The elasticity of marital fertility in three sub-Saharan African countries: a decomposition analysis

Abstract

Many studies of fertility have reported a wide range of factors to be important determinants. These determinants include proximate factors and male and female background variables. However, most of the research has been based on an analysis of cross-sectional fertility elasticities. Therefore, there are limited attempts to examine the temporal behaviour of especially marital fertility in association to the proximate and background determinants. To fill this gap, this study analysed Demographic and Health Survey data for Kenya, Rwanda and Zimbabwe using the Oaxaca-Blinder decomposition technique to determine the magnitude of marital fertility changes in association with selected socioeconomic factors. The results showed evidence of significant marital fertility transitions characterised by stalling for all three countries. Marital fertility rates were more responsive changes in reproductive behaviours than compositional characteristics. Male variables like community level of education were positively associated with stalling especially in Kenya and Zimbabwe. Despite the small contributions overall, analysing male and female variables improves the understanding of the sources of marital fertility changes in patriarchal societies.

Background and problem statement

Most of sub-Saharan Africa (SSA) continue to have high fertility rates despite evidence of significant fertility transition in some countries starting in the 1970s and 1980s (Askew, Maggwa, & Obare, 2017; Blacker, 2002; Burger, Burger, & Rossouw, 2012; Houle, Pantazis, Kabudula, Tollman, & Clark, 2016; Indongo & Pazvakawambwa, 2012; Letamo & Letamo, 2001; Mbacké, 2017; Mturi & Joshua, 2011; Palamuleni, 2017). In the countries that have experienced long-term decreases in birth rates, their fertility transitions have been characterised by stalling (Bongaarts, 2006; Garenne, 2011; Garenne & Joseph, 2002; Shapiro, 2013; Shapiro & Gebreselassie, 2009). Since the onset of fertility transition in SSA, various scholars have found significant effects of determinants (Bongaarts, 2017, 2015; Bongaarts, Frank, & Lesthaeghe, 1984; Bongaarts & Potter, 1983). Others like Kravdal (2012, 2002), Ezeh and Dodoo (2001), Caldwell and Caldwell (1987) and Caldwell (1977) have investigated the role of the so-called background variables such as education, religious beliefs and gender inequalities in reproductive decision-making. One of the compelling arguments in fertility literature has been is that family planning services which led to widespread adoption of modern contraceptives have been a crucial influence on fertility transition in SSA (Beguy, Ezeh, Mberu, & Emina, 2017; Casterline & El-Zeini, 2017; Hartmann, Gilles, Shattuck, Kerner, & Guest, 2012; Mbacké, 1994; Mbizvo & Adamchak, 1991; Parr, 2002).

An important feature of the fertility transitions of SSA countries has been the rural-urban divide whereby urban populations started to experience fertility decline about 10 years earlier than their rural counterparts (Garenne & Joseph, 2002; Shapiro & Tambashe, 2002; Shapiro & Tenikue, 2015). Urban centres have provided sociocultural and economic environments that promote fertility limitation earlier than rural areas. Family planning services, formal education, employment opportunities and mass media were first available in urban centres before spreading to rural areas, thus leading to rural-urban divide in the onset of fertility transition (Shapiro & Tambashe, 2002). Factors which distinguish urban from rural contexts such as declines of infant and child mortality rates, increased use of modern contraceptives, delayed age at first marriage, education and female labour-force participation have been important in influencing the decline of fertility in SSA (Blanc & Rutstein, 1994; Bongaarts, 2015; Bongaarts et al., 1984; Bongaarts & Casterline, 2013; Garenne, 2004; Handwerker, 1991; Hertrich, 2017; Mensch, Grant, & Blank, 2006; Ortega, 2014; Shapiro & Gebreselassie, 2014). The findings of the past studies have varied across countries and over time (Bulatao & Casterline, 2001). Some literatures reported that the demand for and use of contraception in SSA were not entirely to limit fertility but as a replacement for traditional methods of birth spacing (Caldwell & Caldwell, 2002; Udjo, 1996). In Zimbabwe for instance, Udjo (1996) reported that the fertility transition was not associated with prevalence rates of modern contraceptives. Some studies have analysed men’s influence on the reproductive behaviours of women from SSA countries (Dodoo, 1998, 1993; Dodoo, Luo, & Panayotova, 1997; Dodoo & van Landewijk, 1996). In reviewing the existing literature, one obtains that the decreases of fertility rates in SSA countries have been associated with many of the socioeconomic factors reported in previous studies. However, one can note that many of these studies were based strictly on the cross-sectional design approach employing standard regressions. Furthermore, the fertility literature is dominated by studies which focused on total fertility rates (TFRs) of all women and adolescents. There are few studies which have focused on marital fertility rates which account for most of the fertility in SSA (Ndagurwa & Odimegwu, 2019). This study was therefore designed to investigate the temporal marital fertility elasticities in relation to the various proximate and background variables in Kenya, Rwanda and Zimbabwe. The study sought to address the following research question:

