Tracking how mortality a¤ects fertility along the demographic transtition Luis Angeles July 20, 2015 Abstract The importance of reductions in mortality as a driver of fertility change is not …rmly established among economists as the experience of European countries over the 19th century seems to contradict it. This paper shows how a small e¤ect of mortality on fertility is to be expected during the early stages of the demographic transition - and does not preclude a major role for mortality later on. The underlying mechanism makes use of an important empirical regularity uncovered by demographers but seldom discussed in economics: the absence of parity-speci…c fertility control prior to the transition in fertility rates. After discussing the mechanism, I present empirical evidence using three distinct datasets which strongly supports the existence of a nonhomogenous e¤ect of mortality on fertility. 1 Introduction A certain degree of confusion exists within economics regarding the role of reductions in mortality rates as a driver of fertility transitions. On the one hand, mortality rates have been found to be a statistically signi…cant predictor of fertility rates in panel data analyses covering long time periods. Angeles (2010) …nds this in a panel of up to 118 countries over the period Adam Smith Business School (Economics), University of Glasgow. Glasgow G12 8QQ, UK. Email: [email protected] 1 1960-2005, and the same is true of Herzer et al. (2012), who analyze 20 countries over the whole 20th century, and Murtin (2013), covering 70 countries over the period 1870-2000. In all cases the e¤ect is positive - reductions in mortality leading to reductions in fertility - and the concept of fertility being referred to is gross fertility - the total number of births per woman. On the other hand, non-statistical analyses of the fertility transition in Europe are often interpreted as evidence against the importance of mortality as a driver of fertility change. In particular, Oded Galor has pointed out repeatedly that mortality rates started falling in most European countries long before any decreasing trend in fertility could be detected (Galor 2005a, 2005b, 2012). Furthermore, a clear break in the time trend of fertility rates can be observed in most European countries between the 1880s and the early 1900s - from long-term stagnation to rapidly falling fertility. No such change can be observed for mortality rates, which are falling more or less steadily throughout this period. Being the …rst and most heavily studied of all demographic transitions, the European case cannot possibly be brushed aside. This paper argues that the European experience and the outcomes of panel data analyses should not be seen as contradictory. Far from it, I believe the two sets of results …t comfortably together once we use the correct theoretical framework to interpret them. The central message of this paper is that the e¤ect of mortality rates on fertility is di¤erent at di¤erent stages of the demographic transition. The absence of an e¤ect during the early stages of the European transition should not surprise us, as it is exactly in accordance with the adequate framework. While I am not the …rst to make this point, the analysis in this paper improves over previous e¤orts both on the theoretical and on the empirical side. On the theory side, Cervellati and Sunde (2015) and Strulik and Weisdorf (2014) have explored potential mechanisms able to explain how the e¤ect of mortality on fertility may change as the demographic transition progresses. While ingenious, neither of these two papers engage with 2 all the empirical evidence collected by demographers on the subject of the demographic transition - a criticism which, to be fair, applies to the vast majority of economic papers in this area. The mechanism I put forward corrects this, and has the added bene…t of being far simpler than those already in existence. On the empirical side, I present new evidence showing that, as a rule, reductions in mortality rates have a non-homogenous e¤ect on fertility along the demographic transition. The e¤ect is small and potentially di¢ cult to detect in the early stages, turning much larger and unequivocally positive later on. Cervellati and Sunde (2015) began to build the empirical case for this pattern by using cross-sectional analysis of long-run changes for 47 countries over the period 1940-2000. Here I provide much wider evidence using panel data analysis over three di¤erent datasets. The …rst one covers only European countries and tracks them from as early as the year 1750 up until 1949, when their demographic transition was arguably at a very advanced stage. The second one uses the dataset of Murtin (2013), which covers 70 developed and developing countries since 1870, and extends the time coverage up to the year 2010. The third one, …nally, tracks the vast majority of countries in the world over the period 1960-2010. In all cases the results corroborate my theoretical analysis and the insights from the demographic literature. Some …nal remarks are in order regarding the e¤ects of mortality reductions on net fertility - the number of children surviving to adulthood per woman. Angeles (2010) estimates this relationship directly and …nds that lower mortality rates decrease net fertility after a lag of 10 to 20 years. On the other hand, the results of Herzer et al. (2012) suggest the opposite. These authors do not consider net fertility rates directly, but their estimated e¤ect on gross fertility is not large enough to produce decreasing net fertility when mortality falls.1 Some authors have focused on the ambiguity surrounding the e¤ect on net fertility to challenge the importance of falling 1 Indeed, ceteris paribus, lower mortality increases net fertility as more children survive to adulthood. For net fertility to fall, gross fertility needs to fall by a large enough margin. 3 mortality rates, under the argument that "it is the reduction in net fertility and thus population growth that is most relevant from the viewpoint of the theory of economic growth" (Galor 2012, p. 7). In this paper I will also delve into the question of how mortality rates a¤ect net fertility. While the question is obviously of interest, I do not subscribe to the view that the e¤ect on net fertility is of larger importance for economists than the e¤ect on gross fertility. If standard growth models have a place for net fertility, and not for gross fertility, that is only because of the simplifying assumptions that the creators of these models have judged adequate to make. Even if we are only interested in economic growth, it is not di¢ cult to argue that a society where women give birth to six children, and four of them die, is quite di¤erent from one where women give birth to two children and none of them dies - despite them having the same rate of net fertility. In the …rst case women may be forced to abandon the labour market, they will be less motivated to accumulate human capital, and less resources will be available per child - important socioeconomic phenomena which will not fail to have an impact on growth. Thus, how many children each woman gives birth to is a question of comparable importance to how fast the population grows or how many adult children each woman brings about. 