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 257 A comparison of labor force participation for the CPS and DYNASIM showed that DYNASIM was relatively more successful at estimating the average labor force aggregated over years for age-sex categories than it was at predicting year-to-year variability, especially for nonwhite males. The same was true, with larger discrepancies, when comparing DYNASIM and the PSID data for work experience. When this analysis was disaggregated by age group (and aggregated by year), the differences between CPS and DYNASIM increased. For instance, for the age group 16â19, DYNASIM estimated that 45 percent of nonwhite males had labor force experience, while the CPS estimated that 59 percent did. When considering the percentage of labor force participants without work experience, the largest difference was that between the CPS and DYNASIM for white females aged 16â19: DYNASIM estimated 20.3 percent and the CPS estimated 5.9 percent. With respect to years with earnings, DYNASIM was least successful in modeling women. For example, for nonwhite women it estimated that 8 percent would have no years with earnings, while the PSID estimated that 19 percent would have none. According to Hendricks and Holden, the problem was that participation probabilities were dependent on several variables, including participation in the preceding year. These variables did not, however, adequately differentiate, over a long period, the degree of labor force attachment. Thus DYNASIM did not predict enough women with little or no labor force participation or with very strong labor force participation. With respect to variations in earnings, DYNASIM and the PSID showed systematic differences between mean annual earnings of workers by race and sex. The DYNASIM figure was lower than the PSID figure for nonwhite males, and it was higher for females of both races. The discrepancies were on the order of 15 percent. Standard deviations of mean annual earnings of workers by year showed substantial differences between the two models. Hendricks and Holden (1976a:30) note that âin each year the standard deviation of simulated earnings from the simulated mean is 10 to 15 percent lower than the corresponding value from the PSID distribution.â On the other hand, with respect to year-to-year variations in individual earnings, DYNASIM had 27â83 percent more variability than the PSID. In their paper, Hendricks and Holden include suggestions for rectifying some of the inadequacies pointed out by the above comparisons. This is a good example of how model validation can result in an improved model. HENDRICKS AND HOLDEN (1976B) Hendricks and Holden (1976b) examined the utility of using DYNASIM in projecting Old-Age, Survivors, and Disability Insurance (OASDI) costs. In their study DYNASIM was used as a method for developing a longitudinal population sample that was used to test assumptions made by Social Security Administration actuaries in deriving their projections. Hendricks and Holden