Men’s Migration and Women’s Fertility in Rural Mozambique

Abstract

Labor migration profoundly affects households throughout rural Africa. This study looks at how men’s labor migration influences marital fertility in a context where such migration has been massive while its economic returns are increasingly uncertain. Using data from a survey of married women in southern Mozambique, we start with an event-history analysis of birth rates among women married to migrants and those married to nonmigrants. The model detects a lower birth rate among migrants’ wives, which tends to be partially compensated for by an increased birth rate upon cessation of migration. An analysis of women’s lifetime fertility shows that it decreases as the time spent in migration by their husbands accrues. When we compare reproductive intentions stated by respondents with migrant and nonmigrant husbands, we find that migrants’ wives are more likely to want another child regardless of the number of living children, but the difference is significant only for women who see migration as economically benefiting their households. Yet, such women are also significantly more likely to use modern contraception than other women. We interpret these results in light of the debate on enhancing versus disrupting effects of labor migration on families and households in contemporary developing settings.

Introduction and Conceptualization

Profound and pervasive societal changes in rural sub-Saharan Africa are rapidly reshaping its reproductive landscape. Rural fertility, although still high compared with fertility in other parts of the world and in the subcontinent’s urban areas, has shown signs of decline in many sub-Saharan countries over the past two decades. Even in the rural settings where no decrease in fertility rates is yet noticeable or where fertility decline appears to have stalled, survey data point to growing desires for postponing births and reducing family size and to rising contraceptive use (Feyisetan and Casterline 2000). Along with such factors as rising education and increasing access to contraception, a major engine driving the changes in reproductive intentions and behavior is the transformation of rural marriage. Labor mobility, which is a massive and growing phenomenon across the subcontinent (Adepoju 2004, 2006; Agadjanian 2008), plays an important role in this transformation. The macroeconomic restructuring underway in most sub-Saharan nations and the economic opportunities, failures, and imbalances that it generates, continues to push millions of rural dwellers into migration, often separating family members for prolonged periods of time. The same macroeconomic forces, however, make economic returns to migration less stable and predictable, thus increasing uncertainties surrounding migrants’ and their families’ future and consequently straining spousal relationships and undermining marital commitments.

Whereas relatively little research has been done on the impact of migration on marital forms and relationships, the connection between migration and fertility has been extensively studied in sub-Saharan Africa and in similar contexts. However, most studies have focused on fertility of migrants in places of migration destinations, especially in urban areas (e.g., Brockerhoff 1995; Brockerhoff and Yang 1994; Chattopadhyay et al. 2006; Lee 1992; White et al. 2008). Typically, this literature has entertained three main hypotheses on the migration-fertility nexus: that migration is associated with a drop in fertility immediately before and after the move (the “disruption” hypothesis); that migrants adapt their fertility preferences and behavior to those dominant in destination areas (the “adaptation” hypothesis); and that migrants are select subgroups of origin-area populations with distinct reproductive propensities (the “selection” hypothesis) (Goldstein and Goldstein 1983).

In the much more sparse literature on the effects of temporary and seasonal labor migration of men on fertility of their wives left behind in rural areas, the disruption perspective seems to predominate, not least because selectivity of temporary migration and the effects of fertility patterns in places of labor migration destination are difficult to establish. Thus, studies typically conclude that spousal separation because of migration decreases exposure to conception, leading to lower marital fertility, at least in the short term (e.g., Bongaarts et al. 1984; Bongaarts and Potter 1979; Clifford 2009; Lindstrom and Giorguli Saucedo 2002, 2007; Massey and Mullan 1984; Menken 1979; Timaeus and Graham 1989; Van de Walle 1975). The fertility-depressing effect of the separation between migrants and their marital partners could be particularly pronounced in high-fertility and low-contraception settings. However, Millman and Potter (1984) argued that the magnitude of the fertility-depressing effects may also vary depending on the duration of separation and on the length of postpartum amenorrhea. It has been also argued that labor migration may help sustain high marital fertility and even lead to a rise in fertility because the effects of spousal separation are counterbalanced by migration-induced erosion of traditional forms of child spacing, especially of postpartum abstinence (Cleveland 1991; Omondi and Ayiemba 2003). Finally, it is also possible that high frequency of intercourse in the period immediately following men’s return from migration may increase conception probabilities among their wives relative to those among other women (Millman and Potter 1984). Notably, like the predominant view on the migration-fertility association, these perspectives see migration as disrupting the normal functioning of marital relationships and practices.

In this study, we first subject these views to statistical scrutiny by using retrospective survey data from rural Mozambique, which is a high-fertility setting with a long tradition of mass male labor migration. In this exercise, we examine how the rate of birth in any given year is affected by husband’s migration status in the same or the preceding years. In addition, we also estimate the effects of migration on lifetime fertility. In these analyses, we assume, as studies on male seasonal/circular labor migration and fertility typically do, that migration influences fertility rather the other way around. It is, of course, possible that fertility may in fact trigger migration by creating a need to provide for the offspring (or conceivably, discourage migration if child rearing requires the father’s direct involvement). However, in traditional patrilineal settings with a well-established tradition of male migration, like the one considered here, this reverse direction of causality does not seem plausible. Rural men often start labor migration before marriage—in fact, often to amass money for bridewealth payment—and the dominant gendered division of household labor places the responsibility for immediate child care entirely on the mother’s shoulders.

