Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
Goals for SIPP A necessary prerequisite for any decision to change the design, content, or other features of a survey is to set out clearly the survey's goals. The history of SIPP reveals some changes in the goals for SIPP over time, particularly in the extent to which SIPP has been viewed as a very broad versus a somewhat narrower vehicle for informing social science research and policy analysis. The record also makes clear that the Census Bureau has achieved greater success to date in meeting some of SIPP's original goals than others. Currently, views about the goals for SIPP on the part of users and observers exhibit considerable diversity, although some common themes are present. We tour the landscape both historical and contempo- raneous as background for our conclusions and recommendation for SIPP's goals for the future. THE DEVELOPMENT OF SIPP AND ITS GOALS As new and expanded government social welfare programs were introduced in the late 1960s and early 1970s, the inadequacy of the then available statistical base about household economic resources and the need for and use of assistance programs became apparent. SIPP traces its origins to an interagency committee sponsored by the Office of Management and Budget (OMB) that was concerned with improving the personal and family income estimates derived from the Current Population Survey (CPS) March income 26
GOALS FOR SIPP 27 supplement. Some improvements had been achieved, but CPS estimates still fell far short of independently derived estimates, even after imputation for nonresponse. The interagency committee recommended in the early 1970s that a research program be developed on the best way to improve the collection of income data from households. At the same time, policy analysts, in attempting to provide more rel- evant information to federal policy makers, were troubled by the recognized deficiencies of available income statistics. During this period, several fed- eral agencies sponsored the development of a major new policy tool large computer microsimulation models designed to evaluate alternative designs for government tax and transfer programs (see Citro and Hanushek, 1991a: Ch. 4~. Based for the most part on the March income supplement to the CPS, these models had to contend with the limitations of this data source: incomplete reporting of income amounts, particularly from property and welfare; income and employment data that represented annual totals with no information available on intrayear fluctuations in economic circumstances that could make some families eligible for government programs for part of the year; scanty and suspect information on the numbers and characteristics of program participants, and, especially, on persons and families receiving benefits from more than one program; data on family composition and char- acteristics that were reported, not for the prior income year, but for March (the interview month); and the absence of information on asset holdings and taxes, which are needed to determine program eligibility and also to fully characterize the economic status of the household sector. The Income Survey Development Program Impetus for a new survey of income and program participation arose in the Social Security Administration, where one of the first policy models devel- oped from the CPS March supplement was housed, and in the Office of the Assistant Secretary for Planning and Evaluation (ASPE) in the Department of Health, Education, and Welfare. ASPE established a working group to guide the planning of a new survey on behalf of the whole department, and the Census Bureau participated in the planning process because it was ex- pected to be the data collector. A major testing and research program, the Income Survey Development Program (ISDP), was initiated in 1975 to help with the basic design of SIPP, and a number of research organizations were enlisted to work on venous issues.2 1 Key references on the evolution of SIPP are David (1983) and Hunt (1985). This section is adapted from Committee on National Statistics (1989: Ch. 4). 2For an annotated bibliography of articles, papers, and memoranda authored by the Income Survey Development Program staff, contractors, and consultants over the program's 7-year history, see David (1983).
28 THE SURVEY OF INCOME AND PROGRAM PARTICIPATION The major goals for a new survey were not explicitly stated, but they appear to have been: · to improve the depth, breadth, and quality of data on family and personal income; · to obtain detailed information on eligibility for and participation in a wide range of government assistance programs, including information on program exits and entrances over the short term; · to provide an opportunity-through topical modules asked in addi- tion to the recumng core questions to obtain timely information on emerging concerns of social welfare policy, broadly defined; and · to link survey responses with administrative records to evaluate and enhance data on income and program participation. Although the focus of the effort was on improving understanding of the population potentially eligible for government income support programs, there was also considerable interest in improving data on the income and asset holdings of people at the upper end of the income distribution and in improving the capability for modeling tax as well as transfer programs (Scheuren, 1975~. Thus, the largest survey fielded by the ISDP- the 1979 ISDP research panel, which obtained monthly data from a sample of about 11,500 families, who were interviewed six times over an 18-month period- oversampled both high-income and low-income families. Moreover, one of the four research centers sponsored by the ISDP was focused on improving the measurement of assets and net worth. The other three centers worked on improving the measurement of cash and in-kind income, exploring the use of SIPP for microsimulation modeling, and developing improved impu- tation techniques for handling survey nonresponse. Early SIPP Goals When the ISDP was fully launched and plans were being made to start up SIPP, a 1980 interagency memorandum described the major goals of SIPP as follows:3 (1) to extend the scope and precision of policy analyses for a wide range of federal and state tax and social welfare programs; (2) to improve current estimates of income and income change, includ- ing annual and subannual estimates, by source of income; and (3) to broadly assess the economic well-being of the population. 3The memorandum was signed by lohn J. Carroll, representing the Social Security Adminis- tration (U.S. Department of Health and Human Services); George E. Hall, representing the Census Bureau (U.S. Department of CornTnerce); and Wray Smith, representing the Office of the Assistant Secretary for Planning and Evaluation (U.S. Department of Health and Human Services) (see Kasprzyk, 1988a).
