Improving Information for Social Policy Decisions -- The Uses of Microsimulation Modeling: Volume II, Technical Papers (1991)
Chapter: STARTING DATABASES
totally revised. The model was also made more efficient for use on a mainframe computer by separating the creation of demographic and labor force histories from the jobs, pensions, and government program benefits simulations. The latter simulations are executed using a synthetic longitudinal file and can be rerun using new specifications at relatively low cost because the longitudinal history file does not have to be recreated.
The Pension and Retirement Income Simulation Model (PRISM), designed by ICF, Inc. (now Lewin/ICF, Inc.), was developed in 1980 to analyze alternative national retirement income policies for the President’s Commission on Pension Policy. Lewin/ICF added a significant policy analytic capability to PRISM in 1986, when it developed, in a joint effort with the Brookings Institution, a submodel to simulate the future utilization and financing of long-term care for the elderly.1 Like DYNASIM2, PRISM creates synthetic longitudinal histories of demographic and labor force events and devotes significant effort to modeling the development of retirement income sources, including social security, pensions, and assets, over time.
This chapter presents a comparative description of DYNASIM2 and PRISM as an introduction to the structure of a dynamic simulation model that notes important similarities and differences in modeling approaches. Other dynamic simulation models exist, but are not discussed here because DYNASIM2 and PRISM are the two dynamic models that are most often used for policy analysis. Table 1 provides a summary of the basic features of DYNASIM2 and PRISM. The descriptions of these models rely heavily on existing documentation. For DYNASIM2, documentation includes Johnson, Wertheimer, and Zedlewski (1983), and Johnson and Zedlewski (1982). For PRISM, documentation includes Kennell and Sheils (1986, 1990). Additional information about PRISM was found in Anderson (1990). Further insights about the concept of dynamic modeling were obtained from Orcutt et al. (1980) and Zedlewski (1990).
STARTING DATABASES
Dynamic simulation models start with a database, generally cross-sectional, that provides a random sample of U.S. households and information about their incomes, employment, and demographic characteristics. Both DYNASIM2 and PRISM use the March Current Population Survey (CPS). Both take advantage of exact matches that have been made between the CPS and social security earnings records by the Social Security Administration (SSA) to obtain lengthy histories of taxable earnings levels and quarters of coverage for individuals in the CPS sample. DYNASIM2 uses an exact match of social security records with the 1973 CPS, and PRISM uses an exact match of social security records with the 1978 CPS, which, in turn, was exactly matched to the March and May
