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.
Achieving SIPP's Goals In this chapter we amplify our views on SIPP's goals, indicating priority areas for research and development and the implications that we see for modifications or improvement to the survey content. Chapters 4-6 discuss, respectively, the implications for SIPP's design, data collection and pro- cessing, and data products. We also consider in this chapter the role of both the Current Population Survey (CPS) March income supplement and admin- istrative records in the achievement of SIPP's goals. IMPROVING DATA ON INCOME AND OTHER ECONOMIC RESOURCES Since at least the time of the Great Depression and World War II, a major goal of the federal statistical system has been to measure the economic resources of the population and of family and household units. There has been continuing keen interest in assessing how resources have changed over time for these units and in characterizing the distribution of economic re- sources in the population as a whole and for important subgroups. Resources are important to measure because they represent the potential ability of people and households to consume goods and services in order to attain a level of economic well-being. Conceptually, economists might agree that consumption should be examined directly as a measure of eco 43
44 THE SURVEY OF INCOME AND PROGRAM PARTICIPATION nomic well-being. However, because of difficulties in measuring con- sumption and because many government programs provide cash payments (i.e., potential rather than actual consumption), the United States histon- cally has relied on measuring economic resources as a proxy for economic well-being. (Economists relate economic resources to consumption follow- ing the Haig-Simons accounting identity that defines flows of income as the sum of changes in net worth and current consumption; see Bradford, 1986.) For many years, the focus of official measures in the United States has been one component of economic resources, namely, regular cash income before taxes.2 Since the early 1980s, the Census Bureau has supported research on experimental income measures that subtract out most taxes (which represent a use of resources that makes them unavailable to support con- sumption) and that take account of other kinds of resources, including ma- jor types of in-kind benefits (Bureau of the Census, 1982, 1988b, 1990a, l990b, l991b). To date, the Census Bureau's work has not extended to the valuation of asset holdings (other than equity in one's house) or the contn- bution of nonmarket, nongove~nment resources (e.g., the value of home production or income from the underground economy).3 1As is always the case with attempting to assert a general point, there are exceptions. The first one is the difference between what a unit actually consumes and how it values it. The issue here is the utility from consumption and the attempt to make interpersonal comparisons of this utility through such devices as equivalence scales. However, equivalence scales are controversial enough, both conceptually and practically, that it does not seem useful to put forward the measurement of the utility of consumption as a goal for a federal statistical program such as SIPP. 2Anderson (1988:Ch. 7) documents the efforts in the 1930s to develop reliable measures of the unemployed population and people eligible for government assistance, which ultimately led to the CPS as a vehicle for collecting employment and income data. See Goldfield (1958) for a history of income data in the CPS and the decennial census through 1955 and Weluiak (1990) for an updated history of income questions in the CPS. With regard to more direct measures of consumption, the Bureau of Labor Statistics has a long history of conducting consumer expenditure surveys: major surveys were conducted in 1888-1891, 1901-1903, 1918-1919, 1935-1936, 1950, 1960-1961, and 1972-1973, and a con- tinuing Consumer Expenditure Survey (CEX) was initiated in 1979 (see Goldberg and Moye, 1985). The survey data have been used to study aspects of economic well-being, but their primary use has been to develop the market baskets that are the basis for the consumer price index. The current continuing CEX has a small sample size: 6,800 consumer units for, the quarterly interview survey component and 6,000 consumer units for the 2-week diary survey component. 3Currently, the Census Bureau publishes 14 experimental income measures in all. The taxes that are excluded from one or more of these measures are social security payroll taxes, federal and state individual income taxes, and property taxes on owner-occupied housing. The valua- tions of in-kind benefits that are included in one or more measures cover food stomps, school lunches, Medicaid, Medicare, employee health insurance premiums, and subsidized housing. Some of the measures include estimates for realized capital gains, and one measure includes the net imputed return on equity in owner-occupied housing.
ACHIEVING SIPP'S GOALS 45 We agree with Smeeding (1992) and others who argue that the Census Bureau should adopt a broad definition of economic resources to use as a guide for developing improved measures. We note that SIPP was originally designed and has been striving to achieve this goal. In comparison with the March CPS, SIPP's measures of many of the components of cash income (including returns to assets) and in-kind benefits are much more detailed, are obtained for more disaggregated time periods, and can be aggregated for a greater variety of economic units (individuals, households, families, and subunits eligible for such programs as AFDC and SSI). Tables 3-1 and 3-2 list nonasset and asset income sources collected in SIPP and for- what time periods information is obtained about recipiency and amounts for each source. Unlike the CPS, SIPP obtains direct measures of taxes and the value of asset holdings. In comparison with administrative records sources of in- come data (e.g., IRS tax returns), SIPP provides a wealth of relevant demo- graphic and socioeconomic detail for analysis. Also, SIPP includes close to the entire population, not just taxpayers or program beneficiaries, and close to all sources of income and other economic resources, not just those sub- ject to taxation or considered in the calculation of program benefits. SIPP has also attained a high level of quality in measures of many types of income and related variables (see further discussion in Citro, 1991; Jabine, King, and Petroni, 1990; Singh, Weidman, and Shapiro, 1988; Vaughan, 1988~. Rates of nonresponse to basic income and asset recipiency and labor force items were very low in the 1984-1986 SIPP panels; see Table 3-3. For example, fewer than 1.5 percent of respondents failed to say whether they received income from social security, unemployment compensation, or food stamps or whether they owned savings accounts or shares of stock. Nonresponse rates for income amounts were somewhat higher in the 1984-1986 panels: for example, 4-10 percent of recipients failed to provide income amounts from wages and salaries, social security, unemployment compensation, or food steps; and 14-17 percent failed to provide amounts of self-employ- ment salary or draw (Table 3-3~. In comparison with the March CPS income supplement, however, SIPP has achieved markedly reduced rates of nonresponse for amounts of virtu- ally all types of income. For example, about 8 percent of recipients of Supplemental Security Income (SSI) in the 1984 SIPP panel failed to pro- vide an amount compared with 20 percent in the March 1985 CPS (see Table 3-4; SIPP nonresponse percentages are shown on an average monthly basis for each quarter of 1984~. As another example, 7-9 percent of earners failed to provide an amount for wages and salaries in the 1985 SIPP panel compared with 17 percent in the March 1986 CPS; see Table 3-5. Overall, regular money income estimated for 1984 from the SIPP included 11 percent imputed values due to missing responses, while the corresponding figure from the March 1985 CPS was 20 percent imputed values; see Table 3-6.
46 THE SURVEY OF INCOME AND PROGRAM PARTICIPATION TABLE 3-1 Nonasset Income Sources in SIPP Time Period Covered for Income Source Recipiency Amount Wages and salaries (before deductions, including tips, bonuses, overtime pay, commissions, and, for armed forces members, cash housing allowances and other special pay) Self-employment income (before deductions) Earned income tax credit Social security for self (or self and spouse, total before deductions) Social security for one's children U.S. government railroad retirement State unemployment compensation Supplemental unemployment benefits Other unemployment compensation (Trade Adjustment Act benefits, strike pay, other) Veterans' compensation or pensions Black lung payments Workers' compensation State temporary sickness or disability Employer or union temporary sickness Payments from own sickness, accident, or disability insurance policy Federal Supplemental Security Income State Supplemental Security Income Aid to Families with Dependent Children General assistance Indian, Cuban, or refugee assistance Foster child care payments Other welfare Child support payments Alimony payments Pension from company or union Federal civil service or other federal civilian pension U.S. military retirement pay National Guard or Reserve forces retirement State government pension Local government pension Income from paid-up life insurance policies or annuities Monthly Monthly Twice a panel Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Twice a panel Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly Monthly
ACHIEVING SIPP'S GOALS TABLE 3-1 Continued 47 Time Period Covered for Income Source Recipiency Amount Estates and trusts Monthly Monthly Other payments for retirement, Monthly Monthly disability, or survivor Income assistance from charitable Monthly Monthly group Money from relatives or friends Monthly Monthly Lump-sum payments Monthly Monthly Income from roomers or boarders Monthly Monthly National Guard or Reserve pay Monthly Monthly Incidental or casual earnings Monthly Monthly Other cash income not included Monthly Monthly elsewhere Medicare Every 4 most N.A. Medicaid Monthly N.A. CHAMPUS health insurance Monthly N.A. CHAMPVA health insurance Monthly N.A. Military health insurance Monthly N.A. Current employer or union health Monthly N.A. insurance Former employer hearth insurance Monthly N.A. Other health insurance Monthly N.A. Food stamps Monthly Monthly WIC (women, infants, and children Monthly Monthlya nutrition program) Energy assistance Every 4 most 4-mot total School lunch Every 4 mos.b N.A. School breakfast Every 4 mos.b N.A. Public housing Twice a panel N.A. Subsidized housing Twice a panel N.A. G.I. Bill Monthly Monthly Other VA educational assistance Monthly Monthly College work study Every 4 most 12-mo. total Pell Grant Every 4 most 12-mo. total Supplemental Educational Every 4 most 12-mo. total Opportunity Grant National Direct Student Loan Every 4 most 12-mo. total Guaranteed student loan Every 4 most 12-mo. total JTPA training Every 4 most 12-mo. total Employer educational assistance Every 4 most 12-mo. total Fellowship/scholarship Every 4 most 12-mo. total Tuition reduction Every 12 most 12-mo. total Other educational financial aid Every 4 most 12-mo. total continued on next page
48 TABLE 3-1 Continued THE SURVEY OF INCOME AND PROGRAM PARTICIPATION NOTES: Information that is available by month or every 4 months is obtained in the core interviews; information that is available twice a panel or every 12 months is obtained in topical modules. The 1990 panel is used as the basis for the table entnes. Major differences between the 1990 and earlier panels include: The 1984 panel included a topical module (in wave 6) on employee fringe benefits, includ- ing health insurance, life insurance, use of company vehicle, expense account, meals, and lodging; Military health insurance coverage was not directly ascertained prior to the 1990 panel; The earned income tax credit was not ascertained prior to the 1990 panel; Prior to the 1987 panel, residence in public or subsidized housing was ascertained only at the initial visit; and Separate types of educational assistance were not identified in the 1984 panel, but the total amount of such assistance was ascertained every 4 months. aAmounts of WIC vouchers are imputed by the Census Bureau from information supplied by the U.S. Department of Agnculture. bRecipiency information includes whether school lunch or breakfast was free, reduced pnce, or full pnce. N.A., Not applicable or not available. SIPP has also obtained more complete reporting of many types of in- come than has the March CPS when aggregate amounts from the two sur- veys are measured against independent sources. For example, average monthly food stamp benefits by quarter reported in the 1984 SIPP panel amounted to 83-90 percent of the corresponding quarterly totals from independent sources, whereas total food stamp benefits in 1983 from the March 1984 CPS amounted to only 71 percent of the annual total from independent sources. Similarly, average monthly SSI benefits by quarter from the 1984 SIPP panel averaged 90-99 percent of the quarterly totals from independent sources, whereas total SSI benefits in 1983 from the March 1984 CPS amounted to only 85 percent of the annual total from independent sources; see Table 3-7. At the same time, SIPP's success in measuring each and every one of the components of cash and in-kind income and asset holdings has not been complete. Some measures are subject to appreciable error: for example, average monthly Aid to Families with Dependent Children (AFDC) benefits by quarter from the 1984 SIPP panel averaged only 76-86 percent of the quarterly totals from independent sources. These figures are not a marked improvement over the March 1984 CPS, for which the total AFDC benefits in 1983 amounted to 76 percent of the annual total from independent sources (Table 3-7~. Comparisons of SIPP aggregates with independent sources for another income type-state unemployment compensation show widely varying rates of completeness by quarter (from 107 percent of the independent total in the third quarter of 1983 to only 73 percent in the third quarter of 1985~; see Table 3-8.
