The Treasury

Global Navigation

Personal tools


The Ageing of the New Zealand Population, 1881-2051 - WP 03/27

7  Prospects for continued gains in life expectancy

Whatever method is used to model uncertainty, all population projection methods require assumptions about underlying trends. In the absence of convincing arguments to the contrary, most statistical agencies and academic demographers assume that fertility will decline slightly in OECD countries such as New Zealand where it is currently high, and will increase slightly in countries such as Italy or Japan where it is currently very low. Similarly, “preferred” or “median” migration assumptions are usually based on some slightly modified version of the status quo[4]. Assumptions about mortality trends are, however, subject to much greater debate.

On one side of the debate are those who place a high weight on medical arguments and intuitions about plausible ceilings for human life expectancy. Influential advocates for this position are, for instance, Fries (1980) and Olshansky, Carnes, and Cassel (1990). Statistical agencies have traditionally accepted these views, and have based population projections on mortality assumptions that entail a slowing or complete halt in life expectancy improvements during the projection period. Statistics New Zealand follows this approach, as illustrated by Figure 14.

Figure 14 – Assumptions about female mortality rates used by Statistics New Zealand in 2001-base population projections
Assumptions about female mortality rates used by Statistics New Zealand in 2001-base population projections
Source: Customised tabulations from Statistics New Zealand.

On the other side of the debate are demographers and statisticians who argue for a continuation of historical trends. The most striking demonstration of the persistence of these trends is the series for “best practice” life expectancy assembled by Oeppen and Vaupel (2002: 1030). Best practice life expectancy is the level achieved by the lowest-mortality population in the world. It has been increasing by 2.43 years per decade for women and 2.22 years per decade for men, for a full 160 years, with extraordinary regularity, and no sign whatsoever of a slow-down. (Fitting straight lines to the graphs gives an R-squared of 0.992 for women and 0.980 for men.) Individual countries do move towards or away from the best-practice level, but over the last 40 or so years, there has been a tendency for industrialised countries to converge towards it. Cross-country variance in life expectancy has fallen substantially (Oeppen and Vaupel 2002: Supplementary Material, White 2002: Table 3).

Proponents of the argument for a continuation of historical trends point to the repeated breaching of putative biological limits to life expectancy. A striking example was the claim, made by prominent American scientist Louis Dublin in 1928, that the upper limit to life expectancy was 64.75 years; unknown to him, non-Maori New Zealand females had breached this limit 7 years earlier. A more recent example was the claim by Olshansky et al. (1990) that life expectancy at age 50 could not exceed 35 years; Japanese women surpassed this figure six years after the claim was made (Oeppen and Vaupel 2002: 1030).

Table 3 – Life expectancy at birth in 2051
Female Male
2001 life expectancy + improvement of 2 years per decade 91.0 86.1
Statistics New Zealand’s median mortality assumption 86.5 82.5
Statistics New Zealand’s low mortality assumption 88.0 84.5

Source – Information on mortality assumptions obtained from the documentation for the National Population Projections (2001(base) - 2051) accessed from the Statistics New Zealand website in July 2003.

Assuming continued gains at something like historical rates leads to large divergences from conventional life expectancy assumptions by the end of the projection period. Table 3 provides an illustrative example. Males and females are both assumed to experience gains in life expectancy at the rate of 2 years per decade. Under this assumption, New Zealand women would have a life expectancy of 91.0 years in 2051, and New Zealand men a life expectancy of 86.1 years. Neither figure is implausible: Japanese women currently have a life expectancy of 85 years. However, as Table 3 shows, Statistics New Zealand’s median mortality assumption and low mortality (ie high life expectancy) assumption both entail significantly lower life expectancies.

Table 4 – Statistics New Zealand projection results for 2051
Projection series Total population in 2051 (thousands) Population aged 90 and over in 2051 (thousands)
Medium fertility; high mortality; net migration of 5,000 4,721 109
Medium fertility; low mortality; net migration of 5,000 4,891 169
Difference between low mortality and high mortality series 4% 55%

Source – Calculated from Statistics New Zealand (2002a: Table 8.2)

When life expectancy is already high, further gains only lead to a small proportional increase in the size of the total population. They do, however, lead to large increases in the size of the oldest age groups. Table 4 illustrates this effect, by comparing results from two Statistics New Zealand population projections that differ only in the mortality assumptions. The total population is only 4% larger in 2051 under the low mortality assumption, but population aged 90 and over is 55% larger. The oldest age groups are of special interest for fiscal projections because of their very high health costs. Ministry of Health data indicate, for instance, that government health expenditures per woman aged 65-69 are approximately $3 thousand per year, while expenditures per woman aged 95 and over are $25 thousand per year.

Unfortunately, choosing which historical trend to extrapolate into the future is not straightforward. Different results are obtained for New Zealand, for instance, if male and female life expectancies are modelled separately or jointly, or if the trend is estimated from all available data or from data since World War II. Oeppen and Vaupel (2002) and White (2002) suggest that national projections could have two components: a projection of an international trend, and a projection of the individual country’s divergence from this trend. Lee and Carter (1992) advocate projections that extrapolate trends in logged age-specific mortality rates.

None of these approaches is clearly superior to all the others. They are all likely, however, to yield projected age structures with significantly more older people than is currently expected.


  • [4]For some examples of fertility and mortality assumptions, see the following: Statistics New Zealand at; the United Nations Population Division at; Lee and Tuljapurkar (1994); and Lutz, Sanderson and Scherbov (Lutz, Sanderson and Scherbov 2001).
Page top