Forecasting vital rates (cont.)
Can knowledge improve forecasts?
The basic tool for forecasting is extrapolation of past trends. In fact, the only basis we have for forecasting at all is the assumption that future trends will in some respect be a development of trends in the past.
While all forecasts are based on extrapolation from the past, a considerable debate exists about whether forecasts should be based on a theoretical understanding of the determinants of population trends or simply on the mathematical extrapolation of past trends. The first approach seems logically preferable. However, limited evidence exists that it performs better in practice.
One problem with including explanatory variables in the modelling of trends is that, even when it is known that A affects B, it is often difficult to specify how much change in B will result from a given change in A. Moreover, it may just shift the forecasting problem one step back to trying to predict the explanatory variable. For example, even if we understood exactly how changes in standards of living affected fertility or mortality (and we do not), it is unclear that it is possible to forecast the standard of living of a population in the future any more accurately than one can forecast fertility and mortality directly.
Thus, the case for incorporating explanatory variables into the modelling of trends is strongest when long lags exist between the exposure and the effect. For example the educational attainment of cohorts of teenage girls has been used to predict their subsequent fertility in developing countries, cigarette smoking among young adults has been used to predict their mortality in middle and old age, and HIV prevalence has been used to predict subsequent AIDS mortality.
Trends in tobacco use and lung cancer death rates* in the US
Source: American Cancer Society, Cancer Facts & Figures 2013
Asking women about their fertility intentions has proved of limited use for predicting fertility. Many demographers now believe that fertility responds to successive period influences rather than to enduring preferences for a particular family size formed in childhood. Nevertheless, survey data on the fertility intentions of women who already have already started their families can be of value in forecasting how many of them will go on to have an additional birth in the short-to-medium term.