Association between multidimensional poverty intensity and maternal mortality ratio in Madagascar: Analysis of regional disparities

2.1 Study setting

Madagascar, an island nation in the Indian Ocean, exhibits geographical and climatic conditions that profoundly influence the livelihoods of the population. Although endowed with opportunities in fishing, agriculture, and tourism, the country remains highly vulnerable to natural disasters, particularly cyclones and floods, which recurrently destroy crops, housing, and infrastructure. These climatic shocks exacerbate the challenges of accessing clean water and sanitation, disproportionately affecting coastal and rural regions.

Socioculturally, perceptions of wealth and poverty vary across contexts: urban areas associate wealth with material possessions, whereas rural communities link it to productive assets like livestock. This divergence shapes living conditions and housing standards. Demographically, 63.4% of the population is under 25 years of age, with 75% residing in rural areas, a demographic profile that presents both developmental potential and challenges in education, employment, and healthcare.

Despite government and international efforts to alleviate poverty through initiatives such as the poverty reduction strategy paper and the national development plan, 75% of the population lived in poverty in 2018, with marked regional disparities. Climate-vulnerable regions face acute deprivation, while others benefit from more favorable natural or economic resources. Persistent gaps in education, healthcare, and basic services necessitate tailored strategies to reduce inequality.

2.2 Data sources and study period

This study draws on data from Madagascar’s Third General Population and Housing Census (GPHC-3), conducted in 2018 by the National Institute of Statistics (INSTAT). As a comprehensive and nationwide census, it covered the entire resident population, regardless of age, gender, or socioeconomic status. The census collected detailed information at both the individual and household levels, encompassing demographic, educational, economic, and health-related variables. The dataset provides regionally disaggregated data across all 22 administrative regions of the country, offering a robust basis for spatially nuanced analyses. The reference period for most variables corresponds to the 12 months preceding the census, which ensures temporal relevance for examining maternal health outcomes and related socioeconomic conditions.

2.3 Study design and population

The study followed a cross-sectional descriptive design, based on the full GPHC-3 dataset. The analysis focused on two distinct but complementary populations.

To estimate maternal mortality, the study considered all reported deaths of women aged 15–49 during the reference period. As maternal mortality was identified through household reports, the census provided a comprehensive and regionally disaggregated dataset, allowing for spatial analysis of maternal deaths in relation to socioeconomic indicators.

Multidimensional poverty was assessed at the household level to reflect the structural deprivations influencing maternal health access. Given the collective nature of poverty indicators (education, health, and living standards), the household was the most appropriate unit of measurement. Individual-level maternal mortality data were then aggregated by region and linked to corresponding regional multidimensional poverty index (MPI) estimates. This alignment enabled geographic comparisons and the exploration of socioeconomic inequalities associated with maternal health outcomes.

2.4 Components and weighting of multidimensional poverty indicators

To measure multidimensional poverty in Madagascar, we applied the Alkire-Foster methodology, which allows for the identification of households experiencing simultaneous deprivations across several critical domains of well-being. This method is well-suited to the Malagasy context, where poverty is not limited to income scarcity but includes overlapping deficits in education, health, and basic living conditions.

The MPI used in this study draws upon nine indicators grouped into three dimensions: health, education, and living conditions. Each of these dimensions contributes equally to the composite index, accounting for one-third of the total weight. However, the indicators within each dimension are weighted unequally to reflect both their empirical importance and theoretical relevance, in line with international standards and national priorities.

As shown in Table 1, the health dimension comprises two indicators. The first measures whether the household lacks access to clean drinking water. The second considers whether at least one child under the age of 18 has died within the household. Each of these indicators carries a weight of one-sixth, contributing equally to the health dimension and jointly representing 33.3% of the total index.

The education dimension also includes two indicators of equal weight. The first assesses whether no individual aged 11 or older in the household has completed at least 5 years of schooling. The second captures whether at least one child of primary school age (6–10 years) is not attending school. These two indicators are each weighted at one-sixth, bringing the total weight of the education dimension to 33.3%.

The living conditions dimension is composed of five indicators, each contributing equally within this dimension. These include the use of dirty cooking fuel, lack of access to adequate sanitation facilities, lack of electricity, poor quality housing materials (floor, roof, or walls made of rudimentary materials), and limited ownership of essential assets. Specifically, households that do not own a car, van, or similar motor vehicle and possess at most one of the following items – bicycle, motorcycle, radio, refrigerator, telephone, or television – are considered deprived in this last indicator. Each of these five indicators is weighted at one-fifteenth, collectively contributing one-third to the overall index.

