Abstract

Understanding how mortality and fertility are linked is essential to the study of population dynamics. We investigate the fertility response to an unanticipated mortality shock that resulted from the 2004 Indian Ocean tsunami, which killed large shares of the residents of some Indonesian communities but caused no deaths in neighboring communities. Using population-representative multilevel longitudinal data, we identify a behavioral fertility response to mortality exposure, both at the level of a couple and in the broader community. We observe a sustained fertility increase at the aggregate level following the tsunami, which was driven by two behavioral responses to mortality exposure. First, mothers who lost one or more children in the disaster were significantly more likely to bear additional children after the tsunami. This response explains about 13 % of the aggregate increase in fertility. Second, women without children before the tsunami initiated family-building earlier in communities where tsunami-related mortality rates were higher, indicating that the fertility of these women is an important route to rebuilding the population in the aftermath of a mortality shock. Such community-level effects have received little attention in demographic scholarship.

Introduction

A central line of inquiry in population research assesses whether, when, and why fertility changes in concert with mortality. Investigators have hypothesized that exposure to mortality shapes fertility intentions and behaviors of an individual or couple through its impact on expectations about the survival prospects of children. A mortality effect may arise because the death of a child causes replacement of the child, or because expectations about future mortality cause hoarding (Montgomery and Cohen 1998; Palloni and Rafalimanana 1999; Preston 1978; Rosenzweig and Schultz 1983).

Demographic theory is less well suited to describe fertility responses to large-scale mortality events, such as war and natural disasters (Hill 2004). To the extent that a one-time event does not change expectations about future child survival, a fertility response to a temporary mortality increase cannot be attributed to a hoarding motive.

Instead, it may be that the replacement motive extends beyond those women whose children die and operates through social groups, such as extended families, networks, or ethnic groups. When mortality shocks are location-specific, the local community is likely to be a salient group in which a broader replacement and rebuilding motive may operate.

To provide empirical evidence on the extent of both individual and community-level mechanisms, we study the effects of mortality exposure on fertility after a large-scale unanticipated natural disaster: the 2004 Indian Ocean tsunami, which killed over 170,000 people in the coastal areas of Aceh and North Sumatra, Indonesia. This work builds on previous studies that have documented fertility shifts after earthquakes, hurricanes, famines, economic crises, terrorist attacks, war, and genocide (e.g., Agadjanian and Prata 2002; Caldwell 2004; Finlay 2009; Heuveline and Poch 2007; Lindstrom and Berhanu 1999; see also the review in Lee 1997).

Using a sample representative of the pre-tsunami population, we document a positive association between exposure to the 2004 tsunami and subsequent fertility at the aggregate level. We provide evidence that the fertility increase can be attributed to tsunami mortality and that it is driven primarily by two groups of women. First, mothers who lost a child in the tsunami were significantly more likely to give birth again after the tsunami relative to mothers whose children survived. These births account for about 13 % of the aggregate increase in fertility due to the tsunami. Second, where local area mortality was higher, women who had not borne children before the disaster were also significantly more likely to give birth after the disaster relative to similar women in communities with lower tsunami mortality.

Several features of the disaster provide leverage to address these questions. First, because the force of the water on land varied with topographical and hydrological features, the tsunami’s impact on mortality was idiosyncratic even within small areas (Frankenberg et al. 2011; McAdoo et al. 2007; Umitsu et al. 2007). Second, in contrast to deaths from war or famine, tsunami-related mortality occurred almost entirely within a few hours of the precipitating earthquake, making it possible to pinpoint timing. Third, the tsunami was completely unanticipated. Thus, tsunami-related mortality can legitimately be treated as a mortality shock at the local area level.

It has been a challenge to establish a causal link between mortality and fertility in previous research, largely because of data constraints. We use data from a rich longitudinal survey conducted in coastal Indonesia that was designed to address this question. The survey that provides the baseline for the Study of the Tsunami Aftermath and Recovery (STAR) was collected 10 months before the tsunami as part of SUSENAS, a large-scale nationally representative household socioeconomic survey conducted annually by Statistics Indonesia. Follow-ups were conducted annually for five years after the tsunami. The survey includes detailed pregnancy and birth histories, combined with information about the disaster collected at the individual and community levels. The survey encompasses heavily damaged communities with high mortality (on average, about 30 % of residents perished in these communities) as well as nearby communities where the direct effects were much more muted. Data from STAR allow us to relate the experience of mortality within the local area to the fertility of women who were living in that area at the time of the tsunami. This empirical approach is more powerful than relying on temporal variation alone.

The research advances understanding of both replacement fertility of individual women and population rebuilding in the context of high-mortality disasters. In recent years, similarly sudden large-scale events have generated significant mortality shocks in Haiti, Myanmar, Japan, China, and India, among others, and are likely to recur given rising population densities in areas increasingly vulnerable to environmental crises (Marshall and Picou 2008; Vos et al. 2010). By drawing direct comparisons between estimated effects at the community level and estimates of replacement at the individual level, the research provides new evidence on the relative importance of these two effects in population dynamics following disaster.

Fertility in Response to Macro Shocks: Evidence

Population scientists have long studied the demographic consequences of large-scale macro shocks. Effects on fertility have been observed at both the aggregate and individual level. War has received the most systematic attention (e.g., Rindfuss and Sweet 1978). Several studies have documented significant declines in fertility, either overall or for more-affected relative to less-affected subgroups during conflicts accompanied by major social upheaval (Agadjanian and Prata 2002; Blanc 2004; Caldwell 2004; Heuveline and Poch 2007; Lindstrom and Berhanu 1999). In some instances, the end of the conflict is accompanied by a fertility increase. Caldwell (2004), for example, documented a fall followed by a rise in fertility for Russia, Spain, and Germany in the context of major disruptions before the 1960s. Famines are characterized by a similar temporal fertility pattern, as evidenced by studies from the Netherlands, China, and Bangladesh (Ashton et al. 1984; Stein and Susser 1975; Watkins and Menken 1985). Isolating proximate mechanisms and disentangling whether fertility increases represent fundamental shifts in fertility desires or simply the realization of deferred reproduction are complicated when the precipitating events occur over multiple years and involve shifting spatial boundaries.

Other research considers spatially and temporally more discrete events, such as natural disasters and terrorist attacks. Results are mixed. Using a theoretical framework guided by work in psychology on stress and attachment, Cohan and Cole (2002) analyzed rates of marriage, birth, and divorce before and after Hurricane Hugo in affected and unaffected South Carolina counties. Rates for each of these outcomes rise and then fall, leading the authors to suggest that exposure to a life-threatening event prompted significant actions and measurable changes with respect to close relationships. A similar approach and conclusion was reached with respect to fertility in and around Oklahoma City in conjunction with the 1995 bombing of the Alfred P. Murrah Federal Building (Rodgers et al. 2005). By contrast, despite predictions of a post–September 11 baby boom in the United States (Morin 2002; Scelfo 2002), natality data indicated no such increase (Martin et al. 2003). It is important to note that none of these events have caused large-scale loss of life, and thus may provide only limited insights into behavioral responses to the death of a substantial fraction of the population.

Relatively few studies have sought to isolate the impact of large-scale mortality on subsequent fertility. The most comprehensive study involves the long-term impact on fertility of excess mortality in Cambodia during the years of Khmer Rouge control (1975–1978), when some 25 % of the population died as a result of war-related violence and disease (Heuveline 1998; Heuveline and Poch 2007). Using retrospective birth histories collected in 2000 for women (ages 15–74), the authors documented a sharp decline in the total fertility rate between 1975 and 1978, a near doubling between 1978 and 1980 to levels above the prewar rate, and then a decline. The period of most dramatic increase occurs shortly after Vietnam took control of the country and the Khmer Rouge–imposed genocide was abruptly halted, leading the authors to conclude that the fertility increase was a response to the high levels of mortality.

Two other studies have explored fertility in the aftermath of high-mortality disasters. Finlay (2009), using cross-sectional surveys, considered fertility for three earthquakes, each with death tolls of 15,000 or more. Comparing fertility before and after the earthquake for residents of areas affected by the earthquake with fertility of residents of areas that were not affected revealed greater post-disaster increases in fertility in affected areas. The same approach was adopted using census data to examine the impact of the 2003 Bam earthquake in Iran. The authors documented a fertility decline in 2004, followed by a rise in 2005–2007 (Hosseini-Chavoshi and Abbasi-Shavazi 2013).

Fertility in Response to Mortality: Theory

A considerable body of theoretical work posits that fertility levels are partly a response to mortality levels in the broader community and to couples’ own experience of child mortality. These ideas have roots in demographic transition theory as well as in theories from psychology, sociology, and economics.

Fertility in Response to Mortality Outside the Family

The most explicit theoretical link between child mortality occurring outside the family and fertility decisions—or what Preston (1978) referred to as “extrafamilial effects”—involves the concept of “insurance” fertility, whereby parents bear more children than they ultimately want to have in anticipation that some will not survive (Cain 1981; Montgomery and Cohen 1998; Portner 2007; Preston 1978; Rosenzweig and Schultz 1983). A few studies have linked fertility timing to declines in child mortality at the level of the village (Atella and Rosati 2000; LeGrand et al. 2003) or the social network (Sandberg 2006). Of course, insurance effects are irrelevant for disasters unless the event shifts expectations about child survival over the longer term and causes parents to produce “extra” children.1

Mortality outside the family might affect fertility through other avenues as well. One mechanism may work through risk-sharing at the level of the community or ethnic group, as has been described in some agricultural settings (see Conning and Udry 2007; Geertz 1968; Grimard 1997; Suri 2003; Townsend 1994). If community members benefit from the next generation of children, a collective fertility increase might arise in response to child mortality at the community level.