Our study goes beyond establishing a statistical connection between spouses’ separation and the birth of their offspring. In the second part of our analysis, we use data from the same survey to examine how husbands’ migration influences their wives’ reproductive preferences and contraceptive use. To do so, we compare reproductive preferences and use of modern contraceptives among women married to labor migrants and among women married to nonmigrants. At the same time, we explicitly link women’s reproductive intentions to the outcomes of their husbands’ migration. This approach is rooted in a more general conceptualization of fertility decisions as contingent not only on decision makers’ recent/current economic and social constraints but also on their expectations of relative affluence or hardship in the future. We posit that men’s successful migration—that is, migration that results in tangible benefits for women and their households—should encourage reproduction by boosting both women’s economic confidence and their marital optimism. On the contrary, migration failure, by weakening the households economically, may undermine women’s commitment to marriage and their willingness to invest in it through continuing childbearing. Therefore, for this part of the analysis, in addition to comparing migrants’ wives and nonmigrants’ wives, we distinguish between wives of more-successful and wives of less-successful migrants. We define migration success in two ways: on the basis of remittances received from migrant husbands and on the basis of women’s own assessments of the impact of their husbands’ migration on the living conditions in their households. These two approaches are described in the Data and Method section. Importantly, in both cases, migration success is defined from the standpoint of migrants’ wives and other nonmigrating household members: migration success, therefore, is not determined by a migrant’s income per se but instead by the share of that income that the migrant’s left-behind household members receive and by how this share compares against their expectations.

We hypothesize that women married to migrants are more likely to want more children than women married to nonmigrants because of their lower fertility. However, after the number of living children is controlled for, any greater pronatalism of migrants’ wives should be associated with more-successful migration: women should be more willing to make reproductive investments in marital relationships with migrants if these relationships yield tangible benefits and security. In contrast, failed migration would make migrants’ wives more pessimistic not only about the economic prospects of their households but also about the future of their marital unions. This pessimism, in turn, should discourage further reproduction, regardless of the number of children these women already have and of other characteristics.

Finally, we explore possible connections between husbands’ migration status and contraceptive use by their wives, net of fertility intentions. Although the overall contraceptive prevalence should be low among rural women, migrants’ wives could be expected to have even lower contraceptive use than nonmigrants’ wives because of prolonged separations between migrants and their partners. However, we also expect to find differences between wives of more-successful and less-successful husbands. Because migration success is defined by the amount of remittances that the household receives from the migrant and because a large portion of these remittances is transferred in person—that is, when the husband comes home for holidays or when the wife travels to the place of his work—then, all else equal, migration success would be associated with increased interaction between the spouses and, by extension, with a higher frequency of intercourse. Because less-successful migrants—those migrants who do not provide well for their families—are also those who tend to have less interaction with their spouses and therefore lower risks of conception, we expect women married to such men to display a particularly low level of contraceptive use. However, after exposure to intercourse is held constant, the differences across the categories of husband’s migration status should disappear.

We test these hypotheses by using the statistical techniques described in the Data and Method section. In the concluding section, we bring together the results of the three parts of the analysis to reflect on ways in which labor migration shapes fertility behavior and, more broadly, on how migration both disrupts and ensures reproduction of families and households.

Setting

Mozambique is a southern African country of 23 million people. A former Portuguese colony that gained independence in 1975, the country was battered by a brutal civil war for the first decade and a half of its independent existence. Since the war ended in 1992 and economic structural adjustment programs were deployed in the early 1990s, the country has experienced a remarkable macroeconomic growth. Yet, with a per capita gross national income of only $370, life expectancy of 42 years, and a female literacy rate of 33%, Mozambique remains one of the poorest and least economically developed nations of the world (World Bank 2008).

Two nationally representative Demographic and Health Surveys (DHS), carried out in Mozambique in 1997 and 2003, allow for an assessment of trends in fertility and contraceptive use. Table 1 summarizes key fertility and family planning indicators. It shows that the total fertility rate (TFR) declined considerably between the two surveys only in urban areas, but no comparable change took place in rural areas. In fact, the rural TFR registered a minor increase; whether this pattern is real or was somehow influenced by the quality of sampling and data collection in both surveys remains moot. The opposite trends in rural and urban fertility cancelled each other out, and as a result, the national TFR does not seem to have changed, a fact that earned Mozambique a place in the list of sub-Saharan countries where the fertility transition has stalled (Bongaarts 2008). Although the TFR data do not offer signs of declining fertility in rural areas, trends in reproductive intentions and contraceptive use there are indicative of a maturing potential for fertility decline. Thus modern contraceptive prevalence rate (MCPR) among married rural women, although far below the corresponding rate for urban areas, increased from 2.3% to 7.2%. Note that all contraceptives offered through the national health care system are completely free. Even more telling was the change in the percentage of rural women who did not want any more children. Although reproductive intentions stated by survey respondents in sub-Saharan settings are contingent on a variety of conditions and circumstances and are easily changeable (Agadjanian 2005; Johnson-Hanks 2007), these statistics do reflect at least a general preference for fertility regulation and the growing “unmet need” for family planning (Casterline and Sinding 2000).

Table 1.

