Introduction: The need to rethink approaches to population forecasts

Three major groups call on demographers to produce medium- and long-term population forecasts at the national, regional, or global levels - or produce them themselves. They are: other scientists, government and international agencies, and the general public, including private industry. What these consumers of forecasts demand of demographers, or what demographers think that they should demand, has been changing. The types of forecasts demanded are changing, the relevant dimension of forecasts is expanding, and users are increasingly requiring that forecasts include an indication of the degree of uncertainty of the forecast. Because the demands placed on demographers for population forecasts have been changing, it is an appropriate time to rethink some of their basic aspects. In this volume we address what we see as key issues in population forecasting: in what dimensions and at what levels of disaggregation should forecasts be provided? (And, in particular, are the traditional dimensions of age and sex sufficient?) Should population forecasts take note of limits to population or interactions between population and other variables? And how should uncertainty be treated? We believe that, at least in part, these issues are driven by changes in what users of forecasts want from population forecasters.
The reader will note that we have used the term "forecasts" rather than the more common "projections." Demographers claim to produce population "projections," which are correctly computed numerical outcomes of a specified algorithm whose form, initial values, and controlling parameters or transition values are specified by the analyst. By definition, a projection must be correct unless arithmetical or other errors are made. However, users of population projections require population "forecasts." Forecasts are what Donald Pittenger (1980) called a "population projection selected as a likely outcome." Thus although a demographer makes a "projection," the user employs it as a "forecast." Some demographers cling to the distinction and wash their hands of what users do with their "projections" or how they interpret them. But we think that this distinction between "projections" and "forecasts"is false because demographers present only one or a limited number of the many possible projections. On what basis do they choose the projection or set of projections? Surely, on the basis that they judge the projection (or central projection of a set of projections) to be the most likely to occur. This point was made almost 50 years ago by Harold Dorn: it is difficult to see why a demographer would present anything other than the most likely outcome as the preferred middle projection (Dorn 1950). Similarly, although users are told that high and low population forecast variants are not confidence intervals, they are often taken to be so by users. And why should they not? Why else would a high and low variant be reported unless the demographer thought that they indicated the highest numbers and lowest numbers that were possible, although not highly probable? For these reasons, we favor the term "forecast" over the term "projection" where there is any, even implicit, predictive intent.

VIEW CONTENT