• If the fertility rate of a group of women at a given point in time is explained by their use of modern contraceptives, education status, employment and urban residence among other factors, to what extent do their fertility rate decrease or increase in response to changes in the status of these variables?

The concept of elasticity is widely used in economics to describe the fluctuations of demand for goods and services in response to changes in price (Andreyeva, Long, & Brownell, 2010; DeCicca & Kenkel, 2015). In the Mathematical Theory of Elasticity, the focus of a calculation is to estimate the relative shift in the form of an object whose properties are variable due the actions of equilibrating forces from external stimuli (Love, 2013). The current study applies the term elasticity to refer to the magnitude of temporal decrease or increase in the level of marital fertility rates in association with changes in sample distributions and changes in individual- and community-level reproductive behaviours. Methodologically, we differentiate this study from the existing literatures by analysing an age-adjusted indicator of fertility (age-specific marital fertility rates—ASMFRs) which is consistent with total marital fertility rates (TMFRs). Many studies have treated the average number of children ever born (CEB) as a proxy measure for TFRs despite that the two may not be consistently associated. We show proof of this in Figs. 4 and 5 presented in the Appendix. Furthermore, the analysis of ASMFRs provides consistency between the evidence of marital fertility transition (Fig. 2) and the subsequent multivariate estimations. We also differentiate this study by analysing the relative contributions to fertility change of male and female variables in a single multivariate framework. Previous studies Dodoo (1998) and Ezeh (1992) which analysed male and female correlates sought to explain the contraceptive use dynamics while the current aim was to link to actual fertility.

Conceptual framework

The conceptual framework for this study was adapted from the demographic transition theory (DTT), ideational change theory, the model of proximate determinants (PD) and past studies. Almost all the variables analysed in the study were drawn from these theories. The DTT and ideational change theories were relevant for their temporal view of fertility and informed the selection of the background factors. The PD model was instrumental in the conceptual formulation of how the different classes of independent variables affect fertility as well as guiding the selection of the relevant direct determinants. Based on the three theories and existing literature, the conceptual model shown in Fig. 1 guided the empirical multivariate modelling of the temporal change in marital fertility rates. The red arrows denote the theoretical postulations pertaining the relationships between fertility and the proximate and background variables. They also show how fertility is related to the various background determinants from different levels as demonstrated in the literature (Billy & Moore, 1992; Casterline, 1981; Kravdal, 2002, 2012; McNay, Arokiasamy, & Cassen, 2003; Parr, 1992). The solid black arrows indicate the empirical modelling of marital fertility in the current study. The Oaxaca-Blinder decomposition was implemented as a multivariate model to obtain adjusted effects.

Fig. 1

Conceptual framework adapted from the DTT, ideational change and PD theories

Methodology

Data

This study analysed DHS data from Kenya, Rwanda and Zimbabwe. The surveys were collected with funding from the United States Agency for International Development (USAID) and implemented by host countries’ statistical agencies with technical support from the Inner-City Fund (ICF) Macro International Inc., widely cited as ICF International. The collection of DHS data has been undertaken at 5 to 6-year intervals since inception in some countries like Kenya and Zimbabwe and therefore provide consistent source of data for tracking changes in fertility rates over time. The regular time intervals of DHS surveys allow for the investigation of the impact of changing compositional characteristics of populations and their reproductive behaviours on fertility outcomes. The impactful short-term fluctuations in the properties of the macroenvironment like political and socioeconomic instabilities which affect the family planning infrastructure, for instance availability of contraception, can be captured through examining the DHS data.