ACHIEVING SIPP'S GOALS TABLE 3-2 Assets and Asset Income Sources in SIPP 49 Time Penod Covered for Asset Ownership Income Amount Value Interest-beanug bank accounts Joint with spouse Regular passbook savings accounts Every 4 most Money market deposit accounts Every 4 most combined combined Certificates of deposit Every 4 most 4-mot total value twice Interest-earning checking accounts Every 4 most a panela Own Regular passbook savings accounts Every 4 most] Money market deposit accounts Every 4 most combined combined Certificates of deposit Every 4 most 4-mot total value twice Interest-earning checking accounts Every 4 most a panela Other interest-earning assets Joint with spouse Money market funds Every 4 most ~ U.S. government securities Every 4 most combined combined Municipal or corporate bonds Every 4 most 4-mot total value twice Other interest-earning assets Every 4 mos.J a panela Own Money market funds Every 4 most ~ U.S. government securities Every 4 most ~combined combined Municipal or corporate bonds Every 4 most ~4-mot total value twice Other interest-earning assets Every 4 most, a panela Stocks or mutual fund shares Jointly held with spouse Every 4 most 4-mot total Twice a (cash and reinvested dividends panelb separately; value and debt or margin account separately) Own (cash and reinvested dividends separately; value and debt or margin account separately) Rental property Jointly held with spouse (gross and net rent separately; market value and principal owed on mortgage separately) Own property (gross and net rent separately; market value and principal owed on mortgage separately) Held with others (share of net rent; market value, principal, share of net equity separately) Mortgages Jointly held with spouse Every 4 most demo. total Twice a panelb Every 4 most demo. total Twice a panel Every 4 most 4-mot total Twice a panel Every 4 most 4-mot total Twice a panel Every 4 most 4-mot total Once a panel continued on next page
so THE SURVEY OF INCOME AND PROGRAM PARTICIPATION TABLE 3-2 Assets and Asset Income Sources in SIPP Time Period Covered for Asset Ownership Income Amount Value Mortgages coned Own Other mortages Royalties Other financial investments U.S. Savings blonds Checking accounts (no interest) Joint with spouse Own Debt owed on store or credit cards Joint with spouse Own Debt owed on bank or credit union loans (except car and home equity) Joint with spouse Own Any other debt (e.g., medical bills) Joint with spouse Own Individual Retirement AccountsC KEOGH accountsC 401K accountsC Life insurance (including employer) Employer life insurance separately Own home Mortgages, equity, market value Mortgage/rent payments, utilities Vacation/second home Cars/vans/trucks (details on year/make/model and debt owed for individual vehicles) Recreational vehicles (market value, debt owed) Own business Debt owed against own business Capital gains Every 4 most Once a panel Every 4 most l Every 4 most J Every 4 most Twice a panel Twice a panel Once a panel Once a panel Once a panel Once a panel Once a panel Once a panel Twice a panel Twice a panel Twice a panel Twice a panel Once a panel Once a panel Once a panel Twice a panel Twice a panel Twice a panel Monthly Once a panel Twice a panel 4-mot total N.A. combined 4-mot total N.A. N.A. N.A. N.A. N.A. N.A. N.A. Once a panel Once a panel N.A. Twice a panel Twice a panel Twice a panel Twice a panel Once a panel Once a panel Once a panel Once a panel N.A. Once a panel N.A. Once a panel Twice a panel Twice a panel Twice a panel Twice a panel N.A. Once a panel N.A. Twice a panel N.A. Once a panel N.A. N.A. N.A. N.A. Monthly N.A. N.A. Once a panel Once a panel Twice a panel Twice a panel Twice a panel Once a panel Once a panel Twice a panel NOTES: The 1990 panel is used as the basis for the table entries. Major differences between the 1990 and earlier panels include: In the 1984-1986 panels, values for almost all assets and liabilities were obtained twice a panel, because the asset and liabilities and real estate and vehicle topical modules were gener- ally asked twice a panel (see Table 3-13). Subsequently, these modules were asked only once a panel; however, selected assets were also asked about in the program eligibility set of modules. Prior to the 1990 panel, the annual income round-up topical module obtained recipiency and calendar-year income amounts separately for each interest and dividend-bearing asset (e.g., money market deposit accounts, municipal bonds, etc.).
ACHIEVING SIPP'S GOALS TABLE 3-2 Continued 51 aAsset value is asked every 4 months for those respondents who do not know the amount of interest received. bDebt or margin account on stocks is ascertained only once a panel. CInformation on IRA and KEOGH accounts is obtained in the annual income round-up (asked in waves S and 8 in the 1990 panel), which ascertains contributions and earnings in the preceding calendar year, and in the asset and liabilities (wave 4) and program eligibility (wave 7) modules, which ascertain total market value. Information about contributions to 401K accounts is asked in the annual income roundup (waves 5 and 8), while information about total value is asked in the retirement expectations module (wave 4). N.A., Not applicable or not available. TABLE 3-3 Nonresponse Rates for Selected SIPP Core Items, by Panel Question 1984 1985 1986 Labor Force Activity Identification of weeks absent without pay 0.1a 0.1 Identification of weeks with a job or lousiness 2.22.0 2.5 Presence of weeks looking or on layoff 1.01.3 2.0 Identification of weeks looking or on layoff 3.22.4 2.9 Income Recipiency or Asset Ownership Social security 0.60.6 1.0 Unemployment compensation 0.10.1 0.2 Food stoops 0.30.4 0.5 Savings accounts 1.00.9 0.9 Shares of stock 1.31.4 1.5 Income Amounts Hourly wage rate 9.510.4 10.8 Monthly wage and salary 6.27.2 8.4 Self-employment salary or draw 14.016.9 14.6 Social security 8.S9.5 10.0 Unemployment compensation 9.19.7 9.9 Food stamps 3.64.1 4.4 Interest 34.629.8 30.8 (24.2)b(28.9)b (30.2)b Dividends 9.410.5 9.4 (30,7)C(30,5)C (29.1)C aLess than .05 percent bFigure in parentheses is the nonresponse rate on balance in the account. This question was asked of people with savings acccounts who did not provide an estimate of the amount of interest received (e.g., the 34.6% in the 1984 panel). CFigure in parentheses is the nonresponse rate for dividends credited to accounts. SOURCE: Jabine, King, and Petroni (1990:Table 5.5).
52 U U et ~ ~ V~ ~ o ~Cl X C o ~ ~ I ,= ~=, ~ ~ ~ ~ ~ ~ ~ ~, ~ ~r~ ~oo ~In _ II ~ ~_ _ _ j = i1 l E) ', < ~ c _ ~v~ ~ x ~ox E _4 ~ o t_ ~oc _ ~C o ~ =} - ~_ - o C ~o~ ~o mo ~,,- -= X ~ ~ ~ ~ X _ ~ ~ ~ ~- o ~j o=sOD o ~ = ~ X ~ X _ _ C~ ~o ~ ~ C~ I ~ ~ ~ ~ ~ ~ o ~ ° __ ~ __ ~E ,,, . ~ ~ , 3 i, ' - ~ c ~ CC . . o - ._ o - c) cn tL) ~D ,~., o o ._ - ._ . . Cd o V'
53 en C) ~ V o i_ hi ~ , c; ! -O 1 - V, As: Cal To ~ Do tn C.o _ A) C) 1 ~ c: I .= do ~== ! oo i Ct == = Do or ~ ~- C.) PA C) Pa ~5 a C~ -~ ~1 ~ ~ ,~ ~O~ C5 C~ ~s=~ V, _ ~: o C~ C) o z - m ~ crxu~ ~ ~ 0 -oo . .. .. . . .. ~D -~ ~d oo - -oo _ c~ ~ _ _ _ ~ _ _ 1- ) ~u, ~ . .. . 00 ~00 ~ __ ~ O ~ ~ . . . .. oO kD ~ ~ _ _ _ ~ ~O ~ ~ ~ ~ . .. . . . . .. 00 ~00 C ~00 ~ \0 ~ __ ~_ __ t- -}t- ~) - ~ - ON ~ ~oo ~x u~ ~ u~ - - -- oo o~ o c~ o o -- ~ r~oo ~x tn - - - - - c) & ;- c) o c) - ~ - o~ o = ~-2 ~ E c~ E '~ E ~ V - ~ _ ~ c ~ ~ E, ~ a E' O ;~. E c . a c ~o c' c, ~cs ._ c~ o c~ oo eD ;~ D C) - o CO o C~ o O - ; - - ._ Ct o ._ - CO 5= o o C) ~_ ~O ._ _ t5 _ O C) D . . 0\ C) C) & ~ ~ C) O C) ~ U) O ~ C') O 1 3 c~ ~ 4- s: o 0 ~ Z .° _ cc ~ C: O Z x ._ o - c, ~o ct ._ ._ D _} . . V ~: o V,
54 THE SURVEY OF INCOME AND PROGRAM PARTICIPATION TABLE 3-6 1984 SIPP and March 1985 CPS Estimates of Income Aggregates, for 1984 by Type (in billions) Percent SIPP as a Imputed Percentage Type of IncomeSIPPCPS of CPS SIPP CPS Total money income$2,419.7$2,417.4 100.1 11.4 N.A. Regular money income2,414.62,417.4 99.9 11.4 20.1 Earnings1,872.51,906.0 98.2 10.0 18.9 Public and private transfers33S.2300.2 111.6 12.1 20.7 Property income200.9194.9 103.1 23.9 32.4 All other regular money income6.016.3 37.0 15.4 22.3 Lump-sum payments5.1N.A. N.A. 5.6 N.A. N.A., Not available. SOURCE: Jabine, King, and Petroni (l990:Table 10.8). In addition, nonresponse rates for asset values and income from assets are high in SIPP; see Tables 3-3, 3-9. Also, the reporting of many types of asset values and income from assets is far from complete in comparison with estimates from independent sources; see Tables 3-10, 3-11, and the discussion below on improving the measurement of asset information in SIPP.4 In addition, SIPP leaves out some types of economic resources (e.g., many fringe benefits). Finally, SIPP (as is true of household surveys generally, including the March CPS) obtains less than complete coverage of such groups as young black men; see Table 3-12. Population undercoverage may adversely affect SIPP estimates of low income and program participa- tion, given the evidence that undercoverage is worse for people who are economically at risk (see Chapter 7 for further discussion). SIPP needs to make yet further improvements in available information on economic resources, using a broad definition as a guide. Realistically, of course, pnonties will need to be set. Such priority setting should take account of SIPP's focus on the economically at-nsk population and also of the fact that some kinds of resources are inherently harder to measure in household surveys than others. Hence, costs of obtaining information must be balanced against potential benefits, which include the importance of the resource (i.e., the proportion it represents of total resources for all people and for subgroups): a resource that is hard to measure and, based on avail 4It is important to keep in mind that conceptual differences and the existence of errors in the independent estimates may affect the validity of comparisons between aggregate information from surveys and from other sources (the latter typically include program records, tax records, and estimates from the National Income and Product Accounts [NIPA]).