Although demographers have emphasized ties between child mortality and fertility, the literature on resilience in psychology suggests a different motivation for the links between fertility and mortality more generally. Specifically, the experience of mortality may shift preferences toward goals with intrinsic meaning (such as interpersonal connection and community development) rather than extrinsic goals (such as amassing wealth and status) (Vail et al. 2012). The logic suggests that among other things, a renewed investment in family will emerge in response to an awareness of human frailty (Fritsche et al. 2007; Nakonezny et al. 2004).

Related research views fertility as taking on symbolic meaning in the aftermath of population trauma. When crises disrupt individuals’ perceptions of the world as ordered and reliable, births represent renewal and a “return to normal” (e.g., Carta et al. 2012; Norris et al. 2002; Rodgers et al. 2005). Other scholarship describes the pronatalist sentiment emerging from mortality shocks that disproportionately affect (or target) certain ethnic groups (Borneman 2002; Jansen and Helms 2009) or hypothesize its existence in settings where a substantial fraction of the population has perished (Heuveline and Poch 2007).

Behavioral fertility responses to contextual mortality shocks are likely to vary by women’s fertility goals, parity, and the age composition of surviving children. Women who are childless or at lower parities will, on average, desire more children and ceteris paribus will exhibit stronger behavioral fertility response to mortality shocks.

Of course, births in the months following a macro shock may be reduced because of miscarriage. A mortality shock may lower population fertility over the longer term through other avenues as well. Witnessing the deaths of family and friends may induce psychopathologies (Neria et al. 2008; Norris et al. 2002), which may reduce the desire for children, coital frequency, relationship quality, and women’s physiological capacity to carry a child to term (Nakamura et al. 2008; Parker and Douglas 2010; Segraves 1998). How and when the psychological response to mortality produces family-building rather than family disruption is not well understood (Cohan 2010).

Fertility in Response to Mortality of Own Children

Although theory supports a role for contextual effects of mortality, many studies have focused on whether the event of an “own-child” death prompts a couple to conceive again, so that the child who died is “replaced” by one that would not otherwise have been born. Preston (1978) outlined the pathways of individual replacement behavior. Replacement may arise as an artifact of physiology: a child death can increase fertility simply because a woman stops breast-feeding and resumes menstruating. Alternatively, after the death of a child, couples may intentionally try to conceive.

Micro-level research documents these phenomena, finding that both physiologically induced and volitional replacement operate in various contexts, but that neither exhibits large population-level impacts on fertility (Frankenberg 1998; Grummer-Strawn et al. 1998; Hossain et al. 2007; Kuate Defo 1998; Lindstrom and Kiros 2007; Montgomery and Cohen 1998; Palloni and Rafalimanana 1999; Rosero-Bixby 1998). This literature considers settings where child mortality levels are relatively stable and volitional replacement is a response to an event about which parents can form reasonable expectations—an orientation with origins in demographic transition theory. As a result, bias from omitted variables or endogeneity is an oft-referenced issue (Palloni and Rafalimanana 1999).

Studies of unanticipated disasters help put these concerns aside, but individual-level studies are possible for only a few high-mortality disasters. The 2008 Wenchuan earthquake in Sichuan, China is an exception. Thousands of parents lost their only child. In response, the state sponsored fertility programs, which helped a number of women to explicitly “replace” the child that perished (Pinghui 2013; Qin et al. 2009).

The Indian Ocean tsunami provides a similarly powerful context in which to study both “replacement” fertility and a potential response to mortality within the community. The longitudinal data support estimation of whether a woman who lost a child in the tsunami subsequently bore another child, which speaks directly to the question of an individual response. Moreover, STAR provides evidence on the relationship between local area tsunami mortality and subsequent fertility among residents of the area at the time of the tsunami, and thereby allows us to contrast the relative importance of extrafamily and intrafamily responses to a large-scale mortality shock.

Mortality and Fertility During the 2004 Tsunami

The 2004 Indian Ocean tsunami was exceptional in magnitude and scope. On December 26, an earthquake measuring 9.3 on the Richter scale displaced 1 trillion tons of water. The resulting tsunami slammed into the Indonesian coastline, reaching some areas as little as 15 minutes after the earthquake. Aceh and North Sumatra were the provinces hardest hit. In some areas of Aceh, water heights exceeded 15 meters (Umitsu 2011:54, 58) and the water traveled inland via riverbeds for as much as 6 kilometers. About 170,000 people died, and over 500,000 were displaced, losing their homes and livelihoods (Athukorala and Resosudarmo 2005; Doocy et al. 2007; Gray et al. 2014).

The magnitude of the tsunami’s overall impact masks considerable spatial variation, arising from the idiosyncrasies of coastal topography that shaped variation in the waves’ force and extent of inundation (McAdoo et al. 2007; Ramakrishnan et al. 2005; Umitsu et al. 2007). Neighboring communities experienced markedly different degrees of damage as a function of elevation and orientation relative to the shoreline. STAR is designed to capture this heterogeneity.

To assess the impact of mortality on subsequent fertility, we analyze data from communities that experienced substantial tsunami mortality and communities from the same districts (kabupaten) that did not. Communities are excluded from our sample if no area in the district was affected by tsunami mortality, yielding a sample in which comparison areas (those with no tsunami-related mortality) are close enough geographically to be similar to areas that sustained tsunami-related mortality.

Data from the first follow-up wave of STAR regarding survival status of survey respondents provide our primary source of information for classifying communities. Survival status was ascertained by identifying an individual who was a household member at the baseline survey and could confirm other members’ survival statuses. If no original household member could be located, death information was derived from interviews with multiple people who were living in the community at baseline and by cross-checking rosters of the dead maintained in the community. Mortality status was determined for 97.4 % of the STAR sample. Other sources of information are used to cross-validate community classifications, including questions to community leaders on tsunami deaths, water inundation, and destruction; damage observations of survey supervisors; and levels of exposure to the tsunami reported by residents. A community is classified as mortality-affected if one or more STAR respondents died because of the tsunami, unless the number of deaths is very small and no other corroborating evidence suggests that the tsunami acted with any force on the community.2

Table 1 presents descriptive statistics on the tsunami’s impact, estimated separately for communities directly affected by tsunami mortality and for comparison communities from the same districts. As they should be, differences in the mortality rates are stark. In mortality-affected communities, the tsunami killed 29 % of the population. In the communities with no tsunami-related mortality, about 0.5 % died. Within the mortality-affected communities, rates vary by demographic group. About one-third of children under 5 and one-third of reproductive-age women perished, whereas the rate is 19 % among prime-age men.

Table 1

Tsunami impacts by whether the community was mortality-affected

Mortality Zone
Community Affected by Tsunami MortalityCommunity Not Affected by Tsunami Mortality
A. % Dead as a Result of the Tsunami 
 All ages 29.0 0.5 
  Children younger than 5 33.0 1.0 
  Children aged 5–14 30.0 0.4 
  Women aged 15–49 33.0 0.7 
  Men aged 15–19 19.0 0.3 
B. Measures of Physical Damage   
 Satellite-based: Number of pixels indicating a post-tsunami change to bare earth (mean, range 0–9) 5.1 0.8 
 Village leader reports (%)   
  Water contaminated 69 12 
  Damage to roads 68 21 
  Problem with debris 69 13 
C. Subjective Measures of Tsunami Impacts (%)   
 Village leader reports that life in the community changed for the worse 73 34 
 Reported reasons   
  Loss of population 21 
  Increases in price of rice 41 30 
  Increases in other food prices 50 39 
  Fewer business opportunities 43 26 
  Fewer jobs 45 26 
Number of Communities 96 195 
Mortality Zone
Community Affected by Tsunami MortalityCommunity Not Affected by Tsunami Mortality
A. % Dead as a Result of the Tsunami 
 All ages 29.0 0.5 
  Children younger than 5 33.0 1.0 
  Children aged 5–14 30.0 0.4 
  Women aged 15–49 33.0 0.7 
  Men aged 15–19 19.0 0.3 
B. Measures of Physical Damage   
 Satellite-based: Number of pixels indicating a post-tsunami change to bare earth (mean, range 0–9) 5.1 0.8 
 Village leader reports (%)   
  Water contaminated 69 12 
  Damage to roads 68 21 
  Problem with debris 69 13 
C. Subjective Measures of Tsunami Impacts (%)   
 Village leader reports that life in the community changed for the worse 73 34 
 Reported reasons   
  Loss of population 21 
  Increases in price of rice 41 30 
  Increases in other food prices 50 39 
  Fewer business opportunities 43 26 
  Fewer jobs 45 26 
Number of Communities 96 195 

Panels B and C describe destruction to the built and natural environment, using two approaches. Panel B draws on satellite imagery before and after the disaster. Gillespie et al. (2007), using images from the U.S. National Aeronautics and Space Administration (NASA) Moderate Resolution Imaging Spectroradiometer (MODIS) sensor taken on December 17 and December 29, 2004, developed a measure of destruction to ground cover for a 750m grid surrounding each of the center points of the communities surveyed in STAR (based on GPS). In the mortality-affected communities, over half the area of the grid was turned to brown earth, indicating complete destruction of ground cover, but less than 10 % of the grid was turned to brown earth in communities not affected by mortality. Our second approach, displayed in panel C, relies on reports by village leaders about particular types of destruction. Damage to roads, contaminated water, and problems with debris are all more common in mortality-affected communities (affecting about two-thirds of the communities), but they occur in nonmortality communities as well, where rates range from 12 % to 21 % across the indicators.

