Postponement and limitation of childbearing in the transition to low - - PDF document

Draft: 21 Oct. 2017. Please ask the authors for an up-to-date version before citing. 1

Postponement and limitation of childbearing in the transition to low fertility in the developing world

Ian M. Timæus1,2 and Tom A. Moultrie2

1 Department of Population Health, London School of Hygiene & Tropical Medicine; 2 Centre for

Actuarial Research, University of Cape Town. (email: ian.timaeus@lshtm.ac.uk)

Abstract

This paper examines the contribution that postponement of births, as opposed to parity-specific limitation of family size, has made to the global transition to low fertility. It investigates whether the role of postponement varies across sub-Saharan Africa and whether it is important elsewhere. We use survival analysis to model data from 297 fertility surveys in 78 countries and assess changes in period parity progression, the median length of closed birth intervals, and duration-specific fertility. Outside sub-Saharan Africa, fertility transition is dominated by parity-specific limitation. Equally, births are now being postponed in many countries and birth intervals are lengthening. Where progression has dropped in sub-Saharan Africa, it has not done so on a parity-specific basis. Birth intervals are lengthening almost everywhere in the region and, at every level of overall fertility, are longer than elsewhere in the world. A few countries outside sub-Saharan Africa have undergone an ‘African’ transition. Moreover, a few others have experienced an across the board decline in period fertility characterised by neither stopping nor postponement. We conclude that postponement is a family-building strategy of global significance. Nevertheless, fertility transition in sub-Saharan Africa is following a different track from that taken by most other countries in which postponement alone has driven down fertility. The pace of fertility decline in the region will remain slow until women start limiting their family sizes as well as postponing having another birth.

Introduction

This paper examines the transition to low fertility across the developing world. It investigates the extent to which fertility decline has been driven by the parity-specific limitation of family size in different regions of the world. In earlier papers, we have argued that the initial phases of fertility transition in sub-Saharan Africa have been characterised by substantial lengthening of birth intervals (Moultrie and Timaeus 2003; Moultrie, Sayi and Timæus 2012). This, we propose, reflects postponement of births, which we conceptualise as a third motivation for birth control, distinct from both limitation and spacing (Timæus and Moultrie 2008). Postponers have decided neither that they have enough surviving children nor that they want another child once their youngest child is old enough. Rather they are deferring the decision as to whether and when they want more children – they have decided only that they want no more for the

• moment. Their efforts to avoid childbearing reflect not their reproductive histories (the

number, ages and survival of their children), but other circumstances in their lives. This interpretation of women’s motives for adopting birth control accords with evidence from qualitative research in several parts of Africa (Agadjanian 2005; Johnson-Hanks 2004; Towriss et al. 2017). In a recent paper, Casterline and Odden (2016b) have argued that birth intervals have lengthened substantially in the course of fertility transition not just in sub-Saharan Africa but in many other non-Western countries. In a follow-up article, they present evidence

Draft: 21 Oct. 2017. Please ask the authors for an up-to-date version before citing. 2 that indicates that this trend is underlain by the widespread postponement of births (Casterline and Odden 2016a). They conclude: ‘Do the available empirical data support the claim that postponement is a distinctly African phenomenon? Our conclusion is No … there is abundant evidence of postponement in other regions; … But even so, is postponement far more prevalent in Sub‐Saharan Africa? Here the evidence is mixed.’ On this basis, they are sceptical about the claim that was encapsulated in the title of our 2012 article that Africa is experiencing ‘a new kind of transition’. Casterline and Odden’s identification of postponement outside Africa is important. It widens the challenge that our own work has posed to conventional demographic conceptualisations of women’s reasons for adopting birth control. It has never been our position, however, that postponement is restricted to Africa: this is not something that we have investigated previously. What we have argued is that in sub-Saharan Africa: ‘birth intervals are largely independent of mother’s age and parity. By contrast, data from selected less developed countries in other regions, and from Europe early in its fertility transition, exhibit very different patterns’ (Moultrie, Sayi and Timæus 2012). Thus, our hypothesis is that postponement has emerged as a relatively important driver