GOALS FOR SlPP 29 Then, in 1981, the ISDP was abruptly halted when budgetary support was denied both for its completion and for the initiation of the planned SIPP. Despite its premature demise, the ISDP made signal contributions to improved data on the income and economic well-being of the population. The methodological and substantive research that was conducted with the 1979 (and also 1978) research panel generated important new insights of relevance for survey design and for policy. Moreover, the example of the ISDP led the Census Bureau to expand the detail collected in the March CPS income supplement on both cash and in-kind income and to adopt a new format for income reporting that the ISDP demonstrated resulted in more complete data. However, the need for a survey with the unique fea- tures of the SIPP design, especially monthly reporting for samples of house- holds followed over time, remained. A wide range of ambitious expectations for SIPP were generated in the process of enlisting support for the rebirth of the survey in 1982. It appears that at least some participants hoped that SIPP would become a general- purpose social survey, a survey to which all sorts of questions appropriate to households or families might be added. Some expected that SIPP would eventually replace other federal household surveys. Other participants hoped that SIPP could serve more long-run basic research interests as well. The Census Bureau's statement of goals in its April 1982 management plan for SIPP is similar to but somewhat broader ranging than the memo- randum by Carroll, Hall, and Smith (see Kasprzyk, 1988a): The Survey of Income and Program Par~cicipacion (SIPP) is designed to satisfy the need for improved data on the economic situation of persons and families in the United States. Information will be collected on various sources of money and nonmoney income, taxes, and assets and liabilities, to produce improved estimates of income distribution, poverty, and wealth. A major use of the data will be to study the efficiency of federal and state transfer and service programs, to estimate future program costs and cover- age, and to assess the effects of welfare reform proposals, tax reform, social security funding problems, and other proposed policy changes. Despite the long incubation period, SIPP was initiated in a great rush because the enabling legislation called for the survey to be in the field within a year. The questionnaire was extensive, the survey design complex, and many processing and estimation procedures still had to be resolved. This complexity, together with budget cuts and technical problems of pro- ducing estimates from detailed longitudinal data, resulted in a very slow rate of issuance of reports and public-use tapes (see Chapter 6~. Over the last few years, the Census Bureau has made a concerted effort to reduce the time lag between collection and release of data. To accom- plish this aim, the agency has concentrated on two primary goals- that of providing information on actual and potential program participation and its
30 THE SURVEY OF INCOME AND PROGRAM PARTICIPATION concomitants and that of improving income data. With the exception of the variable topical modules (those modules that are held open in each panel for federal agencies to determine content that is most relevant to current policy concerns), the Census Bureau has entertained only those questionnaire con- tent changes that fit within one or the other of the two primary goals, narrowly interpreted. The Bureau intends to maintain this policy for SIPP panels through 1994, but it is reopening the issue of SIPP's goals and content in looking toward the planned redesign of the survey. Assessments of Early Goals Assessments of the success to date of the Census Bureau in achieving van- ous goals for SIPP are an important source of information about possible barriers to meeting current or proposed goals. Although one should expect that many operational problems can and will be overcome, it is important to pay heed to evidence about the range of goals that the survey can reason- ably be expected to accommodate. A major source for the panel on this topic was the interim review of SIPP earned out by the Committee on National Statistics (CNSTAT) (1989~. That report documented problems experienced by the Census Bureau that resulted in a failure to achieve fully all of the goals originally set for SIPP.4 From its survey of uses of SIPP data made by federal agencies, the committee found that SIPP was reasonably successful in meeting the goal of describing the circumstances surrounding participation and eligibility for federal transfer programs (Committee on National Statistics, 1989:37-38~: Data from SIPP have already supported important studies of the character- istics of single and multiple program participants. Longitudinal measure- ment has proved feasible, and analyses are being made of duration of spells and number and correlates of transitions during the 32-month period covered by the full 1984 panel.... SIPP data, including information on assets, have supported studies of program eligibility, and the new integrated eligibility module introduced in wave 7 of Me 1987 panel should strengthen this capability. . . Of course, the goal of providing detailed information on program partic- ipation and eligibility could be expanded in a number of ways.... SIPP could provide information on such topics as client evaluation of program performance and reasons why many eligible people do not participate. SIPP could also provide information on additional social service programs. 4We note that the CNSTAT interim assessment was completed in summer 1989. Subse- quently, more progress has been made toward meeting the survey's goals (e.g., estimates of year-to-year income transitions have been produced). However, the general conclusions of the interim report remain largely valid.
