containing a disabled person. Most of the relationships between characteristics and eligibility transitions observed in the descriptive analysis are maintained when other factors are held constant. In particular: o Households with elderly or disabled members, those on AFDC, and those who were nonwhite were all more likely to become eligible for the Food Stamp Program than otherwise similar households. o Single headed households with children were more likely to become eligible than were other households. o Households with elderly or disabled members, nonwhite households, single headed households, nonearners, and AFDC recipients were all more likely to remain eligible for food stamps than were other types of household. The subsequent study by Lubitz and Carr extended the earlier research on turnover by examining the role of changes in house hold circumstances on the likelihood that a household would enter or leave the FSP.2/ Whereas the initial study identified 2/This section summarizes Irene Smith Luhitz and Timothy J. Carr's Turnover in the Food Stamp Program in 1979: The Role of Trigger Events, February 1985. Washington, DC: Mathematica Policy Research. Prepared for the U.S. Department of Agriculture, Food and Nutrition Service. 27 in Food Stamp Turnover 28 ~·. characteristics of households that are correlated with higher or lower probabilities of transition in and out of the FSP, the subsequent study identified particular events that evidently precipitate changes in program status. For example, the findings of the Carr et al. study include a higher than average program entry rate for households with no earner present. The trigger events analysis estimated the probability of program entry in the months immediately following the loss of earnings. Five potential types of events were defined as trigger events (i.e., events reasonably likely to "cause" or precipitate a change): o Large changes in monthly pretransfer income between one month and the next.~/ Those cases in which the income change resulted in a change in eligibility status were also differentiated from those in which it did not. o A change in the number of employed household members from one month to the next. Multiple-earner were also differentiated from single-earner households. o A change in asset holdings, specifically a decline in assets as a possible trigger of program entrance. o Changes in receipt of Unemployment Insurance, either exhaustion of UI benefits (as an entry trigger) or beginning to receive benefits (as an exit trigger). o Household composition change, defined as a change from having one head to two heads, or vice versa. The methodology employed had two parts. First, a descriptive investigation of the relation between trigger events and transitions was undertaken to track households in the ISDP sample through the course of calendar year 1979 on a month-to-month basis. When a comparison of a household's circumstances in successive months indicated that a trigger event had occurred, the household was further tracked to determine whether a transition (entrance or exit) occurred. The probability that a household experienced a transition after a trigger event, and the ~/The effect of changes in total income on transitions into and out of the program was analyzed; however the analysis was not satisfactory because these changes confounded exogenous changes in the household's economic status (such as the loss of a job) with changes in transfer income brought on by that event. elapsed time between the two events, was then tabulated and analyzed. Second, a multivariate analysis of the effect of trigger events on entry and exit rates was used to isolate the independent efforts of different trigger events. This analysis (based on event history analysis) estimated transition probabilities as functions of explanatory variables in a manner similar to more familiar regression models. The emphasis was on isolating the separate effects of the different trigger ev~nts on transition rates. It is important to note that for many of the trigger events being considered the number of cases available for analysis is quite small. With unweighted sample sizes of less than 10 or 20 for households entering or exiting from the FSP following a trigger event, the analysis cannot support definitive conclusions about the impact of changes in household circumstances on program participation. The principal findings of the Lubitz and Carr study were as follow: o Trigger events were strongly correlated with the probability that a household experienced a transition (entry or exit). A household that experienced a trigger event was far more likely to experience a transition within six months than a household selected at random. o Most instances of entering and exiting from the program in response to trigger events occurred in the same month as the trigger event or shortly thereafter. o The event most likely to precipitate entry into the program was a decrease in the number of earners present in the household. Declines in pretransfer income resulted in households' becoming eligible to participate in the program. Household splitting, and exhaustion of UI benefits were also significant trigger events. o The events most likely to precipitate exit from the program were an increase in pretransfer income and an increase in the number of earners present in a household. Beginning to receive UI and becoming a couple (i.e., moving from one-head to twohead status) were also significant trigger events. o Changes in pretransfer income and in the number of earners were experienced by a large proportion of all households in the course of the year. To this extent, labor market phenomena are more important in explaining turnover than less frequent demographic phenomena such as changes in household composition. 29 30 ~ · . V. MONTHLY PATTERNS OF INCOME RECEIPT The FSP is a monthly program. Eligibility is based on income for a monthly accounting period and benefits are provided monthly. Therefore, survey data with monthly observations provide much more accurate information for describing FSP participating and nonparticipating households and for examining analytical issues concerning FSP design and operations. l.Ji th the monthly observations from SIPP, the household's income, receipt of other transfers, and household composition can be examined for the particular months during which the household reported receipt of FSP benefits. In contrast, the traditional data sources for FSP analyses such as the CPS and the PSID collect data about the past year. Since many low income households experience large income fluctuations from part year employment and earners leaving or entering the household unit, analyses of issues such as the incomes of FSP participants during their period of participation that are based on the CPS and PSID must rely on strong assumptions and are therefore imprecise. This study examined monthly patterns of income receipt and focused on the degree to which the assumptions about intrayear income flows used by analysts with CPS data were valid. One application, in which the assumptions concerning intrayear patterns of income are particularly important, is in the analysis of the effects of proposed program changes on program costs and the distribution of benefits. FNS and other agencies make extensive use of microsimulation models based largely on CPS data for the analysis of potential program changes. Ideally, the microsimulation model would use monthly data in estimating the impact.of program changes. However, prior to SIPP such data were not available and the data base most commonly used in microsimulation was and is the annual March CPS. Since the March CPS does not have information on intra-year income flows and household composition, assumptions about such intra-year patterns need to be made and built into the microsimulation models. The availability of the 1979 ISDP test panel provided the opportunity to test the assumptions underlying the microsimulation estimates which were obtained from the food stamp model used by FNS. The FNS food stamp model--Micro Analysis of Transfers to Households (MATH)--is a CPS-based microsimulation model. As it was originally developed, the model created variables which represent income received in a typical month by r~a ch individual surveyed using the following assumptions: o Annual earnings are assumed to be evenly distributed over the reported number of months worked. Data and Methodology o Annual unemployment compensation is assumed to have been received only during the periods of months looking for work or working part time. o Simulated annual public assistance is allocated according to periods of work and nonwork, but assumed to be a uniform amount per month within those two categories. o The types of income received during a simulation month are assumed predictable by current employment status. o Any unearned income in addition to public assistance and unemployment compensation (e.g., social security, alimony) is assumed to have been received evenly throughout the year. All these assumptions are obviously open to question. The descriptive analysis of patterns of monthly income receipt conducted by Doyle (1984) using the ISDP panel data sheds light on the appropriateness of these assumptions.J! Because the intent of the research summarized here was to improve the assumptions built into the microsimulation model operating on the CPS data bases, certain sources of variability, such as the number and type of jobs held during the year and changes in household composition, were not pursued. Attention thus focused on two sources of variation that can be incorporated into a CPSbased model: o The extent to which husbands and wives tended to work during the same or different periods over a year. o The extent to which types and amounts of unearned income varied over a year, in particular according to work/nonwork status. The data files were constructed from the set of individuals interviewed in Wave 5 with monthly retrospective data for calendar year 1979. Because Wave 5 interviews were conducted in the Spring of 1980, these files closely resembled the March CPS on which the food stamp simulation data bases are built. The labor force participation analysis focused on husband-wife families. The income patterns analysis focused on adults in poor or nearpoor households • .l/This chapter summaries Pat Doyle, An Analysis of Intra-Year Income Receipt. Washington, D.C.: Mathematica Policy Research, July 1984. Prepared for the u.s. Department of Agriculture, Food and Nutrition Service. 31 Findings 32 ~ ·. Due to the lack of longitudinal editing and imputation for nonresponse on income questions (discussed in Chapter. VII)--a potential source of bias in studies of income receipt over time--the study was restricted to households for which all income amounts were completely reported in each wave they were interviewed. The husband-wife data base included 3,283 families with complete data. The poor and near-poor data base included 9,383 individuals, 7,468 of whom were present in the ISDP sample for all 12 months of calendar year 1979. Individuals not present for a period were treated as receving no income for that period. In the construction of transfer program simulation models that operate on the March CPS, it has been assumed that husbands and wives who both reported working for part of the year tended to work during different parts of the year. This follows the theory (called the added worker effect) that when household heads are unemployed, secondary workers such as previously nonworking spouses enter the .iob market in order to maintain customary household income levels. Analysis of the ISDP data provides clear evidence that the added worker effect does not explain the labor force pattern of most husband-wife families. About 40 percent of couples had both spouses working; about 35 percent had the male working while the spouse did not; about 7 percent had the female working while the spouse did not. About 18 percent had neither spouse working. This should not be unexpected given that people are documented to marry persons similar to themselves along a wide range of personal characteristics: ethnic and religious backgrounds, age, level of formal schooling completed, and type of occupation. All these characteristics strongly affect the degree of an individual's attachment to the labor force and lead to a tendency for men with strong labor force attachments to marry women who also work for large portions of the year. In such cases there is necessarily substantial overlap in the periods they work. More detailed findings on labor force attachment of husband-wife families include the following: o The majority of elderly families (head over 65) had no labor force activity. In January, for example, 64 percent did not work; in December, the proportion had increased to 71 percent. o For white families 35 to 40 percent had both spouses working; 30 to 35 percent had only the sample reference person working; 7 to 9 percent had only the spouse working; about 20 percent had neither working. ..... o For nonwhite households there was more monthly fluctuation although no obvious trend over the year. In March, for example, 42 percent had both spouses working; 28 percent had only the reference person working; 26 percent had neither working. In June, 32 percent had both spouses employed; 35 percent had only the reference person working; 26 percent had neither working. o The upper bound estimate of the number of families where the spouse went into the labor force because the reference person lost his or her job was only 15 percent of the total number of families with at least one spouse working sometime during the year. The second component of -the analysis focused on the income patterns in poor and near-poor households. The major focus was on four types of unearned income: o Workmen's Compensation o Unemployment Compensation o Asset Income o Other (Nonwelfare) Unearned Income. Welfare income was given less attention because direct estimates of welfare income are not used in the CPS-based simulation model. Rather, welfare income is simulated from other data on the house-hold. ' On the CPS file, part-year workers can be distinguished from full-year workers and information is available on weeks worked and weeks looking for work. Beyond that, little can be inferred about individual patterns of income receipt within the calendar year. For this reason, the analysis of unearned income patterns from the ISDP summarized here focuses on the work/nonwork period distinction. The patterns of unearned income receipt are based on data on poor and near poor individuals: o Workmen's Compensation. The ISDP data indicate that about half the respondents who reported receiving workmen's compensation (53 percent) reported receipt during the work period. Of these, more than half (57 percent) reported receiving it only during the work period. These results should be interpreted cautiously, however, because the sample was very small; 71 individuals reported receiving it during the year, of which the vast majority (65) were part-year workers. o Unemployment Insurance (UI). According to the ISDP data, 17.6 percent of UI recipients in the sample (poor and near-poor) were full-year workers and 42 percent were part-year workers 33 34 ~·' reporting UI receipt while working. These seem unusually high proportions. However, examination of the extent of the overlap between work and UI suggests periods were ones of transition from work to nonwork or vice versa--a finding that is supported by the difference in UI benefit amounts according to work/nonwork status. Persons receiving benefits during both work and nonwork periods reported average receipts of $191 during the former and $331 during the latter. o Asset Income. More than half (55 percent) of