Parental socio-economic status (SES), which includes income and education, has been extensively examined in relation to children’s health outcomes. Income serves as an indicator of the economic resources available to individuals through capital, employment or wages. Parents with a decent income are more likely to be able to meet their family’s essential needs and provide their children with a higher standard of living through healthy nutrition and adequate healthcare. Additionally, a higher standard of living can act as a protective factor against maternal postpartum depression, the most common postnatal complication in western countries, which likely impacts maternal quality of life, family functioning and children’s development and health (Nakamura et al., 2020). Moreover, income inequality is strongly correlated with infant mortality across all countries, although the causal nature of this association remains to be confirmed in the short term (Avendano, 2012).

The link between women’s education and infant mortality is one of the most consistent and robust findings in public health. A high level of education is likely to improve medical knowledge (Hoffmann et al., 2018), promote healthier practices and provide easier access to economic and social resources (Green and Hamilton, 2019). Empirical evidence supporting the positive link between education and children’s health has been provided in numerous studies. For example, Smith-Greenaway (2020) showed that parental education is independently associated with a reduced risk of low birth weight in their children in Sub-Saharan Africa. In addition, a higher level of mother’s or father’s education is associated with lower infant mortality (Saarela and Finnäs, 2014; Balaj et al., 2021).

The healthy immigrant paradox is attributed to the unexpected health advantages of immigrant groups settled in host countries despite their low SES. Although it has long been discussed in relation to the health of adult immigrants, it can also be observed among children born to immigrant mothers (these children of immigrants are not immigrants themselves, but are sometimes called “second-generation immigrants”). Despite their mothers’ disadvantaged SES and limited use of prenatal care, the native-born children of Hispanic women exhibit comparable or even better health outcomes compared to those born to non-immigrant mothers in the U.S. (Mendoza, 2009; Green and Hamilton, 2019). A similar advantage is documented in perinatal outcomes for African and Latin American mothers in Spain (Farré, 2016; Stanek et al., 2020, 2021).

The “healthy immigrant” effect suggests that individuals who migrate to host countries are selected by either the host countries themselves, or through self-selection among predominantly healthy individuals (Nørredam, 2015; Agyemang, 2019). In the context of children’s health, this implies that there is an intergenerational transfer of health advantages, i.e., an indirect selection whereby healthy immigrant women of childbearing age are more likely to give birth to healthy babies. Evidence supporting this selection has been found in various settings, including France, Belgium, Austria, Spain, the U.S. and Canada, with respect to infant mortality or other birth outcomes such as low birth weight, preterm births and gestational age (Farré, 2016; Guendelman et al., 1999; Hummer et al., 2007; Vang et al., 2015; Sow et al., 2019; Florian et al., 2021, 2023).

The theory of social capital has been proposed to explain the health advantage experienced by immigrants. As described by Putnam (1993), social capital refers to the attributes of a social organisation, including networks, norms and social trust, which benefit both individuals and society as a whole. Coleman (1988) identified three key components of social capital: obligations and expectations, information channels and social norms. These elements collectively shape social relationships and contribute to the concept of social capital.

To quantify social capital, various measures have been employed at both the individual and the community level, although most of the focus has been on the individual level (Kanas et al., 2012; Allen et al., 2014; Siedlecki et al., 2014; Koelet et al., 2017). Thus, participation in charitable activities, affiliation with a political, apolitical or religious group and any situation that enables social ties to be developed and maintained are individual dimensions of social capital.

At the individual level, social capital can be assessed through both formal support provided by midwives, general practitioners and other health professionals and informal support offered by partners, family members or close acquaintances (Nakamura et al., 2020). Formal support is perceived through the lens of the relationships between caregivers and patients, and when these relationships are strong, they can ensure better pregnancy follow-up and improved birth outcomes. Family support also plays a crucial role in pregnancy outcomes (Benza and Liamputtong, 2014) due to its significant influence on the lifestyles and health-related behaviours of its members (Fortin and Le Gall, 2007). Similarly, it has been suggested that the support of the partner is essential to meet the family’s economic needs, and strong paternal involvement contributes to maternal and child health (Fiori, 2020; Nakamura et al., 2020). Evidence supporting this notion was provided by Gaudino et al. (1999), who showed that paternal support during childbirth had a protective effect against low birth weight and infant mortality.

