Examining internal migration effects on short versus long interbirth intervals in Cotonou, Benin Republic

The paper uses primary data collected during the 2018 Fertility and Migration Survey (FMS) in Cotonou. The overarching goal of the 2018 Cotonou FMS was to provide data on migrants’ fertility behaviours. Two questionnaires (available either in Additional file 1 and Additional file 2 or at https://​github.​com/​bolade1227/​fms) were implemented in the 2018 Cotonou FMS: household and women aged 15–49 years. The households were selected using a stratified two-stage cluster design. At the first stage, 56 enumeration areas (EAs) were drawn from the 2013 Benin Republic population and housing census sampling frame. Then, 35 households were randomly selected per EA. Finally, residents in the selected households were enumerated and all eligible women aged 15–49 years were interviewed.

The final sample consisted of 1913 households and 1949 women of reproductive age.¹ The questionnaires obtained information on women’s migration and birth histories, their fertility preferences, family planning practices, and other socio- economic and demographic variables. Information on migration included the respondent’s previous place of residence, duration of residence in Cotonou, and principal reason for residing in Cotonou.

Live births to migrant women that had not occurred in Cotonou were excluded from the sample; this yielded a sample of 3336 live births among 1226 women (with the exception of international migrant women). From among these live births, the sample was then restricted to 2110 live births with information on interbirth intervals. The weighted sample was 2852 live births among 1659 women aged 15–49 years.²

Data were collected using CSPro’s Computer Assisted Personal Interviewing (CAPI) application developed and deployed on Android devices through CSEntry (see www.​csprousers.​org for more details on CSPro). Data collectors were employed and trained on the use of CAPI for data collection. The household questionnaire and women’s questionnaire were designed in French and translated by the enumerator in any local languages where required. Verbal informed consent was sought by the enumerator reading a prescribed statement to the respondent (aged 15–49 years) and recording in the questionnaire whether or not the respondent consented or provided assent on behalf of minors (under 16). Then, the enumerator signs her name attesting to the fact that she read the consent statement to the respondent. The ethics committee approved this procedure. Thereafter, the interview was privately held after the study participant consented to participate in the study. Data were then transcribed back to English for cleaning, screening, and analysis.

The study examined interbirth intervals over women’s reproductive years. Interbirth interval was calculated in months and defined as the time between the current live birth and the following live birth, counting twins as one live birth. The reason of choosing interbirth interval as measure of spacing rather than interpregnancy interval is two-fold. First, the individual female questionnaire did not record pregnancies history. Retrospective data on pregnancies may be more biased by errors due to memory lapses than those on live births, especially when the period of analysis is too long (e.g., over women’s reproductive years). Particularly, in SSA countries where the illiteracy rate is still relatively high, women are more likely to forget a date of pregnancy than a date of childbirth. Second, interbirth interval remains an appropriate measure of spacing when considering the effect of competition or disease transmission from another young child in the family [28].

As in previous studies, interbirth intervals of less than 24 months were categorized as “short”, 24–60 months as “intermediate”, and greater than 60 months as “long”. Both short and long IBIs are considered as more risky than intermediate IBIs [12].

The main independent variable of the study is the migration status. Drawing on the work of Garcia et al. [29], information on the previous place of residence and duration of residence of women in Cotonou was used to define migration. Migrants in this study are women who were living in Cotonou at least 6 months before the survey and who were previously residing somewhere else within Benin Republic. Migrants were then classified into three groups: those who spent less than 5 years in Cotonou, 5–10 years in Cotonou, and more than 10 years in Cotonou.

The literature often cites several socio- economic and demographic variables—education among others—as greatly associated with interbirth intervals as well [15, 30]. Furthermore, scholars who investigated the migration-fertility relationship have repeatedly used socio- economic and demographic variables in order to control for selective migration [19, 22, 23]. The socio- economic and demographic variables used in this study included: religion, education, employment, ideal number of children, modern contraceptive use, and marital status. The baseline control variables for the analysis are: previous number of child deaths, mother’s age at birth index, birth index, having children of both sexes at birth index or not. Finally, interaction terms between reason for migration to Cotonou and migration status were created.