2 The fertility transition and parity-speci…c fertility control The demographic transition is de…ned as the passage from a regime of high mortality and high gross fertility to one of low mortality and low gross fertility. Net fertility is typically low both at the beginning of the process and towards its end, but tends to be high in the middle as declines in mortality often predate those in fertility. The analysis of fertility change has dominated the agenda for the two social sciences most involved in the study of the demographic transition, 4 demography and economics. The reasons for this are many. First, there is considerable agreement over the forces that led to mortality decline, with medical advances, improvements in public sanitation, and general economic development (leading to gains in nutrition and improved hygiene) all playing a prominent role (Kirk 1996, Guinnane 2011). Second, and of especial relevance for economists, the forces leading to mortality reductions appear exogenous to individual control. As the study of individual choice is the most frequent mode of economic analysis, mortality rates tend to be taken as given. Finally, mortality rates tend to behave quite predictably over time; falling steadily rather than experiencing sudden changes.2 None of the above is true when it comes to fertility. The forces leading to its decline are not well understood and still hotly debated, and it seems clear that fertility behavior lies squarely within the realm of individual choice. Furthermore, the most remarkable aspect of time series of gross fertility is their sudden transition from a long-standing period of high and stable fertility rates towards a period of rapidly declining ones. The pattern is present in practically all countries, and researchers in the …eld refer to it as the fertility transition.3 While all of the above is well-known to economists, considerable less attention has been paid to an additional empirical regularity uncovered by demographers and which will be at the heart of the arguments in this paper: the appearance of parity-speci…c fertility control. In demography, a woman’s parity is the number of children she has birthed at a given moment in time. An empirical regularity much-discussed among demographers has been the absence of any relationship between parity and age-speci…c fertility rates before the onset of the fertility transition 2 The adoption of antibiotics, vaccines and methods of malaria prevention around the world starting in the 1940s would be the main exception to this, as they brought about a major decline of mortality rates throughout the developing world (see Acemoglu and Johnson 2007). 3 To avoid any ambiguity, from this point onwards all uses of the terms ‘pre-transition’ and ‘post-transition’will refer to the fertility transition (not the demographic transition). 5 (that is, during the period of high and stable fertility rates). In a typical post-transition society, women who have given birth to, say, three children are far less likely to become pregnant than women of the same age who have given birth to one child. In other words, the probability of pregnancy falls with a woman’s parity once age is controlled for (since women at higher parities tend to be older and natural fertility falls with age). This is not the case among pre-transition societies, where women with six children are as likely to become pregnant as women with one child provided they are of the same age. Thus, the onset of the fertility transition brings about not just a quantitative change in fertility but also a qualitative one. Women do not start having less children at all stages of their lives; they tend to keep high fertility rates at low parities and adopt much lower fertility rates at high parities. Table 1 below illustrates the situation by comparing parity-speci…c fertility rates for women aged between 30 and 35 years old in a society that most likely quali…es as pre-transition (Nigeria in the year 1990) and a society that can be safely described as post-transition (the United States in the years 1985-1990). As can be seen, Nigerian women have a yearly chance of giving birth to a child of about 25% in this age bracket, and the probability is the same at all parities. American women are about as likely to give birth as Nigerian ones if they have no children or one child, but the probability falls from 25% to 11-12% once they reach their second child. The natural interpretation would be that many American women regard two children as an ideal or target family size, and actively reduce their fertility after their second childbirth. [Table 1] Faced with the above evidence, a popular explanation among demographers has been to claim that pre-transition couples had no target level of fertility in mind. In their view, fertility control was not a culturally accepted mode of behavior in these societies, and couples were expected to give birth to as many children as God or nature was willing to send them. In the 6 words of Francine van de Walle, before the onset of the fertility transition, “fertility is not within the calculus of conscious choice”(van de Walle 1986, p. 202).4 While not implausible, this interpretation may well be the reason why economists never seriously engaged with the evidence on parity-speci…c fertility control. After all, it runs in the face of economics’central organizing principle: that people’s behavior is determined by conscious and rational choice. According to demographers, pre-transition societies simply don’t consider fertility as a choice variable in an optimization problem. While demographers may be right or wrong in their theoretical interpretation, the empirical evidence regarding the existence or absence of parityspeci…c fertility control should not be ignored. Economics arguably su¤ers from an excessive richness of theoretical explanations for the demographic transition, and only additional evidence will help us narrow down the relevant options. In this spirit, a …rst contribution of this paper will be to argue that the evidence on the appearance of parity-speci…c fertility control does not need to come at the expense of rational fertility behavior (in the economists’ sense) among pre-transition societies. I hereby o¤er an alternative explanation: that absence of parity-speci…c fertility control is also compatible with a setting in which families optimize fertility decisions and have a target family size, but this target is never reached due to an upper limit in gross fertility rates and a high mortality environment. In other words, families would not 4 We should note that the empirical evidence sustaining this claim is not without problems. The calculation of partity-speci…c fertility rates requirest data on the parity of each woman giving birth. Such data is available for recent periods (as in table 1), so we may con…dently state that today’s post-transition countries feature parity-speci…c fertility control while today’s pre-transition countries do not. On the other hand, it is less certain that today’s post-transition countries did not present parity-speci…c fertility control before their fertility transition as the data for that period is less rich. Demographers have argued in favour of this view using data on the age of women giving birth (but not their parity) - see Coale (1986). This type of analysis has been in‡uential, but can be called into question (Friedlander et al. 1999). 7 reduce their fertility at high parities because they have not reached (or do not expect to reach) their target. To elaborate a bit, recall that in the economists’ standard formulation of fertility behavior it is the number of children surviving to adulthood (in other words, net fertility) which enters into the couples’ utility function. Families target a certain level of net fertility, and gross fertility derives from this choice and the mortality environment. While it is commonly assumed that this desired target is reached, such may not be the case in pre-transition times and that for at least two reasons. First, the level of gross fertility will be bounded from above not just by biological constraints, but also by social and cultural factors. The importance of this last element is worth emphasizing as it is not always su¢ ciently appreciated within economics. In the absence of any inhibiting social or cultural factors, the highest number of births per woman recorded for a population of substantial size is about 12.5 In the vast majority of pre-transition societies, however, births per woman typically average between one half and two-thirds of this maximum, and that despite the fact that couples do not reduce their chances of pregnancy at any parity level. The reason is the existence of social and cultural factors that reduce gross fertility in di¤erent ways. In the European case, this was mainly driven by restrictions for entry into marriage, as fertility rates for unmarried women have always been extremely low. Pre-transition women in Europe married in their mid-twenties (instead of their mid-teens, as in most parts of Asia), cutting their potential fertility by more than a third. An alternative method, common for instance in India, would be to allow early entrance into marriage but to forbid the remarriage of widows - thus ending the fertile life of many women early. In addition to this, cultural norms would also limit fertility within marriage, for instance by prescribing extended periods of breast-feeding or sexual abstinence following a birth. In all these cases fertility rates diminish uniformly 5 The canonical case would be that of the Hutterites, an Anabaptist Christian community used as a standard for high fertility levels in demographic research. 