• f fertility transition in sub-Saharan Africa because parity-specific limitation of family

size has been less prevalent in this region than elsewhere in the world. This paper has two principal aims. First it uses a standardised analytical framework to examine the contributions that stopping, spacing and postponement have made to fertility decline in as many countries as possible. In particular, the paper assesses our hypothesis that parity-specific family-size limitation has played a limited role in fertility transition in sub-Saharan Africa against evidence from developing regions across the

• world. Second, we investigate further whether, as Casterline and Odden (2016a) suggest,

patterns of fertility decline differ within sub-Saharan Africa, with postponement being confined largely to the south of the region while declines elsewhere in sub-Saharan Africa ‘have been driven primarily by limiting rather than postponement’.

Data and methods

Our analysis uses the data from 297 World Fertility Survey (WFS) surveys, Demographic and Health Surveys (DHS), and Reproductive Health Surveys (RHS) from 78 countries to investigate trends and patterns in family formation over the course of their fertility

• transitions. The WFS was conducted in the late 1970s and early 1980s, and collected full

birth histories that allow one to look back a further 15 years. Our analysis therefore covers the period from the mid-1960s through to the present day. This database of demographic surveys includes every DHS in the public domain in early 2017 and all the RHS that collected birth histories. It largely overlaps with that assembled by Casterline and Odden (2016a) but includes a few more WFS and several surveys conducted in 2014–15 and excludes four Multiple Indicator Cluster Surveys and nine WFS surveys of countries that have not conducted a fertility survey subsequently. Nearly half the surveys were conducted in sub-Saharan Africa, but the database includes surveys from most parts of the world apart from Western Europe, North America, Australasia and Oceania (Table 1). It does not include China.

Draft: 21 Oct. 2017. Please ask the authors for an up-to-date version before citing. 3 All the survey data from a single country were combined, preserving the sample design, clustering and weighting aspects of each survey. Each woman’s birth history was split into episodes defined by quinquennial period and the interval since her previous birth (splitting at durations 9, 18, 24 … 72, 84 … 144 months) and her parity and age at the start of each resulting episode was calculated. We then conducted a period-based analysis of fertility. Survival analysis, specifically a Poisson model with exposure offset, was used to model the trend in the log age-order-duration specific birth rates in each

• country. Detailed descriptions of this model are available elsewhere (Moultrie, Sayi and

Timæus 2012; Towriss and Timæus 2017). The key features of the model are that:

• The main effects of parity, age, interval duration and period are modelled using

dummy variables rather than by imposing a specific functional form on the data.

• Differential change in fertility by parity and interval duration is modelled using

continuous variables and their interactions to average across random fluctuations in the data and make it possible to interpolate between surveys to estimate a complete set of rates for older women.

• The age dummies pick up the biological decline in fecundity with age and are

held constant over time to avoid overfitting the data. First births were modelled separately as a function of quinquennial period and women’s age in years using a similar, but simpler, Poisson regression model.

Table 1: Fertility surveys included in the analysis according to date of fieldwork and region

Period Region & Sub-Region 1975- 1979 1980- 1984 1985- 1989 1990- 1994 1995- 1999 2000- 2004 2005- 2009 2010- 2014 2015- 2019 Total Sub-Saharan Africa East 1 1 4 7 9 10 7 12 4 55 Middle 1 2 2 3 3 5 1 17 Southern 1 1 1 2 3 2 10 West 2 3 5 6 10 6 9 15 1 57 TOTAL 5 4 9 16 22 21 22 34 6 139

• L. America & Caribbean

Caribbean 5 1 5 4 6 6 6 4 37 South America 5 4 6 7 6 5 4 37 TOTAL 10 1 9 10 13 12 11 8 74 Middle East & N. Africa 5 2 4 5 3 5 5 4 1 34 Central & Eastern Europe 3 3 South & South-East Asia 7 3 6 7 6 9 8 1 47 TOTAL 27 7 25 37 45 44 50 54 8 297