the ISDP sample of individuals in poor/near-poor households who reported receiving asset income received it throughout the year. For the elderly (over age 60) the proportion was much higher (72 percent). For those who did receive asset income for only part of the year, the ISDP data cannot yield unbiased estimates due to the fact that lump sum recall amounts were evenly allocated over the relevant recall period (three or six months). The amounts of asset income for those who reported rece1v1ng both while working and while not working were very similar in the two periods ($33 and $32, respectively). o Other (Nonwelfare) Unearned Income. This category includes regular sources (such as social security, veterans benefits, and public and private pensions) and irregular miscellaneous income. The patterns of other unearned income receipt by the ISDP poor/near-poor population were very different for the elderly and the nonelderly. Over three-quarters of the elderly reported receiving such income continuously throughout the year versus one-quarter of the nonelderly. The patterns of other income receipt by the nonelderly varied by whether they were regular or irregular sources, it appears that twothirds of the recipients of regular income received it for more than a quarter of the year, half of which received it for the full 12-month period. On the other hand two-thirds of those receiving income from irregular sources received it for three months or less. Finally, for the elderly, whether they were working or not made no difference to the amount received ($325-$330 per month); but for the nonelderly the amount received was higher during periods of work ($452 per month) than periods of nonwork ($348 per month). Data and Methodology VI. PROGRAM PARTICIPATION AND LABOR SUPPLY BEHAVIOR OF FOOD STAMP HOUSEHOLDS Two studies based on ISDP cross-section data deal with the relationships between program participation and labor supply decisions: the work by Czajka (19R1) on determinants of participation in the FSP and the work by Fraker and Moffitt (1985) on the effects of food stamp benefits on the market labor of female heads of household. A. DETERMINANTS OF PARTICIPATION IN THE FOOD STAMP PROGRAM It has been well established that participation in the FSP among the eligible population has been and continues to be less than 100 percent. Policymakers, analysts, and other observers cite as explanations such factors as complex eligibility criteria, benefits being outweighed by costs for some households, the welfare stigma, or lack of program knowledge. But measuring the extent of and understanding the true reasons for nonparticipation depends on adequate data. Before ISDP (with the exception of local area studies) research was limited by the unavailability of subannual income data and any but the most basic assets data. The 1979 ISDP test panel--in particular Wave 2--was especially designed to yield more accurate and detailed income and assets data for the shorter time periods that better approximated those used in administering the FSP. It also made possible a close look at the characteristics of the eligible population and the economic, social, and demographic factors that differentiate food stamp participants from nonparticipants.Jj The analysis by Czajka used Wave 2 of the 1979 ISDP test panel and was done in terms of reference month--i.e., three months ago (month 1), two months ago (month 2), and one month ago (month 3). Reference months were used because the only calendar month common to the entire Wave 2 sample was April 1979. The study was also restricted to those households determined to be eligible: excluding those households reporting receipt of food stamp benefits who were seemingly ineligible. Since households in similar circumstances who did not receive food stamps could not be included among the eligibles, the inclusion of "seemingly ineligibles" among the recipients would yield biased estimates of the effects of household characteristics upon participation. _l/This section summarizes John L. Czajka, Determinants of Participation in the Food Stamp Program: Spring 1979. Washington, DC: Mathematica Policy Research, November 1981. Prepared for the U.S. Department of Agriculture, Food and Nutrition Service. 35 Findings 36 ~·. The analysis had two components. The first looked at food stamp recipiency simply as a function of prewelfare income and total cash income. The second employed two sets of regression equations to estimate the net effe.ct of economic circumstances on the likelihood of the food stamp eligible unit or "food stamp household" within each sample household receiving food stamps. One set of regression equations controlled for the separate effects of prewelfare income, the expected value of the food stamp allotment, employment income, whether unemployed, plus several demographic variables. The demographic variables were included as further measures of need, potential access to other economic resources, and an indirect measure of perceived stigma. The other set of regression equations controlled, in addition, for welfare receipt. The reason for the two formulations is the potential circularity of assuming welfare participation affects food stamp participation when in some cases the reverse may be true. The monthly income patterns reflected in the Wave 2 of the 1979 ISDP data provide strong evidence for the importance of income in determining the probability that an eligible household will participate in the FSP. Participation diminished significantly as a food stamp househpld's prewelfare income (measured relative to the poverty line) rose. The average net participation rate among households with prewelfare incomes as high as three-quarters of the poverty line or higher was only one-quarter what it was among households with incomes less than one-half the poverty line. Since the higher income households constituted more than 40 percent of all eligible households, their low participation depressed the overall participation rate substantially. The overall participation rate2/ ranged from 28 to 31 percent for the three reference months of Wave 2 of the ISDP. ~The participation rates found in the ISDP were substantially lower than the 61 percent found by Beebout (1981) for July 1981, using an administrative count of food stamp recipients and a microsimulation estimate of food stamp eligibles. Possible explanations for this discrepancy include: (1) the Beebout estimates are participation rates for persons while Czajka estimated food stamp household participation rates. Since large households have higher participation rates and weigh more heavily in the computation of a person level participation rate than a household rate, the person rate will tend to be higher; (2) Czajka eliminated seemingly ineligible households f rom his calculations, while Beebout's estimates include them in the numerator but not the denominator. The inclusion of the ineligibles in Beebout's calculations will lead to higher estimates of participation; and, (3) there appears to have been a significant amount of underreporting of food stamp recipiency in Wave 2 of the ISDP. Czajka (1982) discusses possible sources of this underreporting. There is a puzzle, however, in that the participation rate of the 365 households with zero gross cash income did not follow the general pattern. (Note that these were households with no welfare income.) These households--presumably the neediest segment of the population--accounted for 10 percent of all eligibles and had a food stamp participation rate of only 4 to 6 percent for the three reference months covered by the study. The most plausible explanation is that the group included a significant proportion of households that were in fact ineligible for the FSP but were retained in the sample because of the i~perfect replications of the FSP eligibility criteria. That is, some of the zero income households had effective incomes which were not zero and as a result were not eligible to receive food stamps. For example, the group might have included a large number of students living away fran home. It might have included persons whose prospective income was not zero (for_example, persons waiting for a job to begin or persons with relatively large assets--not large enough to disqualify them from food stamps, but enough to allow them to subsist without an income flow for a time). More generally, it might have included persons whose reported income and assets excluded important resources. It is interesting to note that even the extremely detailed monthly data on income and assets from Wave 2 the 1979 ISDP did not resolve the puzzle of these seemingly very poor nonrecipients. Other economic variables appear to be much less important influences on participation than income. Liquid asset holdings did lower the probability of participation--but only marginally. Moreover, too few households held liquid assets of any size for this to have much influence on the overall participation rate. Previous studies have shown welfare recipiency to be a very powerful predictor of food stamp use, and the present study replicates these earlier findings. At the same time, however, it is not clear what this result actually means. Food stamp recipiency may precede welfare recipiency in some instances. In others it may be decided .jointly with welfare. In either situation calling welfare recipiency a determinant of food stamp recipiency may greatly simplify reality. The strong association between welfare recipiency and food stamp recipiency needs to be better understood. Further research exploiting the longitudinal nature of SIPP may shed more light on this relationship in two ways. First, the actual sequencing of receipt of these two forms of transfer income over the two to two and one-half year followup period can be examined. The extent to which one comes before the other versus being received simultaneously would be very informative. Second, household decisions with regard to welfare and food stamps should be modeled jointly, since the household may well consider them simultaneously. 37 38 ., .. Among the sociodemographic variables examined, the number of children under age 16 exhibited the strongest overall relationship to food stamp participation. However, further research is required to identify the source(s) of this relationship: possibly the absolute level of benefits provided to large households or the frequent inability of such households to raise their incomes above the poverty level via the marketplace. A sizable differential in participation by level of educational attainment was apparent, with the less educated (fewer than six grades) showing substantially higher participation rates than the more educated (four or more years of college). One interpretation views education in terms of the potential earnings capacity it implies.. The distribution of eligible households by educational attainment also provides evidence of the possible role of education in determining the size of the eligible population: eligible households had markedly less education than the general population. Participation rates among the elderly were relatively low, in line with earlier results which might be explained by a more extensive accounting of assets. Relatively high participation was noted among blacks and single female-headed households. These might be related to regional or local factors, but the ISDP data do not contain the necessary information to find out. B. THE EFFECTS OF FOOD STAMP PARTICIPATION ON THE MARKET LABOR OF FEMALE HEADS OF HOUSEHOLD Despite the wealth of literature on the effects of transfer programs on labor supply, the study by Fraker and Moffitt (1985) is the first economic analysis of the effects of the FSP on work effort.3/ A model of the household's joint decision regarding market labor and participation in the Food Stamp and AFDC programs is specified and estimated on a sample of 358 female heads of households with dependent children (using lSDP data from Wave 5).4/ The analysis sample for this initial research effort was limited to low-income, female-headed households with dependent 2/This section summarizes Thomas Fraker and Robert Moffitt, The Effects of Food Stamp Participation on the Market Labor of Female Heads of Household. Washington, DC: Mathematica Policy Research, March 1985. Prepared for the U.S. Department of Agriculture, Food and Nutrition Service. 4/These 358 households represent 78 percen~ of all female-headed households with dependent children in the Wave 5 sample. The Model children to minimize the number of programs being modeled and to increase the interpretability of the results.2/ Estimates of the model's parameters are useci to predict how work effort and program participation would be affected by selected hypothetical changes in rules governing food stamp benefit amounts. The effects of work effort and program participation responses on average and aggregate benefit levels are also examined. The model of labor supply and program participation used by Fraker and Moffitt is based on the microeconomic theory of household decision making in the presence of limited household resources. The household budget limitations illustrate the combinations of goods that the household can purchase with its entire income. In this application of the theory, a household is assumed to weigh the tradeoff between leisure and income and then to choose the number of hours of work per month which give it the greatest level of satisfaction (utility). For single women with dependent children (the largest demographic category of food stamp recipients), the Food Stamp and AFDC programs introduce substantial complications into this tradeoff. Household decisions to participate in these programs are modeled as occurring simultaneously with the hours-of-work decision. The empirical specification of the model consists of three equations: 1. A labor supply equation, in which the FSP maximum benefit and the benefit reduction rate6/ are hypothesized to affect work effort via the level of nonlabor income and the effective marginal wage rate. 2. A food stamp participation equation, in which the probability of participation in the FSP depends on the difference in household utility when participating in the program and when not participating. 3. An AFDC participation equation that is analogous to the FSP participation equation. i/The income criteria which was used to limit the sample included: 1) hourly wage of $15 or less for the household head, 2) countable assets of $2750 or less ($4000 or less for households with more than one person and which contain an elderly person), 3) earnings of other adults of $2500 per month or less, 4) unearned, nontransfer income of $1000 per month or less, and 5) transfer income not including AFDC and UI of $1000 per month or less. ~/The maximum benefit is the level of benefits available to the household without income. The benefit reduction rate on earnings within .the program refers to the rate at which benefits are reduced for each dollar of earned income. 39 Data and Methodology Findings The model was estimated on the basis of 358 cases from Wave 5 of the ISDP 1979 test panel. To avoid the severe modeling and estimation problems associated with multiple, interacting programs, the analysis was limited to low-to-moderate-income households with dependent children headed by unmarried, nonelderly, and nondisabled women. These households were categorically eligible for AFDC benefits, either eligible for food stamps or able to become eligible by reductions in hours of work, but unlikely to be eligible for SSI or Social Security. (The Medicaid program was not considered in this study.) · Estimates of the food stamp and AFDC participation equations showed that a program's maximum benefit amount (at zero hours of work by the female head) had a statistically significant positive effect on participation in the program.?/ In contrast, and somewhat surprisingly, a program's benefit reduction rate was not found to significantly affect participation in that program, although the estimates do suggest the possibility of a weak inhibiting effect of the benefit reduction rate on program participation. Nonlabor income (excluding benefits from the Food Stamp and AFDC programs) was found to have a generally insignificant effect on participation in both programs. However, the wage rate (after income and payroll taxes, but not after food stamp and AFDC implicit taxes) had a significantly negative effect on participation in both programs, as expected. Estimates of the hours of work equation indicated that the Food Stamp and AFDC programs' maximum benefit amounts and benefit reduction rates had statistically significant but small impacts on work effort. The findings imply that moderate changes in food stamp (or AFDC) regulations which alter the program's benefit reduction rate or the maximum benefit have only small effects on the work effort of female heads of households. Estimates of the parameters in the three equations were used to simulate the effects on hours of market labor and food stamp benefit amounts of three hypothetical changes in current FSP regulations: 1. Increasing the benefit-reduction rate (i.e., the implicit tax rate on earned income) from .30 to .33 Llrn this study, an estimate is referred ~o as statistically significant if a test of the hypothesis that the explanatory variable has no impact on the dependent variable can be re.iected at the 10 percent level of significance.