ACHIEVING SIPP'S GOALS 55 TABLE 3-7 1984 SIPP and March 1984 CPS Estimates as Percentages of Independent Estimates of Income Recipients and Amounts for Selected Income Types SIPP as a Percentage SIPP as a Percentage CPS as a Percentage of the Independent of the Independent of the Independent Estimate of Average Estimate of Quarterly Estimate of Annual Income Source Monthly Recipients Income Income for 1983 Wage and salary 3rd quarter 1983 N.A. 95.0 99.0 4th quarter 1983 N.A. 94.3 1st quarter 1984 N.A. 93.2 2nd quarter 1984 N.A. 94.4 3rd quarter 1984 N.A. 95.2 4th quarter 1984 N.A. 94.5 Federal Supplemental Security Income 3rd quarter 1983 92.0 89.8 84.9 4th quarter 1983 91.3 93.5 1st quarter 1984 94.8 96.4 2nd quarter 1984 98.2 97.4 3rd quarter 1984 98.3 98.6 4th quarter 1984 98.1 99.2 Social security 3rd quarter 1983 99.2 99.6 91.7 4th quarter 1983 96.3 100.6 1st quarter 1984 97.3 100.5 2nd quarter 1984 97.7 101.1 3rd quarter 1984 97.5 101.3 4th quarter 1984 97.5 101.6 Aid to Families with Dependent Childrena 3rd quarter 1983 78.5 76.2 76.0 4th quarter 1983 79.2 78.5 1st quarter 1984 84.5 85.3 2nd quarter 1984 86.0 86.0 3rd quarter 1984 82.0 80.2 4th quarter 1984 80.7 78.8 Food stamps 3rd quarter 1983 89.5 90.1 71.2 4th quarter 1983 91.0 83.1 1st quarter 1984 90.8 85.2 2nd quarter 1984 90.5 86.2 3rd quarter 1984 90.3 84.6 4th quarter 1984 91.7 83.6 Veterans' compensa tion or pensiona 3rd quarter 1983 89.2 78.9 63.3 4th quarter 1983 89.7 79.9 continued on next page
56 TABLE 3-7 Continued THE SURVEY OF INCOME AND PROGRAM PARTICIPATION SIPP as a Percentage SIPP as a Percentage CPS as a Percentage of the Independent of the Independent of the Independent Estimate of Average Estimate of Quarterly Estimate of Annual Income Source Monthly Recipients Income Income for 1983 Veterans' coned 1st quarter 1984 90.6 78.0 2nd quarter 1984 90.8 74.5 3rd quarter 1984 89.8 76.3 4th quarter 1984 93.3 79 7 aRecipients exclude dependents covered by payments. N.A., Not available. SOURCE: Jabine, King, and Petroni (l990:Table 10.1). able evidence, makes a small contribution to total resources would have lowest priority for any type of measurement work. Income and Related Measures Measures Relevant to Programs In our view, a priority goal for SIPP should be to provide high-quality measures of the income and assets of relevant economic units in order to determine accurately the incidence, distribution, and temporal changes of program eligibility. Relatedly, SIPP should give priority to measuring rel- evant types of income and other resources that are of interest when examin- ing participation in government programs. It is important not to lose sight of SIPP's focus on program eligibility and participation: it is ultimately more important that SIPP measures well the resources that are relevant for people who are economically at risk than those that are relevant primarily for people who are unlikely to become eligible for programs. We have identified several important implications of a focus on pro- grams. First, SIPP should take a different approach to the measurement of assets and asset income one that is designed to obtain good measures on a more frequent basis for use in determining program eligibility but explicitly accepts some loss of detail for the remainder of the population (see below for our suggested approach). Second, SIPP needs to have the best possible measures of the extent of program participation and the amount of benefits received in order to sup- port analysis of program participation rates and to compare the characteris- tics of participants and eligible nonparticipants. To this end, we support research on the use of administrative records data for major assistance pro
ACHIEVING SIPP'S GOALS 57 TABLE 3-8 Comparisons of SIPP 1984 State Unemployment Compensation Estimates with Estimates Denved from Independent Sources (recipients in thousands, amount in millions) SIPP as a Percentage of SIPP Independent Estimatea Independent Estimate Period RecipientsAmount Recipients Amount RecipientsAmount 1983 3rd quarter 3,084$1,287 3,056 $1,259 100.9102.2 4th quarter 2,8781,193 2,784 1,117 103.4- 106.8 1984 1st quarter 2,9821,206 3,608 1,415 82.685.2 2nd quarter 2,212897 2,682 1,079 82.583.1 3rd quarter 1,927762 2,456 949 78.580.3 4th quarter 2,462978 2,590 969 95.1100.9 1985 1st quarter 3,2251,393 3,771 1,470 85.594.8 2nd quarter 2,220927 2,872 1,193 77.377.7 3rd quarter 1,917783 2,633 1,078 72.872.6 4th quarter 1,981854 2,506 1,103 79.177.4 NOTE: Monthly averages for specified quarters. aExeludes federal supplemental compensation. SOURCE: Jabine, King, and Petroni (1990:Table 10.2). grams (e.g., AFDC, food stamps, SSI, unemployment insurance) to evaluate and suggest ways to improve SIPP reports for these sources (see last section of this chapter). Finally, the Census Bureau should undertake a thorough reexamination of the procedures that are used to edit program data obtained in SIPP and to impute missing income and asset information for program recipients. The SIPP record-check study found that substantial underreporting of receipt of AFDC benefits in Pennsylvania was largely because respondents confused AFDC with the general assistance program in that state (Marquis and Moore, 1989:522~. Cantor et al. (1991) provide evidence of confusion among pro- gram names as well. To the extent that improved questions and field proce- dures do not eliminate the problem, we suggest that editing procedures be developed to correct these kinds of reporting errors.s Similarly, there is 5It should be feasible to develop such procedures in many eases. For example, given different eligibility rules, it is likely that families with children who report receipt of general assistance or other welfare in feet received AFDC, while single individuals who report AFDC in feet received general assistance, other welfare, or some other type of assistance. Coder and Ruggles (1988) provide examples of the types of edits that could be carried out to correct obvious errors in reports of AFDC and general assistance in SIPP.
58 THE SURVEY OF INCOME AND PROGRAM PARTICIPATION TABLE 3-9 Item Nonresponse Rates for Asset Amounts in SIPP and the Income Survey Development Program (ISDP) 1979 Research Panel Nonresponse Rate (percent) SIPP Asset Type 1984 1986 ISDP Amount in savings accounts 16.8 22.8 24.9 Amount in checking accounts 13.3 21.2 23.1 Amount in bonds and government securities 25.9 23.1 32.2 Market value of stocks and mutual fund shares 41.5 36.9 65.8 Debt on stocks and mutual fund shares 41.1 38.6 87.3 Face value of U.S. savings bonds 24.9 24.9 35.8 Value of rental property 33.5 31.3 39.9 Value of own business 37.9 41.8 55.3 Debt on own business 28.8 31.5 50.4 NOTE: Nonresponse rates for SIPP are from waves 3 and 4 of the 1984 panel and wave 4 of the 1986 panel. SOURCE: Jabine, King, and Petroni (1990:Table 5.8). evidence that the current procedures for imputing income and assets to program beneficiaries (and low-income people, generally) are deficient.6 Information obtained from record-check studies that compare SIPP reports with administrative data could be used to investigate the characteristics of nonrespondents, to evaluate the effects of current imputation procedures, and to provide guidance on possible improvements. Measures of Taxes and After-Tax Income Another priority goal for SIPP is to develop accurate measures of federal, state, and local taxes. Such measures are needed to calculate after-tax (disposable) income for comparisons of resources for consumption across the population. Such measures are also needed to study eligibility and use of transfer programs that are administered through tax credits (notably, the earned income tax credit). Finally, the availability of good tax measures will permit analysis of the distribution of the tax burden on families with 6The Census Bureau's "hot-deck" imputation procedures do not take account of poverty or receipt of program benefits in imputing specific income and asset values (although they do take account of income levels in broad terms). Hence, the imputations create instances of people who report receipt of benefits from such programs as food stomps but who have imput- ed incomes or asset values that are too high for them to be eligible (see Doyle and Dalrymple, 1987; Allin and Doyle, 1990).
ACHIEVING SIPP'S GOALS TABLE 3-10 1984 SIPP, March CPS, and Independent Estimates of Aggregate Interest and Dividend Income, 1983-1984 and Calendar 1984 (in billions of dollars) 1984 Income Source 1983-1984 Reported (NIPA-based) NIPA-based to IRS Interest Independent estimate 5239.1 $254.6a $176.4 Survey estimates SIPP $115.7 5115.4 CPS Original imputation $104.7 $109.2 Revised imputation $129.5 $138.7 Survey estimates as a percentage of the independent estimate SIPP 48.4% 45.3% 65.4% CPS Original imputation 43.8% 42.9% 61.9% Revised imputation 54.2% 54.5% 78.6% Dividends Independent estimate $63.6 $66.5a $5o.6c Survey estimates SIPP $38.3 $40.3 CPS $29.1 $30.7 Survey estimates as a percentage of the independent estimate SIPP 60. 1% 60.7% 79.7% CPS 45.8% 46.1% 60.6% aPersonal income aggregate from the National Income and Product Accounts (NIPA) ad- justed to the survey universe based on observed relationship between NIPA aggregate and independent estimate for the CPS universe in 1983. bThe revised imputation corrects for bias in the pre-1983 imputation procedure by making use of matched Internal Revenue Service information on interest income. CTotal domestic and foreign dividends received. SOURCE: Jabine, King, and Petroni (1990:Table 10.3). 59 different gross incomes. In this regard, we note that tax burdens can be significant for such groups as the working poor and near-poor who are part of the economically at-risk population and hence of concern to SIPP. Pechman (1985) estimated that the individual income tax rate for the bottom one- tenth of taxpayers was about 4 percent in 198S, compared with 1 percent in 1966, and that social security payroll taxes had doubled for this group over the same period (although recent tax changes have reduced income taxes for the poor and increased offsetting tax credits). For all these reasons, we support the high priority that the Census
60 so Ct AS en C) so: - m m - Ct o he - Ct C) So: 0 1 Cal _ ~ a . _ _, t: O ~ O U. ~ ~ _ Ce 00 _ ax Cal _ ~ Ct - = ~0 '~ S~ O S: ~ Ct ~ O C~ ~ ~ ~Q _ - -I 4 .= ~ O _ - 4 m~ ~ _ I c ~V 1 - ~ V) ~7 0 _ ~0= ! ~_ ~, O _4 ~ - C~ , _ 1 ce o~o ~ 1 -I ~ ~ ~ oo ~ ~ ~ ~ \o _ _ _ o ~ _ 0 oo ~ . . kD U) ~) ~ 1 _ _ O - - O O Z Z o 0 0 - 1 ~ O ~ ~ O t- ~ ~ O ~ U~ ~ U~ 1 x ~ ~ o-0 ~ ~ c~ ~ c~ ~ ~o 1 C~1 C~; O t~ d- ~ ~ ~ c~ - ~ - ~ 1 O ~ 00 ~ _ _ _ _ C-) c~ _ ~ _ I ~ O 0m ~ 0= ._ ~ ~ ~ _ CO _ 1 ~4 s° u, 6 1 C) e s:: CN ~ rL C~) ~ ~ 1< 0 oo ~ ~ ~o tn ~ ~ ~ ~ ~ 0 ~ ~ ~ ~ _ oo ~ ~ x ~ . . ~ u~ ~ 0 1 ~ ~ 1 --o --oo ZZ o o o- ~oo ~ ~ ~ ~ oo ~ ~ ~o ~- ~ ~ oo . . . . . . . . . . . . . o ~ 0 oo oo O cr~ 0 ~ ~ ~ ~ 1 ~ ~ - } - t.~ d oo O t- ~ ~ '= ~ 1 x as _ ~ U~ _ ~ ~ r~ oo ~ O - ~ ~ _ _ C~ _ _ ~g '_ _ o ~o ~ _ ~ V} ~ ~ o . . . . . . . . . . . . . . ax ~ ~ ~ ~ x ~ c~ x 0 - o ~ ~ ~ ~ ~ . . ~ ~ o m) ~ c~ C~ d -~ Z Z X- C~ ~ _ ~ ~ _ os ._ C~ C) O ~Z (_ '- C, '- ,, _ ~8,, . ~ e~ ,_ o -= ._ O D ,_ ~ ~ te _ o _ ~ =oa ~vo =~o.= =oo~ . . V ~ I ~ O Ct O ~ X ~ °° 1 ~ 3 C) C¢ oo ~ - U ~O ~ ~=0 - :: O c) C) Ct .= O ,.5C _ =: ~ C ~O s~ _ ~ oo o O oo _ O O ~ I C~ C) ~ ~ Ct ,C) ~o ~ '- ~ Ce X ~- C:} X _ e._ _ O ~ O - ~ ° C) ~ C 1 _ ~ `: O _ O o ~ CO ~ ~ O ~ Ce _ - C;S d .~: x 0\ ~tV'= O ~ ~ ,) Ct C.) - C) O ~ -O ~ c ~Co _ O ~ CD C;5 ~ ~ ~ - D o O ,= ~ ~ ~ >, O ~ 0 (L) S:: c~ Ct ~ 8 ~ <,, ~ o ~ D 0L2_ co _ ~ V, ~ ~ ~ ~ _ ~ _ _ ~ o 3~33= Z ~ . ~ ~ ~ 5) cn ,_ o ~ C~ o + - ~ 6 _ Ct _ ° ~ ~ ~ _ ~ D~ t;S > V) - _ ~ . . C, - Ct ._ C~ - C'; - . _ . C~ _ C) ~o - .O Ct _ C) ~ v, - o Z 6 z 1 D . . o cr, - oo C~ . . o V,
ACHIEVING SIPP'S GOALS 61 Bureau staff have assigned to developing a tax model for SIPP. We note, again, that this work can benefit greatly from the use of administrative records. SIPP (unlike the March CPS) includes a tax module, but the qual- ity of the information collected for many key items is poor (e.g., filing status and the kinds of forms filed are well reported but not amounts of deductions, adjusted gross income, or taxes paid). By taking advantage of IRS information, there is an opportunity to obtain higher quality estimates of after-tax income and at the same time reduce the level of detail on taxes that SIPP tries to obtain directly from respondents. Currently, the IRS provides the Census Bureau with a limited set of items from the complete Individual Master File (IMF) of tax returns. The Bureau has carried out and plans further matches of these data with SIPP (and March CPS) records for evaluation and methodological research pur- poses (see last section). However, the IMP data have not been subject to the quality control that is applied to a sample of tax returns drawn each year by the Statistics of Income (SOI) Division of IRS. Moreover, the IMP information furnished to the Census Bureau does not include taxes paid or sufficient information to calculate taxes.7 We believe that it is very important for the Census Bureau and the SOI Division to work together to obtain high-quality estimates of after-tax in- come for SIPP. Such cooperation would be very cost-effective for the SIPP program and could also provide useful information to support the tax policy modeling and research that is carried out by the Treasury Department. One possible approach is for the Census Bureau to have the SOI staff use social security numbers (SSNs) to pull the tax records of SIPP respon- dents, clean them in the same manner as is done for the SOI sample, and return the information to the Census Bureau to use in estimating after-tax income. Such an approach would be very powerful in terms of providing accurate tax data, including the filing versus r~onfiling status of all SIPP respondents an important piece of information for tax policy modeling. However, a major obstacle to implementing this approach concerns the pro- tection of confidentiality. Fully adequate procedures would have to be worked out to ensure that there could be no opportunity to identify SIPP respondents for tax enforcement or other purposes. Another possibility is for the IRS to provide samples of tax returns from the SOI files that would be included in SIPP panels (i.e., to develop a multiple-frame sample for SIPP see last section). The linked tax and survey information for these cases would permit much more accurate mod- elin~ of taxes for the regular SIPP respondents than is currently possible. O ~ 7Items released by the IRS to the Census Bureau include type of return (e.g., joint or single), wage and salary income, interest income, and dividend income. Recently, after a lengthy negotiation process, the IRS agreed to provide total adjusted gross income as well.