8 across all parities, and women would experience an average of between 6 and 8 births over their reproductive lifetime.. Add to this the second factor, namely that mortality rates in pre-transition societies could often be as high as 50% between birth and adulthood. As an example, during the second half of the 18th century only 49% of French newborns reached aged 15 (Livi-Bacci 1991, p. 74). Even as late as the period 1950-1955, survival rates to age 15 averaged only 63% in Sub-Saharan Africa (United Nations 2013). Under these conditions, families could expect an average number of children surviving to adulthood of between 3 and 4 - and considerably less if the objective was for the children not just reaching adulthood but taking care of their parents when old. For early modern England, Clark and Hamilton (2006) have shown that, on average, only 2.79 children survive the death of their father over the period 1585-1638. The …gure over-estimates survival rates among the general population, as it is obtained from a sample of people leaving a will at death - that is, the relatively well-o¤. With such poor prospects of surviving, it seems plausible that families would …nd it hard to reach their target number of surviving children - even when this target was a modest 3 or 4 - and rationally decide to give birth to as many children as they possibly can. To summarize, absence of parity-speci…c fertility control is compatible with rational fertility choice as long as the upper bound on gross fertility is binding. The evidence presented in the last two paragraphs suggests that, during the pre-transition period, such was probably the case. On the other hand, the constraint on gross fertility would cease to bind as the demographic transition advances - both because the target number of children declines and because mortality rates are falling. Once families are able to reach their desired family size, parity-speci…c fertility control would appear. Below I discuss how the relationship between mortality rates and fertility would be a¤ected by these considerations. 9 3 Mortality and fertility Mortality has always been regarded as one of the major determinants of gross fertility rates in demographic research. The idea that in a lower mortality environment parents would need to give birth to fewer children in order to reach a target family size was present in demography since the beginning of its development as an academic discipline, following World War II.6 This mechanism, known today as the replacement e¤ ect, advances that parents who su¤er the death of a child take steps to give birth to one additional child than otherwise planned, as they search to compensate for the lost one. The mechanism is typically incorporated in most economic models of fertility choice in a reduced form. In the economists’standard formulation families decide on the number of children they give birth to at a single moment in time (instead of sequentially) in order to maximize a utility function where the number of children surviving to adulthood enters as an argument. Higher mortality rates would thus require a larger number of births for any chosen level of net fertility. Worthy of notice, within this framework the target level of net fertility may or may not be itself related to mortality rates. In a number of economic models it is factors such as technological change, income levels or labour market characteristics which determine the net fertility target, and mortality rates only play a role in determining the required number of births to reach this target (see Galor 2012 for an overview of this literature). On the other hand, economists have also explored models where mortality rates have an e¤ect on net fertility - for instance because parents "hoard" children in order to protect themselves against the possibility of high mortality scenarios (Sah 1991, Kalemli-Ozcan 2002), or because lower mortality increases the return 6 See Guinnane (2011, 598-99). For an account of the development of demography and "Demographic Transition Theory" see Kirk (1996) and Friedlander et al. (1999). Other factors emphasized by Demographic Transition Theory as drivers of fertility rates are the educational level of the population (especially females), urbanization rates, economic development and cultural di¤usion. 10 to investments in children’s education and shift their choices towards more child "quality" (Becker et al. 1990, Ehrlich and Liu 1991). While most theoretical analysis in economics have focused on net fertility choice, it is interesting to note that empirical studies in both economics and demography typically use measures of gross fertility as their dependent variable. A …rst reason for this is the traditional de…nition of fertility transitions as the period when gross fertility rates begin to fall, while the larger availability of data on gross fertility provides a second, more practical, rationale. With gross fertility rates as the dependent variable, the standard replacement e¤ect described above would predict a positive relationship between mortality and fertility at all times (and regardless of whether net fertility is also being a¤ected). The absence of a positive e¤ect during much of the 19th century in Europe may then be regarded as a rejection of the theory. This rationale, however, ceases to apply as soon as we consider couples not reaching their target net fertility due to an upper limit on gross fertility rates. In this case, couples do not change their fertility behavior following the death of a child as this behavior is constant throughout their fertile lifetime: couples plan, from the beginning, to have as many children as possible (having arguably internalized the fact that many of these children will not reach adulthood). In a nutshell, there is no replacement e¤ect. But this is not the end of the story. While the absence of a replacement e¤ect clearly weakens the link between mortality rates and gross fertility, it does not eliminate it altogether. An additional channel for mortality to a¤ect fertility will still remain in place when couples are having as many births as they possibly can. This is because of the existence of what is known as the physiological e¤ ect. As demographers and other social scientists have long discussed, the period following the birth of an infant is characterized by a very reduced pregnancy risk due to the universal practice of breast-feeding and the resulting lactational amenorrhea. For as long as a mother breast-feeds her 11 child, the pregnancy risk remains low.7 This, however, introduces a positive link between infant mortality and fertility as the death of a lactating infant cuts short this low pregnancy-risk environment. As we have described them, couples for whom the upper limit on gross fertility is binding will have as many pregnancies as possible during the time period when such an event is possible. The death of an infant increases the length of this time period. The existence of the physiological e¤ect implies that mortality rates would have a positive e¤ect on fertility even during the early stages of the demographic transition. The e¤ect, however, would turn larger once the upper bound on gross fertility is no longer binding and the replacement effect kicks in. In short, incorporating the evidence on parity-speci…c fertility control into our analysis points towards a non-homogeneous e¤ect of mortality on gross fertility as the demographic transition advances. The e¤ect would be small and positive during the early stages, when the upper limit on gross fertility is binding, and large and positive later on. This rejects the common assumption of a constant-size e¤ect of mortality on fertility that characterizes linear regression models, and explains why the e¤ect may be di¢ cult to capture in 19th century Europe. The analysis so far has not discussed when along the demographic transition will the upper bound on gross fertility cease to bind. Being largely determined by social and cultural norms, this upper bound takes di¤erent values for di¤erent societies. Under such conditions, and following the usual practice in demography, I assume the society-speci…c upper bound on gross fertility equals the (relatively constant) gross fertility rate observed before the fertility transition.8 It follows that the onset of the fertility transition separates the period when the upper bound on gross fertility is binding from the period when it is not, which in turn determines the empirical strategy 7 Besides the purely hormonal changes induced by breastfeeding, mothers will often sleep with their babies and must care for them - thus reducing the chances of sexual intercourse. 