GOALS FOR SIPP Expanding the goal in these directions would require major additions of questions to the survey. Similarly, extending the ability of SIPP to follow the dynamics of program participation over time and to shed light on how people cope with economic setbacks generally would require major chang- es to the survey design, such as extending the length of panels for all or some people in the sample. The committee believes that extensions of the goals of SIPP in one or more of these directions are reasonable to consider for the future and that they should be thoroughly evaluated in the planned second phase of the SIPP study. 31 With regard to a second goal of improving understanding of the income distribution in the United States, the committee concluded that this goal was only partly being met (Committee on National Statistics, 1989:38-39~: On the plus side, SIPP has achieved major improvements in the reporting of many sources of income, data are now available for subannual periods, and innovative studies of intrayear spells of poverty have been earned out. Moreover, the collection of asset data for people at the lower end of the income distribution scale has succeeded.... However, the collection of data on asset holdings and income for high- income individuals has not been materially improved, and the attempt to collect information on taxes has suffered from major nonresponse effects [for some items]. As a result, SIPP has not yet proved useful in the modeling of tax programs, one objective of this goal. Moreover, annual calendar-year estimates, for strengthening certain parts of the national in- come accounts and for providing a basic monitoring series that many users will consult, are not yet being issued. Nor are estimates for multiyear accounting periods available. With regard to a third original goal of providing information on policy issues related to household well-being through topical modules, the com- mittee found a mixed picture (Committee on National Statistics, 1989:39~: Issues that can be analyzed from existing SIPP questionnaires can be and are being studied; for these issues SIPP is a rich data source. [But] . . . the objective of providing policy-relevant information will not be fully met until SIPP procedures for developing and approving additions to the ques- tionnaire are better defined and streamlined.... Finally, the report noted that the prominent role originally envisioned for matches of administrative records with SIPP data has only been partly realized. The expectation was that administrative records would be used to increase sampling efficiency by providing supplementary frames of partici- pants in specific programs or persons with other specified characteristics; to provide additional data (e.g., by matching with social security earnings records to obtain longitudinal earnings histories to add to the SIPP files); and to compare and validate specific items common to both sources, by means of record-check studies. Social Security Administration (SSA) and
32 THE SURVEY OF INCOME AND PROGRAM PARTICIPATION Internal Revenue Service (IRS) records were considered prime candidates for such use. To facilitate matching, SIPP had from the outset a very successful program to obtain and validate social security numbers (SSNs) for sample members of each panel, reporting about a 95 percent success rate, even though no attempt is made to match or find a number for a person who refuses to give one in the interview (Bowie and Kasprzyk, 1987:6-7~. (By contrast, the March CPS does not have a regular program of validating SSNs, and, when matches are attempted, generally reports only an 80 per- cent success rate.) However, to date, there has been only limited use of administrative records for supplementation or evaluation of SIPP data. No supplementary sampling frames have been developed from administrative records for SIPP.5 A match with SSA records was camed out for the 1984 panel and provided to SSA analysts under an agreement that limited its use to SSA staff for a 2- year period. A limited match with IRS tax records was earned out for the 1984 panel to develop weighting factors for reducing the variance of in- come estimates from SIPP (Hugging and Pay, 1988~.6 A single record-check study, which matched SIPP records in four states from the first two waves of the 1984 panel with records from eight state and federal programs (AFDC, food stamps, unemployment insurance, workers' compensation, federal civil service retirement, social security, SSI, and vet- erans' pensions and compensation) was carried out to evaluate the quality of survey reports of these income sources (Marquis and Moore, 1989, 1990a, l990b). The study was a full record check, designed to identify false posi- tive responses (i.e., people who report participation in a program to SIPP but for whom the agency has no record of participation) in addition to false negatives (i.e., people who fail to report their participation to SIPP). The study encountered serious delays.7 Almost 5 years elapsed from its initia- tion in 1984 until detailed research results appeared. Even then, many potentially useful analyses were never undertaken (e.g., analysis of benefit amounts as distinct from recipiency). However, the study did stimulate the Census Bureau's current cognitive research program for improving the SIPP questionnaire (see Chapter 7~. The ISDP had considerably more success with the use of administrative records to evaluate the quality of survey responses and improve question SAdministrative records were used to provide supplementary sampling frames for the ISDP. However, the data were never analyzed because linked data files with appropriate weights could not be produced on a timely basis (see Kasprzyk, 1983; see also Chapter 4). 6Further work on variance reduction using this approach is in progress. 7Delays occurred in part because of the time required for negotiations with state welfare agencies to obtain program records. Indeed, it ultimately proved impossible to obtain records from one of the four states. Problems at the Census Bureau in conducting the Itches and preparing analysis files also contributed materially to the delays.
