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.
EVALUATING THE ACCURACY OF U.S. POPULATION PROJECTION MODELS 309 yield different population trajectories. Neither the projections of the Bureau of the Census nor those produced by the Social Security Administration are unconditional forecasts of future population size or growth, although many users may want to interpret them that way. To obtain accurate unconditional forecasts, one must correctly anticipate future paths in age- and sex-specific fertility, mortality, and international migration. Because neither the Census Bureau nor the SSA's Office of the Actuary is in a position to know these future trajectories, the customary procedure is to publish a series of projections, each embodying alternative assumptions about fertility, mortality, and migration. Publishing a variety of projections also helps emphasize the conditional nature of the forecasts. These alternative assumptions are typically of the âhigh,â âmedium,â and âlowâ variety, and while claims regarding scientific confidence intervals are rarely made, the high and low variants are usually chosen to reflect the outer limits of what is considered likely to materialize. The high and low projections, in other words, are intended to âbracketâ likely futures. Users must then select for themselves which set (among many) of underlying assumptions they consider most likely. These traits of population projection models suggest that, even though the projections are conditional forecasts, most researchers engaged in validating such models do not confront the same degree of difficulty they would if they were trying to validate microsimulation outputs. First, even when the âifâ part of a population projection does not materialize, the projection qua projection is valid unless some computational error has been made in linking population inputs to projection outcomes. Second, unlike microsimulation models, it is not difficult to separate specific sources of uncertainty in population projections. They are ultimately linked to just four model components: the baseline population or assumptions governing fertility, mortality, and international migration. Third, to complement the decennial censuses, there is an ongoing annual effort by the Census Bureau to produce postcensal population estimates disaggregated by age, race, and sex. These annual estimates can and indeed have been used to âtrackâ population projections to see how closely they correspond to the eventual truth. VALIDATING POPULATION PROJECTIONS According to Webster's dictionary, to validate means to âconfirm the correctness of.â In Part I of Volume I, three complementary types of model validation strategies were identified: (1) external validation, or assessment of the validity of a model's estimates compared with measures of the truth; (2) estimation of variance in the statistical sense; and (3) sensitivity analysis. These validation exercises may also be applied to population projection models. Sensitivity analyses would include an examination of how sensitive projected population totals are to variations in underlying conditions of fertility, mortality, and