Variables

The dependent variable was marital fertility which was operationalised as ASMFRs. The ASMFR describes the number of live births per woman at any given age or age group during their reproductive career given that they remain in a union from 15 to 49 years. The TMFR is derived from ASMFRs and refers to the total number of live births that a woman will give birth to by the end of her reproductive career if she remained married and experienced given ASMFRs. The choice of ASMFRs ahead of measures of cumulative parities like children ever born (CEB) was predicated on two related reasons. The first reason was that a measure of fertility like CEB may not consistently reflect the temporal decreases and increases in fertility rates. We explored this and provide a detailed illustration in the Appendix (Figs. 4 and 5). Consequently, an age-adjusted measure of fertility was justifiably preferable. The second reason was that fertility trends of countries which as cited in policy motivations are usually reported based on age-adjusted estimates but the CEB measure is not adjusted for age. In order for scientific inquiry to address both academic and policy questions regarding the correlates of the persistent high fertility rates in SSA, it is pertinent that empirical investigations generate evidence based on measures of fertility that speak to TFRs and TMFRs. The independent variables were contraceptive use, postpartum infecundity, abortion, education, occupation/employment status, husband’s desire for children, religion, household socioeconomic status (SES), place of residence, religious beliefs and age group. The variables education and employment status were analysed for both wife and husband. The relevance of these independent variables was drawn from the reviewed literature and established theories of fertility which showed their importance on fertility rates (Agadjanian & Yabiku, 2014; Bakibinga et al., 2016; Balbo, 2015; Bianchi, 2014; Bongaarts, 2015; Mturi & Joshua, 2011).

Statistical analysis

The analysis of the data involved three main parts. The first part involved the application of a Poisson regression-based direct estimation technique to birth history data for women in unions for the calculation of ASMFRs and TMFRs. The use of Poisson regression for computing fertility rates from DHS data has already been demonstrated (Schoumaker, 2013; Schoumaker & Hayford, 2004). In this study, the expected number of births when married for the duration of the reproductive period was expressed as a logarithm linked to a linear function of age group such that

$$ TMFR=5\times \left(\exp \left[\alpha \right]+{\sum}_{k=20-24}^{40-49}\exp \left[\alpha +{b}_k\right]\right) $$

(1)

where α is a constant term and b is the intercept. Given that TMFR is obtained by multiplying ASMFRs by 5, obtaining age-specific rates using Poisson regression involved exponentiating the sum of the constant term and the regression coefficients for the respective age groups which were obtained from

$$ {\lambda}_i=\exp \left[\alpha +{\sum}_{k=20-24}^{45-49}{b}_k{A}_{ki}\right] $$

(2)

for which A_ki are dummy variables for 5-year age groups from 20–24 to 45–49 years; the 15–19-year age group was used as the reference category (Schoumaker, 2013). The trends in TMFRs were reconstructed based on this same method and were computed to provide a visual aid in the illustration of the extent to which TMFRs dominated national trends in TFRs during critical periods like those of rapid declines and stalling. The first task was carried out in Stata version 14.2 using Schoumaker’s (2013) tfr2 command.

The second part of the analysis pertained to the execution of statistical tests of significance of the trends in the TMFRs using linear regression whereby annual TMFRs for distinct periods were regressed against time in single calendar years. The distinct periods were visually determined based on the trends obtained from the preceding phase. These periods were distinguished based on continuous relatively smooth trend, for example, a period of rapid decline as opposed to one of slow fertility decline or increase in fertility. The technique of applying linear regression to test the statistical significance of the slopes was demonstrated by Garenne (2011). Adapting Garenne’s (2011) formulation, the trends were tested as follows:

$$ \Delta {\mathrm{TMFR}}_{15-49}=\alpha +\beta \times t $$

(3)

where ∆TMFR_{15 − 49} denotes the temporal changes in the marital fertility rates for women aged 15–49 years, α is a constant, β represents the slope or extent of the annual decrease or increase in fertility and t stands for time in calendar years. The tests of slopes were implemented in Microsoft (MS) Office Excel 2016 on the estimates exported from Stata.

