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 260 HOLDEN (1977) Holden (1977) conducted a follow-up validation of DYNASIM's estimates of earnings. Three of the problems identified in Hendricks and Holden (1976a) were that (1) DYNASIM produced too much year-to-year variation in individual earnings and too little variation in average earnings over time, (2) it failed to reproduce differentials in average earnings by years in the labor force, and (3) it produced too few women with labor force attachment. Hendricks and Holden, while successful in identifying these shortcomings, did not indicate their likely source. Holden's (1977) study was restricted to an analysis of DYNASIM's estimates of earnings histories for white females, including the decision to work, the number of hours worked per week, and the hourly wage. His analysis made use of the PSID, using DYNASIM to provide labor force attributes for a 6-year period. These attributes were then compared with the actual values for the PSID sample. The largest absolute discrepancy for DYNASIM with respect to replicating the distribution of years in the labor force was 6 percent for the estimate of the number of women who worked all 6 years. DYNASIM generated forecasts of labor force participation as a function of those for the previous year (i.e., as a first-order Markov chain). To rectify the problem, participation was allowed to be a function of labor force participation for the previous 2 years, without success. Another approach examined was to assume that all women have, independent of events such as marriage and childbirth, an individual propensity to participate in the labor force. The model used to approximate this effect was as follows: assign every woman a percentile x, a random draw between 0 and 1. Then, if x>1âp, let p*=(7pâ2)/5p+(2â2p)x/5p, and if x<1âp, let p*=2px/(5â5p). This separated the probabilities into two uniform distributions with a large gap in between, with an overall mean of p. The results represented a slight general improvement over DYNASIM's Markov model, but some discrepancies remained. The largest discrepancy was that the new model estimated the probability of participating in the labor force for unmarried women age 56â60 who did not participate the previous year as 30 percent, while the actual value from the PSID was 11 percent. The DYNASIM model of hours supplied took the overall mean hours supplied for individuals with specified characteristics and added a stochastic term to match observed variability. Continuity in hours worked over time was maintained through the use of an autoregressive parameter. If important characteristics that affect hours supplied were omitted from the characteristics generating the above means, the result would likely be too much year-to-year variation in an individual's hours in the labor force and too little variation in lifetime hours across individuals. Holden again created an alternative model that attempted to measure the