Abstract

Determining whether power outages have significant fertility effects is an important policy question in developing countries, where blackouts are common and modern forms of family planning are scarce. Using birth records from Zanzibar, this study shows that a month-long blackout in 2008 caused a significant increase in the number of births 8 to 10 months later. The increase was similar across villages that had electricity, regardless of the level of electrification; villages with no electricity connections saw no changes in birth numbers. The large fertility increase in communities with very low levels of electricity suggests that the outage affected the fertility of households not connected to the grid through some spillover effect. Whether the baby boom is likely to translate to a permanent increase in the population remains unclear, but this article highlights an important hidden consequence of power instability in developing countries. It also suggests that electricity imposes significant externality effects on rural populations that have little exposure to it.

Introduction

Electrification brings many benefits to rural communities in developing countries, including increasing employment rates, allowing industrialization, and introducing new technologies for home and market production (Dinkelman 2011; Rud 2012). However, electricity in developing countries is often rationed, unstable, and erratic. According to the World Bank, a typical firm in Africa can expect nine outages per month, each lasting an average of seven hours, causing losses of 7 % in sales (World Bank 2013). By contrast, in OECD countries, a typical firm can expect less than one outage per month, lasting an average of three hours when one occurs. Power outages also cause disruptions outside of work: for those who use it at home, the sudden lack of electric power reduces recreational activities, makes chores harder to complete, and makes studying harder for students. When communities go dark, social activities are also affected: public meeting places cannot be illuminated, public televisions remain turned off, and people’s perceptions of security are reduced.

This study provides evidence that the unintended consequences of such events include increases in fertility. The evidence comes from a month-long power outage affecting the entire island of Zanzibar, Tanzania, between May and June of 2008. To document the implications of this event, I use a panel database of community-level births constructed from maternity ward birth records covering a period of two and half years. The identification relies on the timing of birth, with children born 8 to 10 months after the blackout likely to have been conceived during the event. Because not all areas of Zanzibar have electricity, a second source of identification is the location of the mother’s residence, with mothers in villages with no electricity being unlikely to be affected by the blackout. Using a difference-in-difference strategy, I show that the affected communities in the sample experienced an increase in maternity ward births 8 to 10 months after the event. Additionally, I show that the increase in births happened only in communities that had some electricity: villages without electricity did not experience a baby boom.

Having established a fertility response to the power outage, I show that the estimated impact was remarkably high across all communities connected to the grid, regardless of their level of electrification. In particular, villages with low levels of electrification (below 10 %) experienced increases in births that were as significant as those in communities with higher electricity levels. This finding is important because in communities where most people lack access to electricity, many births occur in households that are not connected to the grid. This indicates that the blackout affects households through an externality effect that operates through the community. For instance, the blackout affected televisions and lighting in communal spaces where most people congregate in the evenings to spend leisure time.

To provide further insights on how the blackout affected fertility, I use data from a time-use survey conducted five months after the event to show how time-use patterns were changed. The blackout caused an average increase in the amount of time spent inside the home for both men and women. A simple descriptive regression of time use indicates that the preexisting amount of electricity in the community was correlated with time-use changes, with men spending more time at home the higher the electricity coverage and women increasing domestic time at all electrification levels. This result should be interpreted with caution: while leisure time spent at home is a possible channel through which the blackout affected fertility, the patterns of births across communities are consistent with many alternative pathways.

The idea that procreation increases when the lights go out is a widely held belief that has rarely been empirically examined. It gained traction in the United States after a one-day blackout hit the Northeast in 1965 and several hospitals reported more births nine months later (Udry 1970). The same questions arose following the New York City blackout of 1975 and the Northeast blackout of 2003 (Pollack 2004). This study is the first to rigorously link blackouts to increases in births, and (to my knowledge) it is the first since Udry (1970) to study the link anywhere in the world. The issue is particularly important in the context of sub-Saharan Africa, where fertility rates are high and blackouts are common and often thought by local policymakers to lead to more births.1 In addition, this study contributes to an understanding of the importance of electricity as a public good in areas with almost-zero levels of electrification. In those areas, which are often rural, the spread of electricity takes time: after electric poles are erected and connected to the grid, dwellings and businesses must make significant investments to connect to the electric wires; electrification rates might show only marginal improvements over time. Although electricity as a private good may remain underutilized, electricity has an immediate effect as a public good: meeting points such as stores or mosques are illuminated, electric water pumps are installed at public wells, and public televisions are turned on. The strong fertility response to the blackout shows that this “public good effect” can be large and important.

The study has two important caveats. First, the blackout studied here is unusual in its length. This helps identification, but more work is necessary to explore whether the fertility effects described in this article are larger than what would be expected during shorter blackouts. At least for sub-Saharan Africa, no hard data describe how common lengthy power failures in rural areas are. It is indeed possible that they do occur with some frequency and are underreported. For instance, although the Zanzibar blackout was well publicized in the international media (such as the BBC), in 2008, local Tanzanian newspapers reported a three-week-long blackout in the Mtwara region that went otherwise unreported. Zanzibar itself fell into a new and even more serious blackout that lasted three months in 2010 (O’Connor 2010).

A second caveat is that it is unfortunately not possible to establish whether power instability increases births temporarily (through a “harvesting effect” in which planned pregnancies are anticipated) or causes an increase in the total population. To the extent that mothers do not adjust their subsequent fertility plan, at least some of the baby boom may translate to permanent increases in the population.2 Even if long-term effects among the population are absent, short-run increases in births could strain health and neonatal services and lead to adverse perinatal outcomes by reducing birth spacing among some mothers. If blackouts increase births, even an isolated and short-lived event—such as the July 2012 two-day blackout in India that affected 600 million people—could have significant population implications simply because of the sheer number of individuals involved (BBC World Service 2012).