8 Assuming otherwise implies that fertility could have been higher in pre-transition times, but couples chose not to have more children. I regard this as unlikely not so much because pre-transition fertility is very high, but because it is very stable. 12 for testing the above hypotheses. Finally, let us turn our attention to the e¤ect of mortality rates on net fertility. When couples have as many births as possible this relationship would certainly be negative, as a woman who loses its n-th born child would need to invest twice the amount of time from her fertile years in order to pass from n-1 to n surviving children.9 As women fully use all their available fertile years, the total number of surviving children would diminish. Once the fertility transition gets underway, however, women have some slack fertile capacity which may be tapped into in the case of a child’s death. The e¤ect of mortality on net fertility would then become at least less negative - and potentially zero if enough slack capacity is available so that the replacement e¤ect is complete. The e¤ect may even become positive if some additional mechanism is in place, such as hoarding or the quantity-quality tradeo¤, which would make net fertility directly a function of mortality rates. The empirical section of this paper takes the preceding analysis to test and looks for a change in the relationship between mortality and fertility following the onset of the fertility transition. While the e¤ect on gross fertility should remain positive and clearly increase in magnitude, the e¤ect on net fertility would be expected to pass from negative to nil - or perhaps even to positive. My preferred interpretation for such …ndings would be the existence of an upper limit on gross fertility which is binding before the onset of the fertility transition, but I note that the demographers’ hypothesis of fertility decisions not being subject to individual rational choice in the pretransition period would be observationally equivalent with the data at hand. Accordingly, my aim is not to establish which of these two explanations is correct but rather to demonstrate the existence of a non-homogenous e¤ect of mortality on fertility. It follows from that result that a small e¤ect of mortality on fertility in the early stages of the demographic transition is perfectly compatible with the claim that mortality reductions play an 9 To be more precise, the woman would need to invest twice the amount of time if the child dies once brestfeeding is completed. If the child dies earlier, the amount of time would increase but would not double, because of the physiological e¤ect discussed above. 13 important role as drivers of fertility decline once the fertility transition gets underway. 4 Empirical analysis I test the existence of a non-homogeneous e¤ect of mortality on fertility using three separate panel datasets: 22 European countries over the period 17501949, 70 developed and developing countries over the period 1870-2010, and as many as 187 developed and developing countries over the period 19602010. Sources and summary statistics are provided in the sections below. The econometric speci…cation I use with all three datasets is as follows: fi;t = i + mi;t 1 + Xi;t 1 + "i;t (1) In equation (1) fi;t is a measure of gross fertility for country i during period t, mi;t 1 is a corresponding measure of mortality and Xi;t 1 a set of additional determinants of fertility. The equation includes country-speci…c …xed e¤ects ( i ) in order to control for di¤erences in overall levels of fertility, as the social and cultural aspects limiting fertility in pre-transition times di¤er from country to country. Finally, a full set of time dummies is added to equation (1) as a robustness check in all subsequent analyses. Time dummies may capture unmeasurable aspects such as changes in cultural norms, thus reducing the scope for omitted variable bias. All determinants of fertility in equation (1) are used with a lag, with the purpose of reducing endogeneity problems. The most visible source of endogeneity bias would be reverse causality, for instance because high levels of fertility may increase mortality as less resources are available per child. Reverse causality is much less of a problem in equation (1) since high fertility cannot a¤ect mortality rates (or other determinants of fertility) retroactively. On the other hand, the possible existence of serially correlated omitted factors still leaves scope for an endogeneity bias in (1), and the results that follow ought to be interpreted with this in mind. 14 In order to capture the non-homogeneous e¤ect of mortality on fertility, equation (1) is estimated, …rst, for all country-year observations taking place before the onset of the fertility transition and, second, for all country-year observations taking place afterwards. This approach allows not just for coe¢ cient to change freely between the two regimes; the e¤ect of all other determinants of fertility is also allowed to change. As discussed below, results clearly vindicate this procedure. 4.1 European countries, 1750-1949 As much of the suspicion regarding the importance of mortality as a determinant of fertility decline comes from observing the European case, it seems adequate to start with it. Since many previous analyses have only covered a handful of countries, the analysis here adds signi…cant value by considering 22 European countries - essentially all independent political entities in 19th century Europe other than Turkey. My data comes from the International Historical Statistics (Palgrave Macmillan Ltd. 2013), and tracks mortality, fertility and marriage rates as far back as the year 1750 (though in most cases since the middle decades of the 19th century). The data o¤ers a coverage starting usually several decades before the onset of each country’s fertility transition and is, to the best of my knowledge, the most comprehensive source using common measures of fertility and mortality for all European countries. The data is available annually, and I use 5-year averages in what follows. With the data at hand, the …rst important part of the analysis is the determination of the onset of the fertility transition, which occurs at di¤erent times for di¤erent countries. For this dataset and the next one, which track most countries from well before the onset of their fertility transition, I follow the classical analysis of Coale and Treadway (1986). This approach, standard among demographers, de…nes the onset of the fertility transition as the …rst year in which the chosen measure of fertility falls to a level 10% below its long-run average before that time, provided it does not rise above 15 that average again. The pre-transition long-run average of fertility is allowed to di¤er from country to country. Coale and Treadway (1986, p.38) give the results of this procedure when applied to 19 European countries, all of which are on my European dataset. I take these dates as they are, and derive the dates for the remaining three countries using the same approach.10 Table 2 below shows summary statistics for the three variables at our disposal: the crude birth rate, the crude death rate and the marriage rate (de…ned as the number of births, deaths and marriages per 1,000 population). The table also reports the time trends for mortality and fertility, both before and after the onset of the fertility transition, controlling for country …xed e¤ects. This calculation con…rms the initial impression, mentioned in the introduction, of a disconnect between the time series of fertility and mortality along the European demographic transition. While both series experience an increase in their average change per period as the demographic transition advances, the change in fertility is far more radical. Notice, however, that fertility is on average declining, albeit at a very slow rate, even before its transition. [Table 2] Table 3 presents the results when equation (1) is estimated separately for pre- and post-transition observations using the crude birth rate as a gross fertility measure, the crude death rate as a mortality measure, and the marriage rate as an additional determinant of fertility. Crude rates have the disadvantage of being a¤ected by the age structure of the population, an aspect for which we cannot control in the present case but which will be addressed with our second dataset. Marriage rates are an important determinant of fertility throughout this period as childbearing outside marriage was highly unusual. 