The last part of data analysis applied the Oaxaca-Blinder decomposition method. The Oaxaca-Blinder decomposition is a regression-based technique for partitioning the difference in a distributional statistic between two groups into an ‘explained’ and an ‘unexplained’ part (Fortin, Lemieux, & Firpo, 2011). The explained part of the decomposition refers to the difference in the outcome statistic which is attributable to group distributional differences (Jann & Zurich, 2008). The unexplained portion arises from differences in how the predictor (independent) variables are associated with the outcome which is represented by the distributional statistic such as the mean (Jann & Zurich, 2008). There would always be this unexplained portion even if the first group was to attain the same average levels of measured predictor variables as the second group. This is because it also captures the effects of unobservable variables reported in the form of a constant (Jann & Zurich, 2008; Sen, 2014). The Oaxaca-Blinder method is frequently used in economics where it was originally developed to examine income differences between advantaged and disadvantaged groups. Taking DHS1 to represent the base survey and DHS2 to represent the succeeding survey, this study decomposed change in the average ASMFR over time as follows:

$$ \Delta \mathrm{ASMFR}={\left\{E\left({X}_{\mathrm{DHS}1}\right)-E\left({X}_{\mathrm{DHS}2}\right)\right\}}^{\prime }{\beta}_{\mathrm{DHS}2}+E{\left({X}_{\mathrm{DHS}2}\right)}^{\prime}\left({\beta}_{\mathrm{DHS}1}-{\beta}_{\mathrm{DHS}2}\right)+{\left\{E\left({X}_{\mathrm{DHS}1}\right)-E\left({X}_{\mathrm{DHS}2}\right)\right\}}^{\prime}\left({\beta}_{\mathrm{DHS}1}-{\beta}_{\mathrm{DHS}2}\right) $$

(4)

where

{E(X_DHS1) − E(X_DHS2)}^′β_DHS2 The explained part, the change in average ASMFRs (means) that was due to group differences in the predictors, that is, the endowment effect weighted by the coefficients of later DHS

E(X_DHS2)^′(β_DHS1 − β_DHS2) The contributions of differences in the coefficients and intercepts, shows expected outcome in earlier survey if it had later DHS survey’s coefficients,

{E(X_DHS1) − E(X_DHS2)} ′ (β_DHS1 − β_DHS2) Be an interaction term accounting for the fact that the differences in the endowments and coefficients of the two successive DHS samples exist simultaneously,

E Mean of the predictor variables [adapted from Jann & Zurich, 2008]

The rationale behind this decomposition method can be applied to the study of fertility changes over time. It was applied in the current study as a counter-factual decomposition whereby the obtained results were conditional. The results were conditional on other factors remaining constant; for example, the result for contraceptive use was conditional on all the other independent variables remaining unchanged between the two successive surveys.

Results

Temporal changes in sample distribution

The three countries have experienced some long-term significant transformations in the compositional characteristics of the married samples (Table 1). The use of modern contraceptives steadily increased for all the countries with prevalence rates highest in Zimbabwe and lowest in Rwanda. All the countries have had decreases in the proportions who were breastfeeding and thus infecund at the time of survey collection with Rwanda showing the largest long-term decrease. These changes signified decreases in childbearing rates. There were limited changes with respect to practice of abortion which remained constant in Kenya and barely changed in Rwanda and Zimbabwe. The lack of increase in proportions reporting abortion suggests that this proximate determinant was not instrumental in the marital fertility transitions of the three countries. The age structure shifts of Kenya and Rwanda pointed at long-term increase in the average age of the married women which emanates from the delay of first marriage. A different picture unfolded in Zimbabwe where the proportion 15–19 years was higher for all post-1990 surveys compared to the ZDHS1988.

Table 1 Changes in the distribution of the married samples of Kenya, Rwanda and Zimbabwe over time