Aside from the blackout literature, the study contributes to an understanding of fertility responses to aggregate shocks. Pörtner (2008) studied the long-run effects of hurricanes on fertility in Guatemala and found that hurricane risk increased fertility but that hurricane events diminished it. Evans et al. (2010) also studied the effect of hurricanes on fertility, finding that the response varied depending on the severity of the hurricane advisory. They hypothesized that hurricanes affect the opportunity cost of procreation, possibly through the allocation of time at home.3 Unlike the previous research, this study contains household-level data on time use and provides additional insights on the possible reasons for a fertility increase, including the effect of changes in time spent at home. In addition, because the study shows the effect of infrastructure shocks, it also contributes to the larger literature on the social impacts of infrastructure (e.g., Dinkelman 2011; Duflo and Pande 2007; Olken 2009; Rud 2012).4 This study does not, however, directly address an important question in the infrastructure and fertility literature: the causal relationship between electrification (as opposed to blackouts) on fertility. The reason is that electrification creates important general equilibrium effects to wages, employment, structure of labor and capital markets, relative bargaining power of women, and so on. These other variables are important determinants of fertility, and they are absent in the case of blackouts.

Background Information

Leisure and Electricity in Zanzibar

Like many other places in the developing world, electricity coverage in Zanzibar is quite uneven, with some areas having high coverage and other areas having little or no coverage. In part, this is driven by the slow process of electrification in rural areas. The Rural Electrification Project (RUREL) began in 1984, with 64 villages electrified between 1984 and 1991. A second phase, completed by 2006, added 77 villages, both in Zanzibar and the secondary island of Pemba. Once a community has access to power, private households need to be able to pay for a hookup to their dwelling, which can be difficult or costly for those dwelling farther from the electric lines. Thus, the process of village electrification is gradual and often slow.

Figure 1 provides a scatterplot of the estimated electricity coverage in 2002 and 2007 for communities surveyed both in the 2002 census and the 2007 Labor Force Survey (described later). Between these two dates, most locations improved their electricity coverage, with several rural villages receiving electricity for the first time. The figure also overlays a frequency-weighted local polynomial smoothed regression. Clearly, most villages either maintained or improved their coverage of electricity, with similar improvements in electrification (approximately 14 %) for villages at all levels of 2002 coverage.

Electricity plays an important but complicated role in Zanzibari society. Aside from its impact on households that are able to add a television set, an electric lightbulb, or some other appliance, electrification dramatically changes the set of amenities available in public spaces. Electric lights are often installed on meeting places (baraza) where adults meet after dusk; outdoor televisions appear in those public places and are switched on for the screening of evening news, soccer matches, or soap operas.5 Thus, electrification has substantial externality effects on the quality of leisure time available, even among those who do not have electricity at home. As the proportion of citizens with a private provision of electricity increases and people acquire private televisions, some of the outdoor activities move back to the private sphere.6

The 2008 Blackout

The Zanzibar blackout started on May 21, 2008, at approximately 10 p.m. and lasted until June 18; it was caused by an accidental break in the undersea cable that connects the Zanzibar island substation to the electricity generators on mainland Tanzania (for more details on the blackout, see Online Resource 1). The event was Zanzibar’s longest recorded time without power to that point.

The blackout caused economic damage: those employed in occupations using electricity reported a steep drop in earnings and hours worked, and birth weights of children conceived before or around the blackout were significantly lower compared with those of other children (Burlando 2014).7 The power outage only marginally affected other aspects of daily life: it had no impact on work and earnings of households engaged in activities not relying on electricity, little to no effect on prices of consumer goods, did not cause significant public health problems, and (given that significant use of electric cookers, fridges, air conditioners, and other domestic electric appliances is rare in this population) had little impact on cooking patterns. Finally, the evidence seems to indicate that the blackout did not cause significant out-migration from the troubled island: evidence from the post-event household survey indicate that only 3.1 % of respondents left the island during the blackout—a smaller proportion than those that left the island the previous month.8

Use of petrol-run generators was limited before the blackout (as electricity provision had been quite stable previously) and remained so throughout the period. The price of generators shot up 2- to 10-fold due to restricted supply, and remained high throughout. Moreover, running costs were also very high—reportedly in the order of 35–40 US dollars a day (BBC World Service 2008). A household survey collected five months after the blackout suggests that 7.2 % of workers reported using generators during the blackout.

Data

I estimate community-level fertility rates through birth records from the Mnazi Mmoja maternity ward in Zanzibar, which delivers approximately 25 % of all births in the island. Records from the maternity ward were collected in July 2009 and cover the period between January 2007 (one year and four months prior to the blackout) and May 2009 (11 months after the end of the event).9 The records include the date of birth (but not the date of conception) of the child, and the name, home shehia (i.e., town), number of prior pregnancies, age, and admission date of all expectant mothers. Only children born in the facility were included in the database; a large fraction of (especially rural) children are born at home, and a smaller minority are born in one of the other six public and private maternity wards in the island.