Finally, community leaders were asked how life had changed in the year after the tsunami. For the mortality-affected communities, 73 % of leaders report a downturn in quality of life, and fully 21 % of them explicitly mention population loss as a problem; price increases and reductions in economic opportunity are other common responses. In the communities unffected by mortality, about one-third of leaders felt that life became worse. Population loss is not the problem, but rising food prices and fewer jobs and business opportunities are.

In combination, these indicators capture an important aspect of the tsunami. Although mortality and the most devastating physical destruction were concentrated in a subset of communities along the west coast, the geographical reach of disruption is much broader. Our comparison communities are affected by the disaster, but not by the mortality it caused.

We turn now to macro-level evidence of fertility change. To document patterns of fertility before and after the disaster, we calculate age-specific probabilities of live birth for each quarter-year between 2000 and 2009 for reproductive-age women interviewed in the baseline survey.3 The rates pertain to the period before the tsunami (January 2000–December 2004) and to the period after the tsunami (January 2006–December 2009). We exclude births in 2005 because fertility during most of that year will not be attributable to behavioral decisions made in response to the disaster (and was complicated by miscarriages after the tsunami) (Hamoudi et al. 2014).

Figure 1 displays these rates, stratified by period and mortality zone (dashed lines indicate communities experiencing tsunami mortality). For the 2000–2004 period, the estimated total fertility rates are 2.18 and 2.74 in the areas with and without tsunami-related mortality, respectively. After the tsunami, fertility rates by zone change in opposite directions. In the mortality-affected communities, fertility increases, particularly for women aged 20–34, and the overall rates for 2006–2009 are 2.67 and 2.52, respectively, in the mortality-affected versus comparison communities. The difference between the increase in the mortality-affected communities and the decrease in communities that did not experience tsunami-related mortality is 0.71, which is statistically significant at the 5 % level. (The bootstrapped standard error is 0.21.4) The difference-in-difference for the underlying ages (displayed in the inset table of Fig. 1) confirms that the largest changes in mortality-affected communities occur for women between the ages of 20 and 34.

Fig. 1

Age-specific fertility rate (ASFR) and total fertility rate (TFR) differences by period and presence of tsunami mortality in the community. Average quarter-year age-specific probabilities of live birth presented for the pre-tsunami period (January 1, 2000–December 31, 2004) and the post-tsunami period (January 1, 2006–December 31, 2009) for women in 283 communities. In each quarter-year, women from the baseline survey who are alive and aged 15–49 contribute an observation. Pooled data contain 176,862 observations for 6,363 women. Average annual TFR estimated as TFR = 5 × 4 × Σi(ASFRi), where i indexes five-year age intervals

Fig. 1

Age-specific fertility rate (ASFR) and total fertility rate (TFR) differences by period and presence of tsunami mortality in the community. Average quarter-year age-specific probabilities of live birth presented for the pre-tsunami period (January 1, 2000–December 31, 2004) and the post-tsunami period (January 1, 2006–December 31, 2009) for women in 283 communities. In each quarter-year, women from the baseline survey who are alive and aged 15–49 contribute an observation. Pooled data contain 176,862 observations for 6,363 women. Average annual TFR estimated as TFR = 5 × 4 × Σi(ASFRi), where i indexes five-year age intervals

At the population level, a 0.7 increase in total fertility rate (TFR) for the four-year period after the tsunami constitutes a large effect. Similar magnitudes are reported by Finlay (2009), who identified an effect of roughly one-fifth of a child ever-born per woman, observed one year and four years after major earthquakes in Pakistan and Turkey, respectively. By comparison, the birth increase attributed to the Oklahoma City bombing was less than one birth per 1,000 women over the three years after the event (Rodgers et al. 2005).

To what extent can the change in fertility be attributed to a response to mortality? Couples who lost children may want to replace them. However, mortality outside the family may also influence reproductive choices. Theory suggests the strongest response to mortality is likely to occur within units of social salience for respondents. Because the impact of the tsunami was location-specific and because the community (desa or kelurahan) is highly salient in Indonesian society, we focus on responses to community-level mortality.

Methods

Empirical Issues in Linking Mortality and Fertility

Establishing a causal effect of mortality on fertility poses several challenges. Detailed data on the timing and locations of births and deaths of children are essential, but they may not be sufficient to disentangle associations that reflect micro- versus macro-level forces (Guha-Sapir and Below 2006; Montgomery and Cohen 1998; Palloni and Rafalimanana 1999; Sandberg 2006). Perhaps most problematic, deaths are typically not random events. The antecedents of mortality—economic conditions, health endowments, and the health-service environment—are also typically correlated with fertility (Olsen 1980). Moreover, reverse causality poses difficulties because birth spacing and parity affect infant mortality (e.g., Palloni and Tienda 1986). Without adjustments for these processes, associations between mortality and fertility are likely upwardly biased.

Some of these challenges can be addressed by examining fertility in the context of mortality shocks. The timing, location, and magnitude of some disasters are unexpected and can be located precisely in time and space. As a result, the behavior preceding and following the disaster is easier to distinguish and attribute to unexpected mortality. Some disasters occur idiosyncratically, which reduces the likelihood that child deaths are correlated with factors associated with prior fertility choices and outcomes. When the precipitating event is unexpected and short-lived, fertility changes are unlikely to reflect tempo effects of fertility delays.

Apart from Heuveline and Poch (2007) studying the Khmer Rouge, no other analysis of fertility in the wake of humanitarian crisis explicitly ascribes fertility changes to the mortality generated by these events. Several difficulties hamper doing so. In the context of disaster, facets of family and community life other than mortality change rapidly as well. Because these may also influence fertility, it is difficult to attribute a fertility response to the death of a child rather than some other contemporaneous process, such as loss of livelihoods, dependence on government aid, or reduced contraceptive access (Hapsari et al. 2009; Hill 2004). In addition, data on the same population before and after exposure to the disaster are rare.

Our study develops an empirical strategy that capitalizes on the longitudinal and multilevel nature of the STAR data. We analyze the fertility of individual women in STAR with information collected in both the pre-tsunami baseline and five annual post-tsunami survey rounds. Each post-tsunami survey provides detailed information about the mortality of household members since the pre-tsunami baseline. The most detailed information on fertility comes from complete pregnancy histories asked of reproductive-age women in the second follow-up round, with updates in each subsequent round. For pregnancies ending in live births, each child’s survival status is updated at the time of the interview. Information on date and age of death is available for each child who died, along with whether the death was related to the tsunami.

Recontact rates in STAR are high. Among reproductive-age women who survived the tsunami, 93 % were surveyed in the 2010–2011 follow-up (STARF), and the rates are similar for women regardless of whether their community was affected by tsunami mortality and in spite of the high level of devastation and dislocation caused by the tsunami.

Response to Death of Own Children

We begin with the question of how the death of a child affects the mother’s fertility. Detailed fertility histories collected in the second post-tsunami follow-up along with any updates collected in each of the three subsequent waves are used to create a dichotomous indicator, Bic, for whether a mother i who was living in community c at the time of the tsunami had at least one live birth between January 1, 2006, and December 31, 2009.5 To compare the post-tsunami fertility of mothers who lost a child with the fertility of mothers who did not, the model includes an indicator variable Mic, which takes the value 1 if the mother lost at least one child in the tsunami and a value of 0 otherwise.
Bic=γ0+γ1Mic+γ2Xic+vc.
(1)

In each model, the woman’s characteristics, Xic, include factors likely to predict both fertility and vulnerability to child mortality in the context of the disaster. These characteristics are measured before the disaster using the baseline pre-tsunami survey, and they include women’s parity, age (specified with a piecewise linear spline with knots at five-year intervals), and education (measured with dichotomous indicators capturing completion of junior high and completion of high school). To control for individual experiences in the disaster, we include an indicator of whether the respondent was swept up in the water or saw family or friends struggle or disappear in the water.

As in many societies, fertility in the region is patterned by socioeconomic status. For this reason, we also include household per capita expenditure and indicators of homeownership and landownership (all measured in 2004, before the disaster).

Means of the key variables related to fertility, child mortality, and exposure to the tsunami are provided in Panel A of Table 2. The first column aggregates women regardless of child mortality experience. Subsequent columns present the differences in these variables for women who lost a child relative to those who did not. Overall, among the 2,301 women who were mothers at the time of the tsunami, just over 5 % lost a child during the disaster, and just over 10 % of women were exposed to the tsunami waves. Exposure rates are much higher for women who lost a child in the tsunami than for those who did not. Overall, 27 % of women gave birth to another child after the tsunami. Although the rate is higher for those who lost a child, relative to those who did not, the difference is not statistically significant. With respect to other demographic and socioeconomic features (prior fertility, age, and education), women who lost a child are no different from women whose children survived. These results are consistent with the idea that apart from losing a child as a result of the disaster, our two groups of women are largely similar.

