Le traitement de l'incertitude dans les perspectives démographiques

Often the predicted trends are extrapolations of trends observed before the projection's initial date, while reality may present an unexpected return of past trends or irregularities. Uncertainty being related to the impossibility to accurately predict the future, errors in the predictions are the manifestations of such uncertainty and come from the lack of accuracy which affects the extrapolation of the components of demographic change. The lack of accuracy increases as the date of the beginning of the projection gets more remote. Errors tend to increase with the length of the projection. Most of the authors of official demographic predictions construct several time series based on a combination of scenarios. However, they do not provide the probability that the real outcome will be inside the interval defined by the high and low scenarios. This paper gives an outline of the measures used to evaluate the accuracy of the perspectives and determine their advantages and limitations. It also proposes to examine the errors ex-post by means of a method which breaks down the effect age-period-cohort, or to apply a chronological series model to the projection errors detected in variables from earlier projection so as to establich a confidence interval for these variables and use it in the new projection. Finally, it suggests to work with stochastic models or to use a method of population probabilistic projections based on experts' opinions regarding future trends in fertility, mortality and mobility as well as the level of uncertainty linked to these trend