Object Description
Page/Item Description
Title | Part 1 |
Full-text | ft~ United States ~~i~g 1 Department of ~ Agriculture Food and Nutrition Service Office of Analysis and Evaluation Food Stamp Research: Results from the Income Survey Development Program and the Promise of the Survey on Income and Program Participation r • 1 I ,. .. I I J : I l- PREFACE This report was prepared by Mathematica Policy Research under contract no. FNS 53-3198-5-20 from the U.S. Department of Agriculture, Food and Nutrition Service, Office of Analysis and Evaluation. This report represents a team effort in which a number of individuals made important contributions in addition to the authors. We gratefully acknowledge their assistance. We particularly want to acknowledge the sustained help and encouragement of Steven Cole of the Food and Nutrition Service. The authors would also like to thank the individuals from Mathematica Policy Research who helped with this report: John Czajka, Pat Doyle, Tom Fraker and Irene Lubitz contributed important advice and materials; and Lucia Wesley and the secretarial staff typed and formated the manuscript. March 1986 Sharon Long Harold Beebout Felicity Skidmore CONTENTS . I. II. Ill. IV. v. OVERVIEW • ••••••••••••••••••••••••••••••••••••••••• 1 A. Promise of SIPP............................... 1 B. Background. • • • • • • • • . • • • • • • • • • • • • • • • • • • • • • • • • • • 1 C. Objectives of this Monograph.................. 5 D. Organization of this Monograph................ 5 THE SIPP AND ISDP SURVEYS ••••••••••••••••••••••••• 6 A. The History of SIPP••••••••••••••••••••••••••• 6 B. The 1979 ISDP Research Panel.................. 7 C. The Structure and Contents of SIPP............ 10 ECONOMIC STATUS OF TH.E FOOD STAMP TARGET POPULATION. • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • 11 A. Asset Holdings of Food Stamp Participant and B. Nonparticipant Households ••••••••••••••••••• The Pattern and Adequacy of Multiple 11 Program Benefits............................ 14 c. Multiple Program Benefits During the Course of a Year................................... 20 THE DYNAMICS OF FOOD STAMP RECIPIENCY ••••••••••••• 24 MONTHLY PATTERNS OF INCOME RECEIPT •••••••••••••••• 30 VI. PROGRAM PARTICIPATION AND LABOR SUPPLY BEHAVIOR VII. VIII. OF FOOD STAMP HOUSEHOLDS•••••••••••••••••••••••• 35 A. Determinants of Participation in the B. Food Stamp Program •••••••••••••••••••••••••• The Effects of Food Stamp Participation On The Market Labor of Female Heads of Household ••• ISSUES IN THE USE OF PANEL DATA ••••••••••••••••••• 35 38 45 A. Analytical Issues............................. 45 B. Data Issues................................... 58 ISDP RESEARCH AND THE PROMISE OF SIPP ••••••••••••• 65 REFERENCES. • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • 6 6 LIST OF FIGURES APPENDIX FIGURE A- SURVEY WAVES-- SIPP THROUGH 1984 •••• A-1 , .. LIST OF TABLES APPENDIX TABLE A - SUMMARY OF THE METHODOLOGY OF MULTIVARIATE STUDIES ON THE DYNAMICS OF FOOD STAMP PROGRAM PARTICIPATION•••••••••••••••••••••••••••• A-4 APPENDIX TABLE B - SUMMARY OF THE METHODOLOGY OF MULTIVARIATE STUDIES ON THE DYNAMICS OF AFDC PROGRAM PARTICIPATION•••••••••••••••••••••••••••••••••• A-8 I. OVERVIEW A. PROMISE OF SIPP A major new data collection effort by the Census Rureau has enormous promise as the base for analyses of who the Food Stamp Program is serving and how effectively the program is achieving its purpose of raising the food purchasing power of low-income households. This data collection effort is the Survey of Income and Program Participation (SIPP) which, in the 1984 panel, is following a sample of more than 20,000 households, collecting detailed data on economic and household characteristics on a month-by-month basis over two and one-half years. The wealth of longitudinal behavioral data potentially available from SIPP for addressir~ questions related to eligibility and participation in food stamps and other programs is much richer than any previously available data. ~or the first time the actual behavior of households and individuals can be traced over time and potentially valuable new insights gained about the monthly patterns of household composition, income change, receipt of program benefits, and program turnover. Such new insights should in turn lead to improved targeting of program benefits, better estimates of program cost, and improved measures of the adequacy of program benefits in fulfilling program objectives. This ability to address a more comprehensive set of program questions using SIPP should provide a basis for better informed policy decisions. B. BACKGROUND Prior to SIPP, the analysis of the Food Stamp Program (FSP) as well as other income transfers and human resource programs was limited by the lack of income data. The maiority of policy research relied on the March Income Supplement to the Current Population Survey (CPS) and the Integrated Quality Control System data base (IQCS), although some studies used the smaller Michigan Panel Study of Income Dynamics (PSID). March CPS. The March CPS, also known as the Annual Demographic File, is an annual survey of a nationally representative sample of about 60,000 households and three times as many persons. The March survey collects a substantial amount of information for households, families, and individuals, including: o Employment status, occupation and industry, hours worked o Earnings, other income, program participation o Age, race, sex, marital status, ethnicity, education. However, the income data available from the CPS has several limitations since the primary focus of the survey is to obtain estimates of the size and characteristics of the labor force. 2 First, the income measures from the CPS are for the previous calendar year (i.e., the March 1985 CPS asked questions about income received in 1984). This can lead to severe recall problems for the survey respondents. Second, the survey asks about income received from groups of income sources, where the income groups are not necessarily the combinations needed in addressing policy issues. Further, there is little redundancy in the income questions, reducing the extent to which the accuracy and completeness of the data can be checked. Finally, the CPS is not truly a longitudinal file since households are not followed for long periods of time, rather, the CPS is a series of successive cross-sectional files. IQCS. The IQCS is an administrative file used to estimate the amount of food stamp and Aid to Families with Dependent Children (AFDC) benefits issued in error on a state by state basis. The data is obtained from reviews of a sample of 150 to 1,200 cases (i.e., program participant households) in each state over a six month period on a continuing basis. The full sample consists of approximately 45,000 cases. As a byproduct, the IQCS provides a source of detailed monthly data on the characteristics of participants in the FSP and AFDC, including: o Employment status, work registration o Earnings, other income, program participation o Assets; shelter, medical care and dependent care expenses o Age, race, sex, marital status, ethnicity. Although the IQCS provides all the detailed information needed to determine the household's eligibility for food stamps, the file is limited to food stamp and AFDC participant hous~holds. Consequently, issues dealing with the household's decision to participate in the FSP cannot be effectively addressed using the IQCS. Furthermore, since the IQCS is not a longitudinal file it is not possible to examine the dynamics of household participation in the FSP. PSID. The PSID, like the CPS, is an annual survey of a nationally representative sample. However, the PSID is a truly longitudinal data base with a primary focus on providing information on the income of families, particularly low-income families. Begun in 1968 with a sample of 5,000 families, the PSID provides annual information for families and individuals which includes: o Employment status, occupation and industry, hours worked .-· · o Earnings, other income, program pat;:_ticipation o Assets o Age, race, sex, marital status, ethnicity, education. While the PSID is a longitudinal file covering a long time period, the sample of households is relatively small compared to the March CPS or IQCS. SIPP. While the CPS, IQCS, and, to a lesser extent, the PSID were the only data bases capable of supporting a large portion of policy analyses, as the policy questions asked became more detailed and sophisticated, major weaknesses in the use of these data for policy analysis became apparent: o The March CPS and the PSID measure economic data and household composition on an annual basis, whereas eligibility and benefit calculations for most federal programs are based on a monthly accounting period. o Asset holdings and liabilities are not measured in the CPS and are measured on an annual basis in the PSID. o Complicated program rules and insufficiently detailed data in the March CPS and the PSID makes the identification of households eligible for program assistance difficult. o The IQCS contains information for program participant households only. o The information on participation in other, non-food stamp, programs is limited in the March CPS, the PSID, and the IQCS. o It is known from comparison with program records that the CPS data underestimate transfer income, retirement and disability income, unemployment compensation, and property income. The inability of the March CPS, the PSID, and the IQCS to provide the income and household composition data needed in addressing the more complex policy questions led to increasing appreciation of the need for more detailed and more frequent information on assets, income flows, noncash transfer income, household composition, and participation in government programs for both program participant and nonparticipant families, households, and persons. Although the CPS could potentially have been expanded to fill this need, rather than contemplating the major changes that would have been necessary in the survey instrument and procedures of the CPS--for which collecting income information is in any case a secondary goal--the decision was made to design a major new Survey of Income and Program Participation. 3 4 The primary advantages of SIPP over previous data bases are: o The monthly reference period allows need and eligibility to be assessed on the same accounting period as is used in program administration. o The longitudinal data allows questions related to the dynamics of income receipt and program participation to be addressed. o The richness of detail of the data allows a variety of issues, such as participation in multiple assistance programs, to be examined. o The potential matching of SIPP with administrative data would allow a wider variety of administrative issues to be addressed. The promise of SIPP has already been demonstrated in a series of research studies on the Food Stamp Program based on the test survey for SIPP. The 1979 test panel, known as the Income Survey Development Program (ISDP) Research Panel, is very similar in content and structure to SIPP. The results from the set of food stamp studies based on the ISDP data, as presented in the body of this report, provide important new information on the economic status and behavior of FSP recipient households.~/ However, there are serious difficulties to be surmounted which were indicated by the ISDP experience and are becoming increasingly clear as SIPP starts to be used. Whereas before SIPP the call was for more and better data, it is now becoming clear that data collection efforts of this size and complexity present analytical, data preparation and data management problems that are orders of magnitude more difficult than any faced before. Discussion has turned from problems of inadequate data to problems of how best to use the complex data. The experience in using the ISDP data on the FSP research studies reported in this paper indicate ways that the problems of data preparation and data management as well as the conceptual problems of longitudinal data can be handled. A key ~/In addition to the research reported here, the ISDP data proved very useful for providing rapid information to FNS on very specific policy issues. For example, the ISDP was used to examine the adequacy of the earned income and child care deductions in the food stamp benefit formula. SIPP should be useful in this context as well. rationale for this monograph is the hope that by sharing these ISDP experiences prospective SIPP research will benefit from the wide professional critique. C. OBJECTIVES OF THIS MONOGRAPH There are two basic objectives of this monograph. The first is to report the findings from a number of studies of FSP households all of which were based on the ISDP test panel. These studies are important in that they represent the best current information on questions such as how different asset limits would impact FSP eligibility or how FSP program exit rates are associated with factors such as finding a job. However, they are preliminary as more definitive research can soon be conducted based on the larger samples, longer time period, and more tested procedures of SIPP as well as on the more refined analytical procedures that are emerging from the earlier efforts. The second objective is to provide a basis for learning from the ISDP experience. The research reported here can provide important insights into the complexities which arise in the use of longitudinal data files such as the ISDP and SIPP. Maximizing the benefits of the ISDP-based work is contingent upon the sharing of the approaches used, results obtained, and limitations of this research with both the policymaking and research communities. D. ORGANIZATION OF THIS MONOGRAPH This monograph is divided into two major sections. The first section provides an overview of the SIPP and ISDP surveys (Chapter II) and briefly summarizes the existing food stamp research which uses the ISDP test panel--Chapter III presents research on the economic status of the food stamp population, Chapter IV summarizes studies on the dynamics of food stamp participation, Chapters V summarizes research on the monthly patterns of