62 THE SURVEY OF INCOME AND PROGRAM PARTICIPATION TABLE 3-12 Coverage Ratios for SIPP and CPS Samples: March 1984 and March 1986 March 1984 Male Female CPS SIPP CPS SIPP Age Group BlackNonblackBlack NonblackBlack NonblackBlack Nonblack 16- 17 .9485.9390.9650 .9504.8672 .96741.0374 .9557 18-19 .9133.8955.9302 .9667.8763 .9094;9021 .8825 20-21 .7465.8865.8862 .9214.8190 .9139.8698 .9664 22-24 .6561.8615.6433 .8144.8483 .8845.7929 .8838 25-29 .8029.9065.7419 .8461.9069 .9278.9205 .9283 30-34 .7054.9079.8701 .8957.8487 .9499.9335 .8855 35-39 .7677.9109.7294 .8711.8441 .9465.8489 .9022 40-44 .9043.9286.8770 .8868.9793 .9376.8652 .9446 45-49 .8630.9215.7576 1.0039.9500 .96781.1315 .9930 50-54 .8418.9595.9355 .9378.9048 .9791.7172 .9718 55-59 .8302.9604.9864 .9267.8943 .9476.8494 .9361 60-61 1.0034.9622(.9265 .9637.9675 .91331.0557 .9801 62-64 .8591.9261 .9352.9329 .9500[ 1.0363 .9345 65-69 1.0990.9335f.9589 .94001.0704 .9506 .9839 70-74 ( .8942( .9289l 9457[ 1.0186 (.9450[ 1.0492 .9178 75 79 .9206 .9184 80-84 1.0135.9266.9733 1.0358.9804 .9384.9380 .9517 85 and .7831 ~.9759 Total 16 .8350.9193.8460 .9095.9012 .9407.9172 .9330 and over NOTE: Coverage ranos (see text for definition) are similar for other months. In any event, the burdensome tax module in SIPP could be scaled back to include only those items (e.g., filing status, type of return, number of de- pendents) that facilitate linkage with IRS records. Valuation of In-Kind Benefits Still another high-pnonty goal that the Census Bureau staff have set to improve income and related statistics from SIPP, which we endorse, is to develop measures of the value of in-kind benefits.8 Noncash transfers, both 8In-kind benefits are sometimes literally in kind, such as the provision of a public housing unit to a low-income family; more often, they are rendered in the form of coupons (e.g., food stamps) or cash payments to third parties (e.g., reimbursement for heating, child care, or
ACHIEVING SIPP'S GOALS 63 March 1986 Male _ CPS SIPP Female - CPS SIPP Black Nonblacka Black . . 0.9215 0.9739 ~ 0.8973 0.9713 0.7435 0.8915 0.7955 0.7976 0.8685 0.7424 0.7229 0.8893 ~0.9259 0.7276 0.8946 [ 0.5786 0.7869 0.7559 0.9094 0.8270 0.8501 0.9390 , 0.8669 0.9627 0.8608 0.9349 0.9377 0.9479 0.9186 0.9826 0.9211 0.9461 ~0.8987 0.7999 0.9639 0.8736 0.9897 0.7097 0.9162 1.0598 0.8574 0.8830 1.1076 0.8596 0.9367 0.9805 1.0065 0.9792 1.1712 0.8679 0.9596 ~ 0.9953 Nonblack Black . Nonblacka Black Nonblack 0.8891 0.9542 0.8620 0.8821 0.7809 0.8804 0.7913 0.9138 0.8659 0.9389 0.6960 0.9037 0.9446 ( 0.8555 0.9783 0.8344 0.9511 0.9483 0.8587 0.9298 0.8968 0.9398 0.8651 0.9488 r 0.8970 0.9528 0.9174 0.8665 0.9695 0.9506 0.9661 1.0014 0.9986 1.0929 1.0029 1.0208 0.9876 0.1342 1.0013 0.8732 0.9812 0.8788 0.9497 0.9970 1.0340 0.9695 0.9225 0.9517 1.1853 1.0689 1.1253 1.2688 1.0364 0.9912 r 0.8502 0.8172 0.9259 0.797 0.9252 0.8932 0.9485 0.8892 0.9788 aCPS coverage ratios are for whites. Hispanic persons may be classified as either white or black: most probably fall in the white category SOURCE: Jabine, King, and Petroni (1990:Tables 10.12, 10.13). public and private, have expanded dramatically over the past 20 years. The food stamp program did not even exist at the national level in 1970; by 1990 it was annually providing more than $15 billion in benefits (Congres- sional Research Service, 1991:198~. Spending on Medicare and Medicaid has also grown enormously since 1970. Some of this increase reflects rising prices for health care services, but there has also been a major expan- sion in the receipt of health benefits. Privately provided fringe benefits, such as health insurance and pension coverage, also have become much more common, and they are now estimated to represent as much as one- fourth of employers' total labor costs (Levitan and Gallo, 1989:14~. medical bills). The key distinction is that in-kind benefits can be used only to obtain the designated good or service and not for other consumption of the recipient's choice.
64 THE SURVEY OF INCOME AND PROGRAM PARTICIPATION The difficulty comes in valuing in-kind benefits.9 Some publicly pro- vided benefits, including food stamps and public housing, are relatively easy to value, since the cost of the benefit to the government (its "market value") and the value to the recipient ("fungible value") are known to be close.l° Much more problematic is the valuation of benefits from Medicare and Medicaid. As noted by Census Bureau staff (Bureau of the Census, l991a:89: There is great controversy about the valuation of benefits from Medicare and Medicaid, caused in part by the fact that market values of benefits can exceed the poverty line (intended to cover all basic needs) and are always a very large proportion of the poverty line. A second problem is that per- sons can [appear to] gain income by moving into a risk class with a high level of medical need (e.g., the aged or the disabled). There are fewer problems with the valuation of employer-provided medical benefits be- cause employees have chosen to receive part of their compensation in this form. For several reasons because access to medical care is an issue in its own right, the public's willingness to subsidize medical care exceeds its willingness to provide cash income, and the tax-exempt status of em- ployer-provided health insurance premiums is a major benefit to many people- it is clearly important that SIPP continue to regularly collect complete in- fo~mation on the health insurance unites) in households and the existence of health insurance coverage. It is also important that the Census Bureau continue to investigate alternative ways of attempting to incorporate a value for health insurance coverage (public and private) into the SIPP measure of income. We note, however, that some analysts have expressed caution on this matter. For example, Watts (1991:2) argues that it is difficult to know how to allocate the incidence of public medical care transfers "between direct beneficiaries, their extended families, suppliers of medical services, and the common weal." Watts suggests that, instead, there might be a deduction from income for people lacking any health insurance coverage. Another conservative alternative might simply be for SIPP income statistics to show 9See Bureau of the Census (1982) for seminal work by Smeeding on valuation methods for in-kind transfers; see also the proceedings of a conference that was held on this topic (Bureau of the Census, 198Sa). loin addition to food stamps and public housing, the March CPS expanded income estimates currently include the value of school meals. For SIPP, there seems no reason not to include home heating assistance as well. 1 1 Government spending for medical care for low-income people is currently 55 percent greater than government spending on cash assistance (Congressional Research Service, 1991:Table 6). 12For a description of the methods currently used to value benefits from Medicare, Medic- aid, and employer-provided health insurance for inclusion in the March CPS expanded mea- sures of income, see Bureau of the Census (199lb).
ACHIEVING SIPP'S GOALS 65 routinely, for each income class, the proportion of people lacking health insurance (see, e.g., Aaron, 1985). Smeeding (1992) suggests that other nonpension employer-provided fringe benefits may add appreciably to the resources of employed people almost as much, by his calculations, as the value of employer-provided health in- surance.l3 However, some of these benefits (e.g., employee subsidies for education) raise the same problems of valuation as public subsidies for education and health care, while others (e.g., employee meals furnished in lieu of pay) are likely to prove hard to measure in terms of their incidence, let alone their value. One strategy would be to occasionally-include a topical module in SIPP on fringe benefits,~4 and use the results together with relevant data from other sources (e.g., the Survey of Consumer Fi- nances) to determine those fringe benefits that appear to be most wide- spread and potentially most important for an expanded income measure. Research on measurement and valuation methods could be targeted on these benefits. One type of in-kind benefit provided by both the public and private sectors that is of growing policy importance is subsidies for child care. Currently, SIPP obtains measures of out-of-pocket costs for child care in the child care topical module. However, there are no questions about govern- ment or employer-provided subsidies for such care even though more and more parents are apparently opting to use some form of paid child care.l5 It would be useful for SIPP to obtain information about private and public child care subsidies, particularly since the Omnibus Budget Reconciliation Act of 1990 established a new child care and development block grant, which provides funds to states to subsidize child care services for working poor families (Congressional Research Service, 1991:173~. Such informa- tion would be in keeping with SIPP's focus on programs and would provide data that could be used if SIPP subsequently provides a broader income measure that includes valuations for child care subsidies. lain Smeeding's analysis, such fringe benefits include life insurance, short-and long-term disability insurance, accident insurance, dental insurance, employee subsidies for education, discounts on goods and services, employee meals furnished in lieu of pay, and profit-sharing and thrift benefits. Not included in his analysis are travel and entertainment allowances, other perquisites (e.g., free use of company cars, club memberships, tickets, etc.), and child care subsidies. 14Wave 6 of the 1984 SIPP panel included a fairly comprehensive module on fringe bene- fits, but there has been no such module in subsequent panels. 150n the basis of CPS and SIPP data, the distribution of primary child care arrangements used by employed mothers for children under 5 in 1977 was 56.7 percent by relatives (includ- ing care by mothers who worked out of their home), 29.4 percent by nonrelatives, and 13.0 percent in day care or nursery school; the comparable figures in 1987 were 45.9 percent by relatives, 28.5 percent by nonrelatives, and 24.4 percent in day care or nursery school (O'Connell and Bachu, l990:Table C).
66 THE SURVEY OF INCOME AND PROGRAM PARTICIPATION Static and Dynamic Measures Related to Family Characteristics A basic goal of SIPP should be to collect reasonably detailed demographic and socioeconomic information (e.g., ethnicity, family composition, marital and labor force status) that permits comparative analysis of the distribution of income and other economic resources for different types of people and families, households, and other economic units. SIPP should support both cross-sectional analyses and longitudinal analyses that examine changes in resources across time as they are associated with changes in other character- istics, such as family composition. The SIPP questionnaire has generally included a wealth of demographic and socioeconomic information: many items are tracked on a monthly basis in the core interviews, and others are asked occasionally in topical modules. Items of this type should be reviewed periodically to determine whether all of the detail is still needed or, conversely, whether some additional detail would be helpful in order to add explanatory power to SIPP. The weekly information on employment status may be a candidate for reduction in scope, for two reasons. First, the monthly CPS, not SIPP, is the nation's labor force survey. Second, there is some evidence that the de- tailed labor force questions, which come at the beginning of the interview, may reduce the likelihood of respondents answering the income questions (Cantor et al., 1991~. Labor force data clearly need to be collected in SIPP because labor force characteristics are important contextual variables for analyses of income, program participation, and related topics. However, the level of detail needed for such analyses may be considerably less than that currently collected. Conversely, the addition of a few simple questions about the family of origin for example, parents' education, number of siblings, and whether one grew up in an intact family-could enhance SIPP's ability to estimate the distribution of economic risk within the population and contribute to the understanding of persistent poverty and welfare participation. (These kinds of questions were asked in the 1986-1988 panels, but they were subse- quently dropped.) To date SIPP has not had great success in developing good, understand- able indicators of changing characteristics to use to present longitudinal statistics on income and other resources in reports for example, how to characterize the annual income of a "family" in which the spouses divorced midyear (see discussion in Chapter 6~. A related problem, confronted gen- erally by researchers working on issues of income distribution, is determin- ing an appropriate set of equivalence scales to use for comparative analysis of income as a measure of well-being for example, to answer the question of how much more income is required to provide an equivalent level of
ACHIEVING SIPP,S GOALS 67 consumption for a four-person rather than a two-person family. We discuss some interim solutions to these problems for SIPP publications on income and related measures and outline areas for research in Chapter 6. New Forms of Income and Other Economic Resources As we have noted, economic resources are not static over time. Thirty years ago, the government provided hardly any in-kind benefits to low- income people; in fiscal 1990, in-kind benefits (medical care, food, hous- ing, education and training, services, and energy assistance) exceeded ex- penditures for cash aid by a ratio of 2.8 to 1 (Congressional Research Service, l991:Table 6~. Similarly, fnuge benefits have clearly become an increasing share of a worker's total compensation over time. There is an inevitable tension in any income survey between the need to add new questions to reflect the changing composition of economic re- sources and the need to maintain the comparability of time-genes data. We strongly suggest that the Census Bureau strive for flexibility in SIPP's mea- surement of economic resources and, as Smeeding argues, be more innova- tive and willing to experiment in this area than has been done in the March CPS or the initial phases of SIPP. One specific mechanism to achieve flexibility and innovation without unnecessarily disrupting the core survey is to make use of topical modules. In addition to the topical modules in each panel that are open for federal agencies and other outside users to specify questions on topics of emerging policy interest, there could well be another module that is for the use of the Census Bureau analysis staff to develop new measures that relate to the core subjects in SIPP, namely, income and program participation (We explore this idea in the section on topical modules, below.) Recommendation Recommendation 3-1: Priorities for improved income and re- lated measures from SIPP should include: · enhancing the quality of income and related measures that are relevant to program eligibility and participation; · developing measures of taxes and after-tax income; valuing (or otherwise taking account of) in-kind benefits; . and · constructing income and related measures that take account of family characteristics and changes in families over time. SIPP should also keep up to date with respect to new forms of income and other economic resources.