There were subdued increases in female educational attainment especially for Kenya and Rwanda. Although the proportions without education decreased over the long-term from 82% having primary or less in the KDHS1989 to 63% in the KDHS2014, more than half of the married women in Kenya did not have secondary education as of the latest survey. The situation was worse in Rwanda where only 10 and 3% had secondary and tertiary education by the latest RDHS survey. Zimbabwe showed better educational attainment for women than Kenya and Rwanda. Secondary educational attainment more than tripled from the ZDHS1988 (20%) to ZDHS2015 (60%) although tertiary education status at just 6% as of the latest ZDHS survey was still very low. There were greater proportions of men with secondary and tertiary education as well as greater representation in the skilled and semiskilled occupations compared to women in all the countries. Most of the women in all the study countries were from rural areas. This means that the national trends of high TMFRs of these countries have most likely been influenced by the fertility rates of rural areas. The proportions of women with husbands who desired more children increased in Zimbabwe while Rwanda showed some increase in the proportion of women reporting that their husbands desired less children. Kenya experienced a long-term decrease in the proportion of women reporting that their husbands desired more children. This trend is potentially conducive for promoting the decline in marital fertility especially considering that Kenya’s 2003 period which saw a slight increase in women whose husbands desired more children was the time of stalled fertility transition.

Trends in fertility rates and significance of slopes

The trends of fertility rates are presented in Fig. 2. The onset of marital fertility transition was followed by a period of rapid declines in TMFRs. These periods of rapid declines lasted for less than 10 years in Kenya and Rwanda and about 13 years in Zimbabwe. Kenya and Rwanda experienced rebounds in fertility rates to mark the end of the periods of rapid falls of fertility rates while Zimbabwe entered a period of slow decline. Compared to Kenya, Rwanda had a longer period of stalling before resuming fertility transition.

Fig. 2

Trends in TMFRs and TFRs

The stalling of marital fertility transition in Zimbabwe occurred about 22 years after the onset of fertility decline and was short-lived relative to those of Kenya and Rwanda. The trends of TFRs of the three countries were roughly related to those of TMFRs but at much lower rates of fertility, highlighting the role of non-marriage in suppressing national-level fertility rates. The tendency for TFR trends to follow the trajectories of TMFR trends also highlight the fact that marital fertility plays a dominant role in shaping transitions of fertility in SSA countries as well as perpetuating high fertility rates. The statistical tests of the strengths of the distinct episodes of the TMFR trends returned significant results as reported in Table 2.

Table 2 Tests of slopes of the trends showing the marital fertility transitions of Kenya, Rwanda and Zimbabwe. Statistical significance was at p ≤ 0.05 and is represented in the table by 95% CI (confidence interval)

The time coefficients shown in Table 2 for period 1 confirm the rapid annual declines in TMFRs with Zimbabwe showing the largest decreases. The rebounds of period 2 experienced in Kenya and Rwanda represented significant increases of TMFRs with Kenya having the steepest rising trend although it was for a shorter interval of time compared to Rwanda. The overall analysis based on the entire periods covered by all the DHS surveys confirmed that all the three countries have indeed undergone marital fertility transition albeit to varying magnitudes.

Decomposition of inter-survey differences in marital fertility rates

The decomposition of the temporal change in the average of ASMFRs using the Oaxaca-Blinder technique partitioned the difference in the levels of the fertility into three components measuring the conditional contributions of endowments, coefficients and interaction. The contributions were conditional because they reflect the extent to which the mean ASMFR would have increased or decreased if the country experienced a change in the respective component alone while remaining unchanged in the other components. An important consideration when interpreting decomposition results is the sign of the results. The estimation model was set to obtain the difference in the mean ASMFRs between two successive surveys by subtracting the estimate for the latest survey from that of the preceding survey. Consequently, a positive difference denoted a decrease in fertility while a negative difference represented increase in fertility. The positive conditional percentages indicate that a variable’s decomposition component contributed to the observed increase or decrease in the average rate of marital fertility.

Figure 3 presents the aggregate results for endowments, coefficients, interaction and the percentage-points inter-survey decrease and increase of marital fertility. The inter-survey decreases in marital fertility in Kenya and Zimbabwe for period 1 were associated with changes in reproductive behaviour as highlighted by the positive conditional percentage for coefficients. The endowment effect played a role in reducing the magnitude of the decreases for period 2 in Kenya and Zimbabwe, but also suppressed the extent of increases during stalling periods which were periods 3 and 4 respectively. The results for Rwanda showed a different pattern from Kenya and Zimbabwe; the increase in marital fertility for the first inter-survey period was supported by all the decomposition components. The third inter-survey period for Rwanda which was the first to be marked by fertility decrease showed that both endowments and coefficients contributed to this decline. The coefficient effect would have resulted in a larger magnitude of decline between the RDHS2005 and RDHS2010. The generally small percentages for endowments compared to coefficients imply that change in reproductive behaviour played was more prominent in affecting marital fertility transition than compositional changes in the countries’ populations.