At the community level, most measures assess the socio-ethnic and economic characteristics of the neighbourhood or area. The most commonly used socio-ethnic dimension is the concentration of immigrants from the same origin in the neighbourhood (Muchow and Bozick, 2022). Van Kemenade et al. (2006) referred to a similar concept when discussing social networks based on ethnicity. This entails relying on close-knit networks of individuals from the same cultural background, which promote immigrants’ economic and social integration (Van Kemenade et al., 2006; Gill and Bialski, 2011), as well as improving their understanding of the healthcare system (Albertini Früh et al., 2017) and their well-being (Van Kemenade et al., 2006; Zhao et al., 2010). The impact of these networks on children’s health has also been documented. As shown by Zeitlin et al. (2011), there is an inverse relationship between the proportion of immigrants in the neighbourhood and preterm births among women from Sub-Saharan Africa, and a positive correlation among women born in France. In a cross-sectional ecological study, social capital at a level below the state (trust, association membership) was found to be strongly associated with higher infant mortality rates among Blacks (Kawachi et al., 1997).

The link between the socio-economic dimension of neighbourhoods and children’s health has been the subject of numerous studies. Clausen et al. (2006) showed that the risk of pregnancy complications was higher in residential areas of Oslo East with lower SES. Furthermore, Spencer et al. (1999) and Bundred et al. (2003) found that the most deprived neighbourhoods or areas in England (in the north) had higher levels of low birth weight. Even more detailed analyses conducted in the U.S. have confirmed the link between neighbourhood socio-economic level and children’s health. In a multilevel cross-sectional analysis, Buka et al. (2003) put these links into perspective by showing a significant decrease in average birth weight as the level of economic disadvantage in the neighbourhood increased, and only for African-American mothers. The findings of Calling et al. (2011), which report an increase in infant mortality in neighbourhoods with high levels of deprivation in Sweden, support the hypothesis that neighbourhood factors are significantly associated with infant health.

Various mechanisms may explain the links between the socio-ethnic and economic characteristics of neighbourhoods and adverse health outcomes in children. Numerous studies have suggested that residential segregation underlies these associations. Indeed, highly segregated areas tend to experience elevated unemployment and crime rates (van San, 2004; Bolt et al., 2010; Balascio et al., 2023), poverty (Schaut, 2001; van San, 2004; Musterd et al., 2008), social tensions, riots and insecurity (Schaut, 2001), which can expose residents to chronic or acute stressors and depression. Additionally, these areas are often sites of racism and discrimination (Osypuk and Acevedo-Garcia, 2008; Balascio et al., 2023), which can induce stress and influence health behaviours (Janevic et al., 2021). Such psycho-social factors have been linked to preterm birth and pregnancy complications (Misra et al., 2001). Another potential mechanism is inadequate language proficiency in highly segregated areas (Bolt et al., 2010), which may hinder understanding of health promotion messages and affect communication between immigrant women and healthcare providers during prenatal and postnatal visits.

Several studies have examined the interactions between caregivers and patients, with a specific focus on discrimination and its role in contributing to health disparities.

A recent United Nations analysis of Black women’s experiences during pregnancy and childbirth in the Americas concluded that systemic racism in medical systems is one of the main reasons they are more likely to experience serious complications (UNFPA, 2023). Numerous ethnographic studies in France have documented discrimination against women of foreign origin in hospitals at the time of gynaecological and prenatal consultations and during or after childbirth (Fassin, 2000; Saurel-Cubizolles et al., 2017; Ouanhnon et al., 2022). This can contribute to inequalities in birth outcomes. Increased preterm births, low birthweight and low gestational age outcomes among women who have experienced discrimination (Rich-Edwards et al., 2001; Rosenberg et al., 2002; Mustillo et al., 2004; Giurgescu et al., 2011; Daalen et al., 2022) are illustrative of the link between (the perception of) discrimination and adverse birth outcomes. Indeed, racial discrimination or discrimination linked to migration origin can act on children health’s through several mechanisms. It can, via the psychological stress and immunodepression it engenders, stimulate the production of corticotropin-releasing hormone in the placenta, which may in turn increase the risk of preterm delivery (Rich-Edwards et al., 2001). Moreover, discrimination in hospitals can exacerbate tensions during prenatal or postnatal consultations, which can lead to the refusal of care during pregnancy, and thus have a negative impact on pregnancy follow-up. Finally, risky behaviours (smoking, alcohol and psychoactive substance use, etc.) during pregnancy may be a response to the chronic stress generated by racial discrimination (Rosenberg et al., 2002; Brondolo et al., 2009). These risky behaviours, as well as indicators of child health such as preterm births, low birthweight and gestational age, are important predictors of infant mortality (Evans and Alberman, 1989; Reichman and Teitler, 2006; Farré, 2016; Štípková, 2016; Ballon et al., 2019; Kihal-Talantikite et al., 2019).