GOALS FOR SIPP 33 wording and interviewer training procedures (see Kasprzyk, 1983; Logan, Kasprzyk, and Cavanaugh, 1988~. The ISDP used forward record-check studies in which people included in independent samples from administra- tive sources (including IRS and federal and state program records) were administered the ISDP interviews. Such a design eliminates the need for record matching as such and hence reduces processing time; however, it permits identifying only false negative responses, not false positive responses. The committee's report noted major impediments to the use of admin- istrative records, not only in SIPP but also in other survey programs (Com- mittee on National Statistics, 1989:51-52~. Such linkages involve different agencies with different rules, regulations, and objectives. Moreover, at present, linkages are greatly constrained by legal requirements and ethical considerations of protecting the confidentiality of individual responses. VIEWS ABOUT SIPP'S GOALS To obtain input for our study panel's consideration of future goals for SIPP, we consulted a wide range of users and observers. One source was the 1989 report of the Committee on National Statistics, which included information from an extensive series of interviews with federal agency staff about their uses of SIPP and their wish lists for the future. We organized the Conference on the Future of SIPP in April 1991, which invited researchers and policy analysts from both government agen- cies and academia to discuss the usefulness of SIPP in a wide range of subject areas, assess SIPP's comparative advantage vis-a-vis other data sources, and make recommendations for improving SIPP for future research and policy use (the list of paper authors and invited discussants is provided in Appendix B) The topics covered in the conference, which gave particular attention to longitudinal uses of SIPP, included: child care and child sup- port, employment and labor force transitions, extended measures of well- being, health and disability, income transitions for the elderly, interactions of family composition and income change, modeling program eligibility, poverty status and transitions, and program participation dynamics. (The conference papers have been published in a 1992 special issue of the Jour- nal of Economic and Social Measurement [JESM][Vol. 18, Nos.1-41.) Another source of information for the panel resulted from a specific part of our charge to consider ways in which SIPP might be used in con- junction with the March CPS and administrative records to develop im- proved statistics on income. We commissioned a paper on this topic that was also published in the special issue of JESM (Smeeding, 1992) and obtained reviews of the paper from several analysts (Slater, 1991; Watts, 1991~. We also had available pertinent research and planning documents from Census Bureau staff.
34 THE SURVEY OF INCOME AND PROGRAM PARTICIPATION The CNSTAT Report The interim review of SIPP presented the following conclusions regarding the goals that the Census Bureau should adopt, at least in the short term (Committee on National Statistics, 1989:40~: During this start-up period, the committee believes that SIPP should be narrowly focused on . . . [two goals]: improving data on program eligibil- ity and participation and improving information on the income distribution of the United States. The committee thus agrees with the narrow focus now being given to the program by the Census Bureau in its efforts to speed up the release of data.... In considering priorities within [these goals], the committee agrees that the prime focus of SIPP should be on the population economically at risk [emphasis added]. SIPP is appropriately designed to obtain information on people who are currently poor or in need of government assistance and also on people in the near-poor and middle-income range who, if they experienced an event such as loss of a spouse, loss of a parent, or loss of a job, could be at risk of economic deprivation.... In considering the intensity with which . . . topics should be covered for example, the range of detail and the longitudinal duration" the requirements for identifying and analyzing program eligibility and participation should predominate. The committee is not recommending that SIPP be limited to people on the lower half of the income scale, but that it should give priority to changes that will enhance information on the less well-off part of the population.... The committee believes that SIPP should not be viewed as a general- purpose survey. In future years, after the start-up period, the goals, priori- ties, and focus of SIPP may be expanded. The committee is very dubious, however, that SIPP could ever become a general-purpose survey.... Finally, the committee believes that there are a number of steps that are imperative to take in the short term to improve the capability of the SIPP program to meet its priority goals, even with the narrow interpretation Mat they have beers given by us and by the Census Bureau. These steps include restoring the sample size to that originally intended for the SIPP, develop- ing estimates based on combined panels, and improving the quality and timeliness of the SIPP data products. Conference on the Future of SIPP Participants in the Conference on the Future of SIPP expressed a wide range of views regarding the goals of SIPP.8 Most participants were cognizant of 8Participants also expressed views on SIPP's design and data products, which are discussed in later chapters. We note, of course, that the conference participants do not necessarily represent the full community of SIPP data users, although we believe that they offered a variety of perspectives (see Appendix B).