income receipt, and Chapter VI present studies dealing with program participation and labor supply behavior of food stamp households. The second section of the monograph presents a discussion of the analytic and data issues which arise in research using panel data (Chapter VII). The final chapter summarizes the ISDP research and the promise of SIPP. ~ · ' 6 II. THE SIPP AND ISDP SURVEYS A. THE HISTORY OF SIPP A longitudinal survey of the scope of SIPP had never been designed or fielded on a nationally-representative basis. As a result, a major development effort--called the Income Survey Development Program (ISDP)--was authorized in 1975 by the Office of the Secretary of the U.S. Department of Health and Human Services (DHHS) to precede the fielding of SIPP. The ISDP program sponsored extensive research into ways in which the measurement, collection, and processing of income, transfer program, and wealth data could be improved. It also undertook a series of test panels to try out alternative data collection and processing methods, the last of which (the 1979 ISDP Research Panel) was sufficiently large to provide reliable national estimates of many individual and household characteristics. The intent was to fully exploit the experience of the 1979 ISDP Research Panel in order to produce an optimal strategy for fielding the first and later waves of SIPP. Subsequent events prevented this intent from being fully realized. The ISDP program was halted in 1982. Recognizing that the 1979 ISDP test panel was a valuable resource in itself, several government agencies, including the U.S. Department of Agriculture, Food and Nutrition Service (FNS), funded the work necessary to release public use versions of the cross-section files. In addition, the FNS funded the longitudinal linking of the data across waves, a data access system, and the FSP research discussed in this paper. In the midst of the efforts to render the data usable for methodological and public policy research, the Census Bureau obtained the funding needed to conduct the full SIPP. Faced with stringent deadlines for fielding the initial wave, the Census Bureau relied heavily on the ISDP in developing questionnaires, data collection strategies, and processing systems. As a result, SIPP is very similar to the 1979 ISDP test panel-sharing several of the characteristics that led to the complexities in the ISDP data processing five years ago. The ISDP experience and associated research has thus acquired unique importance as a way to learn not only about the promise but also the potential pitfalls of using the SIPP data. B. THE 1979 ISDP RESEARCH PANEL The 1979 ISDP Research Panel consists of a nationally representative samp~e of approximately 7,500 households.~/ The sample design oversampled both low and high income households in order to improve the precision of measurement at both ends of the income distribution. The individuals in the initial sample households were followed for approximately one and one-half years. The survey consisted of six three-month rounds (or waves) of interviewing using a technique called staggered interviewing. That is, the sample of households was divided into three groups of equal size (called rotation groups) and the survey was administered to the individuals in one group each month. The first rotation group received the Wave 1 interview during February 1979 with subsequent waves occurring every three months thereafter. Figure 11.1 illustrates the outcome of this survey technique. Note that the third rotation group, although surveyed every three months, was not administered the Wave 4 instrument during the fourth round of interviewing. Instead, they were administered the Wave 5 questionnaire with the Wave 6 survey following three months later. As a result, for one-third of the sample there are five rather than six waves of data. The survey instruments used in the 1979 ISDP Research Panel included five core modules and a series of supplemental modules which differed by wave. The five core modules were administered during each of the first five waves of interviewing and covered the following topics: o Household composition and characteristics, and person characteristics at the time of the interview o Recipiency of income from assistance programs in each of the three months prior to the interview o Earned income and employment status for each of the three months prior to the interview l/This is actually a subset of all households in the original design. The total sample initially included approximately 8,300 households from the area frame sample and 3,000 households from samples drawn from program records. The 7,500 households are those households from the area sample successfully interviewed in early 1979 and subsequently reinterviewed throughout 1979 and early 1980. 7 00 Reference lob nth 1978 Nov Dec 1979 Jen Feb Mar "pr May Jun Jul "ug Sep Oct Nov Dec 1980 Jen Feb Mer "pr Mey FIGlflE 11.1 SURVEY W"VES -- 1979 I SOP RESE ... RCH P"'£l l---------------------------------------------1979----------------------~-~~-~~~-~~~~------------l l---------------------198o-----------------l Feb Mllr "pr ""'~ Jun Jul "ug Sep Oct Nov Dec Jen Feb Mllr "pr Mil~ Jun ~ tJ = P11nel Rote"tlon Group 1 --> ,-----~ b • Penal Rote"tlon Q-oop 2 c • Panel Rotation Group 3 " b " b c II --- b c -------> " c " b -- " b c Ill ------5 core modules,---------- b c ------> " attitudes. c " b a b c I IV ----------------------------------~~~=:!::~:::~~--------------------------: ___ 1_______ : : ~------->1 c v ---------------------------------5e~~~~b~~e~;~~~~-------l__ b c , _____ ,C 5 core ooodu las, b d lsebll I ty, personal history (...,..rlage, jobs, educ., migration) ----------------------------------------- ----------------------attitudes, reasons tor---------------------------- nor>-pert lclpetlon In "FOe, Med lcald. Food Stamps, SSI • 5 core modu las non-lncCJIIeproducIng assets, 11 11 fecyc le" earnIngs, past- secondary educ., child cere, school breektest end school I unch. (2/3 sample) c a c a b " b b 5 core modules, net worth, pension cCN"erZ~ge. VI ------> 1-c----------- c a c " b " b b Annual lnc~e roundup, cepltel gains lind losses, ta>c:es, I n-kl nd I ncoooe (trlnge benetlts, services, Incl. WIC, Enar(jy "ss 1t). o Unearned income for each of the three months prior to the interview o Characteristics of children under age 16 at the time of the interview. Each of the supplemental modules was administered once during the survey, with Waves 2 through 5 containing at least one of the supplemental sets of questions. Figure I I.1 also summarizes the contents of each of the five waves. As implied above, the reference period for most of the data collected was the three calendar months preceding the interview month. Hence, as can be seen from Figure II.1, the calendar period covered for each wave varies by rotation group. Wave 2 was particularly important for FSP research. W"i th supplementary funding from FNS, in addition to the five core modules, as noted, extra questions were asked on Wave 2 on workrelated expenses, shelter costs, and asset holdings which allowed the identification of FSP eligibles as well as participants. This is the only source of nationally-representative integrated data on (1) income, (2) household composition, (3) work-related expenses relevant to allowable deductions against earned income, (4) asset information relevant to the resources test, and (5) shelter and other expenses allowable as deductions in computing FSP benefits. Such information does not exist for any other ISDP wave; current plans call for it to be collected in the course of SIPP. However, the SIPP data will be more difficult to use as it is scattered through different topical modules to be administered in different waves. The dispersion of the FSP eligibility information throughout the different topical modules of SIPP is the result of the competing needs of the various federal agencies and programs and the academic community. Given the limited time available within each topical module and across the set of modules, it is necessary to choose among questions which are intended to aid in program planning (e.g., questions concerning FSP eligibility) and questions which address important social issues (e.g., questions concerning retirement decisions or health). Although the 1979 ISDP permitted more comprehensive anrl extensive analyses of the FSP than had previously been possible, 1979 was not a typical year for the FSP. In that year, the structure of the FSP changed dramatically as the Food Stamp Act of 1977 was implemented. The provisions of that Act included a reduction of the net income eligibility standard, tighter restrictions on deductibles, an increase in the asset limit, and, most importantly, the elimination of the food stamp purchase requirement. Following the elimination of the purchase requirement and 9 the implementation of the other provisions of the Act, participation in the FSP increased rapidly, rising by about 3.4 million people over the following year. Given these changes, 1979 cannot be considered a representative year for the FSP. Furthermore, the more recent changes in the FSP due to the provisions of the Omnibus Reconciliation Act of 1981, the Food Stamp and Commodity Distribution Amendments of 1981, and the Food Stamp Act Amendments of 1980 and 1982, suggest that there is a need for FSP analyses using the more recent data of SIPP. C. THE STRUCTURE AND CONTENTS OF SIPP The structure of SIPP, as noted, bears a close resemblance to the structure of the 1979 ISDP test panel.2/ First, SIPP has a similar instrument design--which combines a set of core modules on levels of monthly economic well-being and changes in these levels over time with a set of topical modules which vary by wave. Some of these are fixed and assigned to particular waves; others are variable and designed in response to particular government agency program and policy analysis requirements. Second, SIPP has a similar wave, rotation group, and reference period structure, although spread of over a four-month rather than a three-month period. Thus, each SIPP wave has four rotation groups, which are interviewed (in rotation) every four months. For each interview month the reference period is the previous four months (that is, four months ago, three months ago, two months ago, and last month). The sample of households is larger and the period of time covered is longer for SIPP than for the 1979 ISDP test panel. Through 1984, about 20,000 households were interviewed under SIPP, 5,000 each month. In 1985, a second panel of households was introduced bringing the total sample size of SIPP up to about 35,000 households. Each individual in the original sample frame will be interviewed every four months for a period of two and one-half years. Appendix Figure A shows the content and reference periods of SIPP for 1984, 1985, and most of 1986. 2/This description draws heavily on Roger A. Herriot and Daniel Kasprzyk, The Survey of Income and Program Participation. SIPP Working Paper Series no. 8405. Washington, DC: U.S. Bureau of the Census, 1984. '. ~ III. ECONOMIC STATUS OF THE FOOD STAMP TARGET POPULATION Included in the ISDP based research on the economic status of food stamp research are studies 'Yhich address the asset holdings of low-income and food stamp households (Bickel and MacDonald, 1981), the adequacy of benefits from food stamps in conjunction with other income maintenance programs (MacDonald, 1983, 1984), and the pattern and adequacy of multiple program benefits over time (Doyle, 1985; MacDonald, 1985). In this chapter, the studies dealing with each of these topics are summarized in turn. A. ASSET HOLDINGS OF FOOD STAMP PARTICIPANT AND NONPARTICIPANT HOUSEHOLDS Eligibility for food stamps is restricted to households which, in addition to meeting other requirements, have assets below a statutory limit. Such rules are designed to limit participation in the FSP to households without holdings of assets which can be readily liquidated and are therefore potentially available for the purchase of food. In 1980, Congress set the limit at $1,500 for most households; households of two or more, at least one of whom is age 60 or older, are allowed up to $3,000 in assets.~/ Liquid assets, such as bank accounts, stocks, and bonds, are counted toward the limit. In addition, a portion of the value of some vehicles is counted. A homeowner's house and lot are not counted~ Certain other nonliquid or employment-related assets, such as personal effects and tools of a trade, are also excluded. Before the 1979 ISDP test panel there were no nationally representative data on assets available in enough detail to assess the dimensions of current asset holdings by low-income households. Wave 2 of the 1979 ISDP test panel provided the necessary data for such an assessment. The report by Bickel and MacDonald presents findings from this data base on the types and value of assets held by FSP participants and several categories of nonparticipants.l/ In addition, in order to assess the sensitivity of food stamp eligibility to the resource test, Bickel and MacDonald simulate the impact of altering the FSP asset limits. ~/Although the $1,500 limit was increased briefly to $1,750 under the Food Stamp Act of 1977, the $1,500 and $3,000 asset limits were first established in 1971. 2/This section summaries Gary Bickel and Maurice MacDonald, Assets of Low Income Households: New Findings on Food Stamp Participants and Nonparticipants. Washington, DC: Food and Nutrition Service, U.S. Department of Agriculture, 1981. 11 Data and Methodology Findings 12 The Bickel and MacDonald work is of particular interest for two reasons. First, Wave 2 of the 1979 ISDP test panel contains more detailed information on assets than any contemplated wave of SIPP. Second, in contrast to SIPP, the 1979 ISDP Wave 2 contains the integrated assets and expense information necessary for accurate determination of the pool of program-eligible units. Wave 2 of the 1979 ISDP test panel provided information on assets of all household members, including savings accounts checking accounts, cash on hand, certificates of deposit, stocks, bond, cars, trucks, other vehicles, homes, insurance policies, all types of income-producing properties (rental, commer~ial, industrial, farm), undeveloped land, and other assets such as trusts, estates, and mortgages. Certain types of assets were not included in the data collected. However, for most households, pension and retirement funds were the only potentially significant omission. Additionally, sufficient other information on households' incomes and characteristics was collected in order to distinguish between households eligible for food stamps, whether or not they actually participated, and households not eligible. The sample used for this study included about 7,200 households. The analysis focused primarily on the actual food stamp eligibility unit or "food stamp household" within each sample household. All households were classified first according to the FSP income eligibility test and then the resource test. They were further classified into participants and nonparticipants according to whether they had reported receiving food stamps in any of the previous three months. (Three months was used to provide a broader measure of participants, given the relatively rapid entry and exit rates for the food stamp caseload.) Food stamp participants and three nonparticipant groups were defined to permit comparisons of asset holdings. The FSP nonparticipant groups were: 1. FSP eligible nonparticipant households 2. Low-income households which were ineligible for food stamps on the basis of their assets 3. Households which were ineligible for food stamps on all other grounds. The limits on assets have a major impact on FSP eligibility. Over twelve million persons who otherwise would have been eligible for food stamps had assets high enough to exclude them from eligibility. Based on this finding, Bickel and MacDonald estimated the savings in potential benefits to be $2.