Using the name of the home shehia, I matched birth records to the 2002 census data and the 2007 Zanzibar Labor Force Survey. The 2002 census administered a short-form questionnaire that collected demographic information for every household of Zanzibar. One of every four enumeration areas (EAs) was also selected to receive an expanded (or long-form) questionnaire component that collected household-level information on basic asset ownership and dwelling characteristics on all EA residents, including whether the residence had a source of electricity. Thus, the census collected asset ownership information for approximately one-quarter of all households. The Census Bureau made the long-form microdata available to me, but not the short-form data. I used these microdata to derive the percentage of the village connected to the electric grid, the average village wealth (as measured by a household-level index of asset ownership), and the average size of the household (which was also reported in the data file). Because the long-form survey is representative only at the district level and not at the village level, the constructed average is only an approximation of the true village average.10 Although the information covered by the census is limited and somewhat outdated by the time of the blackout, it has the benefit of covering almost all villages in Zanzibar.

The 2007 Labor Force Survey (LFS) is a more recent, complete description of household and worker characteristics. Because it was implemented on few enumeration areas, however, it covers fewer and less representative communities. I derived average household characteristics by village and then matched these characteristics to each community. In total, 76 LFS communities were successfully matched.

Table 1 provides some summary statistics from the two matched data sets. Villages sampled from the census are fairly representative: 26 % of villages did not have any electricity (close to the census average), were slightly ahead in terms of asset ownership (with an index value of 0.39 relative to the census average of 0.17), and had a similar rate of citizens connected to the electricity grid (29 % relative to the census average of 25 % to 38.9 % among villages with some level of electricity). On the other hand, the characteristics of the village matched to the LFS are systematically different, with these villages being larger (as deduced from the higher birth rates) and generally much higher levels of wealth (as measured in 2002). Thus, although this sample provides a more recent set of village covariates, it is also less representative of the average community.

The final data were reformatted so that each observation is a village-week, and the main outcome variable of interest is the number of births in a week in a village. Dropping incomplete birth records and villages with four or fewer positive birth weeks leaves 14,500 usable observations, with 70 villages matched to the LFS and 125 matched to the census.

Note that many shehias are represented in the data through a limited number of births. Figure 2 shows how often shehias have at least one birth per week and reveals that shehias with relatively few birth events are more common than villages with many birth events. Such shehias are likely to be somewhat remote or to have a small population to begin with, whereas shehias with a significant presence in the facility are more likely to be urban and located near the maternity ward. To the extent that the relative frequency of maternity ward attendance is time-invariant, the econometric methods I present in the next section should be able to account for this.

Econometric Strategy

The first objective of the study is to estimate the impact of the blackout on the number of births of children conceived during the blackout. The strategy for accomplishing this objective is to compare the size of cohorts conceived in electrified shehias between May 21 and June 18, 2008, with that of other cohorts after factors such as seasonality and population growth have been taken into account. Denoting the log of the number of births reported by a town or village v during week t by yvt, the baseline regression is
formula
(1)
where BBvt is a dummy variable for whether the births occurred among the “blackout baby” cohort of children (a) conceived during the blackout and (b) conceived in villages with electricity; measures the percentage change in births per week; and and are village fixed effects and month and year fixed effects, respectively. Because Zanzibar is a small island of 640 square miles located close to the equator, weather patterns do not differ greatly across shehias, and the location-invariant fixed effects () should capture seasonality quite well. Provided that the fixed effects correctly capture any time invariant heterogeneity across shehias, that fertility rates co-move similarly across villages over time, and that across-shehia spillover effects from the blackout were small, the coefficient correctly captures the difference-in-difference effect of the blackout on the fertility of couples exposed to the blackout. Because different villages may be growing at different rates, I also provide estimates using a more rigorous set of village-specific time trends in addition to the seasonality controls.
To more formally test that the blackout was the sole cause of the fertility increase, I next consider the following:
formula
(2)
where NEvt identifies those cohorts that were conceived during the blackout in villages with no electricity. Here, the coefficient indicates the degree to which shehias that were not electrified experienced an increase in births. To the extent that the blackout was not felt in these areas and to the extent that all seasonality has been captured by time fixed effects, this coefficient should be zero.11 A positive coefficient on NEvt, on the other hand, would suggest that the identification strategy is suspect either because time-varying controls do not correctly capture all seasonality or because there are cross-village spillover effects.

Performing the preceding analysis is impeded by two limits to the data. First, it would be ideal to determine whether a village had electricity in 2008, when the blackout happened. Lacking this information, I rely on data from the 2002 census and (as a further check) on the more limited 2007 LFS data. Second, I do not observe the date of conception but only the date of birth. One solution is to replace the date of conception with the expected date of conception, and thus consider any child born 40 weeks after any date in which the blackout was ongoing as “exposed” to the blackout. This measure might underreport the actual fertility effect if the blackout also affected the rate of premature or delayed births. An alternative measure, which is adopted here, attributes any birth occurring 8 to 10 months after the blackout to being the result of a conception during the blackout. This measure, which is described in detail in Online Resource 2, captures premature and delayed births, and (to the extent that it captures births that were not affected by the blackout) it provides a lower estimate of the true effect.

