The forecast data were then compared with the actual data for each two-year period from 1980-1981 through 1988-1989. Given the use of actual growth rates for the period 1979-1989, it is not surprising to find that forecasts of the workforce derived from the model closely track the observed data for both the biomedical and behavioral science workforces. The model forecasts an increase of roughly 25 thousand, from 63 thousand to 88 thousand, for the biomedical workforce. The actual increase was 23 thousand. Similarly, for the behavioral science workforce the model predicts a growth of 19 thousand, and the actual increase was 17 thousand.
The findings are not as impressive for the forecasts of required new entrants. The model vastly underestimated the number needed for the biomedical workforce in 1980-1981, but it tracked reasonably well for the remainder of the period. The average absolute value of the relative forecast error is slightly less than 10 percent.
In the behavioral sciences, the observed number of new entrants increased between 1980 and 1985 and then decreased between 1986 and 1989, while the forecast number increased steadily over the entire period. This resulted in a strong trend in the forecast error, which understates needs by about 10 percent in 1980-1981 and eventually overstates needs by 27 percent in 1988-1989. The average absolute value of the relative forecast error was almost 13 percent. The discrepancy reflects, in part, a difference between the actual and forecast workforce growth rates. The model used the average workforce growth rate for the entire 1980-1989 period. The actual growth rate within the 1980-1989 period varied: it was higher in the first half of the decade than it was in the second half.
The strong trend in forecast errors for new entrants to the behavioral science workforce generated a similar trend in forecast errors for the median age of that workforce. The model consistently understates this age, and the gap between actual and forecast values grows over time. For the biomedical science workforce, the forecasts of median age were very similar to the observed median ages.
The relatively poor performance of the model for the behavioral sciences for a period of slowing growth is instructive. These models are useful in estimating demand that is inherent in the demographic structure of the system (for example, demand changes caused by an uneven distribution of ages that results in variation in retirements). They cannot, however, accurately forecast changes that result from factors not included in the model. Since such factors (e.g., economic conditions) can never be completely taken into account, these models provide, at best, a baseline from which to project the demand for new entrants, given assumptions about changes in overall demand. Where these models are useful is in exploring the sensitivity of demand for new entrants, given assumptions about workforce demand and the mobility characteristics of the workforce.