-9 billion in 1981 (19R1 dollars). Several other general patterns in the asset holdings of FSP participant households and others are noteworthy: o Nearly one-half (49 percent) of FSP participant households had no countable assets at all, while 91 percent had countable assets of $500 or less. o FSP participant households had far fewer assets than households not eligible for the program. When all assets except homes were considered, 37 percent of participant households had no assets, and 90 percent had less than $1,500. By comparison, 3 percent of FSP ineligible households had no assets, and only 21 percent had less than $1,500. o FSP participants tended to have few liquid assets. Fifty-one percent of the participant households had no liquid assets, 93 percent had $500 or less. There were no participant households in the sample with stocks or bonds. o FSP participant households had few assets in any other specific asset categories. Approximately one-half (51 percent) had no car, only 36 percent owned homes, 9 percent had life insurance policies, and 2 percent had rental property. No FSP participant in the sample reported any farm or business interests, undeveloped land, mortgages, royalties, estates, or trusts. Ry contrast, 87 percent of FSP households ineligible on the basis of either assets or income owned a car, 70 percent owned their homes, and 43 percent owned life insurance. o Among households eligible for food stamps, those who actually participated tended to have slightly fewer assets than those who did not. Thirty-seven percent of FSP participants, compared to 30 percent of FSP eligible nonparticipants, had no assets (excepting homes). o Households disqualified from food stamps solely on the basis of their assets were a relatively well-off group in terms of the types and value of assets they owned. Two-thirds (68 percent) of these households had total assets (not including homes) in excess of $5,000. Since the data for this report were collected in the Spring of 1979, both the income and asset limits in the FSP have changed. Thus, the figures in this report may not represent the current holdings of food stamps household. More recent data from SIPP is needed in order to obtain up-to-date information on the impact of the asset limit on FSP eligibility. 13 Data and Methodology 14 B. THE PATTERN AND ADEQUACY OF MULTIPLE PROGRAM BENEFITS The debate about whether and/or how to limit federal expenditures for income transfer programs has again focused attention on questions about the adequacy and equity of benefits available from the mix of social insurance, cash welfare, and in-kind programs that help maintain income security. With this debate comes disagreement about tradeoffs hetween the extent of income security provided and its costs. It has been argued that the gradual expansion and extension of income transfer programs has made multiple benefits available that provide more than is necessary for adequate support, and that it has reduced the incentive to work. Opponents reply that without multiple benefits there would be gaps in program coverage, and that the current system does not adequately provide for all.3/ Once again, this issue could not he properly examined before the 1979 ISDP test panel became available s:i:nce the ISDP provided, for the first time, concurrent information on monthly participation in a variety of income transfer programs, along with income information for the same period. The work by MacDonald on multiple program participation and the adequacy of benefits approached the issue by examining the relationship between income and poverty status4/ before transfers and the relationship including transfer income for recipients of all major income support programs. The paper includes comprehensive descriptions of the incidence of multiple ]/This section summarizes Maurice MacDonald, Multiple Benefits and Income Adequacy for Food Stamp Participant and Nonparticipant Households. Washington, DC: Mathematic Policy Research, February 1983. Prepared for the U.S. Department of Agriculture, Food and Nutrition Service. Summary results fro~ an extension of the analysis to include the impacts of nutrition and housing benefits are also included. See Maurice MacDonald, Multiple Benefits and Income Adequacy: Impacts of Nutrition and Housing Benefits. Madison, WI: Institute for Research on Poverty, August 1984. ~/The official Office of Management and Budget (OMB) 1979 annual poverty line was used as the household con ~nmption need standard in evaluating the adequacy of benefits. Three month incomes below one-quarter of that level were classified as inadequate to meet the households needs. (For more information on the official poverty threshold and for poverty statistics see U.S. Department of Commerce, Bureau of the Census, Current Population Reports, Series P-60.) benefit recipiency, the effects of benefits on poverty status, and the characteristics of recipient and nonrecipient households for different programs. It then focuses on the effectiveness of the FSP in alleviating poverty within the context of multiple benefits. The research reported here used Wave 2 of the 1979 ISDP test panel to examine the actual reported receipt of multiple benefits. It analyzes the incidence and effects of receiving benefits from one or more of the six most important cash and inkind transfer programs: Food Stamps (FSP); Public Assistance (PA) including Aid to Families with Dependent Children and other welfare programs; Unemployment Insurance (UI); Old Age, Survivors, and Disability Insurance (OASDI); Supplemental Security Income (SSI); and Medicaid. The combined effect of a somewhat larger list of transfer programs on the adequacy of benefits as measured by national poverty guidelines is also examined. In estimating the effects of transfers on benefit adequacy, transfers were added sequentially to market (pretransfer) income in the same order that these programs count income from other programs in determining benefit entitlements: social insurance (OASDI and UI) first, then cash welfare (PA and SSI), followed by food stamps (FSP).2_/ (Medicaid was omitted from the income count since it is seen as meeting special rather than normal income requirements.) The sample used for this study included about 7,200 households. Figure 11I.1 displays categories of households, to explain how the analysis of multiple benefit recipiency was organized. As can be seen, the primary focus is on households that were eligible for food stamps in Spring 1979. Sufficient information on income types, assets, and relevant expenditures was collected by ISDP to classify all households into food stamp eligible or ineligible groups • .2/There is a substantial literature on the evaluation of in-kind benefits (e.g., Smeeding, 1982, 1984; Manser, 1981) which proposes several different methods: market value, recipient or cash equivalent value, government cost, and poverty budget share. Government cost, since it includes administrative costs, would be likely to yield the highest valuation, while the poverty budget share would yield the lowest since it places a limit on the value of in-kind benefits. The approach used by MacDonald to value food stamp benefits was market value. 1'5 16 FIGURE II I. 1 CATEGORIES OF HOUSEHOLDS (IN MILLIONS) FOR MULTIPLE BENEFITS ANALYSIS, SPRING 1979 All Households ~~----------------- (80.6) ------------------ Food Stamp-Eligible (12.4) Food Stamp Ineligible (68.2) Food Stamp Participants Food Stamp Nonparticipants Poor (3.8) Not Poor (4.8) (64.4) Poor ( 3.3) SOURCE: I Not Poor (1.5) (7 .6) Poor (5.5) I Not Poor (2.1) MacDonald, M. Multiple Benefits and Income Adequacy for Food Stamp Participant and Nonparticipant Households. Washington, D.C.: Mathematica Policy Research, February 1983. Findings In Spring 1979, there were 28 million households who reported receiving benefits from one or more of the six major assistance programs studied. Less than one-quarter of these had benefits from two or more major programs. The other three-fourths received benefits from only one program. A closer look shows that about two-thirds of all households reporting benefits from any of the major programs received only OASDI. The principal findings of the analysis of the receipt of multiple benefits were: o Social insurance recipients (recipients of OASDI and UI) frequently do not have multiple benefits and are usually ineligible for welfare. Only 17 percent of OASDI recipient households and 34 percent of UI recipient households received benefits from other programs. o In contrast, multiple ben_efits were widespread among food stamp, Medicaid, and cash welfare recipients. In each of these programs, over 80 percent of recipients obtained benefits from at least one other program. About two-thirds of food stamp participant households had three or more of the six major benefits (most frequently, food stamps, Public Assistance, and Medicaid). o Many households below the poverty line received few or no benefits. Of pretransfer poor households who remained poor when all transfer income was counted, 38 percent did not receive any of the six major benefits, 32 percent received only one henefit, and 27 percent received food stamps only. Based on the analysis of the adequacy of the program benefits it was found that receipt of multiple program benefits did not lead to high income. The observed effects of multiple benefits on benefit adequacy include the following: o Of all pretransfer poor households, over one-half remained poor, 40 percent were removed from poverty by OASDI and/or UI, cash welfare and food stamps together removed another 5 percent. o Food stamps alone reduced poverty by 1 percent. If all eligible households had participated in the FSP, pretransfer poverty would have been reduced by 2 percent. o Among the pretransfer poor households who received food stamps, most (80 percent) remained poor after all transfers were counted. Of the 14 percent of food stamp households with money incomes below one-half of the poverty line, less than one-third still had incomes below that level after food stamp benefits were counted. , ., 18 Perhaps a more informative way of examining multiple benefit effects is to look at actual dollars. The dollar difference between each household's quarterly income and the quarterly equivalent of the poverty threshold for that household when added together for all poor households is defined as the poverty income gap. In general, all three major types of transfers (i.e., social insurance, cash welfare, and food stamps) had greater percentage effects in reducing the poverty gap than they did in reducing poverty counts. This supports the finding that most of the effect of transfers was to help those with incomes below onehalf of poverty. Among food stamp recipients, social insurance appeared less effective than did cash welfare in reducing the poverty income gap. In particular, it was found that: o Among all households nationwide, social insurance closed over 50 percent of the Spring 1979 $20 billion poverty income gap, cash welfare reduced it by 11 percent, and food stamps reduced it by another 4 percent. After all benefits had been considered, one-third of the original gap between pretransfer income and the poverty line remained. o Among FSP participant households, social insurance reduced the Spring 1979 gap between the pretransfer incomes of FSP participants in poverty and the poverty line by 23 percent, cash welfare reduced it about one-third, and food stamps closed about one-fifth. After all benefits had been considered, about one-quarter of the gap between incomes and the poverty line remained. o In contrast, for FSP eligible households who did not participate, over one-half the poverty income gap remained after cash benefits were added. Figure III.2 summarizes the results of the multiple benefit analysis by showing the Spring 1979 distribution of FSP eligible households relative to the official poverty line as they appeared (1) before government transfers, (2) after major cash transfers, and ( 3) after major cash transfers and food stamp transfers. Before government benefits were considered, 80 percent of FSP participants fell below national income poverty guidelines. This included 64 percent below one-half of the poverty line and an additional 16 percent between 50 and 100 percent of the poverty line. After all major transfer programs were considered, 64 percent of FSP participant still remained below income poverty guidelines, including 4 percent of FSP participants who remained below one-half of poverty. It is important to note once again that the FSP has changed considerably since Spring 1979. Cons~q.uently, more recent data from SIPP is needed to reflect the current adequacy of program benefits. Percent of 30 Households FIGURE III.2 EFFECT OF MULTIPLE BENEFITS ON INCOME AS A PERCENT OF THE POVERTY LINE, 1979 F'ood Stomp Porticlpant.s ~._----~r-----------T-----------~------ Pretransfer Income Percent of 3D Households With Welfare and Social Insurance With Welfare, Social Insurance, a!'ld Food Stamps F'ood Stamp EJI;ible Nonpartidpants ~----------,-----------------T---------- Pretransfer Income With Welfare and Social Insurance (but ~ Food Stamps) Household Income as a Percent of the Poverty Line ~over 130% ~ 100-129% §§ 50-99% ~Under 49% 19 20 A later supplementary analysis of multiple benefit receipt used Wave 4 of the 1979 ISDP test panel to include the School Lunch . Program and housing subsidies and Wave 6 to include the Special Supplemental Food Program for Women, Infants, and Children (WIC) (MacDonald, 1984). While this analysis is not strictly comparable to the main analysis primarily because using Waves 4 and 6 necessitated focusing on pretransfer poor households rather than FSP eligible household, this difference is not a major limitation on the analysis since the pretransfer poor define closely the subset of households who are eligible for most of the benefits of interest. The principal findings of this analysis were: o About 77 percent of both FSP and school lunch recipients reported receiving benefits from another program--compared with 88 percent for cash welfare. o When only cash benefits are counted, 11.2 percent of households were poor in late 1979 (the time of Wave 4). Adding the market value of food stamps, school lunches, and housing subsidies reduced the percent poor by 1.4 percentage points._§/ o The WIC data from Wave 6 indicate that approximately 1 percent of all households and 3 percent of all pretransfer poor households reported receiving WIC. Of the pretransfer poor WIC recipients, 38 percent reported receiving food stamps in Wave 4 and 62 percent reported both food stamps and school lunches. There were no households with school lunch only in Wave 4 who reported receiving WIC in Wave 6. C. MULTIPLE PROGRAM BENEFITS DURING THE COURSE OF A YEAR The studies presented in the previous section considered the receipt of benefits from multiple assistance programs at any point in in the interview wave. By using the five waves of the ISDP test panel, the patterns of transitions between multiple benefit categories over the year (Doyle, 1985) and the receipt ~/In estimating the value of benefits from food stamps, school lunch, and housing benefits, MacDonald ue ed the ·market value approach. Data and Methodology Findings of multiple benefits for the 12 months of calendar year 1979 (MacDonald, 1985) could be considered.?