GOALS FOR SIPP 35 the need to keep the scope of the survey and the questionnaire within bounds. However, they usually wanted to expand SIPP to provide more information for their particular area of interest. Often they asked for more extensive topical modules, asked more frequently (e.g., on health and disability and child care) thereby contradicting the strong recommendation in the CNSTAT report that SIPP focus on the core income and program participation data. There was little agreement on which topical modules were most important, although the survey could not hope to accommodate all of the suggestions made. The diversity of interests expressed at the conference is understandable given that we asked each of the participants to focus on a particular topic, without considering tradeoffs with other topics. Also, the enthusiasm for topical modules may in part stem from the fact that topical module data have generally proven easier to process and analyze than core data-both by the Census Bureau itself, which to date has produced more reports from the topical modules than from the core (see Chapter 6), and by other ana- lysts. Nonetheless, some common themes emerged from the conference: · Many conferees expressed interest in having SIPP provide more and better information about families and children, reflecting increased policy concern with the economic and social hardships being experienced by chil dren and the need for additional information to understand fully family circumstances given high rates of family breakup. Specific suggestions included: extending the length of SIPP panels (to 4 or 5 years) to provide more information about the socioeconomic consequences of family com position change over the short to medium term; following children who move to a nonsample household (e.g., chil dren in a single-parent family at the start of a panel who go to live with the other parent); -following children and other family members who move to institu- tions; -adding questions about children's health status and utilization of health care; adding questions about children's income and receipt of Supplemen tal Security Income (SSI) benefits; adding questions about nonresident kin, income flows among kin, and the financial circumstances of noncustodial parents; adding questions about intrafamily and intrahousehold sharing of Income; adding more detailed questions about child care, including participa tion in federal programs such as Head Start and preschool child care for nonemployed mothers;
36 THE SURVEY OF INCOME AND PROGRAM PARTICIPATION . - adding questions about dependent care provided by family members (both for children and elders); and adding questions to permit complete identification of the relation- ships among all household members (needed, in part, to make a more accurate determination of eligibility for transfer programs). The conferees generally expressed strong interest in exploiting the potential of administrative records to supplement and enhance the data from SIPP. Specific suggestions included: adding supplemental samples of program beneficiaries drawn from administrative records; obtaining additional and higher quality information on selected top- ics from administrative records matches (e.g., with employer, social security, and IRS records); and routinely publishing comparisons of data from SIPP, the March CPS, and administrative records. · A number of participants voiced support for adding supplemental measures of socioeconomic hardship and deprivation, which, in some cases, would also improve measures of program eligibility. Specific suggestions included: asking about episodes of hunger; ascertaining consumption for housing and other basic necessities; expanding questions on disability, both in the core and topical mod- ules; and adding questions on access to credit. Overall, there was clear, if implicit, agreement with the recommen- dation in the CNSTAT report that SIPP should focus on the economically at-risk population. Although the conference participants expressed the de- sire to maintain the ability to analyze the entire income distribution and, specifically, to obtain improved information on after-tax income, no one supported turning SIPP into a sunrey of wealth or the wealthy. Instead, the emphasis throughout the discussions was on people eligible for assistance programs and on families and children at risk of economic hardship. Many of the suggestions listed above stem from this orientation, as does the sug- gestion made by some participants to oversample low-income people, in addition to increasing SIPP sample size generally, which was supported by all participants. The Role of SIPP in Improving Income Data An Outside Perspective In the paper commissioned by the panel, Smeeding (1992) addresses issues of definition and measurement of income and the role of SIPP, the March