Properly regressing and interpreting Eqs. (1) and (2) requires that several econometric issues be considered. First, the underlying outcome of interest—number of births per week per village—is measured in nonnegative integers and has a large number of zero values. An improvement over the standard linear regression would involve estimating the model using a Poisson regression, and this is the method adopted here. Having assumed a lognormal model, the estimates are interpreted as percentage changes. Second, errors are likely to be nonstandard. To correct for possible autocorrelation in the errors, all Poisson regressions report bootstrapped standard errors. Third, records cover only births occurring at the specific facility and therefore exclude a large fraction of home and other hospital deliveries.12 If the number of hospital deliveries is seasonal and varies across villages, the time fixed effects and village-specific time trends will account for these. However, to the extent that the fertility response to the blackout is not independent of the choice of delivery location, the results I present should be interpreted as the blackout effect on the subpopulation that is likely to deliver at a hospital. If fertility outcomes for this subpopulation are more responsive to the blackout, the regressions overestimate the overall fertility effect for the population at large. It is therefore cautious to interpret the magnitude of the coefficients as an upper bound of the overall fertility effect.13

Results: Blackouts and Births

Average Fertility Effects

Table 2 reports the difference-in-difference estimates of the impact of the blackout on the size of the blackout baby cohort. Columns 1–3 report from Eq. (1) for the sample of villages matched to the census using an increasingly complete set of control variables. The difference-in-difference coefficient for the baseline specification is 18.4 %, falling to 15.7 % when quadratic time trends are included in column 2. In column 3, shehia-specific time trends are added to control for differences in population growth rates across communities. The coefficient does not change significantly, remaining close to 0.17 and indicating that the exposed cohort of children born 8–10 months later was 17 % larger than expected (95 % confidence interval = (8.7 %, 25.1 %)). These estimates seem large, but three considerations are in order. First, as discussed previously, 17 % could be considered an upper bound of the average treatment effect. A lower bound could be constructed by assuming that the blackout affected only the fertility of those who attend health facilities. In that case, the average fertility effect on the total population would be 0.17 × 0.61 = 0.104, or 10.4 % (where 0.61 is the proportion of pregnant Zanzibaris who give birth in hospitals). Second, the estimated population increase is not numerically large. A 17 % increase corresponds to an average of 0.25 additional maternity ward births per week across villages with electricity, which translates to 0.25 × 11 weeks = 2.75 more children per electrified shehia born at the facility. In other words, the estimated size of the baby boom in the health facility is 2.75 × 92 = 253 births (where 92 is the total number of electrified shehias in the sample).14

Column 4 separately identifies affected cohort sizes for those born in villages with and without electricity coverage in 2002. The coefficient β2 should be zero if the regression controls correctly capture seasonality and if spillovers from electrified villages to nonelectrified villages are small or nonexistent. The coefficient β2 is estimated at a small and statistically insignificant 0.022, meaning that villages with no reported electricity in 2002 did not experience an increase in fertility. The coefficient β1 is a smaller but significant 0.15.

Using more recent electrification data, columns 5 through 8 repeat the exercise for the sample of villages that were matched to the 2007 LFS. The coefficient in column 5 indicates a similar increase of 15.9 % in cohort size, which again corresponds to approximately 0.28 more births per week. This estimate falls to 12.7 when I include quadratic time trends (column 6) and shehia time trends (column 7). Finally, column 8 again splits the cohort born 8–10 months later into those born in villages with and without electricity; as in column 4, there is no evidence that fertility increased in places with no electricity, increasing confidence that the regression is indeed capturing the effect of the blackout and not some other extemporaneous or cyclical event. However, the difference-in-difference coefficient loses statistical significance (p value = .20).

Table 3 replicates the results by disaggregating the affected cohort dummy variable BBvt into three dummy variables covering four-week periods (early, middle, and late) and running Eqs. (1) and (2) on these dummy variables. Most of the gain in births is concentrated among those born 9 or 10 months later, although some coefficients lack precision. This concentration of births 9–10 months post-blackout suggests that the blackout’s unusual length was an important factor driving the overall fertility increase.

Robustness Tests and Other Outcomes

An important concern of the preceding analysis is that the difference-in-difference regression is capturing some other unobserved factor, such as differential growth in birth numbers across shehias with and without electricity, that is not controlled for by the period fixed effects or time trends. Figure 3 plots the coefficient estimates from Eq. (2), where the exposed cohorts are those conceived before the blackout.15 All the coefficients are close to zero, indicating that in any 12-week period before the birth of the cohort under analysis, the growth rates in births across shehias with and without electricity were similar. To the extent that the short time span available is representative, this validates the “parallel assumption” of the difference-in-difference specification.16

Perhaps the most important unresolved issue is whether the baby boom is likely to translate to a permanent increase in population. As a partial check, I run a falsification test on the difference-in-difference model by letting the exposed cohort be the one born 11 or 12 months after the blackout. The estimated coefficient, shown in column 1 of Table 4, is close to zero; that is, the number of births returned to the predicted level shortly after the baby boom. The short span of data available after the affected cohort thus provides inconclusive evidence. Indeed, a definitive assessment would require analyzing fertility some years into the future.17

Another way to make some headway on this issue is to explore the age and fertility structure of the affected cohort of women. Under the assumption that total fertility increases if either birth spacing falls or women become pregnant earlier, the concern that the increase in population is permanent is heightened if the composition of women giving birth 8 to 10 months later is younger. Table 4 thus explores the compositional differences across cohorts.

Column 2 reports the effect of the blackout on the number of births from first-time mothers, and shows that births among this group increased 22 %. It also means that women with a first pregnancy were overrepresented among the baby boom cohort. This could potentially indicate that women become pregnant earlier. To further study this issue, I regress the average age of all women from village v giving birth in week t on whether the birth belonged to the affected cohort (column 3) and do the same for first-time mothers (column 4). On average, affected mothers are marginally younger, but this result is small and statistically insignificant. In addition, first-time mothers are not younger and are, in fact, marginally older on average. Column 5 considers directly the number of births from teenage mothers. As with first pregnancies, it is quite clear that teenagers were particularly affected by the blackout, with births increasing 22.6 % for this group. It is quite possible that many of these teenage pregnancies might not have happened in the absence of the blackout; however, it is hard to speculate whether these young women will end up having more children as a result of this early pregnancy.