Table 2

Sample characteristics of mothers at the time of the tsunami and all women aged 15–49

A. Mothers at the Time of the TsunamiB. All Women Aged 15–49
Difference: Mother Experienced Child Death – Did NotDifference: Mortality Affected – Other Communities
MeanSEDiff.SEMeanSEDiff.SE
Tsunami-Related Deaths 
 % Mothers lost ≥1 child in tsunami 5.5 0.47       
  Number of children died | ≥1 died 1.4 0.06       
 % Community population died     4.2 0.19   
  % Died in mortality-affected communities     14.7 0.67   
Tsunami-Exposure: Saw Waves 10.4 0.64 43.0 5.8 11.0 0.50 22.5 3.1 
Post-Tsunami Fertility 
 % Women had birth post tsunami 27.1 0.93 7.1 5.1 31.2 0.74 4.5 2.3 
Characteristics of Women: Pre-Tsunami 
 Number of children born pre-tsunami 2.8 0.03 0.2 0.2 1.6 0.03 –0.1 0.1 
 % Who lost ≥1 child pre-tsunami 7.4 0.54 –3.1 1.8 6.7 0.49 –4.4 1.4 
Characteristics of Women at Time of Tsunami 
 Age 34.4 0.16 0.9 1.0 29.2 0.15 –0.6 0.4 
 Education: % completed 
  <High school 13.4 0.71 8.7 5.3 9.3 0.46 –3.0 2.0 
  Some high school 49.6 1.04 –0.1 6.3 53.0 0.80 –7.2 3.2 
  >High school 37.0 1.01 –8.6 6.3 37.7 0.77 10.2 4.0 
Sample Size 2,301    3,936    
A. Mothers at the Time of the TsunamiB. All Women Aged 15–49
Difference: Mother Experienced Child Death – Did NotDifference: Mortality Affected – Other Communities
MeanSEDiff.SEMeanSEDiff.SE
Tsunami-Related Deaths 
 % Mothers lost ≥1 child in tsunami 5.5 0.47       
  Number of children died | ≥1 died 1.4 0.06       
 % Community population died     4.2 0.19   
  % Died in mortality-affected communities     14.7 0.67   
Tsunami-Exposure: Saw Waves 10.4 0.64 43.0 5.8 11.0 0.50 22.5 3.1 
Post-Tsunami Fertility 
 % Women had birth post tsunami 27.1 0.93 7.1 5.1 31.2 0.74 4.5 2.3 
Characteristics of Women: Pre-Tsunami 
 Number of children born pre-tsunami 2.8 0.03 0.2 0.2 1.6 0.03 –0.1 0.1 
 % Who lost ≥1 child pre-tsunami 7.4 0.54 –3.1 1.8 6.7 0.49 –4.4 1.4 
Characteristics of Women at Time of Tsunami 
 Age 34.4 0.16 0.9 1.0 29.2 0.15 –0.6 0.4 
 Education: % completed 
  <High school 13.4 0.71 8.7 5.3 9.3 0.46 –3.0 2.0 
  Some high school 49.6 1.04 –0.1 6.3 53.0 0.80 –7.2 3.2 
  >High school 37.0 1.01 –8.6 6.3 37.7 0.77 10.2 4.0 
Sample Size 2,301    3,936    
Some community-level aspects of the tsunami, such as loss of natural resources and sources of livelihood, have the potential to affect subsequent fertility. If these factors are also correlated with whether women lost children, failure to include them will bias estimates of Mic. To address this concern, we estimate a model taking into account all community-level characteristics that are fixed during the post-tsunami study period, 2006–2009, and affect post-tsunami fertility in a linear and additive way:
Bic=γ0+γ1Mic+γ2Xic+μc+vic.
(2)

In this specification, μc is a community-level fixed effect that absorbs community-level tsunami impacts and pre-tsunami resources.

A concern that has received considerable attention in the literature is whether an observed fertility response after a child’s death arises because of a conscious (a volitional effect) or inadvertently via increased fecundability after breast-feeding stops (a physiological effect). To investigate this question, we consider two additional pieces of evidence.

First, we assess whether deaths of children who have aged out of the period when children are typically breast-fed predicts subsequent fertility. The median duration of any breast-feeding is about 20 months, but supplements are typically introduced within a few months of birth (CBS 2008). Children aged 2 and older are not breast-fed enough to prevent ovulation in their mothers. We extend Model 2 by replacing the indicator of child death (Mic) with two indicator variables that capture (1) the loss of a child who was under age 2 (<1 % of women) and (2) the loss only of children age 2 and older (4 % of women). A positive coefficient on the indicator of at least one death to a child under age 2 may arise either because of a physiological response to the cessation of breast-feeding or intentional behaviors to increase family size. By contrast, a positive coefficient on the indicator of death(s) of only older children is almost certainly not working through breast-feeding behavior.

Our second approach adds an interaction between the loss of a child and an indicator that a woman’s parity exceeded two children at the time of the disaster. Previous research predicts a more muted response to child death among higher-parity women who may have already exceeded their desired family size.

Finally, to explore variation by socioeconomic status, we interact loss of a child with whether the mother completed high school. In all the empirical models, standard errors take into account clustering at the community level as well as arbitrary forms of heteroskedasticity (Huber 1967).

Relationship Between Community-Level Mortality and Post-Tsunami Fertility

We turn next to the association between post-tsunami fertility and community-level mortality. An association between fertility and exposure to mortality at the community level is likely to mask substantial heterogeneity in responses across women. As noted earlier, the relationship may differ for mothers who experienced at least one child death relative to those mothers who did not. In addition, the fertility-mortality relationship likely varies with parity. Our model is specified as follows:
Bic=β0+β1McDic+β2McNic+β3Xic+μc+ζic.
(3)

As before, M indicates exposure to mortality; but here, Mc is a community mortality rate, measured as the percentage of STAR respondents killed by the tsunami. Dic is an indicator variable for women who experienced the death of at least one child in the tsunami, and Nic is an indicator for women who had not borne a child at the time of the tsunami. Estimates of and measure the differential effect of variation in community-level mortality on each of these two groups of women relative to the excluded group: women whose children survived the tsunami. Assuming that there are no unmeasured characteristics that are correlated with unobserved heterogeneity, ζic, estimates in Eq. (3) can be given a causal interpretation.

In the model, comparisons are drawn between women who were living in the same community at the time of the tsunami, and thus the women are likely to have much in common. However, their own experience of the tsunami differed, at least for those who lost a child. Our estimates attempt to take this into account, and we interpret as measuring the difference between the impact of a higher level of community mortality and a lower level of community mortality on the post-tsunami fertility of mothers who lost a child relative to those who did not. A parallel comparison is drawn for , which measures the impact of different exposures to community-level mortality between women who had not yet borne a child at the time of the tsunami and women who were already mothers but did not lose a child.

For these models, the sample of mothers is supplemented with women with no previous births at the time of the tsunami (total N = 3,936 women). Characteristics of the sample are summarized in Panel B of Table 2.6 Overall, about 4 % of the population was killed in the tsunami. The sample of all women is similar to the sample of mothers with respect to exposure to the tsunami and post-tsunami fertility. Parity at the tsunami and exposure to child mortality before the tsunami are a little lower, and the sample is younger and a little better educated, as one would expect with the addition of women who had not yet had children. With respect to the differences in women from mortality-affected versus mortality-unaffected communities, women in the mortality-affected communities were more likely to have been exposed to the tsunami and are more likely to have completed a high school education (for which we will control), but they are otherwise similar.

Results

The Effect of Own-Child Death on Maternal Fertility

Evidence of a behavioral fertility response to the death of a child is explored with Models 1 and 2. Table 3 displays linear probability estimates of a birth between 2006 and 2009 for women with at least one pre-tsunami birth. The first column presents ordinary least squares (OLS) results; column 2 provides estimates from a model that includes community fixed effects. Conceptually, the fixed-effects model comparisons are drawn between mothers who were living in the same community at the tsunami, one of whom lost a child and the other did not (controlling the characteristics Xic in the table).

