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.
EVALUATIONS OF MICROSIMULATION MODELS: LITERATURE REVIEW 263 equations. With the macroaligned model, average annual forecast errors typically exceeded 5 percent. Hayes recommended new macroadjustment factors for federal income taxes, transfer income components (i.e., food stamps and unemployment and workers' compensation), and home values. Hayes also conducted a small sensitivity analysis to determine the degree of interaction among MICROSIM's components. Adjustment factors for modules projecting births, divorces, female labor force participation, and other transfers and unemployment were raised and lowered by 10 percent, holding other alignment factors constant. For unemployment, adjustments of plus and minus 5 percent were examined to test a linearity hypothesis. The resulting percentage changes in 29 output variables from MICROSIM were examined with respect to a proportionality hypothesis (that elasticities are invariant with respect to the magnitude of change) and a symmetry hypothesis (that elasticities are invariant with respect to the sign of the change, holding the magnitude constant). Hayes concluded that, based on the magnitudes of the cross-elasticities, it could be inferred that only a moderate degree of interaction existed among MICROSIM modules. In addition, his analysis uncovered some counterintuitive results. For example, increases in female labor force participation resulted in an increase in births. Also, when the birth rate increased, so did the divorce rate. This type of sensitivity analysis can be extremely informative as to both the degree of interaction and the hypotheses mentioned above, but, in addition, the âblack boxâ quality of microsimulation models is reduced through this method of shaking the box to see what is inside. The methods Hayes used are worthwhile and can be applied generally to microsimulation models in a variety of areas. Finally, a question can be raised as to the effects on these findings of using a reverse statically aged input data set, rather than what would have been used in practice. This problem was partially examined by Hayes, but it is clear that mimicking the actual modeling situation, when possible, would be preferable. On the other hand, Hayes's study provides a technique that might be helpful in situations where an input data set cannot be obtained. JEFFERSON (1983) Jefferson (1983) performed an external validation study of DYNASIM's Family and Earnings History (FEH) model. Mean earnings and earnings distributions from the 1978 DYNASIM simulated population were compared with data on actual earnings from two sourcesâthe 1978 earnings data from the May 1979 CPS and the Social Security Administration's (SSA) 1977 administrative statistics (data for 1978 were unavailable). The 1978 DYNASIM population was produced by simulating 1973 through 1978 under the base scenario for the Twenty-first Century Data Bases Project.