Only a few studies in the literature deal with the context of the country of origin. One such study is Kinge and Kornstad (2014), who focused on infant mortality differentials according to migration origin in Norway. Using the mortality rates prevailing in the mother’s country of origin as a proxy for the socio-sanitary characteristics of the country of origin, they found that immigrants from countries with high infant mortality rates have a higher infant mortality rate in Norway. According to them, the attributes of the country of origin are likely to influence the factors associated with infant mortality in the host country.

This finding is consistent with the observation of Väisänen et al. (2022) that women from the poorest countries have the worst reproductive health. This means that “the inequalities observed in a global scale in countries’ economic outcomes may reproduce themselves as reproductive health inequalities among migrants living in wealthy countries” (Väisänen et al., 2022). Differences in cultural and environmental factors between the country of origin and the host country, perceptions of healthcare services and a general lack of knowledge about the host country’s healthcare system may explain these disparities.

This study uses data from the permanent demographic sample (EDP). The EDP is France’s largest socio-demographic panel. It is compiled and regularly updated by INSEE (the French Institute for Statistics and Economic Studies) from a number of files, including birth and death certificates and census data. Eligibility for the sample is based on date of birth. Prior to 2004, the sample consists of individuals born between 1 and 4 October, and from 2004 onwards, it consists of individuals born in the first four days of the first month of each quarter. Our sample contains 687,535 births from 1990 to 2020, which are representative of all births in France. The number of deaths under age five in the sample is 3024. The implementation of the longitudinal analysis is heavily based on the tools developed by Bocquier et al. (2019) for longitudinal data management and analysis within the framework of the event history analysis method in Stata.

To provide our dataset with a longitudinal structure, we have converted from a wide format (individual-level data) to a long format, with each row representing one follow-up period per individual. This person-period file structure documents each event experienced by an individual in a single entry, containing the individual’s identifier, birth date, event type, event date and residency status (binary variable indicating whether or not the individual lives in France at that time). Key residence events, or events that modify the residence status, include births, deaths and end-of-observation.

To ensure temporal ordering and avoid simultaneity (a time gap should separate each event from the next), we have converted dates to milliseconds, assigning distinct times for each event type. This is made possible by the time precision of the data or the precision of the recorded dates, which are in day-month-year format. For instance, birth dates are set to 06:00:00, while death dates are set to 18:00:00. This avoids any confusion between individuals who were born and died on the same day. Thus, babies who died on the same day they were born will always die 12 hours after they were born. All individuals are right-censored as of 31 December 2020, 00:00:00, ensuring a uniform end date across the dataset.

The outcome variable is death occurring before the child’s fifth birthday. This is the probability of survival to the age of five (5q0). The independent variable is the parental origin measured by the country of birth of the parent (father or mother). Six categories of origin are identified: Sub-Saharan Africa, North Africa, Eastern Europe, other European regions, Asia and an “other” group including America, Oceania and other regions. These categories are compared with French-born or native mothers.

The mixed parenting variable allows for the distinction of four categories of children: those born to two immigrant parents, those born to two native parents, those born to an immigrant mother and a native father and a final category of children born to an immigrant father and a native mother. For simplicity, the last two categories are considered to indicate mixed parenting, given the methodological ambiguity surrounding the measurement of mixed parenthood or mixed unions (Rodríguez-García, 2015). For greater precision on this variable, we have combined it with migration origins.

By focusing on the variable associated with the support of the father during the child’s birth, we can use the availability of the father’s date of birth on the birth certificate. In other contexts, this has been used as a proxy for measuring paternal recognition (Castro-Martin, 2010). While this variable may capture a certain level of support, we think it is insufficient to encompass all dimensions, albeit indirectly. Hence, we combine it with another variable indicating whether the child’s name includes only the father’s name, only the mother’s name or both parents’ names. This enables us to construct a comprehensive variable representing the father’s support. We assume that if the father’s name and date of birth are listed on the birth certificate, this indicates that the father provided some level of support to the child at the time of birth. Consequently, we derive a binary variable indicating whether or not the father was providing support to the child.

Our methodology for determining the proportion of immigrant children born within the municipality (PICM) is based on the approach used by Muchow and Bozick (2022) to quantify social capital through co-ethnic proximity. Indeed, immigration patterns and the distribution of immigrant communities can vary significantly across municipalities. Certain municipalities may have higher concentrations of immigrants, potentially creating distinct community dynamics, including stronger or weaker social support systems. By using the municipality as the spatial unit, the analysis is sensitive to these local variations in immigrant population density and the associated effects on social capital and access to resources. In addition, this proportion may be specific to the years preceding each birth or to the years that each child is exposed to the risk of dying.

Thus, by aggregating the data from the EDP at the municipal level, taking into account the year of exposure to the risk of death for each child, we obtain the PICM indicator as follows: P I C M = n N (1) where n is the number of births to immigrant women in the municipality (of birth) during the year of birth and N is the total number of births in the same municipality in the same year.