Fertility and contraception in Mozambique, Demographic and Health Surveys, 1997 and 2003

Indicator	1997	2003
TFR, All	5.6	5.5
TFR, Rural	5.8	6.1
TFR, Urban	5.1	4.4
MCPR, Married Women, All	5.1	11.7
MCPR, Married Women, Rural	2.3	7.0
MCPR, Married Women, Urban	16.6	23.2
% Married Rural Women Wanting No More Children, by Number of Living Children
0	1.1	1.0
1	0.8	4.1
2	5.9	9.5
3	13.5	17.6
4	16.7	24.6
5	38.6	40.6
6 or more	52.6	58.8

Notes: TFR = total fertility rate; MCPR = modern contraceptive prevalence rate.

Sources: Instituto Nacional de Estatística and Macro International (1998, 2005).

Since colonial times, Mozambican men, especially from the country’s south, have worked in South African mines (CEA/UEM 1997; Crush et al. 1991; First 1983), and this legal migration flow, established through generations, has continued in the post-independence era (De Vletter 1998; Harries 1994). Political and economic changes associated with the civil war and with the subsequent post-conflict transition have amplified international migration. In particular, a growing number of seasonal and commuter migrants have been crossing into South Africa to work there, often illegally, after the end of the apartheid regime in the early 1990s and the loosening of border controls between the two countries (Crush 1997). Mozambicans have come to constitute one of the largest migrant groups in South Africa (Adepoju 2003).

In parallel to international migration, migration within Mozambique, particularly from rural to urban areas, has also been growing rapidly. Limited and controlled by the colonial regime, rural-urban migration, especially to Maputo (Mozambique’s capital), increased with Mozambique’s independence and the civil war that soon followed (Jenkins 2000; Knauder 2000). After the war, the structural adjustment policies, which further undermined traditional subsistence agriculture and magnified socioeconomic imbalances, have spurred new migration flows (Knauder 2000; Wenzel and Bannerman 1995). Today, both internal and illegal international migratory moves often fall short of fulfilling the promise that generates them because migrants rarely manage to secure decently paying jobs at their destinations. Despite the drastically diminished returns, the economic downturn, and rising xenophobia in South Africa, the migration flow goes on unabated as rural economies continue to stagnate (De Vletter 2007).

Data and Method

Data

Our study uses data from the survey Men’s Migration and Women’s HIV/AIDS Risks carried out by the Center for African Studies of Eduardo Mondlane University (Mozambique) and the Center for Population Dynamics of Arizona State University in 2006. The sample for the survey was drawn from the population of rural married women aged 18–40 residing in 56 villages of four districts of Gaza province in southern Mozambique. The population of the four districts is traditionally patrilineal and largely monoethnic. Low-yield subsistence agriculture, the mainstay of this arid area’s rural economy, has created fertile ground for massive labor migration of men to South Africa and, to a lesser extent, other places in Mozambique. In each of the four districts, 14 villages were selected with the probability proportional to size. In each selected village (or in a randomly picked section thereof if a village was big), all households with at least one married woman were canvassed and separated into two lists: those with at least one woman married to a migrant and those with no such women. These two lists were used as sampling frames: from each list, 15 households were randomly selected. In villages where one of the lists had fewer than 15 households, additional households were randomly drawn from the other list to complete the target village sample size of 30 households. In each selected household, a woman was interviewed; in households classified as migrant, a woman married to a migrant was interviewed. The nonresponse rate was less than 5% and was almost entirely due to the unavailability of selected women rather than to their refusal to participate in the survey. The sampling procedure resulted in a total sample of 1,680 women (420 per district, 30 per village), of which just over 40% were married to migrants, more than three-fourths of whom were working in South Africa. Note that this sampling approach was not intended to estimate the levels of migration in the study area. Instead, migrants’ households were oversampled to achieve a more or less balanced representation of both types of households and to allow for sound comparisons between them.¹

The survey collected detailed demographic and socioeconomic information, including pregnancy histories; husband’s migration history (starting in 2000, the year of particularly devastating floods in southern Mozambique that left a uniquely strong imprint in popular memory, and therefore could serve as an easy point of reference for the respondents who had difficulties linking their life events to calendar years); household material conditions; and information on HIV/AIDS awareness and prevention, women’s social networks, and gender attitudes. In parallel with the individual women’s survey, a community survey was carried out for each village included in the sample. The community survey focused on village economic and social life, out-migration, and HIV/AIDS issues.

Although the study area is characterized by high HIV prevalence—with surveillance-based estimates in Gaza province reaching as high as 27% (Ministry of Health of Mozambique 2008)—the coverage of HIV testing was still very limited. Thus, only about 18% of survey respondents had ever been tested for HIV. The survey did not ask the respondents directly about their serostatus, and only one woman spontaneously responded that she was HIV positive to the question on worrying about becoming infected by the husband. Notably, however, nearly 90% of those who were ever tested said that they were worried about contracting HIV from their partners (a higher percentage than among those who had never been tested), which implies that almost all ever-tested respondents did not think of themselves as seropositive.

Method

For the analysis of the influence of husband’s migration on fertility, we employ an event-history approach. We use logistic regression to fit a discrete-time event-history model in which the rate of birth in year t is the event of interest.² Husband’s migration status in the previous year, t − 1, is the primary predictor: we choose to lag the effect of husband’s migration to ensure proper temporal ordering. To examine whether a man’s transition from migrant to nonmigrant status leads to changes in his wife’s yearly rates of births, we also conduct a three-way comparison of women whose husbands never migrated during the observation period, women whose husbands had been migrants before but were not migrants in year t − 1, and women whose husbands were migrants in year t − 1.

