Population projections should not be treated as predictions. They illustrate
growth and change in the population which will occur in the projection period based on a
chosen set of assumptions about the level and direction of underlying values or rates of
the components of population change. These specified assumptions may or may not be
correct. By definition, a projection must be correct if its assumptions are valid.
The time series approach to forecasting the number of births presented in this
thesis is a useful supplement to other birth estimate approaches such as the cohort
component method. This method is currently used by the Australian Bureau of Statistics
to estimate the number of births for Australia. The cohort component method implicitly
relies on judgments about the future rates of growth of the components of population.
While the judgmental methods explicitly rely on an assumed rate of population growth,
these rates for projections for Australia are formulated on the basis of an assessment of
past demographic trends, both in Australia and overseas, and their likely future
movements. However, no statement is made on the probability that such rates will
actually be realized. Obviously, the birth projections are largely sensitive to the rates
upon which they rely. In recognition of the uncertainty of these rates and the sensitivity
of projections, it is more often sensible to provide alternative projections in order to
supply users with a possible range of options under different assumptions. As we have
seen in Sections 8.2 and 8.3, the Australian Bureau of Statistics provides two sets of
birth projections for Australia. The first set has two projection series generated without
overseas migration. The second set has three projection series generated with overseas
replacement for the cohort component method but rather as a complement to it. The
cohort component method yields point projections only. In contrast, the time series
approach yields both point and interval forecasts (see Section 7.4) based on historical
data. The interval forecasts provide information about the precision of the point forecasts
that is reflected by the widths of the forecast intervals. It may help decision-makers in
planning developments for the future according to the degree of certainty of these
forecasts.
9.2 Extension
As we have seen from Sections 8.2 and 8.3, the inclusion of overseas migration
in the evaluation of birth projection affects the results of the projection. We have already
discussed the closed-system cohort survival model and single-region open-system cohort
survival model in Section 2.1. Since the open-system model takes the effect of migration
into consideration, it seems to be more realistic than the closed-system model. However,
the drawback is that since only net migration rates are used, these rates do not give
information about the origins and destinations of migrants. Therefore the information
which is important for predicting migration itself and other demographic variables is not
available. Moreover, an apparent small net migration flow may be the results of large
gross flows in and out. Knowing these flow in and flow out figures is always useful in
improving predictions for migration and for other demographic variables. Another
weakness of the open-system model is the failure to recognize surviving and migrating
infants. In fact, the data collection of origins, destinations, gross flows and migrating
infants (or fairly common use of corresponding rates) can be achieved easily. This
model to the multi-region cohort survival model (Rogers (1966,1968,1971,1973)). It is
worthwhile to note that in the multi-region model, the female populations and survival
rates are all region-specific and the migration and fertility rates are specified by both their
origins and destinations.
We have shown in Section 3.1 that univariate time series models for birth data
can be obtained from the stochastic discrete renewal equation for births. This renewal
equation is developed from the deterministic single-region open-system model (see
Section 2.2) which subsequently modified to allow the fertility, survival and migration
rates varying with time (see Section 3.1). Since the probabilistic time series models and
the deterministic multi-region model are known as two different approaches for
forecasting births, we believe that the relation between these two types of models is
worth further investigation. Multivariate time series models for births for each region
may also be developed from the multi-region model and, as a result, the number of births
for each individual region and the total number of births in the regions being considered
can be forecast. It is that also possible to forecast not only the total number of births, but
also their distribution among these regions. In our case, we may obtain the forecasts of
births for each State and Territory and for Australia as a whole. These pieces of
information relating to the size and distribution of births help planning and developments