Each household’s deprivation score was computed by summing the weights of the indicators for which it was deprived. This score reflects the proportion of weighted deprivations experienced by the household. Based on their total score, households were then categorized using thresholds commonly applied in multidimensional poverty research. Households with a deprivation score equal to or greater than 33.3% were classified as multidimensionally poor. Those with scores between 20 and 33.3% were considered vulnerable to multidimensional poverty. A more severe category, defined as extreme multidimensional poverty, was assigned to households with deprivation scores equal to or exceeding 50%.

This structure allows for a nuanced understanding of poverty by distinguishing not only who is poor but also the depth and severity of deprivation. The computation of the MPI, including the headcount ratio and intensity measures, is presented in the following section.

2.5 Computation of the MPI

The MPI used in this study builds on the two standard components of the Alkire-Foster framework: the headcount ratio (H) and the intensity of deprivation among the poor (A). These two elements provide complementary information. While the headcount ratio identifies the proportion of the population classified as poor, the intensity captures how many weighted deprivations poor households experience on average.

The headcount ratio, denoted H, is calculated by dividing the number of individuals living in multidimensionally poor households (q) by the total population (n). It expresses the incidence of multidimensional poverty:

The intensity of poverty, denoted A, corresponds to the average deprivation score among those identified as multidimensionally poor. It is computed by summing the individual deprivation scores of poor households and dividing the total by the number of poor individuals. This measure reflects the breadth and severity of deprivation among the poor.

The final MPI is the product of these two components:

This composite index thus integrates both the prevalence and depth of multidimensional poverty within the population. By design, it enables comparisons across regions and groups, and provides a basis for analyzing associations with other variables, as developed in the subsequent analytical section.

2.6 Study variables

This study focused primarily on two main variables: the intensity of multidimensional poverty and the MMR.

2.7 Dependent variable: Maternal mortality (denoted y)

The maternal mortality variable represents the MMR, defined as the number of maternal deaths per 100,000 live births among women of reproductive age (15–49 years). In our model, this numerical rate variable was calculated at the regional level and served as the dependent variable. The MMR estimation followed standard epidemiological protocols, encompassing deaths during pregnancy, childbirth, or within 42 days after delivery termination, in line with the WHO definition.

2.8 Independent variable: Multidimensional poverty intensity (denoted x)

The main explanatory variable is the intensity of multidimensional poverty, derived from the Alkire–Foster methodology. This variable captures the average proportion of weighted deprivations experienced by individuals classified as multidimensionally poor. The MPI was constructed using nine binary indicators grouped into three equally weighted dimensions: health, education, and living conditions. These indicators were operationalized from GPHC-3 data using standard definitions – for example, households were considered deprived in health if they lacked access to improved drinking water or reported the death of a child under 18.

The weighting scheme (1/3 per dimension; 1/6 or 1/15 per indicator) and the deprivation thresholds (≥33.3% for poverty, ≥50% for extreme poverty) follow the standard Alkire–Foster method. As such, our approach ensures both methodological rigor and comparability with existing MPI-based studies. Full details on indicator definitions, data sources, and scoring are provided in the dedicated section on the components and weighting of the MPI.

2.9 Statistical analysis

Data analysis was conducted using R software version 4.4.2, which facilitated data processing and adjustment of the statistical model. A simple linear model was constructed to examine the relationship between the intensity of multidimensional poverty (x) and the MMR (y). In this analysis, critical thresholds were defined to assess the quality of the model’s fit and the strength of the relationship between the variables.

A coefficient of determination R ² ≥ 30% is considered acceptable, which means that the model explains at least 30% of the data variability. A correlation coefficient r > 50% indicates a strong relationship, highlighting a significant linear correlation between the variables studied. Lastly, a threshold of p < 5% ensures statistical significance, confirming that the results obtained are unlikely to be due to chance. These criteria are applied to validate the relevance and robustness of the findings in this analysis. The model can be expressed by the following equation:

where β0 is the intercept, representing the value of y when x = 0, β1 is the coefficient of the independent variable x, measuring the impact of the intensity of multidimensional poverty on the MMR, ϵ is the error term that captures unobserved factors that influence y but are not explained by x.

2.10 Evaluation of results

Evaluation of the results included the analysis of coefficient statistics, providing estimates, standard errors, as well as t and p values for each explanatory variable. A residual analysis was also conducted to verify the assumptions of normality and homoscedasticity, which are essential for the validity of linear regression models.