The husband’s migration history data go back seven years from the time of the survey (or to the start of the current marital union if it started less than seven years before the survey), and we can look at that time span only. Although this is a limitation of the analysis, keep in mind that in a setting such as this, recent births tend to be more accurately reported than births that occurred in the more distant past. Another advantage of focusing on recent births is that we can assume that sociodemographic characteristics such as education or religious affiliation as reported at the time of the survey did not change during the observation span; we therefore include these characteristics as time-invariant controls in the event-history models.³

Both models control for time-varying characteristics—such as woman’s age, her work outside the home, and prior pregnancies—and characteristics that are, or are assumed to be, time-invariant, such as the aforementioned education and religious affiliation, as well as a woman’s previous marital experience. In addition to the individual-level covariates, we include an indicator of community-level trends in married men’s labor out-migration. Lindstrom and Giorguli Saucedo (2002) found in Mexico that a higher level of male migration from a community was associated with higher individual probabilities of birth in any given year, with controls for other factors. The measure that we use is based on the assessment of trends in married men’s out-migration from the community in the decade preceding the survey made by community leaders as part of community survey interviews. It is a dichotomy that takes the value of 1 if the interviewed community leader reckoned that married men’s out-migration had increased in the past 10 years and the value of 0 if he/she thought otherwise. Although this measure is impressionistic (that is, no exact data exist at the village level), we assume that it does reflect general changes in the levels of male labor migration in surveyed communities.

The general event-history model can be expressed as follows:

$$\text{logit}(p_{\mathit{ijt}})={\beta }_{0j}+{\beta }_1{M}_{\mathit{ijt}-1}+{\mathbf{\beta }}_{\mathbf{2}}{\mathbf{S}}_{ij}+{\mathbf{\beta }}_{\mathbf{3}}{\mathbf{V}}_{\mathit{ijt}-1}+{\beta }_4{C}_j,$$

where p_ijt = P[Y_ijt = 1|M_ijt−1, S_ij, V_ijt−1, C_j]; Y_ijt is 1 if woman i in community j experiences a birth in her tth year of exposure, and 0 otherwise; β_0j is the intercept that is allowed to vary randomly across the j communities; β₁ and β₄ are coefficients; and β₂ and β₃ are vectors of coefficients. M_ijt−1 is husband’s time-varying migration status (dichotomy or trichotomy, depending on the model); S_ij and V_ijt−1 are vectors of woman’s time-invariant and time-varying characteristics, respectively; and C_j is the community-level trend in male labor migration. The random intercept adjusts the models for the fact that women are clustered in communities, and therefore women in the same communities may share some unobserved characteristics that may affect the association of interest.⁴

For the analysis of lifetime fertility, we fit a Poisson regression model of the total number of children ever born. In this model, the number of years in the past seven years (or since marriage if it occurred fewer than seven years before the survey) that a husband spent in migration is the predictor. This model includes the following woman’s characteristics as covariates: age, marital experience, education, number of years in gainful employment, and membership in organized religion. The trend in married men’s labor out-migration from the community is also included in the model.

To examine the links between husbands’ migration and wives’ fertility preferences, we consider the following outcomes: woman’s intention to stop childbearing versus otherwise; her intention to have more children versus otherwise; and her preference to have a (next) child soon (within two years) versus not to have a child soon. (Women who were pregnant at the time of the survey were asked whether they wanted another child in addition to the one they were pregnant with.) Finally, for the analysis of contraception, we look at current use of modern contraceptive methods among nonpregnant women; in this sample, 97% of contraceptors were using either oral or injectable contraceptives. All these outcomes are dichotomies; therefore, logistic regression for binary response is used.

For the intentions and contraceptive use outcomes, we start with models that consider differences between women married to migrants and those married to nonmigrants. We then examine differences in migration success. For that purpose, we subdivide women with migrant husbands into two subgroups that we define as (1) women married to more-successful migrants and (2) women married to less-successful migrants. We use two different approaches to define migration success. The first definition is based on the amount and frequency of remittances and gifts that migrants send or bring to their nonmigrant wives (as reported by the latter). The second definition is more subjective and is based on women’s own assessments of the impact of their husbands’ migration on the living conditions of their households. The two definitions are, of course, related, yet they break down the sample quite differently. This breakdown and the distribution of the outcomes across husband’s migration status and categories of migration success are presented in Table 2. The intentions and contraceptive use models control for age, number of living children, education, woman’s employment outside subsistence agriculture, polygyny, household material status, and woman’s knowledge or suspicion about husband’s unfaithfulness. Because fertility intentions can be affected by women’s assessment of their general health, the intentions model controls for whether a respondent has been diagnosed with a disability or ever diagnosed with a serious illness or a chronic medical condition, such as tuberculosis, cancer, diabetes, hypertension, or asthma. The intentions model also controls for women’s worries about becoming infected with HIV through their husbands. The contraceptive use model also controls for a woman’s desire to have a child soon (within two years) and recent sexual intercourse. All covariates refer to the time of the survey or immediately preceding the survey. As in the event-history models, in all the other statistical models, we employ a random-intercept approach, allowing the intercept of an outcome to vary randomly by village.

Table 2.

