1
Demographic transition in Sub Saharan Africa: accounting and economics
Robert Eastwood and Michael Lipton
1. Introduction: the background
Sub-Saharan Africa (SSA) has experienced, and is projected to experience, rapid population growth – from 183m in 1950 to 863m in 2010 and 1753m in 2050. Yet both the rate of natural increase (crude birth-rate minus crude death-rate, i.e. population growth net of migration) and the dependency ratio peaked around 1985. Projected falls in these variables in 1985-2025 exceed, by about a third and a quarter respectively, the preceding rises from 1950. This mirrors a comparable trajectory in much of Asia, where the peak was some twenty years earlier, and where the fall in the dependency ratio has been linked to a large ‘demographic dividend’ in the form of more rapid economic growth (Bloom and Williamson 1998; Bloom et al. 2000).
This paper asks whether such a dividend can be expected in SSA during 1985-2025, as natural increase
slows and the dependency ratio falls. At the outset, one important difference must be noted. While natural increase and the dependency ratio in both regions exhibit hump-shaped trajectories (Figures 1-2), SSA’s levels lie markedly above Asia’s. And the pace of development may depend not only on the trajectories of these variables but also on their levels: for instance, high youth dependency may slow development by reducing savings.
Demographic transition
Demographers (e.g. Coale 1973, Montgomery 2009) identify four stages, of which only the middle two are part of transition as such: pre-modern equilibrium, with crude birth and death rates (CBR and CDR) both around 35-45; urbanising/industrialising, with CBR unchanged but CDR falling towards 15; mature industrialising, with both rates falling towards 10; and post-industrial equilibrium, with the rates roughly equal at 10 or less. The recent emphasis given by economists to the effects of changes in the dependency ratio – and the balance within it between young and old dependents – has led some to break the transition into three phases (Lee 2003, fig.6):
Phase 1 (c. 25 years): Rising dependency ratio driven by rising young dependency.
Phase 2 (c. 40 years): Falling dependency ratio driven by falling young dependency.
Phase 3 (c. 50 years): Rising dependency ratio driven by rising old dependency.
We adopt this scheme, noting that the start of Phase 2, defined by peak dependency, also approximately coincided with peak natural increase in Asia and SSA (Figures 1, 2). We concentrate on the possible dividend that SSA may reap in Phase 2.
Crucial for the start, and speed, of Phase 2 - i.e. for the move to falling dependency and natural increase - is the shift to falling TFR and hence birth-rates. Malthus (1824), from regional census data from Switzerland and Norway, suggested that ‘extreme healthiness’ accelerated ‘prudential checks’: in today’s language, that lower mortality (we might add: especially among children) induced behavioural change that reduced total fertility and hence CBR. That remained the standard view of demographic transition theory at its formulation (Thompson 1929) and during modern development (Coale and Hoover 1958, pp.12-13). Of the proximate determinants of TFR (Bongaarts 1978, 1982), falling child mortality - CMR, used in this paper to mean per-thousand live-born children who die before age 5 - may quantifiably affect three:
post-partum infecundability, use of contraception, and proportion married (Preston 1978, but see Montgomery and Cohen 1998).
This view of fertility transition has been challenged by theory and evidence that marital age and fertility are determined jointly with other variables, notably female labour supply, within some kind of household optimizing framework (Becker 1960, 1981; Schultz 1981, 2007). This view suggests that (a) factors other than falling child mortality cause fertility decline, notably (prior) female education and female wages (reflecting opportunity-cost of motherhood); (b) without such factors, child mortality decline - which on its own raises natural increase - may fail to induce enough fertility decline to reduce this, let alone to a post-industrial equilibrium of zero (Preston 1978, Doepke 2005). Whether falling young-end mortality raises or lowers the expected (or average) number of surviving children is unclear. The reduced expected cost of rearing a new-born to adulthood induces couples to plan for more surviving children (Tzannatos and Symons 1989; Birchenall and Soares 2009;). Working the other way is the ‘dynastic’ motive: if couples wish to have a given chance of attaining a fixed number (e.g. 2) or more of adult offspring, then a reduction in risk will cause them to plan for fewer surviving children on average (for example, with zero risk, they would plan to have exactly 2). Econometric attempts to resolve such issues have been bedevilled by the need to control for other determinants of fertility, such as female education, and for the endogeneity of infant mortality (either because IMR and TFR are jointly determined by other variables, or because less frequent births change the risk of child mortality). Nevertheless, recent literature using best-practice econometric methodology concludes that a fall in young-end mortality is the main driver of TFR decline (section 2).
Impact of phase 2 demographic transition on growth of income per person: accounting
Projection of this impact in developing countries was pioneered in 1957 for India (Coale and Hoover 1958). Losses in Phase 1, and especially gains in Phase 2, seemed small, because both peaks and subsequent falls in the dependency ratio were under-projected. It was projected to fall only from 52 to 50 from 1986 to 2006, if post-1986 TFR stayed at half the 1956 rates, (ibid. pp.233, 322). In fact India's TFR fell more slowly than projected, not reaching half of its 1956 value until about 2000. This raised the dependency peak (82 in 1965) and the subsequent rate of fall, to 73 in 1985 and 60 in 2005 (UN 2009), so that much larger economic effects from age-structure change now seemed plausible. Kelley and Schmidt (1995) and Bloom and Williamson (1998), disaggregating fertility and mortality, claimed that the Asian tigers, and to some extent South Asia too, had gained a large demographic dividend as a result of age-structure change during Phase 2 and, moreover, that this was the total effect of demographic change.
