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.
DATABASES FOR MICROSIMULATION: A COMPARISON OF THE MARCH CPS AND SIPP 39 Combining overlapping SIPP panels will produce larger sample sizes for cross-sectional estimates. Indeed, the SIPP overlapping panel design was intended for this purpose; however, delayed data delivery has meant that users to date have had very little experience with combining panels. Moreover, the larger 1990 panel has no overlapping panel for the first year of interviewing. However, the 1990 panel has the advantage of oversampling black, Hispanic, and single female-parent households. The Census Bureau's plans to restore the SIPP panel size to 20,000 households, to restore the overlapping panel design beginning in 1991, and to oversample poverty households in the post-1990 census SIPP redesign could all help make the survey more useful for microsimulation modeling of income support programs.14 Yet even these changes will not produce the sample sizes that are provided for microsimulation and other analysis by the March CPS, particularly if the proposal to expand the CPS sample were revived. Data Delivery Another comparative advantage for microsimulation modeling of the CPS data has to do with data delivery. A public-use file from the March CPS is regularly issued about 6 months after the interviews are completed. It takes the modelers several more months to reprocess the file. (The Urban Institute requires about 3 weeks to convert a new CPS to TRIM2 format when there are no major changes in the CPS, including the allocation of annual income to months, correction for underreporting of asset income, and validation of all input variables. Another 14 Restoration of sample size depends upon obtaining sufficient funds, as may also restoration of the overlapping panel design. It should be noted that these particular design choices are not the only ones that could achieve the survey goals. For example, larger nonoverlapping panels could be a viable design. The Census Bureau has asked a panel of the Committee on National Statistics to evaluate the design of the survey and make recommendations for implementation, along with the sample redesign, in 1995.
DATABASES FOR MICROSIMULATION: A COMPARISON OF THE MARCH CPS AND SIPP 40 6â10 weeks is required to complete baseline simulations of all tax and transfer programs. In some years, additional time is required to program major new features of tax and transfer programs.) Hence, a policy simulation can be carried out with household composition data that are only about 1 year old, together with annual income and work experience data for the preceding calendar year. SIPP data files, in contrast, have been issued only after considerable delays. Initially, cross-sectional wave files from the 1984 panel were released an average of 13 months after the last month of data collection. However, the complexity and magnitude of the survey quickly overwhelmed the Census Bureau's data processing resources. Files from the 1984 panel containing topical module data (some of which are needed for income-support program modeling) were released an average of 25 months after the last month of data collection. The full 32-month longitudinal research file from the 1984 panel was released 20 months after the final interviews were completed (Committee on National Statistics, 1989: Table 2â4). The Census Bureau experienced ma ny problems in producing usable files from the 1984 panel. Each wave file was recalled at least once to correct specification and programming errors. Moreover, the frequent changes in the questionnaires that occurred in the initial SIPP panels hampered efforts to develop a smooth-running and error-free data processing system. As a result, the delivery record for the 1985 and 1986 panels deteriorated greatly, with delivery of files for the first waves of these panels lagging by about 3 years after data collection. Currently, the Census Bureau is striving to meet a schedule of releasing wave files within 12 months after data collection. At the present time, all files for the 1984â1987 panels are available, as are several of the wave files from the 1988 panel and the wave one file from the 1990 panel (there are no plans to release data from the truncated 1989 panel). However, even if the Census Bureau meets its delivery targets, the modelers must spend time reprocessing the data. It is likely that SIPP will never be as timely a resource for microsimulation modeling as the CPS. One more pointâalthough more relevant for modeling retirement income rather than income support programsâshould be made with regard to data delivery: missing exact-match files. In the mid-1970s, a very useful file was prepared that matched records from the March 1973 CPS with social security earnings histories for the same individuals, based on social security numbers. The 1973 exact-match file, which was made publicly available, is the primary database for the DYNASIM2 model. A second such match was conducted of the March 1978 CPS with social security records, but this file was not widely distributed. However, the 1978 file was obtained by President Reagan's Commission on Pension Policy, which turned it over to Lewin/ICF, Inc., to use as the basis for a microsimulation model of retirement income. An exact-match file from the March and May 1979 CPS, which was matched in turn to the