Selected reproductive characteristics by husband’s migration status and migration success

Characteristic	Nonmigrant Husband	Migrant Husband	More-Successful Versus Less-Successful Migrant, Based on …
			Remittances		Wife’s Assessment of Impact on Household
			More Successful	Less Successful	More Successful	Less Successful
Number of Children Ever Born (mean)	2.9	2.4	2.4	2.3	2.5	2.3
Number of Living Children (mean)	2.4	2.0	2.0	1.8	2.1	1.8
Wants No More Children (%)	30.8	20	20.4	19.1	19.1	20.1
Wants Another Child (%)	64.3	76.1	75.6	77.3	77.8	74.3
Wants Another Child Within Two Yearsa (%)	48.0	55.3	56.7	52.2	56.4	54.5
Uses Modern Contraceptivesa (%)	16.4	14.5	17.3	7.4	18.6	10.4
% in Sample	58.9	41.1	29.8	11.3	20.6	20.5

^aExcludes currently pregnant.

Before we present the results of the statistical tests, several caveats need to be acknowledged. First, women’s marital status was determined on the basis on their self-reports: every woman aged 18–40 who said that she “currently had a husband” during the canvassing of the sampled villages was eligible to participate in the survey. Although the survey opted for the most inclusive (and culturally acceptable) definition of marital union, it was impossible to verify whether all the women who reported having a husband interpreted the question the same way. Second, we assume that the reporting of birth history, fertility intentions, and contraceptive use was reasonably reliable and that if any misreporting occurred, it did not vary systematically across the husband’s migration categories. This, of course, may or may not be the case. Specifically, migrants’ wives may have underreported contraceptive use with extramarital partners. On the other hand, nonmigrants’ wives who were using contraceptives unbeknownst to their husbands could have been reluctant to report that use to an interviewer, fearing inadvertent disclosure. A third caveat relates to the potential influence of HIV/AIDS on fertility. Given the high HIV prevalence in the region, one can think of a number of ways in which HIV/AIDS could impact the outcomes of interest. Thus, if HIV disproportionately affects migrants and their partners, as some studies suggest (for a review of the recent literature on the topic, see Lurie 2006), then elevated subfecundity and fetal loss typically associated with HIV infection (Waters et al. 2007; Zaba and Gregson 1998) may contribute to the reduction of birth rates among migrants’ wives. Our data do not allow us to estimate this contribution, but given the sampling procedure, we believe that this contribution in this study population should not be large. The sample was composed of women who were in reasonably good health; women who were ill were not included in the study for ethical reasons. Some of the women may have been HIV positive, but because antiretroviral treatment at the time of fieldwork was barely available, most of these women were probably in the relatively early stages of HIV, when the effects of the infection on health in general and on fecundity in particular are usually not pronounced. Likewise, by definition, the migrant husbands of these women were all working and therefore presumably in good health at the time of the survey and in several years preceding it. HIV infection has also been said to depress fertility intentions (see Nattabi et al. 2009 for a review of the literature). However, as we noted earlier, few respondents had ever been tested for HIV, and the share of those who had been infected and knew about it was probably very small. Although we lack direct measures of serostatus, we include worries about HIV infection as a control in the fertility intentions model.

Results

Husband’s Migration and Rate of Birth in a Given Year

The first two panels of Table 3 display the results of two event-history models: (1) a model in which husband’s migration status in year t − 1 is a dichotomy, and (2) a model in which wives of migrants in year t − 1 are contrasted with those of nonmigrants and of past migrants. As we expected, husband’s migration is associated with a significantly lower rate of birth in any given year, net of other characteristics. In Model 1, the rate of having a birth is 17% higher for women whose husbands were present in the village in the previous year, relative to women whose husbands were absent. In Model 2, women whose husbands did not migrate during the entire observation span are significantly different from women whose husbands were migrants in year t − 1, yet so also are women whose husbands had been migrants before year t − 1 but were back in the village in that year. In fact, in the former case, the magnitude of the effect is much smaller than in the latter (rate ratios of 1.15 and 1.42, respectively). Although the difference between the wives of never-migrating men and the wives of ex-migrants is not statistically significant at conventional levels (not shown), the magnitude of the difference is substantial and would probably have been within the conventional significance range if the sample size had been larger. The results, then, point not only to a higher rate of birth among the wives of returned migrants, relative to women married to men who are current migrants, but also possibly to a compensation of births forgone during the time of the husband’s migration after the migration period ends.

Table 3.

Yearly rate of birth, random-effects discrete-time logistic regression (odds ratios)

Covariate	Model 1	Model 2
Husband’s Migration Status
Husband present in year t − 1	1.171 *
Husband had not been migrant by year t − 1		1.149 *
Husband had been migrant but returned by year t − 1		1.417 *
Wife’s Characteristics
Age 21–25	1.511 **	1.514 **
Age 26–30	1.615 **	1.614 **
Age 31 or older	1.237	1.245
Second or higher-order marriage	0.661 *	0.657 *
Number of pregnancies before year t	0.979	0.979
Gainfully employed in year t − 1	1.076	1.077
1–4 years of school	1.174 *	1.179 *
5 or more years of school	1.270 **	1.277 **
Affiliated with organized religion	1.172†	1.171†
Number of Married Men Going to South Africa Increased in Past 10 Years	1.196 **	1.198 **
Model Chi-Square	68 **	69 **
Number of Person-Years	4,674	4,674

Notes: Reference categories are age 18–20, first marriage, no schooling, not employed, not affiliated with organized religion, and number of married men going to South Africa decreased or did not change in past 10 years.