An alternative way to address the concern of a permanent fertility increase is to check birth increases at the other end of the age distribution, among women over the age of 40. Among these women, unplanned births are more likely to represent an unplanned increase in total fertility. Column 6 displays no evidence that the number of births increased for this group. The evidence is thus inconclusive.

Mechanisms: Time Use and Externalities

Why did the blackout affect pregnancy rates? In general, blackouts are transitory and have no impact on future employment, wages, life expectancy, or other long-term household or child characteristics. If fertility is fully determined by these characteristics, power failures would affect births through a “harvesting effect,” in which planned future births are brought forward in time without changing lifetime fertility rates. Alternatively, blackouts could also increase unplanned pregnancies by increasing the rate of unprotected sex in the population. This would require that blackouts reduce the opportunity cost of procreation—perhaps by increasing the amount of time available for sex or by decreasing the overall quality of time devoted to alternative activities.

In practice, the effect of blackouts on the opportunity cost of procreation (through its effect on leisure time) is ambiguous and is likely to depend on whether a household is a direct consumer of electricity. For workers temporarily displaced by the blackout, a decrease in work hours would be met by an increase of time spent at home, other things being equal. For those with domestic electricity, the reduction in domestic amenities—for instance, the sudden lack of television programming—might lead them to more boredom (favoring procreation) but also a substitution of time away from home to alternative activities elsewhere (discouraging procreation). In addition to these direct effects, every member of a community is potentially subject to an indirect effect. For instance, the lack of private or public sources of lighting and the darkening of television screens would discourage public gatherings of friends or social clubs. In addition, to the extent that the lack of outdoor lighting in the evenings heightens fears of thieves and home intrusions, a natural response would be to remain inside the home to protect valuable assets, such as electric appliances, from theft or to avoid interactions with the rest of the community at a time of distress.18

In the remainder of the article, I provide empirical evidence that both time-use effects and externality effects play an important role in the transmission mechanism from blackout to fertility. First, I show that time use changed during the blackout to favor procreation. I then use the results from the time-use analysis to develop a simple empirical model that will be used to test the presence of externalities.

Time Use During the Blackout

I use data from a time-use survey documenting the changes in time use during the blackout on 664 individuals living in parts of Zanzibar with varying degrees of electrification (the data are described in Online Resource 3). I observe time use for each respondent in two periods: in the month before the blackout and during the blackout. Figure 4 shows the average percentage change in time use for five broad categories of time use: work, leisure outside the home, leisure at home, housework, and sleep. On average, work hours (which is often an outside-of-home activity) and time spent outside of the home declined. Domestic leisure, on the other hand, increased by an average of 4 % for men and more than 9 % for women. Sleep and housework patterns did not change much for either gender.

I next show how changes in leisure time are related to own electricity use and to the rate of electrification in the community. Let the amount of leisure time spent at home (net of sleeping or housework) for person i in community or village v be ltic. The effect of exposure to the blackout on the change in log leisure hours during the blackout period for person i, , is
formula
(3)

The model here measures whether individuals connected to the grid (those with domestic electricity, DE) or whose jobs depend on electricity (WE) responded differently to the blackout relative to others. The variable VE measures the percentage of the village with an electricity connection. Absent other unobservable characteristics, the coefficient α3 should capture the net spillover effect of the electrification rate on time use. In practice, however, this coefficient might also capture the effect of other confounding variables that are correlated with the electrification rate.19 The matrix Xiv includes other possible work/leisure shifters, including ownership of domestic leisure substitutes (such as radios, which are generally battery-powered), participation in social and religious activities (participation in savings groups, in other community groups, and regular fasting), and other household and individual characteristics that could be correlated with the shape of the utility function with respect to leisure (e.g., age, education, family size, and wealth).

Columns 1 and 4 in Table 5 estimate the coefficients α1 and α2 for men and women separately. Those connected to the electric grid reported a larger increase in their time spent at home, with the effect being statistically significant for men only. Perhaps surprisingly, when I control for domestic electricity use, there is no additional effect of working with electricity for men or women. Columns 2 and 5 introduce the percentage of the village that had electricity at the time of the 2002 census. The coefficient on overall village electricity is substantially correlated with leisure time for men but not for women. Columns 3 and 6 omit observations from villages without electricity. The amount of variance explained by the regression increases in the restricted sample, but the estimated coefficients are unchanged from the previous two specifications.

In summary, the blackout reduced time spent outside the home and increased time spent in the home for both men and women. The patterns of time-use changes reveal that a possible pathway between power outages and fertility is the increase in domestic leisure. If fertility is driven by a couple’s time spent together, then we might expect that the fertility effect is larger in areas with high rates of electricity, given that the increase in time spent at home is significantly correlated with the electrification rate for men. However, other factors (e.g., quality of time use, boredom, preferences over fertility, presence of community externalities) could also play a role, which is particularly important if the externality effects are large and important. I introduce these other elements in the conceptual framework that follows.

Detecting Externality Effects: Conceptual Framework

Suppose that the short-term probability of a pregnancy that results in a child C from a couple i living in village or community v (with a total Nv couples) is a function of quantity and quality (i.e., boredom) of time, and let that function be the following reduced-form equation:
formula
(4)

It is thus assumed that the procreation function c(t, q) is a function of the number of hours a couple spends at home, t, and the overall quality of their time, q.