Event-history analysis was employed to examine the association between migration and interbirth intervals over women’s reproductive years. The analysis time (in months) is the duration between two successive live births. The IBI variable records the type of event observed: 0 for intermediate IBI (24–60 months of interval), 1 for short IBI, and 2 for long IBI.

Competing-risks models that treat a short IBI as the event of interest and a long IBI as the competing event, and vice-versa were fitted. During her reproductive years, a woman who did not experience short IBI may experience long IBI and vice-versa—considering intermediate IBI as less risky for maternal and child health [17]. In this case, using standard survival analysis like Cox proportional hazard (PH) model presents a number of limitations [31‐33]. As a remedy, Fine and Gray [34] proposed cumulative incidence function (CIF) based PH model to analyse survival data arising from a competing risk setup. They then specified a model for the hazard of the subdistribution of the different events at time t as

where Y^(k)(t) is an indicator function with Y^(k)(t) = 1 if the event of type k (here 1 for short IBI and 2 for long IBI) occurs.

The advantage of modelling the subdistribution hazard, or subhazard, is that one can calculate the CIF from it;

where \( {\overline{H}}^{(k)}(t)={\int}_0^t{\overline{h}}^{(k)}(t)\kern0.5em dt \) is the cumulative subhazard.

Competing-risks regression performed in this manner is similar to Cox regression.

The model is semiparametric in that the baseline subhazard \( {{\overline{h}}_0}^{(k)}\kern0.5em (t) \) (that for covariates set to zero) is left unspecified, while the effect of the covariates X are assumed to be proportional:

For each woman i, the subhazard at time t is assumed to be

The exponential of coefficients produces subhazard ratios (SHRs). A positive (respectively negative) coefficient means that the effect of increasing that covariate is to increase (respectively decrease) the subhazard and therefore increase (respectively decrease) the CIF across the board. The competing-risks models—all models being initially controlled for previous number of child deaths, mother’s age at birth index, birth index, having children of both sexes at birth index or not—used for this study can be summarized as follows.

Statistical analysis was performed using STATA version 13.1 (StataCorp, TX, USA). SHRs were calculated along with 95% confidence intervals.

Nineteen percent of live births occurred after intervals of less than 24 months, 65% occurred after intervals of 24–60 months and 16% occurred after intervals of greater than 60 months. The prevalence of short IBIs was higher among migrant women who spent less than 5 years in Cotonou (29%) than among non-migrant women (19%) and earlier migrant women (18%). Conversely, 10% of births among the most recent migrant women (who spent less than 5 years in Cotonou) were of long interval against 19% for non-migrant women. The proportion of long IBIs was 13% among migrant women who have been residing in Cotonou for 5–10 years and 17% among the earliest migrant women (who had resided in Cotonou for more than 10 years) (Fig. 1).

Overall, 37% of live births were among non-migrant women, 10% among most recent migrant women, 17% among migrant women who spent 5–10 years in Cotonou, and 36% among the earliest migrant women (Table 1). The majority of live births were among women who were employed (61%), reported 4 or more children as ideal number of children (73%), were non-Muslims (85%), and were married (86%). On average, maternal age at birth index was 24 years. The mean number of live births and child deaths at birth index was 2.1 and 0.1 respectively.

Table 2 presents subhazard ratios (SHRs) and 95% confidence interval of SHRs for migration status and other covariates obtained from Fine and Gray [34] competing-risks survival regression models for short and long IBIs.

Findings show that there are significant associations between internal migration to Cotonou and short versus long interbirth intervals. Migrant women who have been residing in Cotonou for less than 5 years have higher hazard rates of experiencing short IBIs (SHR: 1.71, 95% CI: 1.33–2.21) and lower hazard rates of experiencing long IBIs (SHR: 0.64, 95% CI: 0.47–0.87) than non-migrant women in Cotonou (Table 2, Model 1, 1b). However, the hazard rates of experiencing short and long IBIs do not differ significantly between earlier migrant women (who have been living in Cotonou for 5 years or more) and non-migrant women. Thus, migrant women’s and non-migrant women’s birth spacing practices converge over time, which is consistent with the gradual adaptation hypothesis (Hypothesis 1a and Hypothesis 1b).