10 The three additional countries are Bulgaria, Romania and Serbia. It should be noted that the transition years from Coale and Treadway (1986) are calculated using marital fertility rates, while those I derive here use the crude birth rate. These two measures of fertility tend to give very similar transition dates, judging by the cases in which both series are available. 16 [Table 3] Columns 1 and 2 report the results when no time dummies are included in the regressions, while columns 3 and 4 report the results with time dummies. Both sets of results lead to similar conclusions, with some minor di¤erences that I note below. First, and most important, the e¤ect of mortality on fertility is indeed clearly heterogeneous: coe¢ cient experiences a large jump in its value between the pre- and post-transition periods. The posttransition estimate of is well above 1 without time dummies and 0:9 when time dummies are included, while its pre-transition value is between 0:16 and 0:23. Worthy of notice, the e¤ect of mortality on gross fertility is still positive during pre-transition times, reaching statistical signi…cance at the 1% level in column 1 and at the 10% level in column 3. Our discussion in the previous section is thus corroborated not only by the increase in following the transition but also by the existence of a small positive e¤ect even before the transition. Marriage rates have, as expected, a positive and statistically signi…cant e¤ect on fertility in all regressions (marginally so in column 4). Of more interest, the size of the coe¢ cient falls by more than half between the preand post-transition periods. This arguably re‡ects a looser association between marriage life and childbearing following the fertility transition. Indeed, post-transition couples were characterized precisely by limiting their fertility within marriage, while the occurrence of childbirth out of wedlock progressively became more common. A third and …nal observation concerns net fertility. The di¤erence between crude birth rates and crude death rates, the rate of population growth, is the best approximation we have here to net fertility rates. The e¤ect of mortality on this measure can therefore be easily gauged by subtracting one from the coe¢ cients on mortality reported in table 3. It follows that a coe¢ cient above 1 would denote a positive e¤ect of mortality on net fertility, while a coe¢ cient below 1 would denote a negative e¤ect. From this we deduce that the e¤ect of mortality on net fertility is de…nitively nega17 tive in the pre-transition period, as was indeed expected from the previous section. For the post-transition period, on the other hand, the outcome is less clear. The results in column 2 suggest a positive e¤ect on net fertility, but the inclusion of time dummies in column 4 render this uncertain. While the con…dence interval for in column 4 does include values higher than 1, the point estimate of 0.9 suggest that the e¤ect is more likely to be slightly negative. In de…nitive, all of the above results are in accordance with our discussion from the previous section. The usual assumption of a constant-size e¤ect of mortality on fertility all along the demographic transition is soundly rejected, as the existence of an heterogeneous e¤ect in European countries is given solid support. 4.2 Developed and developing countries, 1870-2010 Our second dataset was put together by Murtin (2013) from a variety of sources, and o¤ers observations of crude birth rates, crude death rates, and additional determinants of fertility every 10 years from 1870 until the year 2000. This dataset includes 72 countries, 49 of which are from outside Europe and its o¤shoots. I extend the time coverage of the original dataset by one period by incorporating crude birth rates for 2010 from the World Bank (other variables are not required for 2010 as they enter the equation with a lag)11 . I determine the onset of fertility transitions using the procedure of Coale and Treadway (1986) as described above, unless the country in question is one of the 19 European countries for which these authors originally calculated the transition date - in which case I use their date. Two special cases 11 In addition to this, I extend the time coverage for the United States. Murtin’s dataset only covers this country over the period 1950-2000 for the crude death rate and 1960-2000 for the crude birth rate. Instead, I use data from the International Historical Statistics to cover the period 1900-2000 for crude death rates and 1910-2000 for crude birth rates, plus data from the World Bank for the year 2010. The data for 1900-1990 refers to the white population, while the observations for the years 2000 and 2010 are for the whole American population. The data for the white population reported in the IHS is practically identical to the one used by Murtin over the years 1950-2000. 18 are France and the United States, as the onset of their fertility transitions is placed by the literature well before the year 187012 . I place all available observations from these two countries in the post-transition group. Furthermore, because some developing countries are characterized by very high values of the crude birth rate which may su¤er changes of 10% or more following events such as wars or famines, I have also added the condition that the onset of the fertility transition should be characterized by a crude birth rate below the level of 36 per 1,000 inhabitants. To put this in perspective, the average crude birth rate at the onset of the fertility transition for the 21 OECD countries for which we can provide this data is 29.69, and the highest value among them is 35.7. Table 4 shows summary statistics of the variables at our disposal in this second dataset. As in the …rst one, the available measures of fertility and mortality are the crude birth and death rates, with rather similar average values as previously but somewhat larger variability around the mean - a normal occurrence given the more heterogeneous set of countries we now consider. As additional determinants of fertility we have the level of GDP per capita and female education, measured as the average number of years of schooling among the female population aged 15 and over. In addition to them, we can also include two controls for the age structure of the population: the share of the population aged 20-29 and that of the population aged 30-39. As these two age intervals cover most child-bearing years, these controls largely correct for the e¤ect of the population age structure on crude birth rates. [Table 4] As in the previous section, the last two rows of table 3 report the time trends of fertility and mortality before and after the fertility transition, controlling for country …xed e¤ects. While the pattern of fertility is as expected, with a large acceleration following the transition, we now see that mortality 12 In 1827 for France according to Coale and Treadway (1986), in 1848 for the United States according to Bailey (2009). 