Table 3

Effect of death of own child in tsunami on subsequent birth probabilities: Dependent variable is an indicator variable for birth post-tsunami

Model
Child DiedChild Died Interacted With
OLSFENo. of Children Who DiedChild’s AgeParityEducation
Covariates(1)(2)(3)(4)(5)(6)
Mother Lost >1 Child in Tsunami 0.050 0.099     
 [0.031] [0.036]     
Number of Children Lost in Tsunami   0.069    
   [0.023]    
(1) If Mother Lost ≥ 1 Child Interacted With Age of Children at Tsunami 
 All age >2    0.098   
    [0.045]   
 Some age <2    0.094   
    [0.231]   
Mother’s Parity at Tsunami 
 <3     0.114  
     [0.044]  
 ≥3     0.066  
     [0.056]  
Mother Completed High School? 
 Yes      0.086 
      [0.037] 
 No      0.153 
      [0.103] 
Mother’s Parity at Tsunami –0.001 –0.009 –0.010 –0.010 –0.014 –0.009 
 [0.005] [0.006] [0.006] [0.006] [0.009] [0.006] 
Mother’s Parity ≥3 at Tsunami     0.024  
     [0.028]  
Mother Exposed to Tsunami Waves 0.016 0.065 0.065 0.065 0.064 0.063 
 [0.024] [0.032] [0.032] [0.033] [0.032] [0.031] 
Mother Completed Junior High 0.030 0.037 0.037 0.035 0.038 0.037 
 [0.019] [0.024] [0.024] [0.025] [0.024] [0.024] 
Mother Completed High School 0.062 0.073 0.072 0.070 0.074 0.070 
 [0.025] [0.034] [0.034] [0.034] [0.034] [0.034] 
Sample Size 2,301 2,301 2,301 2,301 2,301 2,301 
R2 .21 .22 .22 .22 .22 .22 
Community Fixed Effects No Yes Yes Yes Yes Yes 
Number of Communities 276 276 276 276 276 276 
Model
Child DiedChild Died Interacted With
OLSFENo. of Children Who DiedChild’s AgeParityEducation
Covariates(1)(2)(3)(4)(5)(6)
Mother Lost >1 Child in Tsunami 0.050 0.099     
 [0.031] [0.036]     
Number of Children Lost in Tsunami   0.069    
   [0.023]    
(1) If Mother Lost ≥ 1 Child Interacted With Age of Children at Tsunami 
 All age >2    0.098   
    [0.045]   
 Some age <2    0.094   
    [0.231]   
Mother’s Parity at Tsunami 
 <3     0.114  
     [0.044]  
 ≥3     0.066  
     [0.056]  
Mother Completed High School? 
 Yes      0.086 
      [0.037] 
 No      0.153 
      [0.103] 
Mother’s Parity at Tsunami –0.001 –0.009 –0.010 –0.010 –0.014 –0.009 
 [0.005] [0.006] [0.006] [0.006] [0.009] [0.006] 
Mother’s Parity ≥3 at Tsunami     0.024  
     [0.028]  
Mother Exposed to Tsunami Waves 0.016 0.065 0.065 0.065 0.064 0.063 
 [0.024] [0.032] [0.032] [0.033] [0.032] [0.031] 
Mother Completed Junior High 0.030 0.037 0.037 0.035 0.038 0.037 
 [0.019] [0.024] [0.024] [0.025] [0.024] [0.024] 
Mother Completed High School 0.062 0.073 0.072 0.070 0.074 0.070 
 [0.025] [0.034] [0.034] [0.034] [0.034] [0.034] 
Sample Size 2,301 2,301 2,301 2,301 2,301 2,301 
R2 .21 .22 .22 .22 .22 .22 
Community Fixed Effects No Yes Yes Yes Yes Yes 
Number of Communities 276 276 276 276 276 276 

Notes: Linear probability models for all mothers at time of tsunami living in mortality-affected districts. Community fixed effects are measured for community of residence at time of pre-tsunami survey. All models include controls for age of mother splines, ln(per capita expenditure), indicator variables for household ownership of land, and ownership of house, all measured prior to the tsunami. Standard errors (shown in brackets) take into account clustering at the community level and are robust to heteroskedasticity. Difference between OLS and FE estimate of impact of child death = 0.049 (SE = 0.020).

As shown in column 1 of Table 3, mothers who lost at least one child are 5 percentage points more likely to have an additional birth in the post-tsunami period. The effect is not statistically significant. With inclusion of community fixed effects, mothers who lost at least one child are 10 percentage points more likely to have an additional birth in the post-tsunami period—an effect that is statistically significant. A Hausman test rejects the hypothesis that the OLS and fixed-effects models are equivalent (χ2 = 34.54, p value < .01). Given that 27 % of women in the sample had an additional birth, the 0.10 effect represents about a 37 % increase in the probability of an additional birth associated with death of one’s own child. If one measures own-child mortality as the number of children that died (column 3), the effect (0.07) is further evidence of a strong fertility response to child death.

Interpreting these results with respect to theory or women’s welfare depends partly on whether the pattern represents volitional behavior. Column 4 of Table 3 presents the results from a specification that includes indicators distinguishing death of a young child from deaths only of children age 2 and older. The coefficient for losing only older children, at a statistically significant 0.098, is nearly as large as the result for losing any children in column 2. For women who lost a child under age 2, the coefficient is 0.094 and is not statistically significant. Given these results, breast-feeding cessation cannot be the sole explanation for the pattern in column 2, suggesting that intentional replacement plays a role.

Comparisons are also drawn between lower parity (≤3) and higher parity (>3) mothers (column 5). Women at lower parities are 11 percentage points more likely to have an additional birth after losing a child than women who did not lose a child. The gap for higher parity mothers is just over one-half the size (6.6 percentage points) and is not statistically significant. All evidence, then, indicates volitional replacement behavior by mothers whose child(ren) died during the tsunami.

The last column in Table 3 considers whether the strength of the response to own- child death varies by women’s educational level—specifically, whether women completed high school. The coefficients on death of a child are 0.153 for women who did not graduate and 0.086 for women who did. The former is statistically significant; the difference between the estimates is not.

The Fertility Effects of Mortality in the Community

We turn now to the models for response to the level of tsunami mortality within the community. Because this question is relevant for women with and without children, we add childless women to the sample. Our models distinguish both women without children and mothers who lost a child, relative to the omitted group of mothers whose children all survived the tsunami.

Results from Model 3 using a linear probability specification are presented in Table 4. Estimates of the coefficient on mortality at the community level for mothers who lost at least one child are reported in the first row of the table. Estimates for women with no children before the tsunami are in the second row of the table. As before, standard errors take into account clustering of the sample and heteroskedasticity of arbitrary form; all models include the individual controls listed in the table.

Table 4

Effect of community mortality on probability of a post-tsunami birth

Fraction of Total Population in Community That Died, by Woman’s Characteristics at Time of Tsunami
ParityAgeCompleted High School
≤3>325–34NoYesFraction of Children That Died
Covariates(1)(2)(3)(4)(5)(6)(7)
Fraction Died in Tsunami Interacted With 
 A. Mother lost at least one child 0.460 0.578 0.056 0.279 0.204 1.077 0.315 
 [0.184] [0.231] [0.325] [0.372] [0.220] [0.311] [0.146] 
 B. Woman childless pre-tsunami 0.468 0.488 –– 1.238 0.400 0.565 0.341 
 [0.181] [0.190]  [0.421] [0.231] [0.345] [0.153] 
Mother Lost ≥1 Child in Tsunami 0.012 0.004 0.088 0.204 0.029 –0.024 0.026 
 [0.043] [0.053] [0.077] [0.106] [0.046] [0.110] [0.041] 
Woman Childless Pre-tsunami –0.150 –0.167 –– –0.224 –0.155 –0.153 –0.146 
 [0.025] [0.030]  [0.051] [0.032] [0.046] [0.025] 
Woman’s Parity at Tsunami –0.008 –0.022 0.013 –0.008 –0.010 –0.001 –0.008 
 [0.005] [0.011] [0.012] [0.016] [0.006] [0.013] [0.005] 
Woman Exposed to Tsunami Waves 0.028 0.026 –0.022 –0.013 0.023 0.049 0.028 
 [0.026] [0.029] [0.039] [0.062] [0.033] [0.043] [0.026] 
Woman Completed Junior High 0.029 0.021 0.071 0.118 0.034 –– 0.028 
 [0.023] [0.029] [0.044] [0.077] [0.024]  [0.023] 
Woman Completed High School 0.092 0.091 0.112 0.162 –– –– 0.090 
 [0.028] [0.034] [0.067] [0.080]   [0.028] 
Sample Size 3,936 3,347 589 1,264 2,675 1,261 3,931 
Number of Communities 283 283 226 268 272 233 280 
R2 .16 .13 .19 .07 .15 .17 .16 
F Test (p value) 
 Row A = Row B .970 .707  .020 .489 .193 .882 
 Rows A and B = 0 .008 .010 .863 .011 .191 .002 .025 
Fraction of Total Population in Community That Died, by Woman’s Characteristics at Time of Tsunami
ParityAgeCompleted High School
≤3>325–34NoYesFraction of Children That Died
Covariates(1)(2)(3)(4)(5)(6)(7)
Fraction Died in Tsunami Interacted With 
 A. Mother lost at least one child 0.460 0.578 0.056 0.279 0.204 1.077 0.315 
 [0.184] [0.231] [0.325] [0.372] [0.220] [0.311] [0.146] 
 B. Woman childless pre-tsunami 0.468 0.488 –– 1.238 0.400 0.565 0.341 
 [0.181] [0.190]  [0.421] [0.231] [0.345] [0.153] 
Mother Lost ≥1 Child in Tsunami 0.012 0.004 0.088 0.204 0.029 –0.024 0.026 
 [0.043] [0.053] [0.077] [0.106] [0.046] [0.110] [0.041] 
Woman Childless Pre-tsunami –0.150 –0.167 –– –0.224 –0.155 –0.153 –0.146 
 [0.025] [0.030]  [0.051] [0.032] [0.046] [0.025] 
Woman’s Parity at Tsunami –0.008 –0.022 0.013 –0.008 –0.010 –0.001 –0.008 
 [0.005] [0.011] [0.012] [0.016] [0.006] [0.013] [0.005] 
Woman Exposed to Tsunami Waves 0.028 0.026 –0.022 –0.013 0.023 0.049 0.028 
 [0.026] [0.029] [0.039] [0.062] [0.033] [0.043] [0.026] 
Woman Completed Junior High 0.029 0.021 0.071 0.118 0.034 –– 0.028 
 [0.023] [0.029] [0.044] [0.077] [0.024]  [0.023] 
Woman Completed High School 0.092 0.091 0.112 0.162 –– –– 0.090 
 [0.028] [0.034] [0.067] [0.080]   [0.028] 
Sample Size 3,936 3,347 589 1,264 2,675 1,261 3,931 
Number of Communities 283 283 226 268 272 233 280 
R2 .16 .13 .19 .07 .15 .17 .16 
F Test (p value) 
 Row A = Row B .970 .707  .020 .489 .193 .882 
 Rows A and B = 0 .008 .010 .863 .011 .191 .002 .025 

Notes: Linear probability models with community fixed effects for all women age 15–49 and living in mortality-affected districts at time of tsunami. See notes to Table 3.