The output from the regression model is a complete series of age-order-duration-specific fertility rates for each five-year period covered by the fertility survey data. Using a methods devised by Rallu and Toulemon (1994), we construct detailed synthetic cohort indices of fertility for each period from these. This entails conducting a multistate life table analysis to calculate the final fertility outcomes that a cohort of pubertal girls would experience by the end of their reproductive lives if they went through life experiencing the age-order-duration-specific rates of the period in question. Thus, the life table analysis provides a complete set of parity progression ratios for the period, together with

Draft: 21 Oct. 2017. Please ask the authors for an up-to-date version before citing. 4 period estimates of the median duration of the birth intervals of each order and of all

• rders. Moreover, by summing the synthetic cohort estimates of births of different
• rders, one can calculate a fully parity-, age-, and duration-adjusted index of total

fertility (PADTFR) analogous to the conventional age-adjusted total fertility rate. The article then uses three heuristic diagnostics to assess the features of the fertility transition in each country. These are based on the insight that stopping, spacing and limitation, defined in a formal but intuitive way, have different and unambiguous effects

• n changes in the duration-specific fertility schedule (Timæus and Moultrie 2013). We

focus on straightforward measures calculated from the estimated fertility rates. In earlier work, we also considered the coefficients of the regression model used to smooth the rates as these allow one to test for the statistical significance of changes and differences in fertility. However, we have noted not only that such indicators are difficult to interpret, but also that even trivial changes in fertility patterns are usually statistically significant given the large sample of births we analyse for most countries. First, we derive period parity progression ratios, showing the proportion of women of a given parity who go on to have another child within 12 years. This diagnostic is based on the idea that limitation can be inferred if plots of progression across parities become increasingly concave over time (Brass, Juárez and Scott 1997). This occurs when progression to the third and fourth birth drops by more than progression to higher-order births, pulling the curve downward. In contrast, if the PPRs decrease linearly with increasing parity, no clear evidence exists of parity-specific limitation. As a summary measures of reduction in progression at smaller and larger family sizes we use progression to the fourth birth, calculated by multiplying the first four parity progression ratios, and progression from the fourth to the eighth birth, calculated by multiplying together the next four ratios. Second, we examine the trend in the parity-specific median length of closed birth

• intervals. In previous research on sub-Saharan Africa, we adopted the median survival

time in an interval as our preferred indicator. In populations in which postponement is more important than stopping, it is theoretically appropriate to consider trends in the duration of open and closed birth intervals combined. In low fertility populations where family size limitation is common, however, it makes more sense to examine closed intervals separately. This indicator is again estimated on a period basis. It is simply the duration since the previous birth by which half of the women who will have another birth within 12 years have had that birth, given the fertility rates of the period. Third, we examine the shape and pattern of the estimated duration-specific hazard rates. In this regard, we follow our earlier work, identifying spacing behaviour by means of a rightward shift in the mode of the distribution; limiting by a decrease in the hazard at all durations; and postponement by a counter-clockwise rotation of the hazard, resulting in hazards of closing a birth interval increasing at long durations. We use the life table probability of closing a birth interval within 5 years (B60) as a summary index of fertility at short and medium durations and the probability of closing an interval at between 5 and 10 years duration among women who have yet to close it at 5 years to summarize fertility in the right-hand tail of the distribution.

Draft: 21 Oct. 2017. Please ask the authors for an up-to-date version before citing. 5 The path toward low fertility in each of the 70 countries is classified according to these three heuristics and the results assessed both to identify regional clustering in the route by which different countries have progressed through the fertility transition and to identify countries that have undergone exceptional transitions.

Results

Figure 1 compares our estimates of parity-age-duration-adjusted total fertility for 1965-9, 1975-9, … 2005-9 with the estimates for the same periods published by the United Nations (UN Population Division 2017). The two series agree closely though the estimates from World Population Prospects tend to be slightly lower in populations with very high fertility, and are a bit higher in populations in which total fertility is 4 and 6 children per woman. One would not expect the two series of estimates to be identical, both because the United Nations’ estimates of total fertility standardize only for age and because they are based on a wider range of sources than fertility surveys. Nevertheless, the congruence between the two series does provide external evidence of the validity of

• ur model of parity-age-duration-specific fertility.