/ Both the study by Doyle and the study by MacDonald used a linked longitudinal file created from the five waves of the 1979 ISDP test panel. This longitudinal file was created for the analysis of the dynamics of FSP participation and is discussed in the context of that work in the next chapter. By using the linked longitudinal file, the patterns of the receipt of benefits by households from multiple assistance programs over the full 12 months of 1979 could be examined. The five assistance programs considered in the studies were: Old Age, Survivors, and Disability Insurance (OASDI), Unemployment Insurance (UI); Supplemental Security Income (SSI); Aid to Families with Dependent Children (AFDC); and Food Stamps (FSP). Eleven program participation categories were defined. When only one of the five ma.ior programs was reported by the household for a month the benefit type was simply that benefit. When more than one benefit was reported, the household was classified by whether or not it received food stamps as well as by the largest benefit amount from the other cash programs. The residual group for each month consisted of households without benefits from any of the five programs analyzed plus recipients of one very infrequent benefit combination of two major benefit programs (AFDC and UI). Doyle combined several multiple assistance categories because of small sample sizes. On average, there were 3,174 households per month in the sample for the Doyle study and 3,205 for the MacDonald study. The study by Doyle examined the extent to which households switched between nine multiple benefit categories and the nature of those transitions. The study found that one-third of the sample of households reported some change in the multiple program category in which they participated during 1979 and onefifth reported three or more transitions during the year. Other findings of interest include: o Households receiving OASDI only were the most stable group, with only 18 percent reporting transitions over the year. UI L/This section summarizes MacDonald, Maurice Serial Multiple Benefits and Monthly Income Adequacy. Washington, D.C.: Mathematica Policy Research, March 1985. Prepared for the U.S. Department of Agriculture, Food and Nutrition Service and a memorandum from Pat Doyle of Mathematica Policy Research the U.S. Department of Agriculture, Food and Nutrition Service. 21 22 recipients were the least stable, with all recipient households reporting at least one transition. o Of those households receiving food stamps, the most stable households were those also receiving OASDI, with almost three-fourths reporting no transitions over the year. Households receiving food stamps only or food stamps and SSI were the least stable. Over three-fourths of both groups of households reported at least one transition. The final food stamp group, households receiving food stamps and AFDC, was somewhat more stable, with 40 percent of all households reporting no transition. o The most frequent transitions for households receiving food stamps only or food stamps in conjunction with some other program were to multiple program categories which included food stamps. For example, over 70 percent of the households receiving food stamps only which 'experienced a transition added participation in OASDI, SSI, or AFDC to their food stamp benefits. o Of those households not participating in any assistance program in January, over one-fourth had begun participating in at least one program by December. For 5 percent of the households, the program they entered was the FSP. o The number of months between transitions is short ( t to 3 months) for households with two or more transitions and approximately 90 percent of such households return to their original multiple benefit category. o For those households observed to have only one transition approximately 50 percent had no benefits in January but began receiving benefits during the course of the year, 25 percent were participating in one program for the full year and supplemented that program at some time during the yeEr, and 20 percent simply discontinued benefits at some point in the year. MacDonald examined average monthly participation rates and average length of participation for eleven multiple benefit categories and the effects of the different benefit combinations on the adequacy of the benefits. The principal findings of the MacDonald study were as follow: o Among those who received any benefits from the five transfer programs for the whole of 1979, those with only OASDI or only Ul were by far the most frequent beneficiaries. Households receiving only food stamps were third in frequency at 7 percent of all 1979 recipients. About 30 percent of those who received benefits from any program received them from two or more programs at some time during 1979, most of whom had benefits from only two programs. o The adequacy of different monthly benefit combinations was evaluated by comparing incomes with and without the cash benefits and the face value of any food stamps they received to a monthly poverty line determined by the net income limit for food stamp eligibility in July 1979.~/ By that standard, sample households would have been in poverty for 35 percent of the months if they had not received assistance. The benefits they actually received were sufficient to reduce the percentage of 1979 months spent in poverty to 25 percent. o Whatever their income without the monthly benefits, receiving any single one of the cash benefits was always more advantageous for the household than receiving food stamps alone. Yet monthly food stamp benefits were sufficient to move about 30 percent of those recipients who would have been below onehalf of the monthly poverty line above that standard. ~/The monthly net income limit for the FSP is less than the official annual poverty line divided by twelve. Data and Methodology 24 ; ' ' IV. THE DYNAMICS OF FOOD STAMP RECIPIENCY The analysis of program turnover is crucial, among other reasons, in order to shed light on the major issues of long-term dependence on public transfers--how long do households typically remain recipients? What are the turnover patterns of different types of households? What factors affect turnover? The ISDP 1979 test panel provided the first nationally representative data with which monthly program caseload turnover rates could be analyzed. Such a task required monthly income and related information necessary to determine eligibility, detailed household characteristics, and monthly observation of program participation status. There are two studies which make full use of the longitudinal nature of the ISDP in looking at turnover in the FSP. The first study (Carr, Doyle, and Lubitz, 1984) generated overall estimates of turnover rates in food stamp participation and compared them with estimates generated from earlier, less representative data. In addition, separate estimates of turnover for important subgroups within the lowincome population were calculated and changes in eligibility status were estimated over time. The wide variation found by Carr et al. in food stamp turnover rates by type of household makes it important to identify what kinds of events cause households to enter and leave the food stamp caseload. The second turnover study (Lubitz and Carr, 1985) pursued this question by estimating the impact of particular events on the probabilities of entrance to and exit from the FSP. The ISDP data made such an analysis possible because it provided the month-to-month linked information needed to identify changes in household circumstances and to associate those changes with program entry and exit. The data base from which estimates of turnover in the FSP were derived was a linked longitudinal household file develoyed from the first five waves of the 1979 ISDP test panel~ Sociodemographic variables such as household composition, ethnicity, and education were included as well as indicators of food stamp receipt and all components of the eligibility determination process. For the purpose of the turnover studies, longitudinal household units were constructed according to the status of the principal person(s) in the unit and complete income data from all five ~/This section is primarily a summary of Timothy J. Carr, Pat Doyle, and Irene Smith Lubitz, Turnover in Food Stamp Participation: A Preliminary Analysis. Washington, DC: Mathematica Policy Research, July 1984. Prepared for the u.s. Department of Agriculture, Food and Nutrition Service. ISDP waves were used. Principal persons are generally the "reference person" in the terminology of the survey, and his or her spouse, if any. (They may be thought of as the heads of the household.) As a result of this approach, the composition and characteristics of units were allowed to change over time. For example, it is possible to observe an ongoing unit which with the passage of time gains or loses a household head or other member. For the purposes of the present analyses, the universe of households consisted of all units headed by a primary sample member. A primary sample member is an individual included in the initial sampling frame, and therefore ·followed throughout the survey to the extent possible. A household could be formed during any month and dissolved during any month. The only sample households excluded from the studies were units formed after the initial interview which were headed by individuals who were not present in Wave 1~ About 7,000 sample household units were retained in the analysis. The analysis in the study by Carr et al. was carried out in two phases~ First, a descriptive analysis provided an overview of the general level of turnover in food stamp eligibility and participation during 1979, the manner in which turnover varied over the course of the year, and the manner in which turnover varied across socioeconomic groups. Several indicators of turnover levels in program eligibility and participation were used. The most important of these were the entry rate and exit rate. For program participation, these are defined as the proportion of all households who did not receive food stamps in one month who were receiving food stamps in the next month and, similarly, the proportion of all households who did receive food stamps in one month who were not receiving them in the next month. Other measures of participation turnover used were the proportion of households that continuously received food stamps, the number of spells of food stamp participation during the sample period, and the average duration of food stamp participation. Analogous turnover measures were defined for program eligibility. Since such descriptive analysis cannot identify the independent effects of individual factors, the second phase of the analysis estimated a multivariate statistical model of program participation and eligibility. The probabilities of entering and exiting from the FSP (or to and from eligibility for the program) were estimated (using event history analysisl/) as functions of household characteristics expecterl to affect eligibility and participation. 2/see Tuma (1982) for a discussion of event history analysis. 25 Findings ; • . 26 There was substantial movement of households into and out of the FSP; the number of households who received benefits from the program over the course of a year was over 70 percent greater than the number who received benefits in any given month. Furthermore, there were significant variations in observed turnover across socioeconomic groups. Specific findings of interest include the following: o Of all households who received food stamps in a given month, 7.3 percent left the program within the next month. o Given that a household did not receive food stamps in a given month, there was an 0.53 percent probability that it would enter the program in the next month. o Of the households who were present in the sample for the full calendar year and reported receiving food stamps at any time, about one-third received food stamps for the entire year. o The probability that a household was a food stamp recipient household at least once in the course of the year was nearly twice (1.74) the probability that it participated in the program in a given month. o There were systematic variations in entry and exit rates across households. The lowest monthly exit probabilities were for households who received AFDC and/or other types of welfare, nonwhite households, households containing an elderly or disabled person, households in which no person was employed, households whose head has relatively little formal education, and households headed by a single person. o The highest probabilities of entrance into the program were for households who received AFDC, households headed by a single person with children, nonwhite households, large households, households in which no person was currently employed, households whose head had little formal education, and households in which an elderly or disabled person was present. The multivariate results tend to support the findings of the descriptive analysis. In particular, after controlling for other factors, entry rates were higher and exit rates lower for households with a nonwhite household head, no earner present, a single household head, elderly or disabled membe-rs, and AFDC recipients. The principal findings of the descriptive analysis of program eligibility are as follows: The Role of Changes in Household Circumstances '.~ o There appeared to be substantial turnover in food stamp eligibility. The probability that an eligible household became ineligible each month was about 17 percent, and the probability that a previously ineligible household became eligible was 6.3 percent. Both these probabilities are substantially higher than the corresponding probabilities of participation change. o The types of households with the highest propensity to become eligible for food stamps were those who received AFDC and other types of welfare, households headed by a single person with children, nonwhite households, households whose head had relatively little formal education, households in which no person was working, and households containing elderly or disabled persons. o The types of households with the lowest propensity to become ineligible were those who received AFDC and other types of welfare, those with a head over the age of 65, those with a single head, those with a single member, those in which no one was employed, and those containing a disabled person. Most of the relationships between characteristics and eligibility transitions observed in the descriptive analysis are maintained when other factors are held constant. In particular: o Households with elderly or disabled members, those on AFDC, and those who were nonwhite were all more likely to become eligible for the Food Stamp Program than otherwise similar households. o Single headed households with children were more likely to become eligible than were other households. o Households with elderly or disabled members, nonwhite households, single headed households, nonearners, and AFDC recipients were all more likely to remain eligible for food stamps than were other types of household. The subsequent study by Lubitz and Carr extended the earlier research on turnover by examining the role of changes in house hold circumstances on the likelihood that a household would enter or leave the FSP.2/ Whereas the initial study identified 2/This section summarizes Irene Smith Luhitz and Timothy J. Carr's Turnover in the Food Stamp Program in 1979: The Role of Trigger Events, February 1985. Washington, DC: Mathematica Policy Research. Prepared for the U.S. Department of Agriculture, Food and Nutrition Service. 27 in Food Stamp Turnover 28 ~·. characteristics of households that are correlated with higher or lower probabilities of transition in and out of the FSP, the subsequent study identified particular events that evidently precipitate changes in program status. For example, the findings of the Carr et al. study include a higher than average program entry rate for households with no earner present. The trigger events analysis estimated the probability of program entry in the months immediately following the loss of earnings. Five potential types of events were defined as trigger events (i.e., events reasonably likely to "cause" or precipitate a change): o Large changes in monthly pretransfer income between one month and the next.