Both for women who lost a child and for women without children at the time of tsunami, the effects of the level of community mortality on bearing a child after the tsunami are strong and positive, and they are identical in size. For each additional 1 percentage point increase in the population killed, mothers who lost at least one child are 0.460 percentage points more likely to have a birth after the tsunami than other mothers. For women who had not borne a child at the time of the tsunami, the estimate is 0.468 relative to mothers who did not lose a child.

The results are striking. In the absence of community mortality, women without children before the tsunami are significantly less likely to have had a child between 2006–2009 than are mothers whose children survived, and mothers who lost a child are no more likely to have reproduced than women whose children survived. However, in communities with higher levels of mortality, both groups of women have appreciably higher fertility.

Additional columns explore whether the community mortality effects are stronger for particular subgroups of women. Results are stratified by parity in columns 2 and 3 of Table 4. For mothers who lost at least one child, a strong fertility response is observed only among those with three or fewer children. For these women, the coefficient is 0.578, compared with 0.056 for women of parity 4 and higher. Column 4 presents estimates for women who were age 25–34 at the time of the tsunami. Women in this age range who were childless pre-tsunami have an especially strong fertility response to higher rates of mortality in the community. Each 1 percentage point increase in mortality is accompanied by an increase of 1.2 percentage points in the probability of a birth. Finally, by schooling attainment, women who completed high school have a stronger fertility response to community mortality than women with less education, particularly for women who lost a child. For these women, the coefficient is 1.077 for high school graduates, relative to 0.204 for those without a degree.

The final column of Table 4 tests the sensitivity of the central result to substituting the fraction of the community’s children that perished for the fraction of the total population that died. Results do not change substantially, which is not surprising given that the correlation in child and total death rates is over 0.9.

Discussion

Mortality shocks arising from natural disasters are not unusual. Worldwide, between 2000 and 2010, 1 in 28 people were affected by some form of mortality-inducing disaster arising from natural hazards (Guha-Sapir and Hoyois 2012). Studying fertility in the wake of these events sheds light on fundamental aspects of population rebuilding and provides a new lens through which to consider the relationship between mortality and fertility.

We document a large increase in fertility after the 2004 Indian Ocean tsunami that is concentrated in communities where mortality from the tsunami was high. In these areas, during the second half of the 2000–2009 decade, total fertility was one-half of a birth higher per woman than would have been expected in the absence of the tsunami.

We find evidence of a statistically significant increase in fertility among women who lost one or more children. The results are consistent with a volitional effort to increase fertility in response to the deaths of one’s children. However, the effect is estimated to be only about 0.10 additional births, which does not come close to replacing the children these mothers lost, let alone account for the large increase in the total fertility rate after the tsunami. What explains these additional births?

Over and above evidence of partial replacement after the death of one’s own child, we also find a powerful effect of mortality at the community level on subsequent fertility. These effects are largest among mothers who lost a child in the tsunami and among women who had not borne a child at the time of the tsunami; they are very similar in magnitude for both groups of women, amounting to almost 0.5 additional births for each additional 1 percentage point increase in mortality in the community. For women who lost a child, the estimate reflects both the impact of that death and the death of children in the community. For women who had not borne a child, the 0.5 additional births reflect the impact of community-level mortality. With additional waves of STAR, it will be possible to determine whether this pattern reflects accelerated timing of births or results in larger completed family sizes.

Based on the fixed-effects estimates in column 1 of Table 4, we estimate there were 9,500 more births between 2006 and 2009 in the study area than would have occurred in the absence of tsunami-related mortality. Of those births, 13 % occur to mothers who lost at least one child in the tsunami, and 87 % occur to women who had not borne a child prior to the tsunami.

The estimated effects are not likely to reflect a motivation of “hoarding” or “insurance” as traditionally theorized in demographic transition literature. A tsunami of this magnitude had not affected mainland Aceh for five centuries, and it is unlikely that most women expected an event of this magnitude to occur again.

Instead, the disaster may have increased broad pronatalist sentiment and a desire to “rebuild” communities. Women without children at the time of the disaster would be expected to respond more readily to this kind of sentiment, and the response might be magnified in closely knit communities that share strong kin ties. In that case, community members may have sped up family-building to replace the villages’ lost children.

Entrance into new unions is also part of the story. Marriage is nearly universal in Aceh, and marriage and childbearing are tightly connected. Births outside of marriage are nearly nonexistent. Because men were more likely to survive the disaster and more likely to repartner in its wake, some new unions with younger women were formed (Burrows et al. 2012). But it does not follow that that the new births are an accidental afterthought of the new unions. Instead, these new unions were almost certainly generated with the intent of childbearing. In this context, intentions related to union formation and childbearing cannot be cleanly separated.

Thus, the fertility increases observed after the disaster likely arose from multiple mechanisms operating simultaneously. For childless women, a desire to contribute to rebuilding families and communities through reproduction may have increased fertility. For mothers who suffered the death of children, the desire to have another child in the aftermath of a child loss raised fertility.

Our study provides a complementary lens on questions central to the study of population dynamics. Importantly, we do not directly address some of the hypotheses in the earlier demographic transition literature linking mortality and fertility decline. The fertility effects of mortality increase and mortality decrease are likely asymmetric for a number of reasons, including differences in how increases and decreases in mortality are perceived (e.g., Montgomery 2000).

Instead, the contributions of this study include the use of several design elements to ensure that the measured fertility changes arise at least partially in response to the loss of children and to the loss of community members. Using data from a population-representative sample established before the disaster and following those respondents (including movers) over time mitigates legitimate concerns about sample-selection bias. In addition to taking into account individual and community attributes measured before the disaster that likely capture preexisting vulnerability to mortality (and may be correlated with fertility behavior), we estimate models that include community fixed effects and draw comparisons between women living in the same community. We thereby reduce the chance that our estimates are contaminated by unobserved heterogeneity. Adjusting for exposure and including community fixed effects reduces the likelihood that we are measuring a fertility response to other changes occurring in the context of the disaster—and instead increases the likelihood that we are measuring a response to the loss of a child and exposure to mortality in communities.

The disaster effects observed here (Fig. 1) are quite large in comparison with events such as Hurricane Hugo, the September 11 attacks, and the Oklahoma City bombing. Instead, they more closely align with the events such as the high-mortality earthquakes in Iran (Finlay 2009; Hosseini-Chavoshi and Abbasi-Shavazi 2013). One clear difference between events in the United States and earthquakes elsewhere is the magnitude of the accompanying mortality increase. This explanation is consistent with the evidence presented here that the mortality shock itself had a sizable influence on fertility in damaged communities. By contrast, the aggregate effect on fertility is substantially smaller than observed after the mortality increases accompanying the reign of the Khmer Rouge in Cambodia (Heuveline and Poch 2007), or after World War II in Western Europe. Among other things, war-related crises tend to last longer; it may be that periods of conflict often generate delayed fertility, whereas natural hazards support a much smaller tempo effect.

Evidence from the Cambodian population suggests that a tempo effect may have been operating in tandem with some of the other contextual mechanisms that we have identified as potentially relevant for the post-tsunami Acehnese. As in Aceh, part of the fertility boom in Cambodia is attributable to women who were young at the time of the Cambodian genocide (Heuveline and Poch 2007).

A further distinguishing factor of the 2004 tsunami for the Indonesian population is the differential mortality for adult women relative to men (Frankenberg et al. 2011). If the fertility response to death emerges in part through family-building—a phenomenon noted as early as Eversley’s (1957) study of eighteenth century England—the effects observed here will be more relevant to mortality shocks that, through direct or indirect means, influence entrance into new unions.

Looking ahead, it will be important to understand the long-term impact of the tsunami on Indonesian families. A five-year follow-up post-disaster is unusually lengthy. It is possible to assess the extent to which families of older women differ markedly as they age out of their reproductive years than they did prior to the disaster. For younger women, we cannot yet assess whether the increase in fertility observed from 2006 to 2009 translates into larger completed family sizes or simply reflects a shift in fertility timing. Answers to this question will emerge as additional data are collected.

Acknowledgments

The authors are grateful for comments from Patrick Heuveline, David Lam, Alberto Palloni, and Jim Walker. This work is supported by the National Institute for Child Health and Human Development (R01HD052762, R01HD051970, R03HD071131), the National Institute on Aging (R01AG031266), the National Science Foundation (CMS-0527763), the Hewlett Foundation, the World Bank, and the MacArthur Foundation (05-85158-000).