68 THE SURVEY OF INCOME AND PROGRAM PARTICIPATION Assets and Asset Income We believe that it is imperative for the Census Bureau to reassess how assets and asset income are measured in SIPP. Asset-related information is needed in SIPP for two purposes. The first is to permit accurate determina- hon of eligibility for means-tested assistance programs, which typically impose a 100 percent tax on unearned income and greatly limit the type and value of assets that people can own and remain eligible for benefits. The second is to provide information, including periodic revaluation of house- hold balance sheets in SIPP, for measuring the economic resources of fami- lies and households under the broad definition outlined earlier (i.e., a defi- niiion that includes resources, in addition to cash income and in-kind benefits, that can be used to support future consumption. The contribution of assets to resources that are potentially available for consumption can be substantial for some subgroups.~7 There are a number of conceptual and technical issues associated with valuing assets for inclusion in an expanded definition of income. In the case of such financial assets as stocks and bonds, one approach is to assign an annuity value that is added to the owner's cash and in-kind income (after first subtracting any reported income from these assets so as to avoid double counting). However, the results from this approach will be sensitive to the period over which the annuity is calculated. Alternatively, people in van- ous income categories could simply be cross-classified by whether their asset holdings are above or below a specified amount. In the case of nonfinancial assets, such as a home, one approach is to calculate the amount of rent the home would bring in the market, net of expenses related to home owner- ship. Another approach, which is used by the Bureau of the Census (199lb:Appendix B) for the March CPS expanded income estimates, is to impute a rate of return to the estimated amount of home equity. 16A third purpose for asset-related information is to provide detailed information for assess- ing the distribution and types of wealth in the United States: we believe this is neither essential nor feasible for SIPP. 17For example, the Bureau of the Census (199lb:Tables D,F,I) estimated that median house- hold income in 1990 was $1,900 higher and the national poverty rate 1.2 percentage points lower under a resource measure that imputed a return to just one type of asset-equity in own home. The differences were most pronounced for people aged 6S and older. Ruggles (199Oa:155) argues caution against estimating a return for the full value of home equity because it may not be feasible to convert all of that value readily to cash, although she generally supports the concept of valuing assets under a broader definition of economic resources. 18See Ruggles (199Oa:149-158) for a discussion of the pros and cons of various approaches for incorporating asset values into a measure of economic resources from the perspective of estimating the poverty population. Ruggles discusses intangible as well as tangible assets, the former including human capital or skills. See also Weisbrod and Hansen (1968) and Moon (1977) on the topic of asset valuation for purposes of measuring poverty.
ACHIEVING SIPP'S GOALS 69 We encourage the Census Bureau to investigate alternative methods of valuing assets so that, ultimately, expanded income measures that are devel- oped from SIPP can take account of asset holdings in some way. In the remainder of this section we discuss how SIPP can best obtain high-quality information about assets and asset income, which is a prerequisite for valu- ation or such other uses as determining program eligibility. Currently, SIPP devotes a considerable amount of space to asset-related items, including such financial assets as bank accounts, stocks, and bonds and such nonfinancial assets as houses and cars (see Table 3-2 for details). The core interviews ask a detailed battery of questions about asset owner- ship every 4 months. These questions require respondents to make fine distinctions (e.g., between a money market deposit account arid a money market fund and between jointly owned and individual holdings). Respon- dents are then asked to provide income amounts for a 4-month period for combinations of asset types (e.g., all types of interest-bearing joint bank accounts). There are no core questions about asset balances or liabilities (e.g., mortgage or credit card debt). Instead, these questions are asked twice or, in some cases, only once ire a panel's life, generally in the topical modules on assets and liabilities (including real property and vehicles) and program eligibility, although a few asset types (such as individual retire- ment accounts) appear in the annual roundup or retirement expectations modules. Nonresponse rates are low for the core asset ownership questions (e.g., about 1% for savings accounts and stocks) but generally high for the ques- tions on 4-month income flows (e.g., 30-35% for interest and 30% for reinvested dividends; see Table 3-3~. In a study of how respondents per- ceive and respond to the core SIPP questionnaire that used cognitive inter- viewing techniques, Cantor et al. (1991) found that few respondents know the amount of interest or dividend income they receive in a 4-month penod, that lower income respondents are offended by the asset questions, and that few respondents consult records but rather provide very rough "guessti- mates" in reply to the asset income questions (see further discussion in Chapter 7) After imputation for nonresponse, SIPP obtains an estimated 80 percent of the dividend income reported to the IRS (compared with 61% in the March CPS) and an estimated 65 percent of reported interest income (compared with 79% in the March CPS, which uses an improved imputation procedure).~9 i9The March CPS estimate of interest income using the old imputation procedure is only 62 percent of the IRS estimate. Both SIPP and the March CPS fall much farther short of dividend and interest income aggregates when the comparison is made to the National Income and Product Aeeounts (NIPA); however, the NIPA estimates require extensive adjustments, which may not be complete, for comparability with household survey estimates (see Table 3-10).
70 THE SURVEY OF INCOME AND PROGRAM PARTICIPATION Nonresponse rates to the questions on value of asset holdings in the topical modules are also very high, although lower than were experienced in the Income Survey Development Program (ISDP): 35-40 percent for value of own business, market value of stocks and mutual fund shares, and debt on these assets; see Table 3-9. After imputation, SIPP obtains higher esti- mates of equity in homes and motor vehicles in comparison with estimates of the Federal Reserve Board because of somewhat higher estimates of gross value and considerably lower estimates of debt in SIPP, but it obtains considerably lower estimates of equity In noncorporate business, value of financial assets, and consumer debt (see Table 3-11; Lamas and McNeil, 1986:Table D-3; Eargle, 1990:Table D-2~.20 Originally, the assets and liabilities module was asked twice in every panel. However, 1-year intervals for assessing balance sheets appeared to be too short to provide reliable measures of change in net worth (see McNeil and Lamas, 1989; David, 1989~. The Census Bureau conducted an expen- ment to determine if reminding respondents at the second administration of the module of their answers the first time would improve the quality of the data, but the results were inconclusive (Lames and McNeil, 1987; Weidman, King, and Williams, 1988~. Beginning in the 1987 panel, the assets and liabilities module (including real property and vehicles) was asked only once. However, the new program eligibility module instituted in the 1987 panel included asset valuation questions, thereby providing two measures per panel for most types of assets (but not most types of debt. The conclusion that we reach from reviewing the bewildering extent and variable quality of asset data in SIPP is that the Census Bureau should rethink the amount of detail and frequency of collection of asset informa- tion from the perspective of its two primary goals. For purposes of assess- ing program eligibility, the core questions on asset ownership and income flows are needlessly detailed. At the same time, the core interviews do not obtain asset valuation information that is required for eligibility determina- tion. For purposes of measunug the contribution of assets to total economic 20It should be kept in mind that a number of differences between SIPP and the Federal Reserve Board data limit the extent to which inferences can be drawn from comparisons of estimates from them. 21We note that SIPP is not alone is experiencing quality problems with the collection of asset data. A number of panel surveys provide estimates of wealth that fall short of those from the Survey of Consumer Finances (SCF), a complete survey of household wealth that includes a household sample together with a sample of high-income households drawn from the IRS Statistics of Income file who agree to participate. Curun, Juster, and Morgan (1989) reach this conclusion from comparing the 1984 SIPP panel and the 1984 round of the Panel Study of Income Dynamics with the 1983 SCF; Juster and Kuester (1991) reach the same conclusion from comparing the 1979 round of the Retirement History Survey and the 1981 round of the National Longitudinal Survey of Mature Men with the 1983 SCF.
ACHIEVING SIPP'S GOALS 71 resources, a periodic revaluation of the household balance sheet is needed. However, the current set of topical modules obtains a complete balance sheet only once in the life of a panel. One useful strategy may be to focus measurement in the core on those assets that are most relevant for program eligibility and most likely to be held by low-income households, such as checking accounts, regular savings accounts, houses, and cans. Measures of their value could be included in the core questionnaire instead of only once or twice a panel in topical modules. For houses and cars, it would suffice to ask whether the respon- dent still lived In the same house and still had the same vehicles, asking about value only if a change had occurred. A complete balance sheet, including valuation of other assets, should be obtained at an early interview perhaps the second or third interview, using the first interview to locate the key financial person in the family and establish information about record-keeping for later use, encouraging re- spondents to consult records. The use of an unfolding scale might also help minimize nonresponse to questions about asset holdings: that is, if a re- spondent answers "don't know" to a question about the value of an asset, a follow-up question could ask whether the value is above or below a certain amount. The balance sheet interview needs to clarify the subtleties of asset ownership (e.g., assets being managed for children's college education, etch. Also, as Slater (1991) urges, the balance sheet in SIPP should be con- structed with a view toward maximum comparability with the aggregate household sector balance sheet that the Federal Reserve Board prepares annually and with the National Income and Product Accounts.22 With regard to measuring asset income flows, questionnaire research may well show that people can more reliably indicate the balance in their savings accounts every 4 months than estimate their interest Income. In this case, imputation using current interest rates could provide estimates of equal if not better quality than asking about interest income directly. An approach of imputing income flows rather than measuring them directly may work for other asset types as well.23 The balance sheet interview can be used to obtain detail on income flows from assets that will provide an independent source of the rate of return imputed to such assets as savings accounts. The balance sheet mea- surement needs to be repeated but at an interval determined by the rate of change of asset values in relation to the level of response error. If response 22Slater (1991) notes that improvements in the Federal Reserve balance sheet and integra- tion with the NIPA are being addressed as part of the U.S. effort to conform to the United Nations System of National Accounts. 23A different measurement strategy may be needed for people out of the labor force who do not report periodic transfer income (i.e., people whose main source of income is from assets).
72 THE SURVEY OF INCOME AND PROGRAM PARTICIPATION error is as high as for the present SIPP, it will be more meaningful to measure assets at longer than yearly intervals. Our proposed redesign of SIPP (Chapter 4) would permit asset reevaluation twice each panel at an interval of 2 to 4 years. Whether or not our particular suggestions ultimately prove feasible, it is clear that the measurement of asset-related information in SIPP is ripe for rethinking and experimentation. Innovative ways need to be found to pro- vide the information that is needed to serve SIPP's primary goals while acknowledging the real limitations on what can be obtained from household respondents. Recommendation 3-2: The measurement of asset-related infor- mation in SIPP should be reassessed in light of SIPP's focus on programs and the economically at-risk population. The collec- tion of asset items should be redesigned and simplified, if pos- sible, to reduce respondent burden and improve the quality of the data needed to serve SIPP's primary goals. Other Resources There are some other types of resources that may be important to consider in developing improved income and related measures from SIPP. For ex- ample, home production, which can take many forms (e.g., home-grown food, housework, and child care), is a significant source of goods and ser- vices for many households. In particular, with so many mothers in the labor force, it may be important to recognize the value of unpaid child care contributed by parents or other relatives that would otherwise have to be purchased in the market. Similarly, it may be important to value the caregiving time that is contributed by relatives for children and for adults who need long-term care services. SIPP has included questions in topical modules about parent- and other kin-provided child care, long-term care providers for respondents age 15 and older, and the time that respondents contribute to assisting people out- side their household (see Harpine, McNeil, and Lamas, 1990~. We believe that continued collection of this kind of information, with the frequency determined by agency needs, could be important for many policy purposes. However, given the difficult measurement issues involved (see Ruggles, 1990a:147-149), we do not suggest that SIPP should go very far in the direction of valuing caregiving time (or other types of home production) for inclusion in an expanded measure of income.24 24We also do not suggest that SIPP attempt at this time to address the issue of income from the underground economy.