Figure 1: Comparison of parity-age-duration adjusted total fertility with the United Nation’s estimates of total fertility for the same quinquennia

While it is infeasible to present and discuss the detailed results for all 78 countries, Figure 2 presents illustrative results for two of them – India and Kenya.1 India

1 Corresponding figures for all 78 countries are presented in Appendix Figures A1-A3.

2 4 6 8 10 Estimated TFR 2 4 6 8 10 UN TFR Africa Europe

• L. America

MENA S&E Asia

Draft: 21 Oct. 2017. Please ask the authors for an up-to-date version before citing. 6 exemplifies a transition to low fertility that is driven by parity-specific birth control. Fertility declined at all durations but by more at long durations than short ones. The plot

• f the period PPRs was initially linear but became strongly concave over the 40-year

period, driven by very large reductions in progression to the third and the fourth births. The median closed birth interval changed little, but perhaps shortened slightly if one can rely on the estimates for the 1960s.

Figure 2: Trends in fertility by duration and progression and interval duration by parity in successive quinquennia

India Kenya

Draft: 21 Oct. 2017. Please ask the authors for an up-to-date version before citing. 7 The fertility transition in Kenya follows a different path. Fertility declined by more at long durations than short ones, producing the flattening of the duration-specific fertility distribution that is characteristic of postponement. This fall in fertility only had a substantial impact on parity progression in the 1990s, although birth intervals began to lengthen a decade earlier. By 2005-9 their median length had risen to 37 months. Notably, the decline in progression with parity in Kenya remained close to linear even once the proportion progressing had begun to drop. Indeed, progression dropped by more at high parities than lower ones. This may be, in part, because women were prevented from progressing as intervals lengthened by the decline in their fecundity with age. In addition, it may result from women `retiring’ from childbearing as they become older, more senior and perhaps more concerned about their health (Bledsoe 2002). What is clear is that no evidence exists in Kenya of parity-specific family size limitation by women with moderately-sized families. As can be seen from the parity progression ratios in Figure 2, progression to the fourth birth in India dropped by far more between 1965-9 and 2005-9 than progression from the fourth to the eighth birth. In contrast, in Kenya, progression from the fourth to the eighth birth dropped by more than progression to the fourth birth.

Figure 3: Reduction in higher- and lower-order parity progression, 78 countries

Figure 3 compares this pair of changes for all 78 countries included in this study. Not

• nly Kenya, but nearly all African countries have experienced as large a decline, or a

larger decline, in progression from the fourth to the eighth birth than in progression to the fourth birth. Apart from a few countries in the lower left corner of the graph in which

Lesotho Namibia South Africa Jordan Morocco Sudan (North) Tunisia Yemen Pakistan Timor-Leste Haiti Mexico

.2 .4 .6 .8 Reduction in progression from 4th to 8th birth .2 .4 .6 .8 Reduction in progression to 4th birth Africa MENA S.E. Asia

• L. America

Europe

India Kenya

Draft: 21 Oct. 2017. Please ask the authors for an up-to-date version before citing. 8 progression has changed very little, the major exception is South Africa where fertility is relatively low. Moreover, inspection of the detailed results for sub-Saharan Africa, (Appendix Figure A1), reveals that they resemble those for Kenya: they are characterised by linear declines in progression with parity that steepen over time. No evidence exists

• f the development of increasingly concave curves by parity that are diagnostic of parity-

specific family-size limitation except in the most recent estimates for South Africa, Lesotho and Namibia. Most countries outside mainland sub-Saharan Africa exhibit the opposite pattern of change, with larger reductions in progression to the fourth birth than progression from the fourth to eighth birth and with the development of increasingly concave curves by parity over time (Appendix Figure A1). A few non-African countries, however, have followed an ‘African’ pattern of decline in parity progression, with relatively large reductions in progression among higher parity women and with no evidence of stopping at a preferred small or moderate family size. These countries include Pakistan, several – but not all – countries in the Arab world, Haiti, and a few smaller island states, namely East Timor, Comoros and Sao Tome and Principe. Mexico was also following this pathway of fertility decline in the period up to 1990 for which we have data.