Fig. 3

Aggregate decomposition results and inter-survey percentage-points differences in the mean of ASMFRs

Table 3 presents the detailed decomposition results for all the inter-survey periods of the three countries. There were 14 inter-survey periods analysed, and to avoid a long monotonous description of all the results, we focus on those highlighting the influence of proximate factors and male and female correlates. It is worth noting that the constants were fairly large for all the countries and that these potentially highlight the complexity of the fertility contexts of SSA countries. The constants provide a measure of the effects of variables which were not included in the analysis including factors that cannot be easily measured. Some of these factors which may have contributed to the constants are explored in the discussion. Looking at the inter-survey periods characterised by decreases in marital fertility, the changes in the coefficients of modern and none users of contraceptives, female secondary and community-level education were the prominent determinants in Kenya, for example, the 1993–1998 period. In Zimbabwe, changes in reproductive behaviour associated with the community level of education of husbands and the women as well as for 20–24-, 25–29- and 30–34-year age groups were important in the decrease of marital fertility. Compositional changes involving increases in educational attainment of women and men at the community level also contributed to the decline of marital fertility in Zimbabwe as evident in the 1994–1999 period. In the case of Rwanda, the fertility-inhibiting reproductive behaviours associated with the increase in community-level female and male education, women in agriculture jobs and women whose husbands had primary education only were important in the decrease of marital fertility in Rwanda. The results (Table 3) also showed that changes in the reproductive behaviours of women in the 20–24- and 25–29-year age groups contributed for the decrease in Rwanda’s fertility. There was also evidence of marital fertility declining in association with decreases in infant and child mortality in Rwanda post-2005.

Table 3 Detailed decomposition results of the conditional percentage contributions of the determinants’ marital fertility changes

The stall of marital fertility transition in Kenya (KDHS1998–KDHS2003) was positively associated with the reproductive behaviour changes of women with primary and secondary education, women involved in agriculture jobs and the 20–24-, 25–29- and 30–35-year age groups. With respect to male-related variables, the fertility behaviours of women who had primary education only and husbands who desired more children also contributed to stalling in Kenya. The results for Rwanda appeared to show a counterintuitive picture for the 1992–2000 period. The changes in the coefficients of women using modern contraceptives, women with secondary education, those involved in agriculture occupations and those who had husbands who desired more children contributed to the 1992–2000 increase in marital fertility in Rwanda. Such results potentially highlight the lack of sufficient family planning services which hinder consistence in the access and use of contraceptives leading to contraceptive use being associated with the increase in marital fertility. The increase in marital fertility between the RDHS2000 and RDHS2005 was associated with the fertility coefficients of women who had never had an abortion, those with primary or no education, women who were not employed, those from the richest SES quantile and the 20–24-, 25–29- and 35–39-year age groups. The compositional changes involving the increased proportion of married women in the 20–24-year age group also contributed to the pre-2005 increase in marital fertility in Rwanda.

Just like Rwanda, Zimbabwe also had two inter-survey periods characterised by an increase in marital fertility, but these involved the latest surveys. In the first inter-survey period of marital fertility increase for Zimbabwe, 2005/2006–2010/2011, the results showed that reproductive behaviours of non-users of contraceptives, women who were not employed, rural residents, Apostolic women and the 25–29- and 35–39-year age groups were positively associated with the increase in marital fertility. The coefficients of women married to husbands with secondary and tertiary education and husbands in skilled occupations were also positively associated with the ZDHS2005/2006–ZDHS2010/2011 increase in marital fertility. The increase in marital fertility between the ZDHS2010/2011 and ZDHS2015 was associated with age composition changes of the female married population which became younger as the proportion 45–49 significantly decreasing while 35–39 years increased. The coefficients associated with being married to a husband with tertiary education and in skilled occupation were also positively associated with the increase in marital fertility in the last inter-survey period for Zimbabwe.

Although most of the marital fertility changes were unexplained, there were notable instances where the empirical model explained considerable proportions of the inter-survey differences in ASMFR. These are shown in Table 4.