After adjusting for socio- economic and demographic variables (Table 2, Model 2, 2b), the association observed between migration and IBIs in Model 1, 1b in Table 2 remains unchanged. But the gap in the hazard of experiencing short and long IBIs between migrant women who spent less than 5 years in Cotonou and non-migrant women in Cotonou has slightly decreased. Migrant women who had stayed in Cotonou for 5 years or more and non-migrant women in Cotonou still do not differ significantly concerning birth spacing practices. Migrant women who spent less than 5 years in Cotonou have a 60% higher risk of short IBIs (SHR: 1.60, 95% CI: 1.23–2.08) and a 33% lower risk of long IBIs (SHR: 0.67, 95% CI: 0.50–0.91), as compared to non-migrant women. The gradual adaptation effect persists, and to a lesser, the positive selection on less risky IBIs.

Finally, migration status was substituted for interaction terms between migration status and reason for migration in Model 3, 3b in Table 2. Hypothesis 3 posited that recent migrant women for job or school reason would have lower risk of shorter interbirth intervals as compared with non-migrant and earlier migrant women. The results first reveal that, concerning the experience of short IBIs, there is a delay in the disruptive effect, with women who migrated for reasons other than job or school having the lowest risks. In comparison with non-migrant women, migrant women have higher risk of short IBIs within the first 5 years following the migration, with the risk for the ones for job or school reason being lower than for the ones for reasons other than job or school (SHR: 1.52, 95% CI: 1.03–2.26 and SHR: 1.59, 95% CI: 1.11–2.28 respectively). Thereafter, from 5 years of stay in Cotonou, the hazard of short IBIs becomes lower among women who migrate for reasons other than job or school (SHR: 0.67, 95% CI: 0.46–0.98 for the ones who spent 5–10 years in Cotonou, and SHR: 0.74, 95% CI: 0.54–1.02 for the ones who spent more than 10 years in Cotonou) than among non-migrant women. Regarding the experience of long interbirth interval, significant differences are observed between only migrant women who stayed in Cotonou for less than 5 years (regardless of the reason for migration) and non-migrant women. Migrant women who spent less than 5 years in Cotonou are 33% less likely to experience long IBIs (SHR: 0.67, 95% CI: 0.43–1.06 if the reason for migration is school or job, and SHR: 0.67, 95% CI: 0.45–1.01 otherwise) than non-migrant women.

Women’s socio- economic and demographic characteristics are associated with interbirth interval, with unexpected changes from Model 2, 2b to Model 3, 3b (in Table 2). Findings from Table 2, Model 2, 2b show that unemployed women are more likely to experience short IBIs and less likely to experience long IBIs than employed women. After taking into account the reason for migration (Table 2, Model 3, 3b), women’s employment status is still significantly associated with short and long IBIs, but in reverse direction. The hazard rate changed from 1.26 (95% CI: 1.07–1.50) in Model 2 to 0.71 (95% CI: 0.56–0.90) in Model 3 for short IBI, and from 0.86 (95% CI: 0.75–0.99) in Model 2b to 1.15 (95% CI: 0.98–1.36) in Model 3b for long IBIs. Also, compared to women with no education, the hazard of short IBI shifted from a 19% lower risk in Model 2 to a 58% higher risk in Model 3 for women with primary education (SHR: 0.81, 95% CI: 0.66–1.00 in Model 2, and SHR: 1.58, 95% CI: 1.20–2.09 in Model 3). Women with secondary or higher education remain less likely to have short IBIs (Model 2 and Model 3) and more likely to have long IBIs (Model 2b only) than their peers with no education. Other variables such as the ideal number of children, religion, marital status, mother’s age at birth index, previous number of child deaths are associated with IBI as well. Reporting an ideal number of children of more than 3, being unmarried, and being Muslim are significantly associated with higher hazards of short IBIs. The higher the previous number of child deaths, the higher the hazard of short IBIs and the lower the hazard of long IBIs. The association between mother’s age at birth index and IBIs is weak.