1. Ethical considerations: Although this study was exempted from formal ethical approval by the Madagascar National Ethics Committee, as it was based on anonymized and publicly available data from the GPHC-3, it adhered to Law No. 2014-006 on the protection of personal data. The study strictly followed the fundamental principles of scientific rigor, data integrity, and respect for privacy. The analyzed information was processed anonymously to ensure participant confidentiality, according to the prevailing ethical standards.

3.1 Regional disparities in multidimensional poverty and maternal mortality

The results (Table 2) highlighted significant disparities in multidimensional poverty and MMR in Madagascar, depending on the place of residence and the region. The MPI, which combines the poverty rate and intensity, ranged from 0.150 in Analamanga to 0.592 in Atsimo Atsinanana. Similarly, the MMR varied from 159 deaths per 100,000 live births in Itasy to 928 in Vatovavy Fitovinany. These marked variations suggest a potential relationship between poverty severity and maternal mortality outcomes.

3.2 Urban–Rural contrasts

A stark contrast was observed between urban and rural settings (Table 2). In urban areas, the MPI was 0.151, reflecting a poverty rate of 31.3% and an intensity of 48.2%. Urban households had a higher likelihood of not being vulnerable to poverty (41.5%) and exhibited an MMR of 312. In rural areas, the MPI reached 0.472, with a much higher poverty rate of 83.7% and a greater intensity of 56.3%. In these settings, only 4.4% of households were not vulnerable to poverty, and MMR reached 426 deaths per 100,000 live births.

3.3 Regions facing extreme poverty and high maternal mortality

As shown in Table 2, some regions showed critical levels of deprivation and mortality. Vatovavy Fitovinany had the highest MMR in the country (928) and an MPI of 0.541, with more than 72.5% of households living in extreme poverty. Melaky and Androy also revealed alarmingly high MPI values (0.588 each), with MMRs of 791 and 243, respectively. In these regions, over 80% of households lived in extreme poverty (80.1% in Melaky, 81.7% in Androy). Anosy (MMR: 566, MPI: 0.553) and Ihorombe (MMR: 626, MPI: 0.532) also demonstrated similarly high burdens.

3.4 Regions with moderate poverty and lower maternal mortality

In contrast, some regions showed more moderate poverty levels and lower MMR (Table 2). Analamanga, the capital region, had the lowest MPI (0.150), a multidimensional poverty intensity of 46.8% and an MMR of 215 deaths per 100,000 births. Boeny and Diana, with MPIs of 0.390 and 0.342, respectively, reported lower MMRs, at 290 and 229 deaths per 100,000 births. In these regions, a higher proportion of households were classified as nonvulnerable to poverty (40.2% in Analamanga, 17.9% in Boeny, and 15.2% in Diana).

3.5 Association between poverty intensity and maternal mortality

A cross-regional analysis revealed a consistent pattern: regions where over 50% of households lived in extreme poverty also experienced elevated MMR. For instance, in Atsimo Atsinanana (MMR: 503, extreme poverty: 81%), Menabe (MMR: 797, extreme poverty: 64.5%), and Betsiboka (MMR: 405, extreme poverty: 65.3%), the intensity of deprivation coincided with some of the highest MMRs nationally (Table 2).

3.6 Statistical modeling: Linking maternal mortality and poverty intensity

The scatter plot in Figure 1 illustrates the positive association between MMR and multidimensional poverty intensity using a linear regression model. The equation MMR = −950.67 + 24.4 × Poverty Intensity produced an R ² of 0.412, indicating that 41.2% of the variability in MMR was explained by the intensity of multidimensional poverty. The Pearson correlation coefficient was r = 0.642 (p < 0.001). Regions such as Vatovavy Fitovinany, Melaky, and Menabe were positioned in the upper-right quadrant of the scatter plot, indicating simultaneous high values of both indicators. In contrast, Itasy and Analamanga appeared in the lower-left quadrant.

Figure 1

Correlation between MMR and multidimensional poverty intensity.

3.7 Residual diagnostics and model validity

The residual diagnostics, presented in Figure S1, provide additional support for the robustness of the regression model. The QQ plot indicated that most residuals aligned closely with the normality line, with only minor deviations at the extremes. The plot of residuals versus fitted values showed no discernible pattern, suggesting a reasonably homogeneous distribution of errors around zero and confirming the assumption of homoscedasticity. The histogram of residuals was centered near zero, with a slight skewness and values ranging approximately from –400 to +400. The Cook’s distance plot identified a few potentially influential observations, some approaching 0.2, but none exceeded the conventional threshold, indicating that no single data point had an undue influence on the model’s estimates.