This probability is also a function of many other observed and unobserved characteristics of the couple, such as their preferences regarding children and family planning, as well as other community-level characteristics (such as social norms, the extent of alternative leisure activities, and availability of contraceptives). I summarize these observed and unobserved variables with the terms and , and for simplicity, I assume that they do not influence the shape of the functions t and q.20

The blackout affects the quantity and quality of time in two ways. First, it affects the household through direct exposure to electricity, ei ∈ {0, 1}. This exposure originates in households whose dwellings or whose jobs depend on electricity (e = 1), and it is absent for those who do not have electricity at home or work (e = 0). Second, the blackout affects all households through the spillover effect from all those with electricity to the rest of the community, . These spillover effects include the shutting down of power in common areas, the darkening of streets, the heightened fear of theft, or simply the reduction in economic activity in the community. A blackout is a shock that, in the simplest terms, eliminates electricity—that is, Ev = ei = 0. With a slight abuse of notation, the effect of this shock on fertility is estimated by differentiating the C function with respect to Ev:
formula
(5)

The first term in Eq. (5), , is the effect of time spent at home on fertility. This term is not directly observed; presumably, it is either positive or (if time at home does not affect frequency of unprotected sex) zero. Inside the parentheses is the effect of the direct and indirect electricity shock on time use—a term that, as discussed, is positive on average. The last term in Eq. (5) indicates the effect of quality of time on procreation, multiplied by the effect of the blackout on this quality of time. Although quality of leisure time cannot be empirically observed, it is reasonable to assume that the blackout worsened it. In that case, this (unobserved) term would indicate that the fertility effect through the quality channel is positive if boredom aids procreation and negative if boredom hinders it.

To bring this model to the data, the community-level birth function can be differentiated: . After applying Eq. (5) and manipulating the equation, I get that
formula
(6)

That is, the short-term increase in the fertility rate is a function of direct effects (summarized by ) and indirect spillover effects (summarized by ). I expect  > 0 because (from the time-use data) and or are arguably nonnegative (and likely strictly positive). The indirect effect is possibly nonlinear, and it is equal to zero if there are no spillover effects from electrification.

Equation (6) generates one testable prediction: as Ev → 0, the predicted increase in fertility is given by . That is, a (local) externality effect can be measured by looking at the fertility increase in villages with very low rates of electricity. As Ev increases (through work or domestic use of electricity), the direct effect becomes more important, whereas the indirect effect has an arguably reduced impact on procreation because the externality effect may be less important.

To test the implications of Eq. (6), consider the following empirical model:
formula
(7)
where the continuous variable VEv (village electricity) indicates the estimated fraction of residences that use electricity. identifies the effect of the blackout on electrified villages that have an electrification rate that is close to zero. That is, identifies the parameter from Eq. (6), the externality effect of the blackout when VEv → 0. estimates the additional impact of electricity coverage on fertility; that is, it identifies the linear average effect of the direct and indirect blackout effects, . The prediction for this parameter is ambiguous. While I expect α1 > 0, it is quite possible that : as private use of electricity increases, spillovers become less and less important. In addition, the shape of the function C might be affected by unobserved characteristics and , which might systematically vary by the degree of electrification in a community. For instance, residents of areas with higher electricity coverage might have better access to family planning or have a lower demand for children. For these reasons, could be zero or even negative.

Evidence and Discussion

Table 6 explores the correlation between the rate of electrification and the population increase (Eq. (7)). I start by looking at the census sample in column 1. The first coefficient describes the effect of the blackout on villages with electrification levels close to zero and is the measure of the blackout’s externality effect. This coefficient is large and statistically significant, at 0.21. This externality effect thus seems large and important. The estimated indicates that the level of electrification has no statistically significant effect on the size of the fertility increase; if anything, the coefficient suggests that increasing electrification somewhat reduces the increase in births. Overall, this suggests that what matters for the fertility response is the presence of any electricity, rather than the amount of electricity present.

The remaining regressions limit the sample size to the 70 villages surveyed by the LFS. Column 2 reports the results for Model 7 using the more recent 2007 village electrification data. The coefficient closely follows those found in the difference-in-difference model reported in Table 2 but is now insignificant. Given that very few villages in the sample have electrification levels of between close to zero, the insignificance is probably driven by lack of power rather than lack of effect. In addition, as with the census sample, the fertility effect for the LFS sample does not vary significantly with the amount of electricity in a given community; the fertility effect is similar across electrification rates.

In columns 3 and 4, I check for heterogeneity in the estimated effects through mechanisms other than the rate of electricity coverage. I replace electricity coverage in the regression with two alternative variables: the fraction of the population that reported owning a television and the fraction of the population working in sectors that use electricity.21 These variables should more directly capture the effect on quality of leisure (through having an unusable television at home) and the loss of work. These alternative regressors are not predictive; none of the coefficients are significant. The coefficients on the interaction with television and on the interaction with work are positive but statistically insignificant.

Taken together, the evidence from Table 6 suggests that the pattern of births in villages with electricity differed during the blackout but that electricity coverage did not matter. Figure 5 provides a visual and nonlinear confirmation of this. The figure plots the estimated coefficient on the exposed blackout baby cohort from Eq. (1) when the sample is restricted to villages below a certain rate of electrification. As one moves left to right, the estimated coefficient includes villages with higher levels of electricity. Panel A includes births from villages with no electricity in 2002. The coefficient is centered on zero when electrified villages are excluded, with a wide standard error. As the level of electrification increases, the coefficient increases sharply and then quickly stabilizes at approximately 0.20. Clearly, small levels of electrification have a large impact on the estimated coefficients. An alternative view of the data is given in panel B, which excludes all villages that were not electrified. Estimated coefficients are large with large standard errors (because of small sample sizes) at low levels of electrification and then decline slightly at higher levels of electrification. Table 7 estimates cohort sizes for villages with different levels of electrification in a single regression. Across the various specifications reported, the highest coefficients are found among shehias with little electricity use and are generally lower among those with high coverage. Nonetheless, these differences are statistically insignificant, and each coefficient is statistically indistinguishable from all others.22 Thus, the fertility increase was indeed strikingly high across the (electrified) portions of the island.