Figure 4: Trends in the median duration of closed birth intervals as total fertility decreases

Figure 4 examines what has happened to the median duration of closed birth intervals. As total fertility has fallen, they have lengthened in most countries and regions of the

• world. Median pre-transition birth intervals are of the order of 30 months (a result

consistent with the pioneering work of Sheps and Menken (1973)) and become noticeably longer once total fertility drops below 5 children per woman. Such diverse countries as

24 36 48 60 24 36 48 60 24 36 48 60 2 4 6 8 10 2 4 6 8 10 2 4 6 8 10

Caribbean EastAfrica Europe MENA MiddleAfrica SEA SouthAmerica SouthernAfrica WestAfrica

Median closed interval (months) Parity-age-duration adjusted total fertility

Graphs by Subregion

Draft: 21 Oct. 2017. Please ask the authors for an up-to-date version before citing. 9 South Africa, Namibia, Zimbabwe, Bangladesh, Indonesia, Colombia, Paraguay, Peru and Ukraine, now have medians that exceed 4 years. Thus, our results for intervals of all

• rders (weighted by the number of births by order in a synthetic cohort for the period)

are broadly consistent with those that Casterline and Odden (2016b) made for the interval to the second birth. Except in Southern Africa, the rate of increase in the median length of birth intervals in sub-Saharan Africa has not been that rapid. It is notable, however, that the lengthening

• f birth intervals outside sub-Saharan Africa has been concentrated in the second half of

the fertility transition. In contrast, birth intervals have been lengthening in sub-Saharan Africa since fertility transition began. Thus, controlling for total fertility, virtually no

• verlap exists between the medians for African countries and those elsewhere. For

example, outside sub-Saharan Africa, with the single exception of Bangladesh, the median closed interval only rose above 3 years after total fertility had fallen to less than 4.5 children per woman. In contrast, all the sub-Saharan African countries in which total fertility has fallen that far, already had median closed intervals that exceeded 3 years by the time total fertility dropped to 4.5 children per women. A few countries have seen no increase in the median length of closed birth intervals. The list includes several countries in the Sahel and Central Africa in which fertility has changed little, but also Sierra Leone and Ethiopia, where there has been a significant drop in total fertility. It also includes India, as we have seen already, and some of the countries outside sub-Saharan Africa in which we found no evidence of parity-specific limitation of family size, including Pakistan and the small island states.

Figure 5: Relationship between the probability of progressing to the next birth between 5 and 10 years since the previous birth and the probability of progressing before 5 years

.2 .3 .4 .5 .6 .2 .3 .4 .5 .6 .2 .3 .4 .5 .6 .2 .3 .4 .5 .6 .7 .8 .9 .2 .3 .4 .5 .6 .7 .8 .9 .2 .3 .4 .5 .6 .7 .8 .9

Caribbean EastAfrica Europe MENA MiddleAfrica SEA SouthAmerica SouthernAfrica WestAfrica

Conditional progression 5-10 years Progression by 5 years

Graphs by Subregion

Draft: 21 Oct. 2017. Please ask the authors for an up-to-date version before citing. 10 Figure 5 looks at the extent to which the probability of having a birth between 5 and 10 years after the previous birth, conditional on not having progressed already, has held up as the probability of progressing by 5 years since the previous birth has dropped over

• time. The probability of progressing at birth intervals of 5 and 10 years is generally

higher in sub-Saharan Africa than elsewhere. Moreover, as progression before 5 years has fallen, bringing down overall fertility, most sub-Saharan African countries have seen an offsetting rise in progression at 5 to 10 years. In contrast, in most countries in the rest

• f the world, progression at 5 to 10 years has fallen alongside progression before 5 years.

