Monday, Nov 3
Lee, R.D. "New Methods for Forecasting Fertility: An Overview." Population Bulletin of the United Nations, 1(1978): 6-11.
This paper reviews some new approaches to the forecasting of fertility that are primarily of relevance for developed countries. It concludes that methods based on leading socio-economic theories give contradictory forecasts and are as yet inadequately tested. The use of data on expectations collected from surveys appears to disguise the problem of predicting fertility change rather than solve it. However, recent work does suggest several useful new methods for prediction based on persistent aspects of the internal structure of fertility variation. First, it is essential to analyze the autocovariance structure of fertility, using techniques such as those advocated by Box and Jenkins. Secondly, if age-specific or parity-specific rates are analyzed, it is essential to take account of their cross-correlations over a period of time. Thirdly, the purely random component of variation can safely be ignored. Fourthly, all the above points hold true whether or not some additional model for trends in the mean level of fertility is used. Fifthly, an advantage of these procedures is that explicit confidence intervals can be supplied to the user, and work published so far suggests that such intervals will be much larger than those usually attached to official forecasts. A second potentially useful approach is the statistical separation of period and cohort effects, each of which can then be projected, possibly with the aid of the time series methods just discussed. In any case, estimates of exogenous change in contraceptive failure rates can be used to adjust cetcris-paribus forecasts, or perhaps to adjust the basic data before their analysis.