19 reductions are far slower following the fertility transition among this group of countries. The explanation is that fertility transitions in most developing countries took place much later than in Europe - with few exceptions during the second half of the 20th century and quite often during the last two or three decades. By this time the important reductions in mortality due to the adoption of vaccines and antibiotics had been largely achieved, leaving less scope for further reductions in the post-transition period. As it turns out, methods for reducing mortality travelled from Europe to developing countries much faster than the practice of limiting fertility. Table 5 presents the results of our empirical analysis when applied to this second dataset. Equation (1) is run controlling only for the age structure of the population in columns 1 and 2, while subsequent columns show regressions where GDP per capita, female education and a full set of time dummies are added progressively. In all cases the di¤erence between the preand post-transition estimates of coe¢ cient added is large - when all controls are takes a value of 0:3 in pre-transition years and 0:8 in post-transition ones. These values are quite in line with those reported previously in table 3.13 [Table 5] As was the case above, the estimates imply that the e¤ect of mortality on gross fertility in pre-transition times was not zero but slightly positive - albeit the e¤ect is statistically signi…cant only at the 10% level when all controls are included. In the post-transition phase, on the other hand, the positive e¤ect of mortality on gross fertility is always statistically signi…cant at the 1% level. Turning to the e¤ect of mortality on net fertility, this was certainly negative in the pre-transition phase as none of the relevant con…dence intervals ever comes close to the value of 1. For the post-transition case, however, 13 I note that none of these regressions contain all 72 countries available. This is the case because: (a) some countries are not observed over the pre-transition period, and (b) some countries have not reached the post-transition period by the year 2010. 20 three of the four con…dence intervals we estimate include the value of 1 - at which point the e¤ect on net fertility would be exactly zero. Thus, while the post-transition e¤ect on net fertility was probably still negative, a nil or even slightly positive e¤ect would be possible. With regards to our control variables, it is interesting to note that the e¤ect of GDP per capita on fertility looks quite unstable over our di¤erent speci…cations. When all controls are included, in columns 7 and 8, the coe¢ cient on GDP per capita is not statistically signi…cant in pre- or posttransition times. Female education, on the other hand, appears to have a more consistent negative e¤ect on fertility. Its coe¢ cient is negative and statistically signi…cant at the 1% level in columns 5 to 7, but not so in column 8 (post-transition times controlling for time dummies). I conclude by noting that, although the number of countries has more than tripled with respect to the previous analysis and they now come from every corner of the globe, the coe¢ cients on mortality do not change dramatically. As for the case of Europe, the change in the e¤ect of mortality on fertility following the transition appears large and in accordance with our priors. 4.3 Developed and developing countries, 1960-2010 The third and …nal dataset I consider comes from the World Bank’s World Development Indicators (all variables except female education) and from the Barro-Lee Educational Attainment Dataset (for female education). This dataset brings two important bene…ts. First, it o¤ers an almost comprehensive coverage of developing countries, with a total of 187 countries for which data is available. Second, it uses Total Fertility Rates as the measure of gross fertility and Child Mortality Rates as the measure of mortality - both superior choices to the crude birth and death rates considered previously.14 14 In particular, the Total Fertility Rate is not a¤ected by the age structure of the population as it corresponds to the number of births that a woman would experience over her lifetime if subjected to all current age-speci…c fertility rates. As a shorthand, we may refer to it as the number of births per woman. The Child Mortality Rate is the number 21 On the other hand, the time coverage is reduced to a maximum of eleven quinquennial observations between the years 1960 and 2010. The main disadvantage of this reduced time span is that for many countries the onset of the fertility transition would lie before the start of the sample. This last feature of the data implies that the Coale and Treadway (1986) methodology cannot be used, as the pre-transition level of fertility is not observed in several cases. Instead, I simply set a threshold level of the Total Fertility Rate and classify all country-year observations with values below that threshold as post-transition. I set this threshold at 5.5 births per woman in my baseline calculations, but I consider both higher and lower values in order to check the robustness of my results. Table 6 presents summary statistics for the variables in this dataset. As in the previous case, two important determinants of fertility which we control for are the GDP per capita of the country and the average years of schooling among the female population aged 15 and over. To this we add the urbanization rate, a factor often emphasized in the demographic literature as indicative of cultural norms. Indeed, migration to the city is often accompanied by a change in modes of behavior, whereby traditional beliefs may be abandoned in favour of more "modern" ones. Finally, the World Bank also provides us with two alternative measures of mortality, the infant mortality rate and life expectancy at birth. Both will be used as extensions to our baseline results so their summary statistics are also reported. [Table 6] The bottom of table 6 reports the time trends of mortality and fertility before and after the onset of the fertility transition. The pattern is quite similar to that uncovered in table 4. The decrease in fertility rates posttransition is, on average, about 0.5 births per decade, a …gure which would of deaths between ages 0 and 5 normalized by the number of live births. 22 be much larger if we were to exclude the countries that experienced their fertility transition long ago. The econometric results using this dataset are reported in table 7 and continue to be favorable to the central hypothesis of the paper. In all cases, the post-transition e¤ect of mortality on gross fertility is positive, statistically signi…cant at the 1% level, and far larger in magnitude than in the pre-transition period. The estimate of appears to stabilize around a value of 15.5 in columns 6 and 8, implying that a decrease in child mortality of 10 percentage points would lead to 1.55 less births per woman in the post-transition period - a large and meaningful e¤ect. The pre-transition e¤ect, on the other hand, is much smaller and eventually disappears when all control variables and time dummies are accounted for. [Table 7] Turning to the set of control variables, the e¤ect of female education appears to be robust and negative both before and after the onset of the fertility transition - which accords well with the results of table 5. The rate of urbanization has also the negative coe¢ cient we would have expected, but the e¤ect appears to characterize only post-transition periods. GDP per capita, …nally, has once again a rather unstable relationship with fertility although the results of columns 6 and 8, when most controls are included, do suggest a positive e¤ect post-transition. The e¤ect on net fertility cannot be immediately deduced from these regressions since, contrary to what was the case when crude birth and death rates were used, net fertility is no longer the di¤erence between these two rates. Instead, let net fertility be de…ned as n = g(1 m15 ) where g is gross fertility (the number of births per woman), m15 is the number of deaths between birth and age 15 per live birth (age 15 being our de…nition of adulthood), and n is net fertility, thus de…ned as the number of adult children each woman brings about. This de…nition of net fertility is roughly proportional to the Net Reproduction Rate, a standard demographic mea23 sure which calculates the number of daughters reaching their reproductive age per woman15 . The derivative of net fertility with respect to m15 is given by @g @m15 (1 = m15 ) g. Since our regressions have made use of mortality rates up to the age of 5, this last expression may be rewritten as m15 ) @n @m15 @n @m15 = g , where m5 is nothing but the child mortality rate responds to coe¢ cient @g @m5 @m5 @m15 (1 @g and @m cor5 as estimated above. Using data from the United Nations, I am then able to calculate the derivative @n @m15 for di¤erent groups of nations over time under the assumption of a constant value for the derivative @g @m5 : By using the value @g @m5 = 15:5 I approximate the e¤ect of mortality on net fertility among post-transition countries. The results are given in table 8 and suggest that, on average, the e¤ect of mortality on net fertility has been positive throughout the world over the last …ve decades in post-transition countries. This resonates well with the …ndings of Angeles (2010), who provides direct statistical evidence of a positive e¤ect on net fertility using Net Reproduction Rates as the dependent variable - albeit without distinguishing between pre- and post-transition observations. Together, these two sets of results support the idea that mechanisms other than pure replacement e¤ects are in place, leading couples to reduced their desired target family size when mortality rates diminish. [Table 8] Tables 9 and 10 o¤er some robustness checks on the above results. First, in table 9, I experiment with di¤erent values of the threshold level of gross fertility separating pre- from post-transition periods. I set this level equal to 4.5, 5, 5.5 (the baseline value) and 6 children per woman. All regressions include the most comprehensive set of controls: GDP per capita, the urbanization rate, female education and time dummies, but their coe¢ cients are not reported for conciseness. Results are very consistent: in all cases 15 Net Reproduction Rates are not reported in the World Development Indicators, which is why a direct estimation of the e¤ect on net fertility is not possible. 