Cohen J E
Demography. 1986 Feb;23(1):105-26.
This paper compares several methods of generating confidence intervals for forecasts of population size. Two rest on a demographic model for age-structured populations with stochastic fluctuations in vital rates. Two rest on empirical analyses of past forecasts of population sizes of Sweden at five-year intervals from 1780 to 1980 inclusive. Confidence intervals produced by the different methods vary substantially. The relative sizes differ in the various historical periods. The narrowest intervals offer a lower bound on uncertainty about the future. Procedures for estimating a range of confidence intervals are tentatively recommended. A major lesson is that finitely many observations of the past and incomplete theoretical understanding of the present and future can justify at best a range of confidence intervals for population projections. Uncertainty attaches not only to the point forecasts of future population, but also to the estimates of those forecasts' uncertainty.
本文比较了几种生成人口规模预测置信区间的方法。其中两种方法基于具有生命率随机波动的年龄结构人口的人口模型。另外两种方法基于对1780年至1980年（含）期间瑞典人口规模每五年一次的过去预测的实证分析。不同方法产生的置信区间差异很大。不同历史时期的相对大小也有所不同。最窄的区间给出了关于未来不确定性的下限。本文初步推荐了估计一系列置信区间的程序。一个主要教训是，过去有限的观测以及对当前和未来不完整的理论理解，最多只能为人口预测提供一系列置信区间。不确定性不仅附着于未来人口的点预测，还附着于这些预测不确定性的估计。