Table 4 Proportions of the changes in ASMFR explained by the model

The results in Table 4 relate to compositional changes involving the indirect factors. The model was able to explain 23% of the 2003–2008 decrease in ASMFR in Kenya. This period marked the resumption of marital fertility transition. The model also explained close to 10% of the decreases in AMSFR for Kenya’s first two inter-survey periods. In Rwanda, 17% of 1992–2000 increase in ASMFR was explained. There were two instances where considerable proportions of changes in ASMFR were explained in Zimbabwe. As shown in Table 4, these were the 1999–2005/2006 decrease and 2010/2011–2015 increase in ASMFR.

Discussion and conclusion

This study found variations in the onset and nature of the trajectories of the marital fertility transitions of the three countries. These variations in the fertility trends of countries reflect differences in the national family planning environments as well as the macro environments affecting fertility (Askew et al., 2017). This potentially implies that previous findings showing marked variations of TFR trends across countries from SSA (Askew et al., 2017; Casterline, 2001) were in fact reflective of inter-country differences in marital fertility. The factors such as mass access to formal education, female labour-force participation and family planning policies have been found to determine the decreases and variability of TFRs in SSA countries (Colleran, Jasienska, Nenko, Galbarczyk, & Mace, 2014; Mbacké, 1994; Murthi, 2002; Siah & Lee, 2015). The findings from this study were consistent with those from past studies; marital fertility transition was more advanced in Zimbabwe where average educational attainment was greater than in Kenya and Rwanda. It is worth noting that education and labour-force participation are integral to the postulations of the demographic transition theory. The DTT states that development enhances labour-force participation and access to formal education which improves adoption of modern contraceptives with fertility-inhibiting effects. The adoption of modern contraceptives can be understood in terms of ideational change theory as cultural resistance to new ideas about birth control are eroded.

This study found evidence of stalling of marital fertility transitions in all the three countries with Kenya and Rwanda experiencing stalls relatively soon after fertility transition started. The stalling of marital fertility transition in Kenya, Rwanda and Zimbabwe relates to the findings from various previous studies which investigated TFRs (Bongaarts, 2008; Garenne, 2008; Gries & Grundmann, 2015; Moultrie et al., 2008; Shapiro & Gebreselassie, 2009). The contribution of the current study is showing that the stalls of TFR transitions in these SSA countries were in fact stalls in marital fertility transitions. It has previously been shown in Zimbabwe that the fertility rates of women who were not in unions continued to decrease in the 2000s with the stalling of TFRs being influenced by rebounds in TMFRs (Ndagurwa & Odimegwu, 2019).

The results from decomposition analysis showed that very small proportions of the changes in ASMFR were explained by the model. When small proportions of fertility decrease are explained, it is because changes in the compositional characteristics of samples were not large enough to have a meaningful effect on fertility rates. In the three countries studied, the increases in employment, education, proportion urban among other characteristics were mostly small. Consequently, it was unable to explain most of the marital fertility decreases as well as stalls. The lack of substantial changes in compositional characteristics can be explained by the fact that Kenya, Rwanda and Zimbabwe did not experience significant economic development over the 1990–2015 period. This lack of development has been cited as a determinant of the stalling of fertility transitions in many SSA countries (Bongaarts, 2006; Shapiro, 2013). Having small proportions of change in ASMFR explained by the model means that most of the TMFR transitions were unexplained. The larger effects of coefficients compared to endowments indicate that the temporal changes in the reproductive behaviours of the different segments of the married women were important in the marital fertility transitions of the three countries. The greater importance of reproductive behaviour changes compared to compositional characteristics found in the current study arguably explains the parallel fertility trends for the different education categories previously reported by Garenne (2012). The longitudinal analysis of fertility rates by education status (no education, primary, secondary and tertiary) by Garenne (2012) showed that the differences in the fertility rates of women were explained by a time lag in the onset of transition. This means that family planning services which influence reproductive decisions and outcomes have been crucial to the marital fertility transitions of SSA countries.