GOALS FOR SIPP 37 CPS, and administrative records in improving the nation's income statistics. The goal of his paper is to present a wealth of broad ideas to stimulate thinking, rather than to discuss specific details. Smeeding comments first of all that the policy context must be kept in mind when determining priorities for improvement of income data. Three developments of particular consequence in this regard are the growing use by the federal government of the tax system to achieve social welfare goals (e.g., the recent expansion of the earned income tax credit to help the work- ing poor); the growing interest in such economic outcomes as welfare de- pendency that can only be measured with longitudinal data; and-the interest in reassessing the current measure of poverty (both the standard of need and the definition of resources to compare to the standard). Smeeding proposes that the Census Bureau develop a concept of "full income," or the ability to maintain a given level of consumption, to use as a template for identifying pnonty areas for improvement.9 The traditional definition of money income is generally confined to payments made in cash to the family on a regular or at least periodic basis: it excludes such "income" as in-kind benefits, lump-sum payments (e.g., gifts), and (net) imputed returns from housing and other assets. The traditional concept also does not subtract taxes or other mandatory contributions such as child sup- port, nor does it include contributions (received or made) of time (e.g., caregiving).~° With regard to SIPP, Smeeding identifies a dozen areas of opportunity and assigns them pnonties. He states four premises that underlie his choices: (1) SIPP should complement the CPS, not substitute for it; (2) SIPP should help improve the CPS but also have its own constituency and hence its own report senes; (3) in general, SIPP should be innovative in fact, expenmen- tal in its approach to most issues, which may include imputation and valu- ation in addition to measurement of venous types of income; and (4) SIPP 9Watts (1991) notes that "full income" is not well defined by Smeeding and could be confused with the term as it is usually understood by economists namely, the maximum income that a worker could achieve if he or she worlced as much as humanly possible (i.e., had zero leisure). Both Watts and Slater (1991) critique specific aspects of Smeeding's concepts but agree that the recommendation for the Census Bureau to adopt a broad definition of income to use as a guide for research and development is sound. In our detailed discussion in Chapter 3 of moving SIPP toward this goal, we use the term "income and other economic resources;" elsewhere, we use "income" as shorthand for a broader definition. 10Smeeding notes that the Census Bureau has invested considerable effort in developing alternative income measures from the March CPS, specifically measures that exclude most taxes and include selected in-kind benefits and returns from assets see Chapter 3.. He com- pliments the Census Bureau on this effort but points to conceptual and measurement problems that remain.
38 THE SURVEY OF INCOME AND PROGRAM PARTICIPATION should serge the federal policy community and perhaps state policy needs as well. Smeeding assigns the highest priority to development of after-tax mea- sures of income from SIPP, followed closely by efforts to value basic noncash benefits, including food, housing, and medical benefits. Next in his pnon- ties are improved measurement (in topical modules) of caregiving, health insurance and health care finance, and disability. Smeeding argues that each of these topics is central to current public policy debates and each is undertreated in current measurement work. Improved data from SIPP will support analysis of these topics in their own right and also contribute to development of improved measures of components of income and need. Next in order, Smeeding supports work to improve the quality of mea- sures of transfer payments through links with administrative records. Next comes work to develop improved equivalence scales for comparing income across different types of families (which have different needs) and work on measunug intrahousehold sharing of income. Smeeding assigns lowest pn- onty to improved measurement of wages and salanes, self-employment in- come, property income, and imputed rent. Census Bureau Perspective About 2 years ago, Census Bureau staff began to develop a concept for improving the nation's income statistics that in some ways would expand and in other ways limit the future role of SIPP. The proposal started with two key premises. The first is that the March CPS and SIPP will continue to coexist, rather than the latter ever replacing the former as the primary source of income information. The second is that the Census Bureau should set aside its traditional approach of generating data files and published reports from individual surveys (e.g., income data from the March CPS are currently published in the P-60 senes, while the new P-70 series is used for income data from SIPP); instead, the goal should be to produce the best income statistics based on all available data sources-including the March CPS, SIPP, and administrative records. Census Bureau staff initially sketched Ollt an ambitious plan to accom- plish this goal. In broad outline, the project would involve using admin 1lTo be responsive to state needs, Smeeding suggests that SIPP oversimple six or seven states with maximum variation in their policy approaches to topics of national interest (e.g., income transfer policy), so that estimates for state-specif~c subgroups could be developed at some acceptable level of accuracy. 12This description of the Census Bureau's plan is based on a presentation by John Coder to the panel, July 30, 1990; see also Coder (1991). Hemot (1988) first outlined the concept of an integrated income statistics system. His paper envisioned a more prominent role for SIPP than
