Population projections: revealing multifaceted sources of their immanent uncertainty

Christina Bohk, University of Rostock

Population projections are uncertain by their very nature, as the prediction of human behavior concerning demographic events, like births or migration, relies on complex individual decision processes. Besides the projection-immanent uncertainty when generating assumptions, every step of a projection process adds various sources of uncertainty. An important question in this context is what a good projection looks like. A popular criterion is accuracy, measured by deviances between projected and observed outcome. Commonly used error measures for accuracy are constructed for deterministic rather than probabilistic population projections. Since error measures for deterministic projections can not easily be applied to probabilistic ones, specific error measures for probabilistic projections need to be developed. Moreover, a comprehensive evaluation of a projection model requires considering other important criteria like usability, consistence, and reliability as well. At first, an overview of uncertainty sources will be presented, followed by a brief survey on error measurements for various evaluation criteria. Then, methods to cope with some of the uncertainty sources will be described in the context of the Probabilistic Population Projection Model (PPPM). These methods are based on certain properties of the system specification, e.g., by assigning occurrence probabilities to flexibly and exogenously generated assumptions for each model parameter via expert judgment. Despite the consideration of relevant uncertainty sources, the presented methods - counter-intuitively - do not necessarily narrow down the resulting confidence intervals. Finally, future research challenges will be outlined.