^†p ≤ .10;

^*p ≤ .05;

^**p ≤ .01

Interestingly, the effect of trends in migration of married men in the community is also statistically significant albeit in the opposite direction: residing in communities with an increasing migration outflow is associated with a higher birth rate in any given year. In addition, education has a positive effect on the rate of birth (not unusual for pre-transitional and early transitional settings), as does membership in organized religion (although the effect of the latter is only marginally significant). At the same time, the effect of woman’s work outside subsistence agriculture is not statistically significant.

The deficit of births associated with husband’s labor migration is manifested in wife’s lifetime fertility. Table 4 shows the parameter estimates and standard errors of a Poisson regression model of the cumulative effect of men’s migration (years spent in migration in the seven years preceding the survey) on their wives’ parity. As shown in the table, each additional year spent by a man in migration is associated with a 0.014 decrease in the logged total number of live births to his wife, net of other factors. This effect is highly statistically significant. The coefficient for the increase of married men’s out-migration from the community in the 10 years before the survey has a positive sign, but unlike the earlier presented event-history models, this effect does not reach the threshold of statistical significance.⁵

Table 4.

Total number of children ever born, random-effects Poisson regression (parameter estimates)

Covariate
Husband’s Migration
Number of years husband was away	−0.014 **
Wife’s Characteristics
Age 21–25	0.641 **
Age 26–30	1.164 **
Age 31 or older	1.544 **
Second or higher-order marriage	−0.132 **
1–4 years of school	0.023
5 or more years of school	−0.057
Number of years in gainful employment	0.009
Affiliated with organized religion	0.076 †
Number of Married Men Going to South Africa Increased in Past 10 Years	0.022
Model Chi-Square	879 **
Number of Cases	1,677

Notes: Reference categories are age 18–20, first marriage, no schooling, not employed, not affiliated with organized religion, and number of married men going to South Africa decreased or did not change in past 10 years.

^†p ≤.10;

^*p ≤ .05;

^**p ≤ .01

Husbands’ Migration and Wives’ Reproductive Intentions

In this section, we present the results of the analysis of women’s desires for more children. Because of the prospective nature of the outcome, we can now use a wider range of covariates and, most importantly, distinguish between the effects of more-and less-successful migration. Overall, about 70% of the women in the sample wanted to have more children, and this intention was much more common among migrants’ wives than among nonmigrants’ wives (see Table 2). Within the former group, more-successful migration was associated with a higher likelihood of wanting another child, but the difference between wives of more- and less-successful migrants was rather modest. The first three columns of Table 5 summarize the results of logistic regression models predicting the intention to have more children. (Because the models predicting the intention to stop childbearing produced results that were very similar to those depicted in Table 5, their results are not shown.) In the models presented in the first half of the table (columns 1–3), the predictor is whether the husband was a migrant at the time of the survey. The baseline model (column 1) demonstrates that women married to migrants are significantly more likely to wish to continue childbearing than women married to nonmigrants. The difference diminishes considerably when the number of living offspring is controlled for, but it is still highly statistically significant. The addition of remaining controls barely changes the difference between the two groups, which remains statistically significant.

Table 5.

Intention to have another child, random-effects logistic regression (odds ratios)

Covariate	1	2	3	4	5	6
Husband’s Migration Status
Husband is currently a migrant	1.768 **	1.369 **	1.310 *	—	—	—
Husband is a more-successful migrant	—	—	—	1.937 **	1.742 **	1.642 **
Husband is a less-successful migrant	—	—	—	1.614 **	1.078	1.059
Wife’s Characteristics
Number of living children		0.483 **	0.524 **		0.479 **	0.520 **
Age 21–25			0.948			0.931
Age 26–30			0.673			0.668
Age 31 or older			0.433 **			0.425 **
In polygynous marriage			0.652 **			0.667 **
1–4 years of schooling			0.953			0.937
5 or more years of schooling			0.868			0.856
Gainfully employed			0.839			0.841
Affiliated with organized religion			0.906			0.902
Diagnosed with disability or chronic illness			1.021			1.023
Currently pregnant			0.415 **			0.414 **
Knows/suspects husband has extramarital partners			0.691 **			0.689 **
Very worried about becoming infected with HIV through husband			1.349 †			1.338 †
Household Material Status Index			1.110			1.093
Model Chi-Square	26 **	307 **	336 **	27 **	308 **	336 **
Number of Cases	1,678	1,678	1,674	1,677	1,677	1,674

Notes: Reference categories are nonmigrant husband; age 18–20; in monogamous marriage; no schooling; not employed; not affiliated with organized religion; never diagnosed with serious illness; not pregnant; does not know/suspect that husband has extramarital partners; and somewhat worried, not worried, or not sure.