In summary, the most important result from Tables 6 and 7 is that births increased significantly in areas where private connections to the electric grid were few and where most exposure was (presumably) through public electricity use. The estimated indirect (externality) effect of the blackout is between 0.20 and 0.25.

The second result from Tables 6 and 7—an absence of a fertility gradient across electrification areas—is consistent with many different and mutually exclusive mechanisms, which unfortunately cannot be separately identified with the data at hand. For instance, the lack of gradient may be a consequence of fertility not responding to the quantity and quality of time spent at home. More plausible is that fertility increases with time spent at home but that the larger expected increase in births in areas with more electricity is counterbalanced by an opposing mechanism, such as a decline in the importance of the public externality effect () or the presence of other unobservable characteristics of the community (such as access to modern forms of contraception or a lower desire for children).

Finally, the coefficients in Tables 6 and 7 and in Fig. 5 may suffer from two biases. The first bias is the result of changes in electrification rates. Many villages that help estimate at low levels of electricity are likely to have had a higher electrification rate by 2008. A significant positive effect of electrification rate on fertility would lead to an overestimate of . However, the coefficient on electrification rate is negative or null, suggesting that this source of bias is likely small. The second possible source of bias originates from the greater likelihood among women in shehias with low electrification rates of delivering their children at home. In that case, the coefficients and could be downwardly biased, perhaps reflecting an underestimate of the externality effect.

Conclusion

This article provides evidence that a blackout can indeed produce a baby boom. Using a particularly well-defined and lengthy power outage on the island of Zanzibar, Tanzania, this study shows that blackout babies born 8–10 months later were more numerous than expected. The study of the blackout provides important and policy-relevant insights about the distribution of fertility effects, with large increases in the number of births found in villages with electricity, regardless of the degree of electrification, and no change in birth numbers in areas not served by the electricity network. Most important, the study shows that births at a health facility increased significantly in villages with few private connections to the grid. These villages are characterized by a public use of electricity (e.g., through public televisions and illumination in front of public spaces). The birth increases in these areas are indicative of important externalities of electricity on fertility in areas with little private use of electric power.

Acknowledgments

I thank three anonymous referees; Dilip Mookherjee, Shankha Chakraborty, Todd Pugatch, Liz Schroeder, and Anh Tran; and seminar participants at Oregon State University, University of Colorado Denver, the Northwest Development Workshop, and the Northeastern Universities Development Conference for helpful comments. In addition, Hajj Mohamed Hajj, Mayasa Mwinyi, Amour Bakari, and the staff at the Zanzibar Ministry of Health and the Zanzibar Office of the Chief Government Statistician all provided excellent fieldwork support.

Notes

1

In 2009, for instance, then Uganda Planning Minister Ephraim Kamuntu commented that the frequency of electricity shortages was causing too many births (BBC World Service 2009).

2

In sub-Saharan Africa, lifetime fertility is influenced by delays in first pregnancy and birth spacing (United Nations 2011); anticipating a birth without adjusting birth spacing could thus lead to one more child in a woman’s reproductive lifetime.

3

Lindstrom and Berhanu (1999), Pörtner (2008), and Rodgers et al. (2005) find that the events they study had a significant long-run effect on fertility. In contrast, Evans et al. (2010) found little or no evidence of a long-term effect on fertility.

4

Within the development literature on blackouts, see Adenikinju (2003) for its effects on firm-level outcomes in Nigeria.

5

See Winther (2008) for a rich and very enjoyable anthropological study of the impact of electricity in rural communities in Zanzibar.

6

An extensive literature has explored the impact of televisions on fertility. Jensen and Oster (2009) and La Ferrara et al. (2012) provided evidence that television programming reduces fertility. These studies suggest that television programming provides information about outside social norms, including smaller family sizes.

7

Burlando (2014) also discussed some aggregate fertility effects, but did not study the impact of village electricity on fertility.

8

This statistic excludes those who permanently left Zanzibar because they could not answer the questionnaire. It is unlikely that this group was large, and I could not find quantitative or qualitative evidence that significant out-migration occurred.

9

Records from preceding years were missing, and the data for June 2009 were not yet ready at the time of collection.

10

There is no reason to expect that the lack of representativeness is correlated with the number of births during the blackout in a way that would bias the analysis. Lacking access to the short survey, it is not possible to construct a measure of village population size, or even determine the proportion of village residents that answered the long-form questionnaire.

11

Even non-electrified rural villages could have felt the effects of the blackout if, for instance, it disrupted the work pattern of residents. This channel is likely to be minor in the study communities: the blackout disrupted jobs that depended on electricity directly, and few rural residents hold these types of jobs.

12

I estimate that approximately 25 % of total births occur at Mnazi Mmoja. According to facilities data from the Ministry of Health, the ward delivers 500–900 children per month, representing 48 % of all children born in health facilities (National Bureau of Statistics (NBS) 2011). It is estimated that 61 % of all children in Zanzibar are born at a health facility (NBS 2011).

13

A final possible source of bias is blackout-induced out-migration from Zanzibar. This bias is not likely important given that I could find no quantitative or qualitative evidence for higher than normal migration (Burlando 2014).

