addition, important flow parameters (e.g., death and retirement rates) are expected to be reasonably stable. This strengthens the expectation of reliable results.
Among the limitations of these models is their reliance on longitudinal databases for estimation of the flows. Such databases are expensive to compile; hence they are relatively scarce. Moreover, such databases are vulnerable to possible response and selectivity biases, which could bias parameters derived from them. For example, if people who change fields are less likely to answer the questionnaire than those who remain in a field, a field mobility parameter estimated from the sample will be smaller than the true parameter. Further, the degree of reliable disaggregation such databases will allow will be limited by their sample sizes. This means that fields that display disparate behavior may be lumped together, and important aspects of behavior for separate fields may be missed. In addition, parameter estimates may vary due to changes in the composition of the heterogeneous sample, even though they might be stable for a particular field taken separately.
Another possible limitation arises from the assumption usually made by such models that their estimates of transition rates are given and not sensitive to market conditions and/or other factors that could affect flows into, out of, and within these workforces.7 If transition rates are influenced by factors not included in the description of the demographic system, projections using them are likely to prove inaccurate. Demographic models can, however, provide "baseline" analyses that reflect the demographic characteristics of the population being modelled. Adjustment to market conditions can then be incorporated in the demographic model. For example, salaries, which adjust to excess supply or demand, will affect transition rates. If the magnitude of this effect is estimated, transition rates that are assumed fixed in the demographic model can then be made endogenous in a model that incorporates this adjustment to market conditions.
Despite the limitations of demographic models, the committee felt that it could illuminate a number of aspects of adjustment in the biomedical/behavioral workforce that were muddied or improperly specified in the earlier models constructed for NIH.