GOALS FOR SlPP 39 istrative records and other sources to assess the extent and nature of errors, such as underreporting and misreporting, in the March CPS and SIPP in- come data. There would then be an attempt to adjust the SIPP data through some sort of multivariate imputation or weighting technique. Alternatively, exactly matched administrative values might be substituted directly in the SIPP records, if the problems of data access could be worked out. Then, the adjusted SIPP data would be used to improve the quality of income data from the March CPS through a related imputation or modeling procedure. The CPS data would retain the advantage of timeliness (assuming that the adjustments would be made from an earlier SIPP file), and the-later SIPP data would provide the advantage of additional subject detail and intrayear amounts. Initial work began on the project with a match of selected IRS data and March CPS records to evaluate the quality of earnings reported by married couples (Coder, 1990~. However, the enormity of the entire effort quickly became evident. The latest paper developed by staff in the Housing and Household Economic Statistics (HHES) Division gives higher priority to making selected improvements to SIPP and CPS income measures and views the development of a fully integrated system of income statistics as a much longer range goal (Bureau of the Census, 1991a). The HHES staff paper agrees with Smeeding that SIPP should comple- ment rather than substitute for the CPS. With regard to SIPP income mea- sures, the paper gives priority (as does Smeeding) to implementing im- provements that have already been undertaken for the March CPS- specifically, developing a model to estimate after-tax income and methods for valuing noncash benefits. The tax model developed for SIPP should be superior to that for the March CPS, because the SIPP model can take advantage of tax information reported in each year's tax topical module (but see Chapter 3 on this point); the CPS model bears no direct relationship to actual tax data. The availability of a SIPP tax model, together with an improved CPS-IRS match that includes additional IRS data, should make it possible to improve the CPS tax model as well. With regard to noncash benefits, the current plan is to apply the CPS methodology to SIPP. Funding will be sought to use updated information to value employer-provided medical benefits (from the 1987 National Medical Care Expenditure Survey) and home equity (from an updated match of American Housing Survey and CPS records). The Census Bureau plan also includes improved publication series and user access to microdata from both SIPP and the March CPS (see discussion _ _ . do the proposals subsequently developed by the Census Bureau staff (see discussion in the text). Herriot suggested that CPS income estimates, adjusted on the basis of SIPP estimates for an earlier year, would be issued as "preliminary" figures for Me most recent year, to be superseded when higher quality SIPP estimates became available.
40 THE SURVEY OF INCOME AND PROGRAM PARTICIPATION in Chapter 61. A regular SIPP report series is being developed that will include both cross-sectional and longitudinal statistics on income and pro- gram participation. SIPP income data will also be included in the CPS P-60 reports, although the intention is for the March CPS to remain the primary source of annual income estimates. Recently, the CPS report series was expanded to include estimates of state-level median income and poverty rates. Reports that provide more detailed analysis of SIPP longitudinal and CPS time series data are also planned. Finally, with regard to assessment and correction of reporting errors in the SIPP and March CPS, which is an essential component of a fully inte- grated income statistics system, the current plan is to treat this as a long- range goal. Initially, a major study to compare SIPP and CPS statistics on income, poverty, the labor force, and other topics will be undertaken, using the 1990 SIPP panel and the March 1991 CPS. Work will also continue on evaluating income reporting in both the SIPP and CPS through exact matches with administrative records. For the March CPS, after completion of the work on wages and salaries, the next step is to study interest and dividend reporting using the same CPS-IRS exact-match file. The scope of potential exact matches with SIPP is much broader: for example, it involves linkages with records for assistance programs such as SSI, food stamps, and AFDC as well as a match with IRS data. In summary, the staff paper asserts (Bureau of the Census, l991a:13~: Our ultimate goal is to provide public access to microdata files that are likely to contain a combination of survey responses, modelled responses, and administrative information. If successful, our efforts could result in a regular series of income and poverty estimates that would be adjusted for nonsampling error. Note that CPS and SIPP estimates would be adjusted to the same totals.... This would represent an enormous improvement over our current survey estimates. However, the paper acknowledges that progress toward this goal will neces- sarily be slow, given budget and staff constraints and also the time and effort required to obtain new sources of administrative data and increase the amount of information from sources to which the Census Bureau now has access (e.g., additional items from IRS records). We gave serious attention to the Census Bureau's proposals to develop an integrated income statistics system. We fully agree that there is a need to improve the nation's statistics on income, poverty, and related measures. Indeed (see below), we view the improvement of data on income and other economic resources as a primary goal for SIPP. We further agree that administrative records can contribute over the long term to improving the income measures from SIPP. However, we conclude that SIPP should be- come the nation's primary source of income statistics and, hence, should have priority for investments to evaluate and develop improved income