A notable finding from the current study was the large constants which provided a measure of the effects of variables excluded from the analysis. Some of these factors are HIV/AIDS and economic crisis in Zimbabwe which affected family planning services. In Rwanda, there were social and political instabilities from the civil war of the late 1990s which arguably led to the rise in fertility. We undertook an exploratory review of the temporal changes in the preferred number of children for the married women from the three countries focusing on periods of stalling. In Ghana and Kenya, the average ideal number of children decreased in the first inter-survey period and remained virtually unchanged for all the subsequent periods. Rwanda experienced decreases in ideal number of children between the RDHS2005 and RDHS2010 only while in Zimbabwe this occurred from ZDHS1988 to ZDHS2005/06. We also explored the proportions of women desiring four or more children and found that most women expressed this desire throughout and this may explain why stalling has occurred at high levels of TFRs. The high demand for children by most women from SSA has been reported in previous research which noted this feature as a prominent characteristic historically (Tabutin, Schoumaker, & Rabenoro, 2004).

This study found complementary small conditional percentage contributions of male and female variables for periods of significant decreases of marital fertility rates, but in instances of stalling, male variables tended to have greater conditional percentages. This means that the decreases of marital fertility in Kenya, Rwanda and Zimbabwe benefited from the congruence of male and female endowment and coefficient effects while stalling was associated with men more than women. These results echo findings from previous research which sought to examine the role of men in the fertility behaviours and outcomes for women from SSA (Dodoo, 1995; Ezeh, 1991, 1992; Hartmann et al., 2012; Kimuna & Adamchak, 2001; Mbizvo & Adamchak, 1991; Ngome & Odimegwu, 2014; Odimegwu, Bamiwuye, & Adedini, 2015). In conclusion, marital fertility transitions are underway in Kenya, Rwanda and Zimbabwe. The major determinants of these transitions lie mostly outside the vector of individual-level socioeconomic characteristics of men and women. While socioeconomic characteristics have been found to explain differences in fertility among women at a given time, it does not necessarily mean that the same variables can account for the temporal decreases or increases in marital fertility which have been experienced in Kenya, Rwanda and Zimbabwe. The importance of these variables remains relevant nonetheless.

Study limitations

One of the limitations of this study was an unbalanced set of independent variables. The lack of comparable variables for household SES, religion and abortion especially for the 1980s’ surveys meant that the study was unable to undertake a full appraisal of roles of these factors throughout the period covered by DHS surveys. Another limitation relates to the derivation of community-level variables which was based on clusters which may not be synonymous with the socio-spatial patterning cultural norms which have a bearing on fertility outcomes. While the use of cluster boundaries to represent communities makes technical convenience and has been used in other studies, it is always associated with the risk of misclassification of individuals into communities they may otherwise not belong to culturally. One should therefore exercise caution when interpreting the results for community-level variables. Lastly, some variables were excluded due to the fact that the decomposition method used is best suited to analysing variables showing notable change over time. Consequently, variables like desire for children which barely changed following the first inter-survey periods were not analysed.

Policy recommendations

The findings from this study suggest that analysing male and female correlates of marital fertility can shed more light on the fertility dynamics of countries. In patriarchal societies like those mostly found in SSA, demographic problems emanate from the social structures which are characterised by unequal division of power and decision-making which are associated with the ability to determine other people’s individual outcomes. It is therefore recommended that demographic research does not dissociate its objects of enquiry like fertility from the socioeconomic environments within which they are nested. The incorporation of variables that are reflective of the wider societal social organisation in future studies of fertility through the use of innovative analytical methods is recommended.

Appendix

Temporal associations of TMFRs, CEB and ASMFRs

An illustration of the advantage of ASMFRs over CEB is presented in Fig. 2. It can be observed that the trends of TMFRs and ASMFRs were parallel which is expected because TMFRs are computed from ASMFRs. However, the trends in CEB were not parallel to those of TMFRs for all the countries. We acknowledge that it can be a critique of the comparisons in Fig. 2 that TMFRs are a pseudo cohort estimate which conceptually assumes that all the women have completed their reproductive careers since the TMFR statistic is a measure of completed fertility of pseudo cohort. To counter this, we also constructed trends of CEB for the last age group 45–49 and observed that cumulative fertility measure still fell short of predicting the true transitions of TMFRs. The results in Figs. 4 and 5 thus provide evidence to support the use of ASMFRs as the dependent variable in a study of marital fertility transition which is understood in terms of trends in TMFRs.

Fig. 4

Inconsistencies between CEB and TMFRs and consistencies between ASMFRs and TMFRs

Fig. 5

Differences between ASMFRs and CEB in depicting marital fertility changes