ACHIEVING SIPP'S GOALS 73 Smeeding and others have argued that SIPP should gather new data, not only on the incidence and degree of caregiving, but on such items as access to support from neighbors and kin, access to credit, and whether the house- hold owns standard appliances (e.g., a washing machine) that make it less dependent on the need to purchase services in the market. Many of these topics have to do with measuring the degree to which families and individu- als are economically secure against venous types of risk. To express the concept another way, these items have to do with measuring the access that people have to sources of economic support above and beyond the types of income and benefits that they currently receive. This type of information could make an important contribution to understanding poverty and differ- ences in various types of behavior across demographic groups (for an analy- sis of economic risk facing the elderly, see Smeeding and Holden, 1989~. We do not give as high priority to developing measures of protection against economic risk as we do to the other areas discussed above for improved income data from SIPP. However, this topic represents a poten- tially important aspect of income and well-being, and we suggest that the Census Bureau support research on how to measure such items as access to credit,25 with a goal of having an appropriate topical module within a few years. We note in this regard that wave 6 of the 1991 panel and wave 3 of the 1992 panel will include a new module on extended measures of well-being. This module has questions on consumer durables (e.g., whether the family has a clothes washer or dryer); living conditions (e.g., whether the house is in good repair and the neighborhood is safe); and ability to meet expenses for basic needs (e.g., whether the family was ever evicted for nonpayment of rent). A few questions on sources of help when in need are also included (e.g., how much help you could expect to get from family living nearby if you were sick), but the module as currently designed focuses much more on measures of hardship than on measures of access to sources of economic support. Recommendation 3-3: SIPP should develop, on an experimental basis, selected measures of economic security against risk, such as access to credit. IMPROVING DATA ON PROGRAM ELIGIBILITY AND PARTICIPATION Since at least the 1960s, policy makers and researchers have wanted infor- mation for understanding, assessing, and planning government social insur 25Measures of access to credit may also serve as a proxy for such intangible assets as knowledge and skills.
74 THE SURVEY OF INCOME AND PROGRAM PARTICIPATION ance and assistance programs. Not until SIPP, however, had a single data collection program existed that was designed to provide the full range of information required for such understanding. Program records provide data on participants, but they lack detailed information about recipients' charac- teristics, they omit information about people who are eligible but do not participate, and the records for one program rarely contain information about other programs in which recipients participate. The March CPS obtains information about participation in many but not all programs and has some of the information needed to determine program eligibility. However, the yearly accounting period of the March CPS greatly limits its suitability for analysis of programs (particularly participation in multiple programs), many of which use a monthly accounting period.26 The CPS also lacks many of the variables needed to determine eligibility-for example, asset values and such expenditures as child care, which can be deducted from income in determining eligibility for many programs. The March CPS also lacks intrayear information on family composition, which enters into the eligibil- ity determination for most programs (e.g., AFDC provides benefits only to certain types of families or subfamily units). SIPP, by contrast, provides detailed intrayear information on a wide range of government assistance programs (see Table 3-1) along with in- come, employment, and family composition, making it a rich source for analysis of such topics as multiple program participation and such events as loss of a job or marriage that may trigger a program entrance or exit. For programs, SIPP obtains monthly reports on recipiency and benefits for all types of cash income assistance both means-tested programs (e.g., AFDC, SSI, earned income tax credit) and non-means-tested programs (e.g., social security, unemployment compensation, black lung benefits). SIPP also ob- tains detailed information for the major types of in-kind assistance (e.g., food stamps, the supplemental food program for women, infants, and chil- dren [WIC], school lunches, energy assistance, Medicare, Medicaid, although public housing is only ascertained sporadically). Finally, SIPP asks at each interview about educational assistance (e.g., Fell grants, Veterans' Adminis- tration educational benefits) and training (e.g., Job Training Partnership Act [JTPA] programs), although information on amounts or extent of benefits for most of these programs is obtained on an annual rather than monthly or 4-month basis. Improved Eligibility Measures Although a vast improvement over the March CPS data, SIPP's information on program eligibility has been a relatively weak component of the survey's 26Thus, families with an annual income too high to qualify for means-tested assistance programs may have had periods of sufficiently low income to be eligible for benefits for some months of the year.
ACHIEVING SIPP'S GOALS 75 program-related data (see Doyle et al., 1987~. In the early panels (1984- 1986), not all needed information was collected for such major programs as AFDC and food stamps. Moreover, the information collected was scattered throughout several topical modules, so that an analyst had to merge files for several waves and assume that information from different interviews was contemporaneous (although, in fact, the information from one wave on, say, child care expenses might not apply to the family situation at a different wave). In response to user requests, the Census Bureau developed a unified program eligibility topical module that has been included in wave 4 or 7 of each panel since 1987 (see below).27 This module includes-eligibility- relevant information for the major assistance programs on countable assets (e.g., homes, vehicles, life insurance), deductible expenses for the prior month (e.g., out-of-pocket medical care costs, dependent care costs, and shelter expenses including rent or mortgage payments and utilities), and existence of a work disability. We strongly urge the Census Bureau to consider including eligibility questions for the major programs in each wave's core interview. Informa- tion on eligibility is a key contribution of SIPP to program analysis: with- out such information, it is not possible to answer such questions as whether an increase in the caseload (as has occurred recently for both food stamps and AFDC) represents an increase in the program participation rate, an increase in the eligible population (due to deteriorating economic condi- tions perhaps), or a combination of both phenomena. It seems odd to have monthly data on program participation and some factors related to eligibil- ity (such as employment status and income) and not to have more frequent data on other factors involved in eligibility. In addition, more frequent information on such expenses as medical, dependent care, and shelter costs could prove useful for developing expanded concepts of income or well- being. Residence in public or subsidized housing needs to be ascertained at least once a year and also whenever respondents move. Another eligibility-related topic on which improved information is needed is family composition. Such programs as AFDC have very complicated rules regarding which household members are considered part of the eli- gible unit and which members' incomes must be considered in determining benefits: for example, some portion of a stepparent's income must be counted, and, if a mother under age 18 who is living with her parents applies for benefits, some portion of her parents' income must be counted as available to her and her child (see U.S. House of Representatives, 1991:567~. Al- though SIPP collects information at each interview about arrivals and de- partures of household members, the relationship of everyone in the house 27However, because the 1989 panel was terminated after three waves due to budget cuts, there is a 3-year gap between the collection of eligibility data in wave 4 of the 1988 panel (spring 1989) and wave 7 of the 1990 panel (spring 1992).
76 TlIE SURVEY OF INCOME AND PROGRAM PARTICIPATION hold to the reference person or household head, and the parent or respon- sible person for each child under 18, this information is not always suffi- cient to determine the eligible assistance unit in households in which no one is currently participating in a program.28 It would be useful for SIPP to obtain more detailed information about household and family relationships, not only to permit more accurate deter- mination of eligible units, but also to support social science research on household and family living arrangements. In particular, it would be useful to ascertain the relationships of children to all adults in the household and to add the category of "partner" to identify consensual unions.29 -Research on questions that could ascertain which members of the household share food expenses would also be useful, as such information is needed to deter- mine eligible units for the food stamp program.30 To make room for additional information on program eligibility in the core questionnaire, some other items would most likely need to be elimi- nated or cut back. We suggested earlier that the weekly information on employment status might be a candidate for reduction in detail. We also strongly urged that some of the asset information currently included in the core (e.g., detailed information on asset ownership and income) be scaled back in favor of obtaining more frequent valuations of asset types that are important for program eligibility. Spell Information For analysis of programs as well as income distnbution, a major focus of policy interest is in time-genes statistics for such periods as months, quar- ters, and years. For program analysis there is another major focus of inter- est, namely, the dynamics of program eligibility and participation (a similar interest extends to the dynamics of poverty). Policy makers are interested in the extent to which program recipients use benefits to cope with tempo- rary economic reverses in the short term versus the extent to which they depend on benefits over a longer hme span. Hence, analysts want informa 28Also, sometimes, the composition of participating units cannot be determined accurately on a month-by-month basis because information about the unit is ascertained only at the time of the interview and not also for each reference month. 29Wave 2 of each panel collects detailed information on relationships among all household members, but such information is not available as household composition changes over the course of a panel. (Note that it would be necessary only to update the information when new people join a household and not to burden respondents otherwise.) 30In most cases, the food stamp recipient unit is the same as the census household; however, subunits arid multiple units within a household are possible, including instances in which all members of a household do not purchase or prepare meals together (Doyle et al., 1987:II.2- O.s)
ACHIEVING SIPP'S GOALS 77 lion to estimate both the length of individual spells of participation and the total time on welfare (or in poverty), counting repeated spells for the same individual (see Ellwood, 1986, for an example of the latter type of analysis with the Panel Study of Income Dynamics UPSIDES. Although less fre- quently examined, spells of program eligibility and how they relate to par- ticipation spells are also of interest. Finally, more understanding is needed of the correlates and consequences of entrances to and exits from programs. Analysis of spells and welfare dependency issues ideally requires ob- servation of program participation behavior over individuals' lifetimes or, at least, large portions of their lives. The PSID, which has followed a sample of families for 25 years, has proven the nation's richest source of longitudinal data for analysis of such topics as welfare dependency and long-term poverty. SIPP was not intended to compete with the PSID or other long-running panel surveys (such as the National Longitudinal Sur- veys of Labor Market Expenence [NLS]), which have followed cohorts of men and women for periods of 15 to almost 25 years); rather, SIPP was designed to provide more detailed information on short spells of program participation and poverty, including spells of less than a year, which the longer term panel surveys do not measure well because their interviews occur at annual or biennial intervals.3~ However, we believe that SIPP can and should improve the data that it provides for analysis of program eligi- bility and participation spells over the short to medium term. Such data are needed to respond to the growing policy concern about welfare dependency and the interest in reforming welfare programs so as to lessen dependency. One essential improvement, as we argued in the discussion of improved income measures, is to maximize the accuracy of inflation about partici- pation during the time span covered by SIPP. As noted above, record 31See Hill (1992) for a description of the PSID, which has conducted annual interviews since 1968 with an initial sample of about 5,000 families (including an oversample of low- income families headed by someone under 60 years of age), augmented in 1990 with a sample of 2,000 Latino families. The PSID is managed by the University of Michigan Survey Re- search Center with funding from the National Science Foundation and other agencies. See Center for Human Resource Research (1988) for a description of the NLS, which is managed by the Center at Ohio State University with funding from the Bureau of Labor Statistics. The NLS includes an older men's cohort of 5,000 men aged 45-99 in 1966 who were interviewed 12 times from 1966 to 1983 and again in 1990; a mature women's cohort of 5,100 women aged 30-44 in 1967 who were interviewed 15 times from 1967 to 1990; a young men's cohort of 5,200 men aged 14-24 in 1966 who were interviewed 12 times from 1966 to 1981; a young women's cohort of 5,200 women aged 14-24 in 1968 who were interviewed 15 times from 1968 to 1990; and a youth cohort of 12,700 men and women aged 14-21 in 1979 who have been interviewed annually since then. See Matthew Greenwald & Associates, Inc. (1991) for brief descriptions of major panel (and cross-sectional) surveys that provide information about the socioeconomic circumstances of the population.