To set this age-structure hypothesis in context, an accounting approach is useful. Figures 1 and 2 show two features of Phase 2 of transition: age-structure change and falling natural increase. Each can be held to yield growth benefits from a simple accounting standpoint. For age structure, a falling dependency ratio implies arithmetically that a given rate of growth of output per worker translates into a faster rate of growth of output per person. For natural increase, the slower it is, the less an economy needs to save in order to maintain total (natural plus reproducible) capital per person, so the higher is sustainable consumption per person. In an economy where natural increase is falling through time, therefore, the amount that has to be saved in order to equip extra workers is also falling, so that the amount left over and available for (sustainable) consumption is growing through time. Expressing this somewhat differently, any natural increase means that a given stock of capital is ‘diluted’ by having to be shared among more workers: so falling natural increase delivers a demographic dividend via reduced dilution.
Accounting thus allows us to calculate two types of demographic dividend, one from improved age structure and one from slower natural increase. These dividends are not strictly commensurate, as one comprises extra growth of output per person, and the other extra growth of sustainable consumption per person. Nevertheless we estimate and compare the dividends for a number of countries in section 4: in SSA the age-structure gains are likely to be some two to three times as large as those from falling capital-dilution.
This accounting, however, has severe limitations. For the age-structure accounting to be the whole story, there must be no effects of Phase 2 on the time path of output per worker. However there may be positive effects either from age-structure itself, if for instance a fall in youth dependency raises the savings rate, or from slower natural increase, since this implies that given savings are spread less thinly, tending to raise capital per worker and therefore output per worker. The first effect fits into the age-structure hypothesis, simply implying that the effects are more than arithmetical, while the second does not. Similarly, for the natural-increase accounting to be the whole story, we must assume that sustainability can be identified with constant capital per person, and thus must neglect technological advance, in particular the possibility that age-structure change might stimulate such advance. The natural-increase accounting also has a normative character, calculating extra savings ‘needed’ for sustainability under given circumstances, but silent on whether transition might make these savings either more or less likely to materialize.
To assess the effect of demographic change on growth in output per person, there is an alternative to the accounting approach: econometric estimation. This literature, reviewed in section 4, strongly supports the age-structure hypothesis. Not only does falling dependency increase growth arithmetically, as described above, but the effect - i.e. the demographic dividend – is more than arithmetical. Moreover, no link from natural increase to growth is found. These findings cannot, however, be taken as final, because of doubts over the robustness of the econometric methodology employed.
Overshadowing the demographic dividend in Phase 2 in SSA, and highly pertinent to any comparison with Asia, are the consequences of SSA’s combination, throughout transition, of much more rapid natural increase (Figure 2) and much lower savings rates. In Asia, not only did the slowing of natural increase raise sustainable consumption per head during 1965-2005, but savings were high enough that the path of development was itself broadly sustainable. This is very unlikely to be the case in SSA over 1985-2025 (section 4). Slower natural increase will indeed raise sustainable consumption per head by reducing the savings that would be needed to sustain capital per person, but this pales into insignificance beside the likelihood that savings will fall far short of the necessary level.
Outline and contributions of this paper
Section 2 states the paths of main demographic variables during the three phases of transition. We analyse natural increase rather than the rate of growth of population, because the causes and consequences of migration would require a major digression. We then enquire whether UN middle projections, used in this paper, correctly predict Phase 2 transition. Being global and univariate, they do not allow for region-specific impacts on fertility of causal variables, such as female literacy and young-end mortality. In particular, where the latter is higher, or falling more slowly, than the global norm, these projections may overstate future fertility decline, and therefore reductions in both the dependency ratio and the rate of natural increase.
Section 3 tracks transition data in SSA, its regions, and its most populous countries. In most cases, natural increase and dependency peaked (bringing in Phase 2) at higher levels than in Asia, and about twenty years later, around 1985. This sets the stage for larger total declines in dependency and natural increase in Phase 2. However, UN middle projections to 2030 show annual declines lower in most of SSA than comparable falls twenty years earlier in Asia and even those projections look optimistic. But the future can be changed. Parts of Asia and Africa experienced fast reductions in young-end mortality, and measures to transmit these rapidly into less pro-natalist behaviour. It is medically possible for much of SSA to do the same. Yet total fertility decline has been sluggish in some large West and Middle African countries, and has stalled in many, but not all, groups and regions of East Africa (Ezeh et al. 2009). Groups and areas slow to reduce natural increase come to represent growing proportions of populations. Most groups and regions will need faster reduction of child mortality, plus measures influencing fertility decisions more directly, for Africa as a whole to reap a large demographic dividend.
Section 4 moves to estimates of the demographic dividend. Giving appropriate Asian comparisons, we evaluate econometric evidence as it affects SSA’s dividend so far, and present alternative accounting estimates for 1985–2025. As is evident from Figures 1 and 2, both the dependency ratio and natural increase in SSA reach a higher peak and then decline more slowly in SSA compared to Asia. Therefore the accounting dividends are for SSA larger in total but smaller per year. Yet, as noted above, arguably more important than the dividends from demographic transition as such are the effects of continuing natural increase, when set in the context of national savings rates. We estimate these underlying rates of capital dilution for populous SSA countries, with comparative estimates for some Asian countries.Section 5 concludes.
The main contribution is to bring together the elements needed for an assessment of the demographic dividend in Sub-Saharan Africa, reviewing past estimates and presenting some new calculations. Since all this must rest on demographic projections, we first set UN middle projections in the context of some determinants of fertility, and of the extent to which policy might modify them. Given the demography, we calculate two sorts of accounting dividend for the region and its populous countries, and assess the influential view that the dividend arises entirely from changes in dependency. We show that, under current policies, adverse effects of ongoing rapid natural increase in much of Sub-Saharan Africa exceed likely gains from its reduction during transition.