14

Assuming that all blackout babies were born at Mnazi Mmoja, 253 is also the lowest bound of the estimated total increase in births. Assuming that only 25 % of blackout births per village were at Mnazi Mmoja (which might be considered an upper bound), the total population increase is 1,012, or approximately 0.084 % of the 1.2 million Zanzibar population.

15

More precisely, I replicate column 4 of Table 2 in which the exposed cohort is not the “blackout baby” cohort but rather the cohort of children born in a 12-week period starting with the sixth week of 2007. Thus, the first coefficient identifies the difference-in-difference estimate on children born between 6 and 18 weeks from the start of 2007, the second identifies those born between 19 and 30 weeks, and so on.

16

As an additional check, I examined the seasonality of births from the 2011–2012 Tanzanian DHS and found no evidence that the “control months” of March 2007 and 2008 were lower than average for the entire country. Tables are available upon request.

17

Although the result in column 1 provides no evidence for or against a harvesting effect, it should alleviate the concern that the baby boom was induced by a permanent and possibly exogenous shift in fertility.

18

A heightened fear of theft was widely reported in qualitative data from conversations with respondents in Zanzibar. In addition to their monetary value, domestic electric appliances confer social status to a family in rural Zanzibar (Winther 2008); as valuable assets, they are owned by the husband and are often received as wedding gifts.

19

For example, suppose that the amount of electricity is correlated with social capital in the community, with more-electrified communities having less social capital. These communities may be less likely to gather together, with more people deciding to stay at home during the crisis.

20

Alternatively, I could write tiv and qiv.

21

These sectors are defined as employing managers, professionals, technicians, clerks, and plant and machine operators.

22

A table with data restricted to the LFS provides similar results and is available from the author upon request.

References

Adenikinju, A. F. (
2003
).
Electric infrastructure failures in Nigeria: A survey-based analysis of the costs and adjustment responses
.
Energy Policy
,
31
,
1519
1530
. 10.1016/S0301-4215(02)00208-2
BBC World Service
. (
2008
, May 30). Melting in Zanzibar’s blackout.
BBC News
. Retrieved from http://news.bbc.co.uk/2/hi/africa/7427957.stm
BBC World Service
. (
2009
, March 11). Uganda blackouts “fuel baby boom.”
BBC News.
Retrieved from http://news.bbc.co.uk/2/hi/africa/7939534.stm
BBC World Service
. (
2012
, August 1). Power restored after huge Indian power cut.
BBC News.
Retrieved from http://www.bbc.com/news/world-asia-india-19071383
Burlando, A. (
2014
).
Transitory shocks and birth weights: Evidence from a blackout in Zanzibar
.
Journal of Development Economics
,
108
,
154
168
. 10.1016/j.jdeveco.2014.01.012
Dinkelman, T. (
2011
).
The effects of rural electrification on employment: New evidence from South Africa
.
American Economic Review
,
101
,
3078
3108
. 10.1257/aer.101.7.3078
Duflo, E., & Pande, R. (
2007
).
Dams
.
Quarterly Journal of Economics
,
122
,
601
646
. 10.1162/qjec.122.2.601
Evans, R., Hu, Y., & Zhao, Z. (
2010
).
The fertility effect of catastrophe: U.S. hurricane births
.
Journal of Population Economics
,
23
,
1
36
. 10.1007/s00148-008-0219-2
Jensen, R., & Oster, E. (
2009
).
The power of TV: Cable television and women’s status in India
.
Quarterly Journal of Economics
,
124
,
1057
1094
. 10.1162/qjec.2009.124.3.1057
La Ferrara, E., Chong, A., & Duryea, S. (
2012
).
Soap operas and fertility: Evidence from Brazil
.
American Economic Journal: Applied Economics
,
4
(
4
),
1
31
.
Lindstrom, D. P., & Berhanu, B. (
1999
).
The impact of war, famine, and economic decline on marital fertility in Ethiopia
.
Demography
,
36
,
247
261
. 10.2307/2648112
National Bureau of Statistics (NBS) [Tanzania] and ICF Macro. (
2010
).
Tanzania Demographic and Health Survey 2004–2010
.
Dar es Salaam, Tanzania
:
National Bureau of Statistics and ICF Macro
.
Olken, B. (
2009
).
Do television and radio destroy social capital? Evidence from Indonesian villages
.
American Economic Journal: Applied Economics
,
1
(
4
),
1
33
.
Pollack, M. (2004, May 23). Blackout babies.
New York Times
. Retrieved from www.nytimes.com/2004/05/23/nyregion/fyi-684430.html
Pörtner, C. (
2008
).
Gone with the wind? Hurricane risk, fertility and education
. Unpublished manuscript, Department of Economics, University of Washington, Seattle.
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, 1990–1999
.
Demography
,
42
,
675
692
. 10.1353/dem.2005.0034
Rud, J. P. (
2012
).
Electricity provision and industrial development: Evidence from India
.
Journal of Development Economics
,
97
,
352
367
. 10.1016/j.jdeveco.2011.06.010
Udry, J. R. (
1970
).
The effect of the Great Blackout of 1965 on births in New York City
.
Demography
,
7
,
325
327
. 10.2307/2060151
United Nations, Department of Economic and Social Affairs, Population Division. (
2011
).
World fertility report 2009
.
New York, NY
:
United Nations
.
Winther, T. (
2008
).
The impact of electricity: Development, desires, and dilemmas
.
New York, NY
:
Berghahn Books
.
World Bank
(
2013
).
Enterprise Surveys
. Retrieved from http://www.enterprisesurveys.org

Supplementary data