In most countries fertility at longer intervals has fallen relatively slowly, leading to some flattening of their duration-specific fertility schedules, but this flattening is considerably more dramatic in sub-Saharan Africa than elsewhere. As one would expect, most countries in which the median closed birth interval has lengthened have experienced smaller reductions in progression at 5 to 10 years, compared with 0 to 5 years, than countries in which the length of birth intervals has not

• changed. There are three interesting exceptions, however, where birth intervals have

lengthened without any evidence of postponement: northern Sudan, Tunisia and Yemen. Closer inspection of the duration-specific fertility schedules for these countries suggests that they have shifted toward longer intervals without changing in shape (Appendix Figure A3). Thus, these are the only three countries in which the evidence suggests that birth intervals have lengthened because women are spacing their births more widely, rather than postponing the next birth sine die. Note that they all three countries are among those countries outside Africa in which progression to the next birth has fallen more at higher parities than at moderate family sizes. In Mexico and Haiti, on the other hand, where there was also no evidence of parity-specific family-size limitation, birth intervals also lengthened, but progression at 5 to 10 years rose. Thus, the decline in fertility in these countries was driven by postponement.

Figure 6: Pathways to low fertility, 1965 – 2010

Based on the results summarised in Figures 3 to 5, combined with a more detailed inspection of the estimates for each country of the type shown in Figure 2 (see Appendix

Draft: 21 Oct. 2017. Please ask the authors for an up-to-date version before citing. 11 Figures A1 – A3), we have attempted to classify the 78 countries in the analysis according to the relative importance of parity-specific family size limitation and postponement of next birth in fertility decline (Figure 6). The countries shown in grey have yet to begin or are only beginning their fertility transition and therefore it is impossible to determine what pathway they are following. The smallest category of countries is those following a ‘classic’ path to low fertility, characterised by parity-specific family size limitation with no increase in the length of birth intervals. It is limited to India and Nepal, together with Azerbaijan in the last part

• f its transition. The countries shaded light green have also undergone a fertility

transition driven by family-size limitation, but birth intervals have lengthened

• somewhat. In the yellow countries both family-size limitation and postponement have

made a substantial contribution to fertility transition, though note that the estimates for Brazil only run up to 1990. In the countries shaded orange, fertility transition has been dominated by postponement, but the most recent available estimates show evidence of the onset of family size limitation. In countries shaded red, which apart from Mexico and Haiti are confined to sub-Saharan Africa, fertility decline has resulted entirely from

• postponement. Note though that, like those for Brazil, the estimates for Mexico refer to

the period before 1990. The countries indicated by blue shading appear to be neither practicing parity-specific limitation nor postponement. The drop in their parity progression ratios follows the African pattern. It declines with parity without showing any evidence of preference for specific family sizes. However, no evidence exists of flattening of the duration-specific fertility distributions of these countries. In the darker-blue shaded countries, birth intervals have not lengthened at all. In the lighter-blue shaded countries, they have lengthened moderately due to wider spacing of births.

Discussion

Our theoretical and conceptual research on postponement was originally stimulated by the discovery that very long birth intervals had emerged during the course of fertility transition in Southern Africa. To many demographers, including ourselves, it seemed implausible that birth spacing, as it is usually conceptualised, could produce such long

• intervals. Our conclusion was that women were not spacing but doing something

different – postponing. In retrospect, this intellectual journey has led both ourselves and

• thers who have become interested in postponement to focus perhaps too intently on

what has happened to birth intervals and not sufficiently on what has happened to parity progression. Once one extends one’s gaze beyond sub-Saharan Africa, however, to regions where parity-specific limitation is undeniably important, it is vital to examine changing patterns of parity progression. In this paper, we attempt to redress this balance. The results presented here demonstrate that fertility transition in most the developing world outside sub-Saharan Africa has been characterised by parity-specific limitation of

