The demographic future is uncertain. Conventionally, this uncertainty is conveyed by different scenarios with specific, stated assumptions about the components of population change (fertility, mortality, and migration). Alternative projection scenarios give an indication of possible uncertainty, although the uncertainty is not quantified. A stochastic or probabilistic approach to projections can potentially help their interpretation by quantifying the inherent uncertainty.
This paper outlines a stochastic method and summarises the results for projections of the New Zealand population from a 2009 base. Uncertainty is modelled from historical data for fertility (total fertility rate), mortality (life expectancy at birth), and net migration, as well as for the sex ratio at birth. Uncertainty in the base population is modelled using expert judgement. Simulations of these parameters give probability distributions around Statistics New Zealand’s deterministic mid-range projection.
The results illustrate that the deterministic scenarios give a poor indication of uncertainty for some key demographic characteristics. Even for other characteristics, the uncertainty indicated by the scenarios is neither consistent across the projection period, nor consistent between characteristics. This largely reflects that the low and high deterministic assumptions are not equivalent to a given probability interval that is consistent among the fertility, mortality, and migration components. Moreover, the uncertainty is rarely symmetrical.
There are few practical obstacles to producing stochastic population projections, other than the additional resource required to formulate measures of uncertainty and produce multiple simulations. A stochastic approach could also be applied to other demographic projections produced by Statistics NZ, aiding their interpretation where uncertainty is even greater (eg ethnic population and subnational projections). However, the method outlined in this paper can be applied to the New Zealand population projections without compromising the current methods or results of those other demographic projections.