^†p ≤ .10;

^*p ≤ .05;

^**p ≤ .01

The last three columns (4–6) of Table 5 present the results of the same models but break down the migrants’ wives into two subgroups based on women’s assessment of migration’s impact on their households. As mentioned earlier, we also fit models with the definition of migration success based on remittances, but because the results are similar to those of the models using the subjective definition, we do not present them in the table. The baseline model shows that women in either migrant-husband category are significantly more likely to want more children than women married to nonmigrants when no other factors are taken into account. When we control for the number of living children, the effect of less-successful migration disappears, suggesting that the differences in reproductive intentions between wives of nonmigrants and wives of less-successful migrants is due largely to the lesser number of children among the latter. At the same time, controlling for the number of children changes the effect of more-successful migration only slightly: it remains strong and highly significant. The addition of all remaining controls barely changes the magnitude of this effect. Notably, wives of more-successful migrants are significantly different not only from wives of nonmigrants but also from wives of less-successful migrants (not shown). When we use the definition of “migration success” based on remittances, the contrast between the two categories of migrants’ wives is not as pronounced but the overall pattern of differences across the three groups is the same as in the case of subjectively defined migration success. These results support our expectation of greater pronatalism among women married to more-successful migrants net of the effects of other factors. Among these other effects, it is noteworthy that education appears irrelevant to reproductive preferences. In addition, women who have to share their husbands—and, therefore, presumably the income that their husbands generate—with other women, either through polygyny or the husbands’ extramarital partnerships, are significantly less enthusiastic about continuing reproduction than women in monogamous unions or whose husbands do not have outside partners.

We also look at the desire to have a child within the next two years, limiting the analysis to nonpregnant respondents. As in the case of the overall desire to have a child anytime, migrants’ wives had a much higher share of those who wanted to have a child within two years of the survey date (Table 2). However, this difference is fully explained by the number of living children and other controls. And so also is the difference between the two subgroups of migrants’ wives: although the patterns of differences are reminiscent of those in the model predicting overall fertility desires, the effects are smaller and not statistically significant (results not shown).

Husband’s Migration and Contraceptive Use

Finally, we look at women’s current use of modern contraceptives. As Table 2 shows, women married to migrants were only slightly less likely to use contraceptives than those married to nonmigrants. However, this modest overall difference masks a rather substantial gap between women married to more-successful and women married to less-successful migrants, with contraceptive use, as we predicted, being higher among the former. The contraceptive “advantage” of more-successful migrants was particularly pronounced when migration success was defined on the basis of remittances but was also substantial when the subjective definition was applied.

Table 6 summarizes the results of three logistic regression models predicting the likelihood of modern contraceptive use among nonpregnant respondents. The first model compares migrants’ wives and nonmigrants’ wives without controlling for recent intercourse. This model suggests a lower likelihood of modern contraceptive use among migrants’ wives (OR = 0.76, p ≤ .10). However, when we control for exposure to intercourse (Model 2), the direction of the association is the opposite (although the coefficient is not statistically significant). Model 3 uses the same covariates as Model 2 but splits migrants’ wives into two subgroups on the basis of migration success, subjectively defined. The results of this model defy our expectations and are quite illuminating: whereas wives of less-successful migrants are now indistinguishable from wives of nonmigrants, wives of more-successful migrants are significantly more likely to use modern contraceptive methods. (A model that breaks down the migrants’ wives category on the basis of remittances produces results that are nearly identical to those in Table 5, and we do not present them separately.) Among other factors, the positive effects of education (a nearly universal correlate of modern family planning) and of knowledge/suspicion of the husband’s infidelity should also be noted.

Table 6.

Current use of modern contraceptives among nonpregnant women, random-effects logistic regression (odds ratios)

Covariate	1	2	3
Husband’s Migration Status
Husband is currently a migrant	0.759 †	1.313
Husband is a more-successful migrant			1.639*
Husband is a less-successful migrant			0.990
Wife’s Characteristics
Number of living children	1.338 **	1.374 **	1.367 **
Age 21–25	1.456	1.338	1.290
Age 26–30	1.045	0.931	0.903
Age 31 or older	0.676	0.585	0.557 †
In polygynous marriage	0.761	0.769	0.791
1–4 years of school	1.550 *	1.487 †	1.471 †
5 or more years of school	2.747 **	2.485 **	2.448 **
Gainfully employed	0.770	0.740	0.753
Affiliated with organized religion	1.342	1.310	1.306
Knows/suspects husband has extramarital partners	1.466 *	1.457 *	1.452 *
Wants to have a child within two years	0.506 **	0.474 *	0.465 *
Had no sex with husband in past four months		0.397 **	0.397 **
Household Material Status Index	1.179 †	1.186 *	1.163 †
Model Chi-Square	81 **	98 **	100 **
Number of Cases	1,411	1,407	1,406

Notes: Reference categories are nonmigrant husband, age 18–20, in monogamous marriage, no schooling, not employed, not affiliated with organized religion, does not know/suspect that husband has extramarital partners, and not pregnant.

^†p ≤ .10;

^*p ≤ .05;

^**p ≤ .01

Discussion

Labor migration is typically seen in demographic research as reducing exposure to conception and disrupting marital childbearing. This view usually fosters a more general conclusion about disruptive effects of labor migration on families. Our study started with a similar general premise and has supplied evidence supporting the fertility-reducing effect of spousal separation through labor migration. However, our results suggest two important corrections to the generally held assumption. First, in settings where the tradition of labor migration is well established and livelihoods of a large segment of the population are dependent on migration remittances, the disruption caused by migration is a common part of life course and family-building trajectories, and in this sense, is normative. The reduction in births associated with husband’s migration is a “price” that households pay for a chance to improve their material conditions. Yet, as our results suggest, migrant households at least partially compensate for the deficit of births accrued through years of migration after the period of migration ends.⁶ In addition, although a migrant husband’s absence is negatively related to his wife’s rate of birth, the aggregate levels of migration are positively associated with the rate of birth. Specific pathways through which aggregate migration trends may be connected to individual fertility outcomes would require a special investigation. However, we can speculate, following Lindstrom and Giorguli Saucedo (2007), that this effect may be related to how migration and its economic benefits strengthen—rather than strain—the family system and the broader social fabric of the community.