• births. Yet, as Casterline and Odden (2016a, 2016b) have pointed out, many countries
• utside Africa have also experienced some postponement of births and lengthening of

birth intervals, especially in the second half of their fertility transitions. In contrast, the

• nly evidence of any parity-specific limitation in sub-Saharan Africa is in the relatively

Draft: 21 Oct. 2017. Please ask the authors for an up-to-date version before citing. 12 low fertility countries of Southern Africa where it is just beginning to be apparent. Instead, where progression has dropped in Africa, it has fallen most at high parities, as in Kenya. This produces either a steepening linear decline in progression with parity or a convex curve. Birth intervals have lengthened almost everywhere. Thus, as we have suggested before, it is the absence of parity-specific family size limitation rather than the presence of postponement that renders sub-Saharan Africa distinctive. In Ethiopia, Sierra Leone, a few countries in North Africa and West Asia, and several small island states including East Timor, parity progression has dropped without any evidence of parity-specific limitation of births. There is no evidence of postponement in any of these countries, though a few of them have seen an increase in birth spacing contingent on the age of women’s youngest child. This pattern of fertility change represents a potential challenge to our, as well as others’, typology of fertility transitions. Fertility has declined abruptly, on a period basis, among women of every parity and at all interval durations. This pattern of change, however, immediately suggests a possible explanation of the fall in fertility in these countries. It may result from the initial take up at a relatively late date, in global terms, of birth control by a population that previously either lacked access to contraceptives or never conceived that they could control their

• fertility. Women started to limit their fertility at whatever family size they had reached at

the time that contraception became available. As younger cohorts build up their families, a more typical pattern of parity-specific family-size limitation is likely to arise. We believe that one of the strengths of the analysis presented in this paper is that it integrates the regression modelling of period fertility based on birth history data with a multistate life table model that calculates the parity progression ratios and durations of birth intervals in a synthetic cohort that experiences the fertility rates of a specific period. This provides a more detailed and comprehensive picture of the process of fertility transition across the developing world than has been available hitherto. It enables us to examine progression and birth intervals by birth order. In addition, however, the analysis calculates how many maternities of each order would occur in a cohort that went through its reproductive life experiencing the fertility rates of a specified period. Thereby, it also makes it possible to examine period trends in progression and the length

• f birth intervals at all orders combined. In contrast, earlier research has focused on the

changes occurring in a particular birth interval or presented unstandardized measures for all intervals in which the distribution of births by order is determined by the history

• f fertility change in the population concerned, not by current conditions (e.g. Casterline

and Odden 2016a; Casterline and Odden 2016b). One limitation of this study is that our data often provide only a partial snapshot of the entire fertility transition in a country. In many Latin American and Asian countries, fertility transition was well underway a decade before they first conducted a fertility

• survey. Moreover, in most of sub-Saharan Africa, one can only speculate as to what may

happen to family building patterns during the second half of their fertility transition as this is yet to occur. Thus, our determinations that a country is characterised primarily by stopping or by postponement might be open to qualification if information existed on that country’s entire fertility transition. Nevertheless, by analysing the WFS data, we

Draft: 21 Oct. 2017. Please ask the authors for an up-to-date version before citing. 13 have documented the early stage of sufficient fertility transitions outside sub-Saharan Africa for it to be clear that postponement is not a feature of the initial stages of fertility transition, that is bound to be replaced by limitation. Rather, outside sub-Saharan Africa, postponement seems to increase in importance as fertility falls to a low level. Thus, our analysis supports Casterline and Odden’s (2016a) conclusion that postponement is a family-building strategy of global significance. We find no evidence, however, to support their conclusion that ‘recent fertility declines in Sub‐Saharan Africa – Rwanda, Ghana, Ethiopia, even perhaps Kenya – have been driven primarily by limiting rather than postponement’. In fact, it is in only in Southern Africa, not further north, that we find any evidence of parity-specific limitation in sub-Saharan Africa. The (literal, and conceptual) map that we have drawn of fertility transition across what was once termed ‘the developing world’ is a complex one. An exception exists to almost every generalisation about the global decline in fertility that one might venture to make. Thus, Africa is not unique: a few other countries have experienced ‘African’ transitions. Nevertheless, the big picture is clear and spatially coherent. The initial stages of fertility transition in sub-Saharan Africa have followed a different track from that taken by almost all the rest of the world that has been characterised by postponement without family size limitation. The pace of fertility decline in Africa will remain slow until African women start limiting their family sizes as well as postponing having another