Notes

1

As an example of a shift in perceptions occurring in the context of mortality increase, Trinitapoli and Yeatman (2011) observed an association between uncertainty about survival driven by the AIDS epidemic in Malawi and the desired timing of fertility initiation among adolescents.

2

In these cases, deaths are thought to have arisen as a result of residents having been elsewhere during the tsunami.

3

For each quarter-year, each woman who is alive and aged 15–49 contributes an observation; the pooled data include 176,862 observations for 6,363 women. The sample includes all age-eligible women who were interviewed in the pre-tsunami baseline survey and therefore represent the pre-tsunami population.

4

Standard errors are estimated by bootstrapping the sample using 10,000 replications (Efron and Tibshirani 1994).

5

Although the data are designed to support analysis of fertility timing using, for example, an event-history approach, the main goal of this research is to measure the extent to which there has been successful population rebuilding five years after the tsunami, which is provided by an estimate of γ1 in Model 1.

6

The number of communities is smaller than shown in Table 1 because the analysis is conditioned on women of reproductive age surviving the disaster and being in interviewed in follow-up surveys.

References

Agadjanian, V., & Prata, N. (
2002
).
War, peace, and fertility in Angola
.
Demography
,
39
,
215
231
. 10.1353/dem.2002.0013
Ashton, B., Hill, K., Piazza, A., & Zeitz, R. (
1984
).
Famine in China, 1958–61
.
Population and Development Review
,
10
,
613
645
. 10.2307/1973284
Atella, V., & Rosati, F. C. (
2000
).
Uncertainty about children’s survival and fertility: A test using Indian microdata
.
Journal of Population Economics
,
13
,
263
278
. 10.1007/s001480050137
Athukorala, P-C, & Resosudarmo, B. P. (
2005
).
The Indian Ocean tsunami: Economic impact, disaster management, and lessons
.
Asian Economic Papers
,
4
,
1
39
.
Blanc, A. (
2004
).
The role of conflict in the rapid fertility decline in Eritrea and prospects for the future
.
Studies in Family Planning
,
35
,
236
245
. 10.1111/j.0039-3665.2004.00028.x
Borneman, J. (
2002
).
Reconciliation after ethnic cleansing: Listening, retribution, affiliation
.
Public Culture
,
14
,
281
304
. 10.1215/08992363-14-2-281
Burrows, M., Frankenberg, E., Katz, P., Sikoki, B., & Thomas, D. (
2012
,
May
).
Family formation in Indonesia after the 2004 Indian Ocean tsunami
. Paper presented at the Annual Meeting of the Population Association of America, San Francisco, CA.
Cain, M. (
1981
).
Risk and insurance: Perspectives on fertility and agrarian change in India and Bangladesh
.
Population and Development Review
,
9
,
688
702
. 10.2307/1973546
Caldwell, J. C. (
2004
).
Social upheaval and fertility decline
.
Journal of Family History
,
29
,
382
406
. 10.1177/0363199004267744
Carta, G., D’Alfonso, A., Colagrande, I., Catana, P., Casacchia, M., & Patacchiola, F. (
2012
).
Post-earthquake birth-rate evaluation using the brief cope
.
Journal of Maternal-Fetal and Neonatal Medicine
,
25
,
2411
2414
. 10.3109/14767058.2012.697945
Central Bureau of Statistics (CBS) (Indonesia), State Ministry of Population/National Family Planning Coordinating Board (NFPCB), & Ministry of Health (MOH), & Macro International Inc. (MI). (
2008
).
Indonesia demographic and health survey 2007
.
Calverton, MD
:
CBS and MI
.
Cohan, C. L. (
2010
).
Family transitions following natural and terrorist disaster: Hurricane Hugo and the September 11 terrorist attack
. In T. W. Miller (Ed.),
Handbook of stressful transitions across the lifespan
(pp.
149
164
).
New York, NY
:
Springer
.
Cohan, C. L., & Cole, S. W. (
2002
).
Life course transitions and natural disaster: Marriage, birth, and divorce following Hurricane Hugo
.
Journal of Family Psychology
,
16
,
14
25
. 10.1037/0893-3200.16.1.14
Conning, J., & Udry, C. (
2007
).
Rural financial markets in developing countries
. In R. Evenson & P. Pingali (Eds.),
Handbook of agricultural economics
(Vol.
3
, pp.
2857
2908
).
Amsterdam, The Netherlands
:
North-Holland
.
Doocy, S., Gorokhovich, Y., Burnham, G., Balk, D., & Robinson, C. (
2007
).
Tsunami mortality estimates and vulnerability mapping in Aceh, Indonesia
.
American Journal of Public Health
,
97
(
Suppl. 1
),
S146
S151
. 10.2105/AJPH.2006.095240
Efron, B., & Tibshirani, R. J. (
1994
).
An introduction to the bootstrap
.
Boca Raton, FL
:
Taylor and Hall
.
Eversley, D. E. C. (
1957
).
A survey of population in an area of Worcestershire from 1660–1850 on the basis of parish records
.
Population Studies
,
10
,
253
279
. 10.1080/00324728.1957.10413223
Finlay, J. E. (
2009
).
Fertility response to natural disasters: The case of three high mortality earthquakes
(Policy Research Working Paper No. 4338).
Washington, DC
:
World Bank
.
Frankenberg, E. (
1998
).
The relationship between infant and child mortality and subsequent fertility in Indonesia
. In M. R. Montgomery, & B. Cohen (Eds.),
From death to birth: Mortality decline and reproductive change
(pp.
316
338
).
Washington, DC
:
National Academies Press
.
Frankenberg, E., Gillespie, T., Preston, S., Sikoki, B., & Thomas, D. (
2011
).
Mortality, the family, and the Indian Ocean tsunami
.
The Economic Journal
,
121
,
F162
F182
. 10.1111/j.1468-0297.2011.02446.x
Fritsche, I., Jonas, E., Fischer, P., Koranyi, N., Berger, N., & Fleischmann, B. (
2007
).
Mortality salience and the desire for offspring
.
Journal of Experimental Social Psychology
,
43
,
753
762
. 10.1016/j.jesp.2006.10.003
Geertz, C. (
1968
).
Agricultural involution: The processes of ecological change in Indonesia
.
Berkeley, CA
:
University of California Press
.
Gillespie, T., Chu, J., Frankenberg, E., & Thomas, D. (
2007
).
Assessment and prediction of natural hazards from satellite imagery
.
Progress in Physical Geography
,
31
,
459
470
. 10.1177/0309133307083296
Gray, C., Frankenberg, E., Gillespie, T., Sumantri, C., & Thomas, D. (
2014
).
Studying displacement after a disaster using large scale survey methods: Sumatra after the 2004 Tsunami
.
Annals of the Association of American Geographers
,
104
,
594
612
. 10.1080/00045608.2014.892351
Grimard, F. (
1997
).
Household consumption smoothing through ethnic ties: Evidence from Côte d’Ivoire
.
Journal of Development Economics
,
53
,
391
422
. 10.1016/S0304-3878(97)00016-3
Grummer-Strawn, L. M., Stupp, P. W., & Mei, Z. (
1998
).
Effect of a child’s death on birth spacing: A cross-national analysis
. In M. R. Montgomery, & B. Cohen (Eds.),
From death to birth: Mortality decline and reproductive change
(pp.
39
73
).
Washington, DC
:
National Academies Press
.
Guha-Sapir, D., & Below, R. (
2006
).
Collecting data on disaster: Easier said than done
(CRED Research Paper No. 221).
Brussels, Belgium
:
Centre for Research on the Epidemiology of Disasters
.
Guha-Sapir, D., & Hoyois, P. (
2012
).
Measuring the human and economic impact of disasters
.
London, UK
:
Center for Research on the Epidemiology of Disasters, Government Office for Science
.
Hamoudi, A., Frankenberg, E., Sumantri, C., & Thomas, D. (
2014
).
Impact of the December 2004 earthquake and tsunami on birth outcomes in Aceh, Indonesia
(Working paper).
Durham, NC
:
Department of Economics, Duke University
.
Hapsari, E. D., Widyawati, , Nisman, W. A., Lusmilasari, L., Siswishanto, R., & Matsuo, H. (
2009
).
Change in contraceptive methods following the Yogyakarta earthquake and its association with the prevalence of unplanned pregnancy
.
Contraception
,
79
,
316
322
. 10.1016/j.contraception.2008.10.015
Heuveline, P. (
1998
).
“Between one and three million”: Toward demographic reconstruction of a decade of Cambodian history (1970–79)
.
Population Studies
,
52
,
49
65
. 10.1080/0032472031000150176
Heuveline, P., & Poch, B. (
2007
).
The Phoenix population: Demographic crisis and rebound in Cambodia
.
Demography
,
44
,
405
426
. 10.1353/dem.2007.0012
Hill, K. (
2004
).
War, humanitarian crises, population displacement, and fertility: A review of evidence
.
Washington, DC
:
National Resource Council
.
Hossain, M. B., Phillips, J. F., & LeGrand, T. K. (
2007
).
The impact of childhood mortality on fertility in six rural thanas of Bangladesh