This cementing role of migration has its limits, however, and this is where our study makes another correction to the commonly held views. Although migrants’ wives are more likely to desire more children than nonmigrants’ wives, their greater “pronatalism” is conditioned on the outcome of migration. We proposed two simple approaches to distinguishing between migration success and migration failure from the nonmigrant wife’s perspective: an objective approach based on the amount and frequency of remittances, and a subjective approach based on wives’ assessment of how migration has affected the living conditions of their households. Both approaches yielded rather similar results, but the subjective definition produced a starker contrast between the effect of more-successful migration and that of less-successful migration. Both subgroups of migrants’ wives were more willing to continue childbearing than wives of nonmigrants; however, although the difference between wives of less-successful migrants and women married to nonmigrants was mainly due to the lower lifetime fertility of the former, the corresponding difference between women married to more-successful migrants and those married to nonmigrants was largely independent of parity. This difference was also net of wife’s education, wife’s employment outside the home, and household economic conditions (none of which, incidentally, were significant determinants of reproductive intentions). Hence, a more positive assessment of the husband’s migration outcomes for the household strengthens his wife’s desire to continue childbearing, arguably by instilling greater optimism about the future and/or greater need for retaining the husband’s attention and therefore the flow of migration-generated benefits. In comparison, migrants’ wives who do not see improvements in their households’ well-being as a result of migration may not be, ceteris paribus, any more motivated to continue childbearing than women whose husbands are not migrants. Notably, these differences were largely impervious to women’s knowledge of their husbands’ fidelity, which itself had a significant depressing effect on fertility intentions.

It is important to note, however, that the differences in childbearing intentions, while reflecting women’s views of the future, are not necessarily indicative of their future reproductive behavior. Obviously, with the data at hand, we are unable to examine how the women’s fertility will match their reproductive preferences. (Such analyses will become possible as subsequent waves of data collection are carried out.) It is noteworthy, though, that we did not find comparably stark differences across the types of husband’s migration status in short-term fertility intentions as we did in overall fertility preferences. Moreover, it is quite intriguing that women married to more-successful migrants were significantly more likely to be using modern contraceptives than either of the other two groups of women, even after we controlled for the number of children, exposure to intercourse, and other factors. We do not have a definitive explanation for this finding. It is possible that more-successful migrants learn about the benefits of contraceptive use in places where they work and then transmit this knowledge to their wives. However, family planning services are generally available, and the benefits of contraception are widely popularized even in rural areas. An explanation based on successful migrant households’ financial advantages not captured by the controls does not seem plausible, either, as pecuniary costs of contraceptives in this setting are nil or minimal. We suggest that the contraceptive advantage of women married to more-successful migrants is linked, like their fertility intentions, to the same sense of marital optimism and commitment that men’s successful migration may foster among their spouses. Increased investment in marriage on the part of these women translates into a stronger determination to avoid untimely conceptions and other reproductive mishaps and therefore leads to greater mastery of reproductive techniques. This is not to argue that wives of more-successful migrants have more autonomy in reproductive and, specifically, contraceptive choices than do other rural women. In fact, men’s migratory success may limit their wives’ decision-making power because they and their children are more likely to depend on their husbands’ earnings than are families of less-successful migrants or of nonmigrant men (see, e.g., Menjívar and Agadjanian 2007). Instead, in a pre- and early transitional setting, where modern contraception is meant to minimize reproduction-related morbidity and mortality rather than to reduce family size (Bledsoe et al. 1994, 1998), higher contraceptive use and greater pronatalism among women who see their husbands’ migration as beneficial are compatible, since both may reflect enhanced commitment to marriage and, more specifically, to the success of marital reproduction.

The findings of our study not only further expand on earlier analyses of the role of migration in family-building and reproductive strategies and behavior but also help foresee the potential implications of changes in the nature and outcomes of labor migration for future fertility trends in rural sub-Saharan settings, such as the one examined here. Unlike earlier studies that stressed social and demographic distortions caused by male migration and their consequences for families and reproduction in the context of uniformly favorable economic outcomes of migration (e.g., Mabogunje 1989; Timaeus and Graham 1989), our study highlights the effects of the deterioration of these outcomes in contemporary labor migration. The increasingly erratic and ever-diminishing returns to present-day migrant employment betray the hopes that continue to push scores of rural men, and increasingly rural women, into migration. As labor migration persists and even grows in quantity but declines in quality, it evolves from a force that strengthens rural families and communities to one that undermines them. The HIV/AIDS epidemic, which has reached a particularly large scale in southern Africa, including Mozambique, may further amplify this effect of migration, not least because labor migrants are commonly perceived as being at higher risk of HIV infection (Agadjanian et al. forthcoming) and because exit from marriage may be an increasingly common response to the perceived risks of infection (Reniers 2008). The migration-catalyzed erosion of spousal commitments to marriage inevitably adds to the fertility-reducing pressures stemming from the spread of schooling as well as other cultural and economic innovations taking hold in rural society. And as the preference for smaller families gains ground, contraception will increasingly change from a means of assuring family stability and reproduction through birth spacing to an instrument of fertility limitation.