GOALS FOR SIPP 41 measures. The CPS should continue to collect data on income (needed to support analysis on labor force topics), but the March CPS income supple- ment can never be designed to provide the same extent of detail or achieve the same quality of reporting as in SIPP (See the discussion in Chapter 3 of the role of the March CPS and administrative records in achieving SIPP's goals.) RECOMMENDATION Despite the diversity of views about goals for SIPP, we agree with the conclusion of the CNSTAT interim report that SIPP cannot and should not be viewed as an all-encompassing general-purpose survey in the area of social welfare policy. Rather, it is essential for the cost-effective operation of the program that it focus on a core set of major goals, which we present here; in the next chapter we discuss how to achieve these goals. We believe that a primary goal for SIPP should be to provide improved data on the income and other economic resources of the population. Such data are vitally needed by government agencies to track a key component of the economy the extent and distribution of resources available to the household sector and to plan, operate, and assess government programs and policies that affect household resources. Despite improvements in the March CPS income supplement, a major survey is needed to obtain detailed and high- quality measures of income and other economic resources (e.g., assets) for periods both shorter and longer than a year, one that includes the necessary corollary information (e.g., family composition and changes in composition over time) to support analysis of important subgroups of the population. We believe that a second primary (and interrelated) goal for SIPP should be to provide improved data on eligibility for and participation in govern- ment assistance programs. Currently, there are a large number of programs that provide income support and other assistance to retirees, disabled people, people temporarily out of work, people whose income is not sufficient to cover their bills for medical care or higher education, and low-income fami- lies with dependent children. Government outlays for these programs are substantial.~3 Administrative records for individual assistance programs provide useful information, but a major survey is needed to collect informa- tion required to determine who is eligible to participate in programs as well as who participates, to determine eligibility for and participation in more 13In fiscal 1988, federal, state, and local governments spent more than $350 billion on social insurance programs (including social security, Medicare, unemployment insurance, and workers' compensation) and close to $175 billion on all forms of cash and in-kind assistance to people with limited income (Bureau of the Census, 1991d:Tables 583, 584).
42 THE SURVEY OF INCOME AND PROGRAM PARTICIPATION than one program at a given time or over time, and to follow the dynamics of program entrances and exits. Within these two primary goals, we believe that SIPP should concen- trate on improving information for people who are economically at risk, as that term is defined by the CNSTAT report (see above). That report sug- gested that a focus on the economically at-nsk population might character- ize SIPP only so long as the survey was in a stan-up phase. However, we conclude that SIPP should retain this focus over the longer term. Despite the many improvements that we fully expect will be implemented as part of the upcoming redesign of SIPP, the survey will still be very much in a developmental phase. Attempting to obtain markedly improved data on income and other economic resources (e.g., assets) for the very well-off, which the evidence shows is difficult for household surveys to accomplish (see Chapter 3), could well compromise SIPP's ability to provide improved information for most of the population. Moreover, it is the population that already receives or is at risk of becoming eligible for government assistance on which good data are most needed to support program analysis and re- lated social welfare research. We believe that SIPP should have a third, but subordinate goal, namely, obtaining additional policy-related information on topics of current interest that relate to the main goals of the survey. The core questions need to remain relatively stable during the period between major redesigns. Yet important changes may occur in the economic circumstances of the popula- tion or in the kinds of policies that are proposed to respond to assistance needs, and, correspondingly, there ought to be a way to obtain timely infor- mation from SIPP in response. The topical modules can serve this purpose, as well as provide a means to obtain information that is included in the survey on a continuing basis but does not need to be asked more than once or twice a panel. Recommendation 2-1: The two primary goals of SIPP should be to provide improved information on the distribution of income and other economic resources for people and families and on eligibility for and participation in government assistance pro- grams. Within these two goals, most attention should be paid to improving the information for people who are economically at risk. A third important but subordinate goal is for SIPP to have a capability to respond to current policy needs for data in topical areas that are related to the core subjects of SIPP.