• birth. Gaining a better understanding of the motivations that underlie African women’s

family building patterns is essential to any attempt to develop appropriate reproductive health care services for Africa. Gaining a better understanding of the consequences of those decisions is vital if we wish to understand their implications for future fertility and population growth, not just in sub-Saharan Africa, but globally.

References

Agadjanian, Victor. 2005. "Fraught with ambivalence: Reproductive intentions and contraceptive choices in a sub-Saharan fertility transition." Population Research and Policy Review 24:617-645. Bledsoe, Caroline H. 2002. Contingent Lives: Fertility, Time, and Aging in West Africa. Chicago: University of Chicago Press. Brass, William, Fatima Juárez, and Anne Scott. 1997. "An analysis of parity- dependent fertility falls in tropical Africa." Pp. xii+453 in The Continuing Demographic Transition, edited by Gavin W. Jones, Robert M Douglas, John C. Caldwell, and Rennie M. D'Souza. Oxford: Clarendon. Casterline, John B, and Colin Odden. 2016a. "Exceptional or global: Evidence for the postponement in the developing world." in Annual meeting of the Population Association of America. Washington, D. C. Casterline, John B., and Colin Odden. 2016b. "Trends in inter-birth intervals in developing countries 1965–2014." Population and Development Review 42:173- 194.

Draft: 21 Oct. 2017. Please ask the authors for an up-to-date version before citing. 14 Johnson-Hanks, Jennifer. 2004. "Uncertainty and the second space: Modern birth timing and the dilemma of education." European Journal of Population 20:351- 373. Moultrie, TA, and IM Timaeus. 2003. "The South African fertility decline: Evidence from two censuses and a Demographic and Health Survey." Population Studies 57:265-283. Moultrie, Tom A, Takudzwa S Sayi, and Ian M Timæus. 2012. "Birth intervals, postponement, and fertility decline in Africa: A new kind of transition?" Population Studies 66:241-258. Rallu, Jean-Louis, and Laurent Toulemon. 1994. "Period fertility measures: The construction of different indices and their application to France, 1946-89." Population: An English Selection 6:59-93. Sheps, Mindel C., and Jane A. Menken (Eds.). 1973. Mathematical Models of Conception and Birth. Chicago: Chicago University Press. Timæus, Ian M, and Tom A Moultrie. 2008. "On postponement and birth intervals." Population and Development Review 34:483-510. —. 2013. "Distinguishing the impact of postponement, spacing and stopping on birth intervals: Evidence from a model with heterogeneous fecundity." Journal of Biosocial Science 45:311-330. Towriss, Catriona A., Donatien Beguy, Alison Wringe, Barwako Hassan Hussein, and Ian M. Timæus. 2017. "Planning a family in Nairobi’s informal settlements: results of a qualitative study." submitted. Towriss, Catriona A., and Ian M. Timæus. 2017. "Modelling period fertility: schooling and intervals following a birth in Eastern Africa." Population Studies. UN Population Division. 2017. "World Population Prospects: The 2017 Revision." New York: United Nations, Department of Economic and Social Affairs.

Draft: 21 Oct. 2017. Please ask the authors for an up-to-date version before citing. 15

Appendix Figures

Figure A1: Parity progression ratios according to order by quinquennium in 78 countries

Draft: 21 Oct. 2017. Please ask the authors for an up-to-date version before citing. 16 Figure A2: Median duration of closed birth intervals according to order by quinquennium in 78 countries

Draft: 21 Oct. 2017. Please ask the authors for an up-to-date version before citing. 17 Figure A3: Duration-specific fertility rates by quinquennium in 78 countries