~/ Those cases in which the income change resulted in a change in eligibility status were also differentiated from those in which it did not. o A change in the number of employed household members from one month to the next. Multiple-earner were also differentiated from single-earner households. o A change in asset holdings, specifically a decline in assets as a possible trigger of program entrance. o Changes in receipt of Unemployment Insurance, either exhaustion of UI benefits (as an entry trigger) or beginning to receive benefits (as an exit trigger). o Household composition change, defined as a change from having one head to two heads, or vice versa. The methodology employed had two parts. First, a descriptive investigation of the relation between trigger events and transitions was undertaken to track households in the ISDP sample through the course of calendar year 1979 on a month-to-month basis. When a comparison of a household's circumstances in successive months indicated that a trigger event had occurred, the household was further tracked to determine whether a transition (entrance or exit) occurred. The probability that a household experienced a transition after a trigger event, and the ~/The effect of changes in total income on transitions into and out of the program was analyzed; however the analysis was not satisfactory because these changes confounded exogenous changes in the household's economic status (such as the loss of a job) with changes in transfer income brought on by that event. elapsed time between the two events, was then tabulated and analyzed. Second, a multivariate analysis of the effect of trigger events on entry and exit rates was used to isolate the independent efforts of different trigger events. This analysis (based on event history analysis) estimated transition probabilities as functions of explanatory variables in a manner similar to more familiar regression models. The emphasis was on isolating the separate effects of the different trigger ev~nts on transition rates. It is important to note that for many of the trigger events being considered the number of cases available for analysis is quite small. With unweighted sample sizes of less than 10 or 20 for households entering or exiting from the FSP following a trigger event, the analysis cannot support definitive conclusions about the impact of changes in household circumstances on program participation. The principal findings of the Lubitz and Carr study were as follow: o Trigger events were strongly correlated with the probability that a household experienced a transition (entry or exit). A household that experienced a trigger event was far more likely to experience a transition within six months than a household selected at random. o Most instances of entering and exiting from the program in response to trigger events occurred in the same month as the trigger event or shortly thereafter. o The event most likely to precipitate entry into the program was a decrease in the number of earners present in the household. Declines in pretransfer income resulted in households' becoming eligible to participate in the program. Household splitting, and exhaustion of UI benefits were also significant trigger events. o The events most likely to precipitate exit from the program were an increase in pretransfer income and an increase in the number of earners present in a household. Beginning to receive UI and becoming a couple (i.e., moving from one-head to twohead status) were also significant trigger events. o Changes in pretransfer income and in the number of earners were experienced by a large proportion of all households in the course of the year. To this extent, labor market phenomena are more important in explaining turnover than less frequent demographic phenomena such as changes in household composition. 29 30 ~ · . V. MONTHLY PATTERNS OF INCOME RECEIPT The FSP is a monthly program. Eligibility is based on income for a monthly accounting period and benefits are provided monthly. Therefore, survey data with monthly observations provide much more accurate information for describing FSP participating and nonparticipating households and for examining analytical issues concerning FSP design and operations. l.Ji th the monthly observations from SIPP, the household's income, receipt of other transfers, and household composition can be examined for the particular months during which the household reported receipt of FSP benefits. In contrast, the traditional data sources for FSP analyses such as the CPS and the PSID collect data about the past year. Since many low income households experience large income fluctuations from part year employment and earners leaving or entering the household unit, analyses of issues such as the incomes of FSP participants during their period of participation that are based on the CPS and PSID must rely on strong assumptions and are therefore imprecise. This study examined monthly patterns of income receipt and focused on the degree to which the assumptions about intrayear income flows used by analysts with CPS data were valid. One application, in which the assumptions concerning intrayear patterns of income are particularly important, is in the analysis of the effects of proposed program changes on program costs and the distribution of benefits. FNS and other agencies make extensive use of microsimulation models based largely on CPS data for the analysis of potential program changes. Ideally, the microsimulation model would use monthly data in estimating the impact.of program changes. However, prior to SIPP such data were not available and the data base most commonly used in microsimulation was and is the annual March CPS. Since the March CPS does not have information on intra-year income flows and household composition, assumptions about such intra-year patterns need to be made and built into the microsimulation models. The availability of the 1979 ISDP test panel provided the opportunity to test the assumptions underlying the microsimulation estimates which were obtained from the food stamp model used by FNS. The FNS food stamp model--Micro Analysis of Transfers to Households (MATH)--is a CPS-based microsimulation model. As it was originally developed, the model created variables which represent income received in a typical month by r~a ch individual surveyed using the following assumptions: o Annual earnings are assumed to be evenly distributed over the reported number of months worked. Data and Methodology o Annual unemployment compensation is assumed to have been received only during the periods of months looking for work or working part time. o Simulated annual public assistance is allocated according to periods of work and nonwork, but assumed to be a uniform amount per month within those two categories. o The types of income received during a simulation month are assumed predictable by current employment status. o Any unearned income in addition to public assistance and unemployment compensation (e.g., social security, alimony) is assumed to have been received evenly throughout the year. All these assumptions are obviously open to question. The descriptive analysis of patterns of monthly income receipt conducted by Doyle (1984) using the ISDP panel data sheds light on the appropriateness of these assumptions.J! Because the intent of the research summarized here was to improve the assumptions built into the microsimulation model operating on the CPS data bases, certain sources of variability, such as the number and type of jobs held during the year and changes in household composition, were not pursued. Attention thus focused on two sources of variation that can be incorporated into a CPSbased model: o The extent to which husbands and wives tended to work during the same or different periods over a year. o The extent to which types and amounts of unearned income varied over a year, in particular according to work/nonwork status. The data files were constructed from the set of individuals interviewed in Wave 5 with monthly retrospective data for calendar year 1979. Because Wave 5 interviews were conducted in the Spring of 1980, these files closely resembled the March CPS on which the food stamp simulation data bases are built. The labor force participation analysis focused on husband-wife families. The income patterns analysis focused on adults in poor or nearpoor households • .l/This chapter summaries Pat Doyle, An Analysis of Intra-Year Income Receipt. Washington, D.C.: Mathematica Policy Research, July 1984. Prepared for the u.s. Department of Agriculture, Food and Nutrition Service. 31 Findings 32 ~ ·. Due to the lack of longitudinal editing and imputation for nonresponse on income questions (discussed in Chapter. VII)--a potential source of bias in studies of income receipt over time--the study was restricted to households for which all income amounts were completely reported in each wave they were interviewed. The husband-wife data base included 3,283 families with complete data. The poor and near-poor data base included 9,383 individuals, 7,468 of whom were present in the ISDP sample for all 12 months of calendar year 1979. Individuals not present for a period were treated as receving no income for that period. In the construction of transfer program simulation models that operate on the March CPS, it has been assumed that husbands and wives who both reported working for part of the year tended to work during different parts of the year. This follows the theory (called the added worker effect) that when household heads are unemployed, secondary workers such as previously nonworking spouses enter the .iob market in order to maintain customary household income levels. Analysis of the ISDP data provides clear evidence that the added worker effect does not explain the labor force pattern of most husband-wife families. About 40 percent of couples had both spouses working; about 35 percent had the male working while the spouse did not; about 7 percent had the female working while the spouse did not. About 18 percent had neither spouse working. This should not be unexpected given that people are documented to marry persons similar to themselves along a wide range of personal characteristics: ethnic and religious backgrounds, age, level of formal schooling completed, and type of occupation. All these characteristics strongly affect the degree of an individual's attachment to the labor force and lead to a tendency for men with strong labor force attachments to marry women who also work for large portions of the year. In such cases there is necessarily substantial overlap in the periods they work. More detailed findings on labor force attachment of husband-wife families include the following: o The majority of elderly families (head over 65) had no labor force activity. In January, for example, 64 percent did not work; in December, the proportion had increased to 71 percent. o For white families 35 to 40 percent had both spouses working; 30 to 35 percent had only the sample reference person working; 7 to 9 percent had only the spouse working; about 20 percent had neither working. ..... o For nonwhite households there was more monthly fluctuation although no obvious trend over the year. In March, for example, 42 percent had both spouses working; 28 percent had only the reference person working; 26 percent had neither working. In June, 32 percent had both spouses employed; 35 percent had only the reference person working; 26 percent had neither working. o The upper bound estimate of the number of families where the spouse went into the labor force because the reference person lost his or her job was only 15 percent of the total number of families with at least one spouse working sometime during the year. The second component of -the analysis focused on the income patterns in poor and near-poor households. The major focus was on four types of unearned income: o Workmen's Compensation o Unemployment Compensation o Asset Income o Other (Nonwelfare) Unearned Income. Welfare income was given less attention because direct estimates of welfare income are not used in the CPS-based simulation model. Rather, welfare income is simulated from other data on the house-hold. ' On the CPS file, part-year workers can be distinguished from full-year workers and information is available on weeks worked and weeks looking for work. Beyond that, little can be inferred about individual patterns of income receipt within the calendar year. For this reason, the analysis of unearned income patterns from the ISDP summarized here focuses on the work/nonwork period distinction. The patterns of unearned income receipt are based on data on poor and near poor individuals: o Workmen's Compensation. The ISDP data indicate that about half the respondents who reported receiving workmen's compensation (53 percent) reported receipt during the work period. Of these, more than half (57 percent) reported receiving it only during the work period. These results should be interpreted cautiously, however, because the sample was very small; 71 individuals reported receiving it during the year, of which the vast majority (65) were part-year workers. o Unemployment Insurance (UI). According to the ISDP data, 17.6 percent of UI recipients in the sample (poor and near-poor) were full-year workers and 42 percent were part-year workers 33 34 ~·' reporting UI receipt while working. These seem unusually high proportions. However, examination of the extent of the overlap between work and UI suggests periods were ones of transition from work to nonwork or vice versa--a finding that is supported by the difference in UI benefit amounts according to work/nonwork status. Persons receiving benefits during both work and nonwork periods reported average receipts of $191 during the former and $331 during the latter. o Asset Income. More than half (55 percent) of the ISDP sample of individuals in poor/near-poor households who reported receiving asset income received it throughout the year. For the elderly (over age 60) the proportion was much higher (72 percent). For those who did receive asset income for only part of the year, the ISDP data cannot yield unbiased estimates due to the fact that lump sum recall amounts were evenly allocated over the relevant recall period (three or six months). The amounts of asset income for those who reported rece1v1ng both while working and while not working were very similar in the two periods ($33 and $32, respectively). o Other (Nonwelfare) Unearned Income. This category includes regular sources (such as social security, veterans benefits, and public and private pensions) and irregular miscellaneous income. The patterns of other unearned income receipt by the ISDP poor/near-poor population were very different for the elderly and the nonelderly. Over three-quarters of the elderly reported receiving such income continuously throughout the year versus one-quarter of the nonelderly. The patterns of other income receipt by the nonelderly varied by whether they were regular or irregular sources, it appears that twothirds of the recipients of regular income received it for more than a quarter of the year, half of which received it for the full 12-month period. On the other hand two-thirds of those receiving income from irregular sources received it for three months or less. Finally, for the elderly, whether they were working or not made no difference to the amount received ($325-$330 per month); but for the nonelderly the amount received was higher