.
Demography
,
44
,
771
784
. 10.1353/dem.2007.0047
Hosseini-Chavoshi, M., & Abbasi-Shavazi, M. J. (
2013
).
Demographic consequences of the 2003 Bam earthquake
.
Canberra, Australia
:
International Conference on the Demography of Disasters, Australian National University
.
Huber, P. (
1967
).
The behavior of maximum likelihood estimates under nonstandard conditions
.
Proceedings of the Firth Berkeley Symposium on Mathematical Statistics and Probability
,
1
,
221
233
.
Jansen, S., & Helms, E. (
2009
).
The “white plague”: National-demographic rhetoric and its gendered resonance after the post-Yugoslav wars
. In C. Eifler, & R. Seifert (Eds.),
Gender dynamics and post-conflict reconstruction
(pp.
219
244
).
Frankfurt, Germany
:
Peter Lang
.
Kuate Defo, B. (
1998
).
Fertility response to infant and child mortality in Africa with special reference to Cameroon
. In M. R. Montgomery, & B. Cohen (Eds.),
From death to birth: Mortality decline and reproductive change
(pp.
254
315
).
Washington, DC
:
National Academies Press
.
Lee, R. (
1997
).
Population dynamics: Equilibrium, disequilibrium, and consequences of fluctuations
. In M. R. Rosenzweig & O. Stark (Eds.),
Handbook of Population Economics
(Vol. 1B, Cap 19, pp.
1063
1115
).
Amsterdam
,
The Netherlands: North Holland
.
LeGrand, T., Koppenhaver, T., Mondain, N., & Randall, S. (
2003
).
Reassessing the insurance effect: A qualitative analysis of fertility behavior in Senegal and Zimbabwe
.
Population and Development Review
,
29
,
375
403
. 10.1111/j.1728-4457.2003.00375.x
Lindstrom, D. P., & Berhanu, B. (
1999
).
The impact of war, famine, and economic decline of marital fertility in Ethiopia
.
Demography
,
36
,
247
261
. 10.2307/2648112
Lindstrom, D. P., & Kiros, G-E (
2007
).
The impact of infant and child death on subsequent fertility in Ethiopia
.
Population Research and Policy Review
,
26
,
31
49
. 10.1007/s11113-006-9018-1
Marshall, B. K., & Picou, J. S. (
2008
).
Postnormal science, precautionary principle, and worst cases: The challenge of twenty-first century catastrophes
.
Sociological Inquiry
,
78
,
230
247
. 10.1111/j.1475-682X.2008.00236.x
Martin, J. A., Hamilton, B. E., Sutton, P. D., Ventura, S. J., Menacker, F., & Monson, M. L. (
2003
).
Births: Final data for
2002 (National Vital Statistics Reports 52(10)).
Hyattsville, MD
:
National Center for Health Statistics
.
McAdoo, B. G., Richardson, N., & Borrero, J. (
2007
).
Inundation distances and run-up measurements from ASTER, QuickBird and SRTM Data, Aceh Coast, Indonesia
.
International Journal of Remote Sensing
,
28
,
2961
2975
. 10.1080/01431160601091795
Montgomery, M. R. (
2000
).
Perceiving mortality decline
.
Population and Development Review
,
26
,
795
819
.
M. R. Montgomery, & B. Cohen (
1998
).
From death to birth: Mortality decline and reproductive change
.
Washington, DC
:
National Academies Press
.
Morin, R. (
2002
, March 31). A 9/11 baby boom?
The Washington Post
, p.
B5
.
Nakamura, K., Sheps, S., & Arck, P. C. (
2008
).
Stress and reproductive failure: Past notions, present insights, and future directions
.
Journal of Assisted Reproduction and Genetics
,
25
(
2–3
),
47
62
. 10.1007/s10815-008-9206-5
Nakonezny, P. A., Reddick, R., & Rodgers, J. L. (
2004
).
Did divorces decline after the Oklahoma City bombing?
.
Journal of Marriage and Family
,
66
,
90
100
. 10.1111/j.1741-3737.2004.00007.x
Neria, Y., Nandi, A., & Galea, S. (
2008
).
Post-traumatic stress disorder following disasters: A systematic review
.
Psychological Medicine
,
38
,
467
480
. 10.1017/S0033291707001353
Norris, F. H., Friedman, M. J., Watson, P. J., Byrne, C. M., Diaz, E., & Kaniasty, K. (
2002
).
60,000 disaster victims speak: Part I. An empirical review of the empirical literature, 1981–2001
.
Psychiatry
,
65
,
207
239
.
Olsen, R. J. (
1980
).
Estimating the effect of child mortality on the number of births
.
Demography
,
17
,
429
443
. 10.2307/2061155
Palloni, A., & Rafalimanana, H. (
1999
).
The effects of infant mortality on fertility revisited: New evidence from Latin America
.
Demography
,
36
,
41
58
. 10.2307/2648133
Palloni, A., & Tienda, M. (
1986
).
The effects of breastfeeding and pace of fertility on mortality at early ages
.
Demography
,
23
,
31
52
. 10.2307/2061406
Parker, V. J., & Douglas, A. J. (
2010
).
Stress in early pregnancy: Maternal neuro-endocrine-immune responses and effects
.
Journal of Reproductive Immunology
,
85
,
86
92
. 10.1016/j.jri.2009.10.011
Portner, C. C. (
2007
).
Gone with the wind? Hurricane risk, fertility and education
. Unpublished manuscript,
Department of Economics, University of Washington
,
Seattle, WA
.
Preston, S. H. (
1978
).
The effects of infant and child mortality on fertility
.
New York, NY
:
Academic Press
.
Qin, L., Luo, S., Li, X., Wang, Y., & Li, S. (
2009
).
Fertility assistance program following the Sichuan earthquake in China
.
International Journal of Gynecology & Obstetrics
,
104
,
182
183
. 10.1016/j.ijgo.2008.12.002
Ramakrishnan, D., Ghosh, S., Raja, V., Chandran, R., & Jeyram, A. (
2005
).
Trails of the killer tsunami: A preliminary assessment using satellite remote sensing technique
.
Current Science
,
88
,
709
711
.
Rindfuss, R. R., & Sweet, J. A. (
1978
).
The pervasiveness of postwar fertility trends in the United States
. In K. E. Taeuber, L. L. Bumpass, & J. A. Sweet (Eds.),
Social demography
(pp.
15
42
).
New York, NY
:
Academic Press
.
Rodgers, J. L., St. John, C. A., & Coleman, R. (
2005
).
Did fertility go up after the Oklahoma City bombing? An analysis of births in metropolitan counties in Oklahoma
.
Demography
,
42
,
675
692
. 10.1353/dem.2005.0034
Rosenzweig, M. R., & Schultz, T. P. (
1983
).
Consumer demand and household production: The relationship between fertility and child mortality
.
American Economic Review
,
73
,
38
42
.
Rosero-Bixby, L. (
1998
).
Child mortality and the fertility transition: Aggregated and multilevel evidence from Costa Rica
. In M. R. Montgomery, & B. Cohen (Eds.),
From death to birth: Mortality decline and reproductive change
(pp.
384
410
).
Washington, DC
:
National Academy Press
.
Sandberg, J. (
2006
).
Infant mortality, social networks, and subsequent fertility
.
American Sociological Review
,
71
,
288
309
. 10.1177/000312240607100206
Scelfo, J. (
2002
).
A 9-11 baby boomlet
.
Newsweek
,
139
(
22
),
42
.
Segraves, R. T. (
1998
).
Psychiatric illness and sexual function
.
International Journal of Impotence Research
,
10
,
S131
S133
.
Stein, Z., & Susser, M. (
1975
).
Fertility, fecundity, famine: Food rations in the Dutch Famine 1944/5 have a causal relation to fertility, and probably to fecundity
.
Human Biology
,
47
,
131
154
.
Suri, T. (
2003
).
Spillovers in village consumption: Testing the extent of partial insurance
. Unpublished manuscript,
Department of Economics, Yale University
,
New Haven, CT
.
Townsend, R. M. (
1994
).
Risk and insurance in village India
.
Econometrica
,
62
,
539
591
. 10.2307/2951659
Trinitapoli, J., & Yeatman, S. (
2011
).
Uncertainty and fertility in a generalized AIDS epidemic
.
American Sociological Review
,
76
,
935
954
. 10.1177/0003122411427672
Umitsu, M. (
2011
).
The geoenvironment and the giant tsunami disaster in the Northern part of Sumatra Island, Indonesia
. In P. P. Karan, & S. P. Subbiah (Eds.),
The Indian Ocean tsunami: The global response to a natural disaster
(pp.
51
64
).
Lexington, KY
:
University Press of Kentucky
.
Umitsu, M., Tanavud, C., & Patanakanog, B. (
2007
).
Effects of landforms on tsunami flow in the plains of Banda Aceh, Indonesia, and Nam Khem, Thailand
.
Marine Geology
,
242
,
141
153
. 10.1016/j.margeo.2006.10.030
Vail, K. E., Juhl, J., Arndt, J., Vess, M., Routledge, C., & Rutjens, B. T. (
2012
).
When death is good for life: Considering the positive trajectories of terror management
.
Personality and Social Psychology Review
,
16
,
303
329
. 10.1177/1088868312440046
Vos, F., Rodriguez, J., Below, R., & Guha-Supir, D. (
2010
).
Annual disaster statistical review 2009: The numbers and trends
.
Brussels, Belgium
:
Centre for Research on the Epidemiology of Disasters (CRED)
.
Watkins, S., & Menken, J. (
1985
).
Famines in historical perspective
.
Population and Development Review
,
11
,
647
675
. 10.2307/1973458