78 THE SURVEY OF INCOME AND PROGRAM PARTICIPATION check studies that make use of administrative program data offer an impor- tant means to achieve this goal. In addition, we recommend (see Chapter 4) that the length of SIPP panels be increased from 32 months to 4 years, so that more spells are observed in their entirety rather than being "censored," there is a longer time span in which to observe behavior prior and subse- quent to program entry and exit, and there is more opportunity to observe repeated spells of participation for the same person.32 We also recommend that SIPP follow children who move out of origi- nal sample households and that it follow both children and adults who move into institutions (see Chapter 4~. Children are an increasingly important object of concern for public policy. They are also more and more likely to live with different people at various times in their childhood, some of whom may not be original sample members: for example, a child in a single- parent family at wave 1 who moves to live with the other parent is now lost to the survey. Information on children's spells of program participation and the impact of programs on their well-being is unnecessarily limited by the current SIPP design. Similarly, institutionalized persons are a sizable com- ponent of the Medicaid caseload and are also part of the caseload for such other programs as SSI (U.S. House of Representatives, 1991:738,1430~; tracking people in and out of institutions and obtaining some information about them while they are institutionalized would fill an important gap in SIPP's ability to support analyses of spells and other aspects of program participation. Finally, we believe it is important for SIPP to obtain some information about respondents' program participation histories prior to the first inter- view for example, whether a current spell of participation as of wave 1 is the first or a repeat spell and the beginning date of the first spell. This type of information is needed to analyze long-term welfare participation, includ- ing multiple spells. Recipiency history was collected in wave 5 of the 1984 SIPP panel, which was not ideal for a number of reasons: attrition meant that no history was obtained for some original sample members and the length of recall was unnecessarily long. Moreover, information was col- lected only for AFDC, SSI, and food stamps. Starting with the 1986 panel, wave 2 included questions on the beginning dates of spells in progress at the initiation of the panel for almost all social insurance and assistance programs. In addition, for AFDC, food stamps, and SSI, respondents were asked about previous receipt of benefits, the beginning date of the first spell, and the total number of spells. Miller and Martini (1992) evaluated the 1986 wave 2 recipiency history 32We recommend other design changes so that increased panel length does not increase the annual number of interviews and hence is cost-neutral for the SIPP program overall (see Chapter 4).
ACHIEVING SIPP'S GOALS 79 questions, concluding that the data were of generally good quality. For example, the distribution of time on AFDC, based on the beginning dates of current recipiency, closely resembled the distribution of time on AFDC found in administrative data. Respondents appeared to correctly distinguish the question on the beginning date of the current spell of receipt from that on the beginning date of the first spell. Miller and Martini suggested word- ing changes to further improve the quality of the data, which the Census Bureau incorporated into the questionnaire for the 1992 panel.33 We en- courage the Census Bureau to continue to evaluate and seek ways to im- prove the recipiency history information in SIPP, which can increase the usefulness of the survey for analyses of longer term program participation. Emerging Programs and Program Changes The coverage of social insurance and assistance programs (both cash and in-kind) in SIPP is impressive. However, for the most part, SIPP does not obtain information about social service programs" for example, Head Start and other state and local programs that provide child care assistance or programs that provide counseling or home health care senices.34 SIPP also does not obtain information about meal programs, other than the major nutrition programs administered by the Department of Agnculture (e.g., Meals on Wheels for the elderly).35 There are many difficulties in obtaining good measures of participation in social service programs, which vary greatly in availability and scope across localities, and we do not suggest that SIPP devote much attention to this area. However, we do suggest that occasional measurement of social service programs in SIPP may be important to provide data to identify important changes in program mix. As an historical example, the food stamp program grew from a pilot program in a few states in the early 1960s to a national program that, by 1975, served more than 7 percent of the population (and more than 60% of the poor population) (U.S. House of Representatives, 1991:1401~. SIPP needs to be able to identify emerging programs that are becoming an important resource for people, particularly those who are economically at nsk.36 33Beginning in the 1992 panel, recipiency history for AFDC, food stamps, and SSI is being collected in wave 1. 34A home health care topical module was included in the 1987-1989 panels. 35The long-term care topical module included in the 1986-1988 panels asked a global ques- tion for respondents aged 15 and older about meals received per week from a community service either in the home or a group setting. 36For brief descriptions of the full range of existing cash, in-kind, and social service pro- grams that provide assistance to low-income people (almost 80 in all), see Congressional Research Service (1991).
80 THE SURVEY OF INCOME AND PROGRAM PARTICIPATION It is also important that SIPP keep up with policy trends and changes in established programs. Two of the most notable today are the increased use of the income tax system to provide benefits to poor and near-poor working families with children and the renewed focus on services in such programs as AFDC for example, training, employment, and child care services that are intended to help improve the skills and employability of welfare recipi- ents and, over time, move them away from long-term dependency on cash assistance. Hence, it is important that SIPP develop mechanisms to track programs that may become a more important source of economic resources in the future and to keep abreast of program developments. The use of topical modules for these purposes seems a particularly promising approach. Recommendation Recommendation 3-4: Priorities for improved measures of pro- gram participation and eligibility from SIPP should include im- proving the range and frequency of information needed to de- termine eligibility for major assistance programs and providing adequate measures of spells of both eligibility and participa- tion. SIPP should also keep up to date with respect to newly important programs and program changes. TOPICAL MODULES The use of SIPP as a vehicle to respond to social welfare policy concerns in areas that are related to the survey's core subjects is an important goal. Indeed, the data from topical modules have proven very popular with a wide range of users; see Table 3-13 for a listing of topical modules included in SIPP panels to date. Nonetheless, the topical modules must necessarily play a supporting rather than a dominant role in the overall SIPP program. We support the continued availability of open, or variable, topical mod- ules in each panel so that federal agencies that have responsibilities for social welfare policy planning and analysis can insert questions of current interest.37 For this component to be most useful, it is critically important that SIPP respond in a timely and flexible manner to agency needs. Hence, we encourage the Census Bureau to do everything possible to streamline procedures for eliciting input from federal agencies and reaching decisions that best accommodate what may often be conflicting agency priorities. 37Fixed modules also play an important role, namely, collecting data on the core subjects of SIPP that do not need to be asked in every interview (e.g., asset holdings).
ACHIEVING SIPP,S GOALS TABLE 3-13 Topical Modules in SIPP, 1984-1990 Panels 81 Panel and Wave Topic 1984 1985 1986 1987 1988 1989 1990 Total number of wavesa 9 8 7 7 6 3 8 Annual income 6,9 5,8 5 5 5 5,8 and retirement accounts Income taxes 6,9 5,8 5 5 5 5,8 Child care 5 6 3,6 3,6 3,6 3 3 Child support 5 6 3,6 3,6 3,6 3 - 3,6 Support for nonhousehold 5,8 4,6 3,6 3,6 3,6 3 3,6 members Disability status of 3 6 3 6 3,6 3 3,6 childrenb Functional limitations 3 6 3 3,6 and disabilityb Use of health careb 3 6c 3c 6c 3,6 3 3,6 Home health care 6 6 3 Long-term care Educational financing 6,9 5,8 5 5 5 and enrollment Eligibility for selected major programs Selected financial assets (see also wealth) 7 4 7 Dependent care costs (see also child care) 7 4 7 Medical care expenses 7 4 7 Shelter costs 4 6 3,7 4 7 Real estate and vehicles (see also wealth) 7 4 7 Work disability (see also functional 7 4 7 limitations) Energy usage 4 6 3 Personal history Education and training 2 2 2 2 2 Education and work 3 Employment 2 2 2 2 2 Family background 2 2 2 Fertility 8 4 2 2 2 2 2 Household relationships 8 4 2 2 2 2 2 Marital status 8 4 2 2 2 2 2 Migration 8 4 2 2 2 2 2 Recipiencyd 5 2 2 2 2 2 Work disability 2 2 2 2 2 Wealth Assets and liabilities 4,7 3,7 4,7 4 4 Real estate property 7 3,7 4,7 4 4 and vehicles Pension plan coverage 4,7e 7 4,7e 4 and retirement expectations 5,8 continued on next page
82 TABLE 3-13 Continued THE SURVEY OF INCOME AND PROGRAM PARTICIPATION Panel and Wave Topic 1984 1985 1986 1987 1988 1989 1990 Work expenses Work schedule Employee benefits Job offers Job training Spells outside the work force s,8 4 6 6 of ~ 6 3 6 3,6 3 , 6 NOTES: Question content for many of the modules (e.g., child care, disability) differs across panels. The 1991 panel has the same modules as the 1990 panel, with the following exceptions: the program eligibility set of modules moves from wave 7 to wave 4; the wealth set of modules moves from wave 4 to wave 7; and wave 6 includes a new module on extended measures of well-being (see text). The wave 6 modules included in 1990 (e.g., child support) are not included in the 1991 panel. The 1992 panel has the following changes from the 1991 panel: the employment and recipiency history modules move from wave 2 to wave 1; the program eligibility set of mod- ules moves from wave 4 to wave 7; the wealth set of modules moves from wave 7 to wave 4; and the extended measures of well-being module is part of wave 3 (and the only module in that wave) What modules will be included in wave 6 is not yet decided. Generally, "variable" topical modules (those determined by the current needs of federal agencies) are included in waves 3 and 6, while "fixed" modules (e.g., annual income and retirement accounts, income taxes, educational financing and enrollment, and the program eligibility, personal history, and wealth sets of modules) appear in other waves (see Committee on National Statistics, 1989:Tables 2-2, 2-3, for identification of fixed versus variable topical modules in the 1984-1990 panels). aThe number of waves in each panel differs due to budget cuts. Also, only two of four rotation groups in the 1984 panel received 9 waves; the other two groups received 8 waves. Also, one rotation group in each of the 1985 and 1986 panels received one fewer wave than the other three groups. bDisability and health care-related modules have different names across panels. CIncludes single question on overall health status. dRecipiency history in 1984 panel covers food stamps, AFDC, and SSI only; in subsequent panels it covers most government programs, with varying detail. eExcludes characteristics of job from which retired; also, pension plan coverage in the second administration of the module is asked only for people who changed jobs or became newly employed since the first administration of the module. includes reservation wage.
ACHIEVING SIPP'S GOALS 83 The Census Bureau should also obtain suggestions from academic re- searchers and other nonfederal users of SIPP data for questionnaire content to include in topical modules. Improvements in policy analysis that can sup- port more informed government policy making ultimately rest on advances in social science research knowledge. Hence, SIPP can serve a useful purpose by offering the opportunity for researchers, as well as agency policy analysts and program planners, to contribute to the specification of variable topical modules (see Chapter 8 for discussion of input mechanisms).38 In addition, the Census Bureau should use topical modules as a compo- nent of its research and development program to enhance the survey's core measures on economic resources and programs. Dedicating a topical mod- ule in each panel for use by Census Bureau analysts (with input from out- side experts) offers one low-cost way for the survey to keep up to date with trends in the composition of income and the mix of programs and to be innovative in providing information that expands knowledge of well-being and program-related behavior. Some questions that are first asked in topical modules may subsequently move into the recurring core part of the inter- view; other questions may not be appropriate for the core but may nonethe- less provide important analytical information. Our recommendation to in- crease the length of SIPP panels and maintain the 4-month reference period (see Chapter 4) will make it possible to include topical modules for this purpose without sacrificing existing modules. Thus, under our proposed design of 12 interviews per panel, one pos- sible sequencing of topical modules that leaves room for a Census Bureau module in addition to variable modules for policy needs is as follows: wave 1, recipiency and employment history wave 2, other personal history modules wave 3, wealth modules wave 4, open wave 5, annual income and taxes wave 6, open wave 7, open wave 8, annual income and taxes wave 9, wealth modules wave 10, open wave 11, annual income and taxes wave 12, open This scheme obtains personal history information early in the life of a panel, provides measures of balance sheets twice per panel and measures of 38We note that Statistics Canada is designing a new longitudinal Survey of Labour and Income Dynamics for which the agency has invited academic researchers to specify the content of topical modules. The researchers are responsible for obtaining funding for the modules.
84 THE SURVEY OF INCOME AND PROGRAM PARTICIPATION income and taxes on an annual basis, and leaves five modules open. An assumption of this scheme is that program eligibility information is asked in the core interview. If, for some reason, this does not prove possible, then it would be important to obtain eligibility measures on at least an annual basis. This could be done by administering the program eligibility modules in waves 6 and 12 and also including selected eligibility questions in waves 3 and 9 39 Such a scheme would leave open 3 topical modules, or one more per panel than under the current design. In developing topical modules on subjects that are of interest for social welfare policy, it is important to keep in mind that SIPP is primarily a survey about income and programs. The temptation to add great detail to the questionnaire on topics that are not central to the core should be re- sisted. There are, and should continue to be, other surveys that focus on such topics as long-term care, educational attainment, and job characteris- tics. SIPP cannot and should not attempt to duplicate the detail in these special surveys. Yet SIPP should also not exclude topics just because they are the focus of other surveys. SIPP's great advantage is that it constitutes a continuing series of panel surveys with a core of demographic and socioeconomic variables that are relevant for many types of analysis. Hence, SIPP can support initial research (with a rich set of explanatory variables) on a topic for which a special survey is subsequently developed; provide a useful time series (in the case of a topic that is covered in more than one panel); and, because its samples are nationally representative, serve to anchor and ex- tend the analysis for special surveys that are limited to specific population groups. For all these uses, it is important that the Census Bureau work closely with analysts in other federal agencies to ensure comparability in basic questions and concepts between SIPP and other surveys. Recommendahon 3-5: The topical module component of SIPP should continued and be strengthened by: · obtaining input from both government agencies and the so cial science research community about topics related to SIPP's core goals to consider for modules; · streamlining the content development process so that timely information can be collected on emerging policy and research issues; and · using some topical modules as a means for the Census Bureau's analysis staff to conduct research on expanded and alternative measures of income and programs. 39The wealth modules scheduled for these waves already include many of the program eligibility questions; it would simply be necessary to add the non-asset-related items (i.e., expenses and disability see Table 3-13).
