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 268 Kormendi and Meguire considered the 50 states and the District of Columbia as mini TRIM2 models, which allowed them to make use of regression analysis in their validation of TRIM2. First, for several outputs, they regressed the forecasts on the âtrueâ values from the CPS and compared the estimated regression coefficients to 1, which they defined as one type of unbiasedness. In addition, they also regressed the logarithm of the ratio of estimated1985 to estimated1979 on the logarithm of the ratio of true1985 to true 1979, which compares the observed forecast percentage change with the actual percentage change. Again, a regression coefficient of 1 indicated unbiasedness (i.e., that the forecasts were not systematically high or low with respect to percentage change). Kormendi and Meguire also used reverse regression to examine whether the forecasts might be improved through correction of the bias. Another application of regression used by Kormendi and Meguire was in the context of a sensitivity analysis. They attempted to attribute the changes from 1979 to 1985 to either administrative or economic/ demographic changes. To do this they ran TRIM2 based on the 1985 CPS on 1979 law, and they also ran TRIM2 based on the 1979 CPS on 1985 law, in addition to the usual runs of the 1979 CPS on 1979 law and the 1985 CPS on 1985 law. They were able to measure the percentage change in forecasts conditional on the law remaining constant, and this was added as a covariate in the regression described above. This provided a way of controlling for variation due to changes in economic and demographic conditions. The forecast errors of TRIM2 for the period 1979 to 1985 were much smaller for benefits than for units. The absolute value of the errors for units generally ranged from 5 to 30 percent, while those for benefits ranged from 1 to 6 percent. The regression of forecasts on CPS for 1979â1985 found that for benefits TRIM2's forecasts were only moderately biased, with a coefficient of 0.87. Similar results were obtained for 1983â1985. However, for units the forecasts and the CPS numbers were discovered to be essentially uncorrelated for both time periods. BETSON (1988) Betson (1988) examined the variability of estimates of the cost of a credit income tax (CIT) from the Kasten-Greenberg-Betson (KGB) microsimulation model as a function of the variability of behavioral parameters. This model was run on a sample of 16,200 unweighted households with children from the 1986 CPS. Betson modeled postreform earnings as a product of the gross wage of the individual, the individual's employment rate, and a linear function of the individual's net wage rate and linearized virtual income. Estimating the variability of postreform earnings necessitated estimating the covariance matrix