24 the coe¢ cient on mortality is close to zero and not statistically signi…cant for pre-transition times, becoming large, positive and statistically signi…cant for the post-transition period. The size of the post-transition coe¢ cient is very stable, around 15.5 for threshold values between 4.5 and 5.5, and only somewhat smaller when the threshold value is raised to 6. This last result is unsurprising as such a high number of children per woman is probably allowing some pre-transition observations into our post-transition group. Table 10 experiments with the other two mortality measures at our disposal, the infant mortality rate and life expectancy at birth. The threshold value of gross fertility is set again at 5.5, and all available controls are used but not reported. The results for the infant mortality rate mirror those obtained for child mortality rates, which is unsurprising as the two variables are highly correlated. For life expectancy we obtain the same qualitative result, with the pre-transition coe¢ cient being not statistically signi…cant while the post-transition one is statistically signi…cant at the 1% level (the sign of the coe¢ cients is negative, as life expectancy is an inverse function of mortality rates). There are, however, some quantitative di¤erences. The pre-transition coe¢ cient of life expectancy, while not signi…cant, is fully one third the size of the post-transition coe¢ cient. Furthermore, for other values of the threshold level of fertility a statistically signi…cant coe¢ cient may be obtained for the pre-transition case (not reported in the table). The result makes sense as life expectancy incorporates adult mortality rates, and lower adult mortality may well have a separate impact on fertility decisions - for instance by shifting female preferences away from early marriage and child rearing and towards longer education and participation in the labour market. 5 Concluding remarks The evidence presented in this paper should leave no doubts as to the importance of mortality decline as a major driving force behind demographic transitions. As a rule, changes in mortality are only weakly associated with 25 gross fertility in the early phase of the process, but this should not surprise us given that families are most likely unable to reach their desired number of surviving children. Once they do so, thanks in large part to lower mortality rates, further decreases in mortality tend to be matched by lower fertility rates; and the e¤ect is both statistically signi…cant and large in magnitude. What is more, lowering mortality rates is quite possibly the single largest cause of declining gross fertility rates. To substantiate this last point, I use the third dataset considered above and perform the following calculation. First, for all countries with the required data, I compute the change in the Total Fertility Rate and the change in the Child Mortality Rate between each country’s …rst post-transition period and 2010 (for countries with a fertility transition before 1960 I compute the di¤erence between the 1960 and 2010 values). I obtain results for 122 countries. For each of these, I compute the percentage of the change in gross fertility that would be accounted by the change in child mortality using a coe¢ cient of 15:5, following the last column of table 7. I discard 9 countries for which the predicted change in fertility is larger than the actual change, and then take the average of all remaining countries. I …nd that, as an average, 40.5% of the change in the total fertility rate experienced since the onset of the transition (or since 1960) can be explained by changes in child mortality rates. Given that no more than 79% of the within-country variation in total fertility rates among all post-transition observations is ever explained in the regressions of table 7, no single factor is likely to be of more consequence than mortality change as a driver of gross fertility. 26 References Acemoglu, D. and Johnson, S. (2007), Disease and Development: The E¤ect of Life Expectancy on Economic Growth, Journal of Political Economy 115, 925-985. Angeles, L. (2010), Demographic transitions: analyzing the e¤ects of mortality on fertility, Journal of Population Economics 23, 99-120. Bailey, A. K. (2009), How personal is the political? Democratic revolution and fertility decline, Journal of Family History 34, 407-425. Becker, G. S., Murphy, K. M. and Tamura, R. (1990), Human capital, fertility, and economic growth, Journal of Political Economy 98, S12-S37. Cervellati, M. and Sunde, U. (2015), The e¤ect of life expectancy on education and population dynamics, Empirical Economics 48, 1445-1478. Clark, G. and Hamilton, G. (2006), Survival of the richest: The Malthusian method in England, 1585-1638, Journal of Economic History 66, 707736. Coale, A. J. (1986), The decline of fertility in Europe since the eighteenth century as a chapter in demographic history, in: Coale, A. J. and Watkins, S. C. (eds.), The decline of fertility in Europe. Princeton: Princeton University Press, 1-30. Coale, A. J. and Treadway, R. (1986), A summary of the changing distribution of overall fertility, marital fertility, and the proportion married in the provinces of Europe, in: Coale, A. J. and Watkins, S. C. (eds.), The decline of fertility in Europe. Princeton: Princeton University Press, 31-181. Ehrlich, I. and Lui, F. T. (1991), Intergenerational trade, longevity, and economic growth, Journal of Political Economy 99, 1029-1059. Friedlander, D., Okun, B. S. and Segal, S. (1999), The demographic transition then and now: processes, perspectives, and analyses, Journal of Family History 24, 493-533. Galor, O. (2005a), From stagnation to growth: uni…ed growth theory. In: Aghion, P. And Durlauf, S. N. (eds.), Handbook of Economic Growth, vol. 1A. Elsevier, Amsterdam. Galor, O. (2005b), The demographic transition and the emergence of 27 sustained economic growth, Journal of the European Economic Association 3, 494-504. Galor, O. (2012), The demographic transition: causes and consequences, Cliometrica 6, 1-28. Guinnane, T. W. (2011), The historical fertility transition: a guide for economists, Journal of Economic Literature 49, 589-614. Herzer, D., Strulik, H. and Vollmer, S. (2012), The long-run determinants of fertility: one century of demographic change 1900-1999, Journal of Economic Growth 17, 357-385. Kalemli-Ozcan, S. (2002), Does the mortality decline promote economic growth?, Journal of Economic Growth 7, 411-439. Kirk, D. (1996), Demographic transition theory, Population Studies 50, 361-387. Livi-Bacci, M. (1991), Population and nutrition : An essay on European demographic history, Cambridge: Cambridge University Press. Murtin, F. (2013), Long-term determinants of the demographic transition, 1870-2000, Review of Economics and Statistics 95, 617-631. Palgrave Macmillan Ltd. (ed.), 2013, International Historical Statistics, Basingstoke: Palgrave Macmillan. Sah, R. K. (1991), The e¤ects of child mortality changes on fertility choice and parental welfare, Journal of Political Economy 99, 582-606. Strulik, H. and Weisdorf, J. (2014), How child costs and survival shaped the industrial revolution and the demographic transition, Macroeconomic Dynamics 18, 114-144. United Nations (2013), World Population Prospects: The 2012 Revision. Available online at: http://esa.un.org/unpd/wpp/Excel-Data/mortality.htm van de Walle, F. (1986), Infant mortaity and the European demorgaphic transition, in: Coale, A. J. and Watkins, S. C. (eds.), The decline of fertility in Europe. Princeton: Princeton University Press, 201-233. Wachter, K. W. (2006), Essential Demographic Methods, mimeo, University of California, Berkeley. 