ACHIEVING SIPP'S GOALS 85 SIPP AND THE MARCH CPS The Census Bureau's plan for improving income statistics (reviewed in Chapter 2) states that the March CPS will remain the primary source for annual income and poverty estimates. Certainly, many users are comfort- able with the March CPS annual reports and data files, finding attractive their large sample size, timely release (within 5-6 months of data collec- tion), and relative ease of use. Although we agree that the March CPS provides a very important his- toncal time series that should and will continue for the foreseeable future, SIPP was launched to improve the depth, breadth, and quality of income and related statistics for the United States. SIPP provides more detail than the March CPS on economic resources, broadly defined, including informa- tion on assets and a greater number of in-kind assistance programs. SIPP has already made important gains in data quality relative to the March CPS: for example, it obtains more complete reporting of most income sources (although there is still room for further gains). The monthly info~ation in SIPP makes possible the construction of improved measures of annual in- come and poverty that include income from people who left the universe during the year (e.g., because they died or moved abroad) and that more accurately represent family composition than is possible with the March CPS. In addition to annual statistics, SIPP supports intrayear and multiyear measures, which are important to better understand economic well-being and the role of government assistance programs. Until now, it has not been possible for SIPP to serve as the primary source of the nation's income statistics because of such problems as small sample size and lack of timeliness. However, with the redesign that we propose, SIPP would provide adequate sample sizes for both longitudinal and cross-sectional measures, particularly for the low-income population. Moreover, technological developments in data collection and processing- specifically, the use of computer-assisted interviewing and an improved database management system (see Chapter 5) would make it possible to process SIPP data on as timely a basis as the March CPS and make the SIPP data files more accessible.40 Hence, we urge the Census Bureau to set a target date-perhaps the year 2000, certainly no later than the year 2005 for the data from SIPP to be of sufficient reliability, quality, and timeliness to be used instead of the March CPS data as the basis for annual (and other) measures of income and poverty. Obviously, the changeover should take place gradually. In this 40The use of computer-assisted interviewing may also effect cost savings that permit a further increase in the SIPP sample size. This increase could be useful for the production of detailed income and poverty statistics and also for analyses of participation in many smaller programs.
86 TlIE SURVEY OF INCOME AND PROGRAM PARTICIPATION regard, we support the Census Bureau's plans to include supplementary SIPP data in the March CPS P-60 report series as an important way to accustom users to SIPP. In addition, we propose an expanded program of cross-sectional reports from SIPP in the P-70 series (see Chapter 6) that would be issued alongside the CPS reports for some years. Once a suffi- cient time series is built up under the new SIPP design and it has proven possible to release the data on a timely basis, it will be appropriate to replace the current P-60 series with the SIPP P-70 reports on income and poverty. At that time, it will also be appropriate to determine the extent to which the income data on the March CPS can be scaled back. The CPS should always include some information on income, which is an important variable for analyzing the labor force data that are the principal focus of that sur- vey.4~ However, the decision on the level of income detail to include should be made on the basis of labor force analysis needs, not on the basis of the requirements for national income statisiics.42 Our recommendation that the Census Bureau set a target date for turn- ing to SIPP as the basis for income and related measures has implications for the concept of an integrated income statistics system as outlined in Chapter 2. The Census Bureau originally suggested the use of administra- tive records to correct SIPP estimates and then the use of corrected SIPP estimates from earlier years to adjust March CPS estimates. We support increased use of administrative records to improve and enhance SIPP mea- sures related to income and programs (see the next section). However, we agree with the Census Bureau's latest position that the goal of using admin- istrative records to correct income estimates for reporting errors in SIPP must be viewed as a long-term and not a short-term objective, as there are many methodological as well as procedural hurdles to overcome. Moreover, if SIPP is to be the prime income survey, it does not make sense for the Census Bureau to use scarce resources to develop extensive SIPP-based adjustments to March CPS measures. Furthermore, we believe SIPP should receive priority over the March CPS for major investments designed to improve measures of income and program participation. For example, resources for administrative record-check studies or for improved income imputation procedures should be devoted to SIPP, not the March CPS. Of course, where improved procedures can readily be implemented for both surveys, that should be done. 41The information that is collected in the March CPS income supplement on prior-year work experience is also important for analyses of the CPS labor force data. 42A reduction in income detail may also ease the data collection burden for the labor force information that is the prime focus of the CPS. At present, interviewers accept relatively high nonresponse rates to the income supplement so as not to increase the chances that a household will refuse to answer the next month's labor force questions (see Citro, 1991).
ACHIEVING SIPP'S GOALS 87 Given that there will be a period before data from the redesigned SIPP are available and a longer period in which SIPP and March CPS income measures are both produced, we favor work that will help users understand differences between SIPP and CPS and implement their own adjustments for their own purposes. Thus, we are very supportive of the Census Bureau's plan to conduct a major comparative study of the 1990 SIPP panel and the March 1991 CPS. We also support incremental improvements to the March CPS. In this regard, we point to a recent Committee on National Statistics panel study (Citro and Hanushek, 1991a) that evaluated microsimulation models for social welfare programs: it concluded (Ch. 5) that the March CPS will remain for some time the database of choice for many of these models because of its large sample size and timeliness. The report sug- gested some modifications to the March CPS (e.g., a few additional ques- tions) that could improve its usefulness for social welfare policy analysis and modeling and, in particular, make it easier for users to develop SIPP- based imputations and adjustments to the March CPS data. Recommendation 3-6: SIPP should become, over time, the pri- mary source of the nation's income statistics in place of the March CPS income supplement. SIPP should receive priority for major investments to develop improved income measures. As there will necessarily be a transition period during which SIPP and CPS income statistics are both published, every effort should be made to increase user understanding of differences and similarities and to effect incremental improvements as ap- propriate in both surveys. SIPP AND ADMINISTRATIVE RECORDS Operating agencies at all levels of government, corporations, and other or- ganizations regularly generate large volumes of administrative records about individuals (e.g., taxpayers, program beneficiaries, employees) and indi- vidual transactions (e.g., payments, eligibility redeterminations). These records can provide a wealth of useful information for statistical purposes, both on a stand-alone basis (i.e., analysis that is limited to the records themselves, such as tabulations of tax returns) and, more powerfully, through linkages with household survey data. Such linkages can provide supplemental data at low marginal cost as well as a means to validate and improve the quality of survey responses. The widespread adoption of computerized systems of record-keeping has greatly enhanced the potential utility of administrative records for statistical research and analysis purposes in recent years. Not all statistical programs can benefit. For some, no records exist
88 THE SURVEY OF INCOME AND PROGRAM PARTICIPATION with which to validate or supplement survey reports by individuals of many types of behavior; for others, relevant records exist but are very difficult and costly to access. In general, there are drawbacks to the use of adminis- trative records, including: · differences in content, format, and recording medium across agen- cies, even for the same program (e.g., not all states have computerized records for such programs as AFDC and Medicaid); · the fact that content (and format) may change without consideration of the needs of statistical analysis (e.g., frequent changes in tax laws have meant that IRS records do not contain consistent information year-to-year for such items as deductible expenses); · different rules, regulations, and procedures across organizations with regard to access to their records for analytical purposes; and · finally, heightened public concern about protecting the confidential- ity of individual replies that has led statistical agencies to take a very strict view about the use of data files that link administrative and survey informa- tion by outside researchers. Yet despite the difficulties of accessing and using administrative records, statistical programs for which relevant records exist cannot ignore their potential to enhance data quality and scope at reasonable cost. SIPP is particularly well positioned to exploit administrative records because of its focus on income and program participation, for which many relevant public and private record systems exist. We see several ways in which administrative records may be of great benefit to SIPP. The first is to match administrative records with SIPP cases in order to obtain additional data items and to extend the available information for each case backwards and forwards in time. For example, social security earnings histories could be appended to each SIPP record for years prior to and including the time covered by the survey. Such data would be very useful for projecting expected future social security benefits and also for studying the lifetime work experiences for various population groups. Also, earnings data could continue to be added for subsequent years to support such analyses as the work efforts of SIPP respondents who experienced a spell of welfare participation. Social security benefit infor- mation could similarly be added for SIPP respondents who were nearing retirement age at the end of a panel's life. In addition to operational problems that affect any use of administrative records (e.g., records may not be available on a timely basis or may not exist in computerized form), the chief difficulty in seeking to augment SIPP records with administrative data concerns access to the resulting linked files. Historically, the Census Bureau has not been willing to release linked files to outside researchers because of concerns about protection of confi
ACHIEVING SIPP,S GOALS 89 dentiality. (Indeed, the file of linked social security and survey data for the 1984 SIPP panel that the Census Bureau prepared for the Social Security Administration [SSA] was made available only to SSA analysts under strict conditions of use.) A panel of the Committee on National Statistics is nearing completion of a study of confidentiality and data access, and its report may suggest ways for the Census Bureau (and other agencies) to provide access to linked files. The second possible type of use for administrative records in SIPP is as the basis for multiple-frame samples for SIPP panels, that is, area probabil- ity samples of households together with cases that are drawn from one or more type of administrative record (e.g., program records, tax records, or employer records). Every case in the multiple-frame sample would be ad- ministered the SIPP questionnaire; in addition, the cases drawn from the administrative framers) could have appended data items from the particular type of administrative record (provided the access problems referred to above could be worked out). Multiple-frame samples can be an efficient way to oversample selected population groups for analysis purposes (e.g., program recipients) if the relevant records are readily accessible.43 We discuss the benefits of this approach for oversampling low-income population groups in SIPP in Chapter 4. We also discuss operational and technical difficulties that can make it difficult to develop multiple-frame samples and produce timely analysis files. We suggest that the cost-effec- tive use of a multiple-frame approach for SIPP will generally require the active cooperation and support of an agency that is interested in adding a component to the SIPP sample that is drawn from its records. Finally, administrative record-check studies can be a very important means of evaluating and improving the quality of survey information. Such studies can lead to improved wording of questions, improved imputation methods for survey nonresponse, and even adjustments to survey responses for reporting errors. Because record-check studies use administrative data for evaluative and methodological research, they can be restricted to use within a statistical agency and hence do not pose the same degree of con- cern about protection of confidentiality.44 Full record-check studies involve matching administrative with survey records after completion of data collection in the survey. Such studies look for matches in the administrative records for all survey records. They can 43It would be difficult at present to develop a national multiple-frame sample for such state- administered programs as AFDC, given the lack of good computerized record systems in many states. 44Such use need not be limited to the agency staff. Other researchers may be able to work with the data under special on-site arrangements: for example, the Census Bureau has allowed researchers to work with confidential data at the agency as special sworn employees.
go THE SURVEY OF INCOME AND PROGRAM PARTICIPATION thus identify "false positives," that is, people who erroneously report par- ticipation in a program (or receipt of a fringe benefit), as well as "false negatives," that is, people who fail to report their participation, and hence can estimate the net extent of underreporting (or overreporiing) of an in- come type or program benefit. The match of the first two waves of the 1984 SIPP panel with records for eight federal and state assistance pro- grams was a study of this type (see Marquis and Moore, 1989, 1990a, 1990b). As another example, Census Bureau staff in the Housing and Household Economic Statistics Division (HHES) are currently conducting matches of IRS records with the 1990 SIPP panel as part of their work to evaluate income data in SIPP and the March CPS and to develop a model for esti- mating after-tax income from SIPP.45 Forward record-check studies involve selecting samples from adminis- trative records and giving the selected individuals the survey interviews. Such studies are more limited than full record checks in that they do not pennit false negatives to be identified; however, forward record checks can be earned out on a more timely basis because there is no need to match administrative and survey records. We support the work that is under way by the HHES staff to match administrative records with the 1990 SIPP panel for research and develop- ment purposes. We also recommend (see Chapter 7) the use of small-scale forward record-check studies (e.g., with records from selected states for such programs as AFDC, SSI, and food stamps) in the implementation of several elements of the SIPP redesign for example, in helping to deter- mine improved wording for important questionnaire items and improved methods to treat nonresponse. Overall, we strongly support an increased role for administrative records in the SIPP program. However, there are many operational and technical problems, in addition to concerns about confidentiality, that impede their ready use, and we do not think that there can be fast progress toward such goals as using administrative records to adjust SIPP responses for reporting errors. Nonetheless, we urge the Census Bureau to seek innovative ways for SIPP to benefit from the extensive information that is available on income and programs from administrative record sources. 45We noted earlier the significant benefits that could accrue to SIPP from working with the Statistics of Income Division staff on ways to develop high-quality tax information for SIPP while at the same time reducing the burden of the tax module in the current questionnaire.