during periods of work ($452 per month) than periods of nonwork ($348 per month). Data and Methodology VI. PROGRAM PARTICIPATION AND LABOR SUPPLY BEHAVIOR OF FOOD STAMP HOUSEHOLDS Two studies based on ISDP cross-section data deal with the relationships between program participation and labor supply decisions: the work by Czajka (19R1) on determinants of participation in the FSP and the work by Fraker and Moffitt (1985) on the effects of food stamp benefits on the market labor of female heads of household. A. DETERMINANTS OF PARTICIPATION IN THE FOOD STAMP PROGRAM It has been well established that participation in the FSP among the eligible population has been and continues to be less than 100 percent. Policymakers, analysts, and other observers cite as explanations such factors as complex eligibility criteria, benefits being outweighed by costs for some households, the welfare stigma, or lack of program knowledge. But measuring the extent of and understanding the true reasons for nonparticipation depends on adequate data. Before ISDP (with the exception of local area studies) research was limited by the unavailability of subannual income data and any but the most basic assets data. The 1979 ISDP test panel--in particular Wave 2--was especially designed to yield more accurate and detailed income and assets data for the shorter time periods that better approximated those used in administering the FSP. It also made possible a close look at the characteristics of the eligible population and the economic, social, and demographic factors that differentiate food stamp participants from nonparticipants.Jj The analysis by Czajka used Wave 2 of the 1979 ISDP test panel and was done in terms of reference month--i.e., three months ago (month 1), two months ago (month 2), and one month ago (month 3). Reference months were used because the only calendar month common to the entire Wave 2 sample was April 1979. The study was also restricted to those households determined to be eligible: excluding those households reporting receipt of food stamp benefits who were seemingly ineligible. Since households in similar circumstances who did not receive food stamps could not be included among the eligibles, the inclusion of "seemingly ineligibles" among the recipients would yield biased estimates of the effects of household characteristics upon participation. _l/This section summarizes John L. Czajka, Determinants of Participation in the Food Stamp Program: Spring 1979. Washington, DC: Mathematica Policy Research, November 1981. Prepared for the U.S. Department of Agriculture, Food and Nutrition Service. 35 Findings 36 ~·. The analysis had two components. The first looked at food stamp recipiency simply as a function of prewelfare income and total cash income. The second employed two sets of regression equations to estimate the net effe.ct of economic circumstances on the likelihood of the food stamp eligible unit or "food stamp household" within each sample household receiving food stamps. One set of regression equations controlled for the separate effects of prewelfare income, the expected value of the food stamp allotment, employment income, whether unemployed, plus several demographic variables. The demographic variables were included as further measures of need, potential access to other economic resources, and an indirect measure of perceived stigma. The other set of regression equations controlled, in addition, for welfare receipt. The reason for the two formulations is the potential circularity of assuming welfare participation affects food stamp participation when in some cases the reverse may be true. The monthly income patterns reflected in the Wave 2 of the 1979 ISDP data provide strong evidence for the importance of income in determining the probability that an eligible household will participate in the FSP. Participation diminished significantly as a food stamp househpld's prewelfare income (measured relative to the poverty line) rose. The average net participation rate among households with prewelfare incomes as high as three-quarters of the poverty line or higher was only one-quarter what it was among households with incomes less than one-half the poverty line. Since the higher income households constituted more than 40 percent of all eligible households, their low participation depressed the overall participation rate substantially. The overall participation rate2/ ranged from 28 to 31 percent for the three reference months of Wave 2 of the ISDP. ~The participation rates found in the ISDP were substantially lower than the 61 percent found by Beebout (1981) for July 1981, using an administrative count of food stamp recipients and a microsimulation estimate of food stamp eligibles. Possible explanations for this discrepancy include: (1) the Beebout estimates are participation rates for persons while Czajka estimated food stamp household participation rates. Since large households have higher participation rates and weigh more heavily in the computation of a person level participation rate than a household rate, the person rate will tend to be higher; (2) Czajka eliminated seemingly ineligible households f rom his calculations, while Beebout's estimates include them in the numerator but not the denominator. The inclusion of the ineligibles in Beebout's calculations will lead to higher estimates of participation; and, (3) there appears to have been a significant amount of underreporting of food stamp recipiency in Wave 2 of the ISDP. Czajka (1982) discusses possible sources of this underreporting. There is a puzzle, however, in that the participation rate of the 365 households with zero gross cash income did not follow the general pattern. (Note that these were households with no welfare income.) These households--presumably the neediest segment of the population--accounted for 10 percent of all eligibles and had a food stamp participation rate of only 4 to 6 percent for the three reference months covered by the study. The most plausible explanation is that the group included a significant proportion of households that were in fact ineligible for the FSP but were retained in the sample because of the i~perfect replications of the FSP eligibility criteria. That is, some of the zero income households had effective incomes which were not zero and as a result were not eligible to receive food stamps. For example, the group might have included a large number of students living away fran home. It might have included persons whose prospective income was not zero (for_example, persons waiting for a job to begin or persons with relatively large assets--not large enough to disqualify them from food stamps, but enough to allow them to subsist without an income flow for a time). More generally, it might have included persons whose reported income and assets excluded important resources. It is interesting to note that even the extremely detailed monthly data on income and assets from Wave 2 the 1979 ISDP did not resolve the puzzle of these seemingly very poor nonrecipients. Other economic variables appear to be much less important influences on participation than income. Liquid asset holdings did lower the probability of participation--but only marginally. Moreover, too few households held liquid assets of any size for this to have much influence on the overall participation rate. Previous studies have shown welfare recipiency to be a very powerful predictor of food stamp use, and the present study replicates these earlier findings. At the same time, however, it is not clear what this result actually means. Food stamp recipiency may precede welfare recipiency in some instances. In others it may be decided .jointly with welfare. In either situation calling welfare recipiency a determinant of food stamp recipiency may greatly simplify reality. The strong association between welfare recipiency and food stamp recipiency needs to be better understood. Further research exploiting the longitudinal nature of SIPP may shed more light on this relationship in two ways. First, the actual sequencing of receipt of these two forms of transfer income over the two to two and one-half year followup period can be examined. The extent to which one comes before the other versus being received simultaneously would be very informative. Second, household decisions with regard to welfare and food stamps should be modeled jointly, since the household may well consider them simultaneously. 37 38 ., .. Among the sociodemographic variables examined, the number of children under age 16 exhibited the strongest overall relationship to food stamp participation. However, further research is required to identify the source(s) of this relationship: possibly the absolute level of benefits provided to large households or the frequent inability of such households to raise their incomes above the poverty level via the marketplace. A sizable differential in participation by level of educational attainment was apparent, with the less educated (fewer than six grades) showing substantially higher participation rates than the more educated (four or more years of college). One interpretation views education in terms of the potential earnings capacity it implies.. The distribution of eligible households by educational attainment also provides evidence of the possible role of education in determining the size of the eligible population: eligible households had markedly less education than the general population. Participation rates among the elderly were relatively low, in line with earlier results which might be explained by a more extensive accounting of assets. Relatively high participation was noted among blacks and single female-headed households. These might be related to regional or local factors, but the ISDP data do not contain the necessary information to find out. B. THE EFFECTS OF FOOD STAMP PARTICIPATION ON THE MARKET LABOR OF FEMALE HEADS OF HOUSEHOLD Despite the wealth of literature on the effects of transfer programs on labor supply, the study by Fraker and Moffitt (1985) is the first economic analysis of the effects of the FSP on work effort.3/ A model of the household's joint decision regarding market labor and participation in the Food Stamp and AFDC programs is specified and estimated on a sample of 358 female heads of households with dependent children (using lSDP data from Wave 5).4/ The analysis sample for this initial research effort was limited to low-income, female-headed households with dependent 2/This section summarizes Thomas Fraker and Robert Moffitt, The Effects of Food Stamp Participation on the Market Labor of Female Heads of Household. Washington, DC: Mathematica Policy Research, March 1985. Prepared for the U.S. Department of Agriculture, Food and Nutrition Service. 4/These 358 households represent 78 percen~ of all female-headed households with dependent children in the Wave 5 sample. The Model children to minimize the number of programs being modeled and to increase the interpretability of the results.2/ Estimates of the model's parameters are useci to predict how work effort and program participation would be affected by selected hypothetical changes in rules governing food stamp benefit amounts. The effects of work effort and program participation responses on average and aggregate benefit levels are also examined. The model of labor supply and program participation used by Fraker and Moffitt is based on the microeconomic theory of household decision making in the presence of limited household resources. The household budget limitations illustrate the combinations of goods that the household can purchase with its entire income. In this application of the theory, a household is assumed to weigh the tradeoff between leisure and income and then to choose the number of hours of work per month which give it the greatest level of satisfaction (utility). For single women with dependent children (the largest demographic category of food stamp recipients), the Food Stamp and AFDC programs introduce substantial complications into this tradeoff. Household decisions to participate in these programs are modeled as occurring simultaneously with the hours-of-work decision. The empirical specification of the model consists of three equations: 1. A labor supply equation, in which the FSP maximum benefit and the benefit reduction rate6/ are hypothesized to affect work effort via the level of nonlabor income and the effective marginal wage rate. 2. A food stamp participation equation, in which the probability of participation in the FSP depends on the difference in household utility when participating in the program and when not participating. 3. An AFDC participation equation that is analogous to the FSP participation equation. i/The income criteria which was used to limit the sample included: 1) hourly wage of $15 or less for the household head, 2) countable assets of $2750 or less ($4000 or less for households with more than one person and which contain an elderly person), 3) earnings of other adults of $2500 per month or less, 4) unearned, nontransfer income of $1000 per month or less, and 5) transfer income not including AFDC and UI of $1000 per month or less. ~/The maximum benefit is the level of benefits available to the household without income. The benefit reduction rate on earnings within .the program refers to the rate at which benefits are reduced for each dollar of earned income. 39 Data and Methodology Findings The model was estimated on the basis of 358 cases from Wave 5 of the ISDP 1979 test panel. To avoid the severe modeling and estimation problems associated with multiple, interacting programs, the analysis was limited to low-to-moderate-income households with dependent children headed by unmarried, nonelderly, and nondisabled women. These households were categorically eligible for AFDC benefits, either eligible for food stamps or able to become eligible by reductions in hours of work, but unlikely to be eligible for SSI or Social Security. (The Medicaid program was not considered in this study.) · Estimates of the food stamp and AFDC participation equations showed that a program's maximum benefit amount (at zero hours of work by the female head) had a statistically significant positive effect on participation in the program.?/ In contrast, and somewhat surprisingly, a program's benefit reduction rate was not found to significantly affect participation in that program, although the estimates do suggest the possibility of a weak inhibiting effect of the benefit reduction rate on program participation. Nonlabor income (excluding benefits from the Food Stamp and AFDC programs) was found to have a generally insignificant effect on participation in both programs. However, the wage rate (after income and payroll taxes, but not after food stamp and AFDC implicit taxes) had a significantly negative effect on participation in both programs, as expected. Estimates of the hours of work equation indicated that the Food Stamp and AFDC programs' maximum benefit amounts and benefit reduction rates had statistically significant but small impacts on work effort. The findings imply that moderate changes in food stamp (or AFDC) regulations which alter the program's benefit reduction rate or the maximum benefit have only small effects on the work effort of female heads of households. Estimates of the parameters in the three equations were used to simulate the effects on hours of market labor and food stamp benefit amounts of three hypothetical changes in current FSP regulations: 1. Increasing the benefit-reduction rate (i.e., the implicit tax rate on earned income) from .30 to .33 Llrn this study, an estimate is referred ~o as statistically significant if a test of the hypothesis that the explanatory variable has no impact on the dependent variable can be re.iected at the 10 percent level of significance. |