28 Table 1 Parity-Specific Fertility Rates: Nigeria and the United States in 1990 Parity 0 1 2 3 4 5 6+ Nigeria 0.272 0.225 0.257 0.244 0.279 0.284 0.233 United States 0.276 0.246 0.116 0.110 0.124 0.148 Notes: each cell contains the fertility rate, defined as the total number of births divided by the number of women, as a function of a woman’s parity. The subjects are women aged between 30 and 35 in both Nigeria and the United States. Data for the United States refers to white women only, and the observation for parity 5 corresponds to women with parity 5 and above. Source: Wachter (2006, p. 280). Table 2 Summary statistics, European countries 1750-1949 Crude Birth Rate Mean 30.17 Std. Dev. 7.79 Min 11.36 Max 50.92 Observations 479 Crude Death Rate 21.08 6.50 8.66 44.54 478 Marriage Rate 15.60 2.65 7.68 25.67 466 Time trends (change per 5-year period, controlling for country fixed effects): Before the fertility transition After the fertility transition Crude Birth Rate -0.177*** -1.053*** Crude Death Rate -0.362*** -0.665*** Table 3 Empirical analysis, European countries 1750-1949 Dependent variable: Crude Birth Rate Pre-transition Post-transition Pre-Transition Post-Transition 0.235*** [0.094 – 0.376] 0.575*** [0.128] 1.448*** [1.253 – 1.641] 0.250*** [0.0871] 0.158* [-0.020 – 0.335] 0.621*** [0.101] 0.904*** [0.521 – 1.287] 0.242* [0.137] no no yes yes 247 22 0.315 193 20 0.768 247 22 0.600 193 20 0.867 One-period lag of: Crude Death Rate Marriage Rate Time dummies Observations Countries R2 Notes: Robust standard errors in square brackets, except for the Crude Death Rate where the 95% confidence interval using robust standard errors is given. The symbols ***, ** and * denote statistical significance at the 1%, 5% and 10% levels. Table 4 Summary statistics, 70 Developed and Developing countries, 1870-2010 Crude Birth Rate Mean 29.78 Std. Dev. 12.22 Min 8.3 Max 59.20 Observations 717 Crude Death Rate 14.65 7.22 3.56 37.80 751 GDP per capita (in logs) 8.51 0.92 6.62 10.87 806 Years of schooling among the female population 3.77 3.38 0.04 13.37 1080 Share of pop. aged 20-29 0.173 0.018 0.103 0.224 1004 Share of pop. aged 30-39 0.134 0.017 0.083 0.205 1004 Time trends (change per 10-year period, controlling for country fixed effects): Before the fertility transition After the fertility transition Crude Birth Rate -0.407*** -1.832*** Crude Death Rate -2.340*** -0.752*** Table 5 Empirical analysis, 72 Developed and Developing countries, 1870-2010 Dependent variable: Crude Birth Rate Pre-transition Post-transition Pre-Transition Post-Transition Pre-transition Post-transition Pre-Transition Post-Transition One period lag of: Crude Death Rate 0.276** 1.262*** 0.449*** 0.919*** 0.376** 0.798*** 0.316* 0.797*** [0.03 – 0.52] [1.10 – 1.42] [0.16 – 0.73] [0.75 – 1.09] [0.07 – 0.68] [0.62 – 0.98] [-0.02 – 0.65] [0.59 – 1.00] 3.338* [1.881] -3.895*** [0.663] 4.589** [1.738] -1.638*** [0.540] -0.750 [0.979] -1.411*** [0.338] 2.086 [1.776] -2.394*** [0.576] 0.874 [1.328] -0.242 [0.307] GDP per capita Female education Controls for pop. age structure yes yes yes yes yes yes yes yes Time dummies no no no no no no yes Yes Observations Countries R2 259 336 237 335 237 335 237 335 67 57 65 57 65 57 65 57 0.155 0.734 0.286 0.786 0.340 0.809 0.491 0.859 Notes: Robust standard errors in square brackets, except for the Crude Death Rate where the 95% confidence interval using robust standard errors is given. The symbols ***, ** and * denote statistical significance at the 1%, 5% and 10% levels. Table 6 Summary statistics, 187 Developed and Developing countries, 1960-2010 Mean 4.21 Std. Dev. 2.04 Min 0.85 Max 9.19 Observations 2131 0.0878 0.0847 0.0024 0.4979 1848 GDP per capita (in logs) 7.93 1.62 3.91 11.75 1664 Urbanisation rate (%) 48.75 25.63 2.08 100 2320 Years of schooling among the female population 5.45 3.31 0.01 13.23 1584 Infant Mortality Rate (under-1 deaths per birth) 0.0577 0.0486 0.0019 0.2502 1834 Life Expectancy at Birth 62.79 11.68 23.73 83.16 2141 Total Fertility Rate Child Mortality Rate (under-5 deaths per birth) Time trends (change per 5-year period, controlling for country fixed effects): Before the fertility transition After the fertility transition Total Fertility Rate -0.089*** -0.227*** Child Mortality Rate (%) -1.954*** -0.576*** Table 7 Empirical analysis, 187 Developed and Developing countries, 1960-2010 Dependent variable: Total Fertility Rate Pre-transition Post-transition Pre-Transition Post-Transition Pre-transition Post-transition Pre-Transition Post-Transition 23.79*** [0.759] 3.070*** [0.960] 19.83*** [0.840] 1.787* [0.909] 15.59*** [0.874] -0.278 [1.251] 15.54*** [0.860] 0.236*** [0.0819] -0.253*** [0.0485] 0.184** [0.0815] -0.0241*** [0.00670] -0.0533 [0.0510] -0.0339*** [0.00369] 0.0465 [0.0882] -0.000472 [0.00675] -0.420*** [0.0483] 0.259*** [0.0577] -0.0200*** [0.00375] -0.201*** [0.0175] -0.141 [0.104] 0.00761 [0.00750] -0.286*** [0.0658] 0.370*** [0.0619] -0.0197*** [0.00363] -0.155*** [0.0264] no no no no no no yes Yes One period lag of: Child Mortality Rate GDP per capita 5.910*** [0.559] Urban rate Female education Time dummies Observations Countries R2 367 947 367 947 297 807 297 807 82 169 82 169 67 132 67 132 0.284 0.693 0.315 0.723 0.530 0.773 0.569 0.791 Notes: Robust standard errors in square brackets. The symbols ***, ** and * denote statistical significance at the 1%, 5% and 10% levels. The Child Mortality Rate is expressed as number of deaths before the age of 5 per child born alive (not per 1,000 children), GDP per capita is in logs, Female education is the average number of years of schooling among the female population aged 15 and over. Table 8 Estimated derivate of net fertility with respect to mortality, groups of countries Developed countries Developing countries Sub-Saharan Africa Asia Latin America 1950-55 9.87 3.09 1955-60 10.14 3.59 1960-65 10.34 3.60 1965-70 10.58 4.75 1970-75 10.66 5.57 1975-80 10.92 6.78 1980-85 10.91 7.34 1985-90 10.90 7.70 1990-95 10.60 8.51 1995-00 10.63 9.09 2000-05 10.81 9.22 2005-10 10.87 9.24 1.63 3.47 4.98 1.93 4.02 5.32 2.17 3.95 5.60 2.41 5.26 6.30 2.60 6.06 7.09 2.84 7.48 7.93 3.12 8.05 8.82 3.46 8.35 9.44 3.80 9.25 9.89 4.23 9.91 10.11 4.49 10.18 10.21 4.88 10.26 10.40 Notes: The entries give the estimate of (2013). as discussed in the text, using = 15.5 and values for is assumed to equal the ratio of under-5 to under-15 mortality rates. , for each region and period from United Nations Table 9 Different definitions of the onset of the fertility transition Dependent variable: Total Fertility Rate Threshold level of gross fertility TFR*=4.5 Pre-transition TFR*=5.0 TFR*=5.5 TFR*=6.0 Post-transition Pre-transition Post-transition Pre-Transition Post-Transition Pre-transition Post-transition 15.75*** [1.066] 0.115 [1.183] 15.48*** [1.032] -0.278 [1.251] 15.54*** [0.860] -0.642 [1.500] 12.27*** [0.716] One period lag of: Child Mortality Rate Observations Countries R2 0.744 [1.083] 422 682 368 736 297 807 217 887 78 117 73 125 67 132 52 136 0.727 0.735 0.660 0.749 0.569 0.791 0.457 0.783 Notes: All regressions control for GDP per capita, the urbanisation rate, female education, and a full set of time dummies. Robust standard errors in square brackets. The symbols ***, ** and * denote statistical significance at the 1%, 5% and 10% levels. Table 10 Other indicators of mortality Dependent variable: Total Fertility Rate Pre-transition Post-transition 1.879 [1.941] 27.34*** [1.634] Pre-transition Post-transition -0.0133 [0.00878] -0.0387*** [0.00907] One period lag of: Infant Mortality Rate Life Expectancy at Birth Observations Countries R2 371 871 386 900 74 132 76 134 0.509 0.764 0.484 0.638 Notes: All regressions control for GDP per capita, the urbanisation rate, female education, and a full set of time dummies. Robust standard errors in square brackets. The symbols ***, ** and * denote statistical significance at the 1%, 5% and 10% levels.
© Copyright 2026 Paperzz