In order to assess the impact of migration origin on under-five mortality (U5M), the Weibull model is used. This model is chosen because of the similarity of the curve of the U5M hazard function to that of a probability density function of a Weibull distribution (Figure S.1, Supplementary material, available online at https://doi.org/p-fdgp-bgc3). The Weibull distribution, which belongs to the family of parametric probability distributions along with the normal and exponential distributions, offers some distinct advantages. As a properly specified parametric model, it helps to address the statistical challenges that arise from limited event occurrences (Schöley, 2019), such as the deaths occurring before one year.

Three models are implemented. We include the logit of PICM in the first (M1) and second (M2) models to analyse the effect of a 1% increase in PICM from PICM = 0.5 on the risk of infant death, with mixed parenting and mixed parenting by region of origin as the main independent variables. According to Ferrari and Cribari-Neto (2004), applying a logit transformation to bounded continuous variables helps to avoid issues related to constant marginal effects in standard linear regression models. The transformation ensures that predictions remain within the [0,1] interval while making the dependent variable unbounded, thus facilitating estimation. We estimate the third model (M3) by including the interaction terms between PICM and migration origins while using PICM as a categorical variable. This allows us to distinguish a low proportion (PICM−) when PICM is less than 0.5 and a high proportion (PICM+) when PICM is equal to or greater than 0.5.

Alongside these variables, we adjust for several other infant and parental characteristics using the propensity score method. These include the child’s sex, multiple birth and region of birth in France, distinguishing between children born in mainland France and those born in overseas territories. The same applies to variables related to parents such as the age of the mother or the father at the child’s birth; the father’s socio-professional category (SPC), ¹ grouped as higher (SPC+) vs lower (SPC−); and parental marital status, indicating whether the parents are or are not married. The mother’s SPC is not considered due to the poorer quality of this variable (26% missing values), which is linked in particular to the uncertainty of the “inactive” and “unknown” categories, which are over-represented at the time of childbirth, when many women are on maternity leave (Niel, 2011).

These variables are introduced into the models in the form of a weighting variable used to balance children of immigrant mothers (CIM) and children of native mothers (CNM) on these characteristics using the propensity score method (Rosenbaum and Rubin, 1983). For more information on this method, please see the online supplementary material.

Figure 3 presents results of the unadjusted (before weighting) and adjusted (after weighting) models

When the hazard ratio is compared before and after weighting, mortality inequalities persist between CIM and CNM. The results with the adjustment for covariates reveal a higher U5M among children born to immigrant mothers, except for those from Europe and Asia. Specifically, children of Sub-Saharan African mothers are twice as exposed to the risk of U5M as children of natives. For children of North African mothers, the risk is 1.18 to 1.57 higher, while for children of the group including America, Oceania and other regions, the risk is 1.55 to 2.55 higher. These persistent disparities, despite the adjustment of covariates, are likely attributable to migration origin (Ma, 2008), indicating its central role in the shaping of U5M inequalities independent of other characteristics, and thus confirming our first hypothesis ( H 1 ).

Table 2 displays the results of three Weibull models (M1, M2 and M3) in which child and parent characteristics are used as control variables.

In terms of the impact of mixed parenting on U5M, its protective effect has not been shown. The risk of U5M for children born to mixed parents remains high compared to that for children born to two native parents. Even when distinguishing between the regions of origin, the mortality risk remains higher, although it decreases for children born to mixed parents, as shown in M2. This result, which does not prove the protective effect of being born to mixed parents, lends support to the second hypothesis ( H 2 ).

Our results provide support for the association between paternal support and the proportion of immigrant children in the municipality (PICM), both independently and with potential interactions with migration origins. These models reveal that paternal support during childbirth plays an essential role in reducing under-five mortality. These results, which are consistent with the third hypothesis ( H 3 ), further underline the importance of paternal involvement in child health.

In terms of the impact of the PICM, the results reveal a remarkably significant association. Specifically, being born in a municipality with a high (vs low) proportion of immigrant children significantly increases under-five mortality (U5M). Through the interaction effect between the proportion of immigrant children and migration origins in M3, it is evident that children of French natives born in a municipality with a high proportion of immigrant children exhibit higher U5M. The results indicate that excess mortality among children of women from Sub-Saharan Africa persists regardless of the type of municipality of birth, whereas excess mortality among children of women from North Africa is only observed among those born in a municipality with a low proportion of immigrant children.

By comparing the weighted and unweighted coefficients, the results indicate that the weighted coefficients partially correct the estimation biases observed in the unweighted coefficients, with greater precision in parameter estimation. This is demonstrated, for example, by a narrower confidence interval in the weighted models than in the unweighted models.