Maternal stress, child behavior and the promotive role of older siblings

Among the most studied risk factors for child health and development are those linked to parental stress [8, 9]. Depending on the characteristics of the stressors (e.g. their intensity, duration or predictability), and on the individual life histories (e.g. early exposure to stress [17, 18]), stressors can trigger adaptive and/or maladaptive responses, allowing individuals to optimally respond to potential threats and/or triggering the disruption of stress coping mechanisms, respectively [19]. When stressors are too intense, or too long, individuals may not manage to cope with them, i.e. the stress response system may remain activated and there can be important negative consequences on individual health [20].

Several stressors can activate the parental stress response system [21‐26], including acute disasters (e.g. terroristic attacks or droughts), psychological and physiological changes linked to maternity, work-related stress and even physical and emotional challenges experienced during everyday life [1, 21, 22, 27]. Parents, for instance, show higher levels of stress when they receive little social support, when their offspring experiences health or behavior problems, in case of conflict among family members, or in the presence of enduring socio-environmental stressors, like deprived environments [23‐26]. Individuals who are satisfied with the environment they live in, in contrast, may experience lower stress levels: a safe neighborhood characterized by high environmental quality (e.g. in terms of air quality), the presence of social areas, meeting places, shops and other infrastructures, for example, may all contribute to increase neighborhood satisfaction and well-being [28‐30] and hence decrease individual stress levels [31]. Similarly, individual stress levels may be lower when the quality of social relationships with neighbors is high, as neighbors can contribute to social integration and individual well-being (e.g. [32, 33]).

When parents are stressed, children have a higher risk of developing social, emotional, behavior and cognitive problems [8, 9, 34‐36]. Maternal stress can have a negative impact on offspring development already in the utero [37‐39], and even before conception [40]. Prenatal stress may have adverse effects on fetal development through an increase of stress hormones during pregnancy [41, 42], placental inflammation [43, 44], or by altering epigenetic regulation through changes in mRNA expression and DNA methylation [45, 46]. The consequences for children can be substantial. Prenatal maternal stress can modify the fetal immune system [47], affect birth outcome [48, 49] and lead to several diseases during childhood, like obesity or wheezing [50, 51].

Prenatal stress can also have negative effects on offspring psychosocial development [52‐54], as well as brain and cognitive development [52, 55‐57]. Moreover, stress during pregnancy might increase the likelihood of behavior problems during childhood [54, 58], including attention deficit or hyperactivity [53, 59], conduct disorders [53, 60], and even increase the occurrence of autism [61], depression and schizophrenia in children [62]. Prenatal stress, for instance, reliably predicts the occurrence of internalizing (e.g. anxiety, poor self-estimation) and/or externalizing (e.g. hyperactivity, aggression) behavior problems in children (e.g. [12, 58, 63‐68]), even when accounting for maternal stress levels after pregnancy (e.g. [69, 70]. These findings are not restricted to extreme prenatal stress or response to acute disasters (e.g. terroristic attacks or droughts), but they may also extend to maternal stress caused by daily challenges, pregnancy anxiety or relationship strain [21, 22].

Maternal stress can also be linked to the occurrence of offspring behavior problems in the short term. Several studies have shown a clear link between the occurrence of parental stress and children’s internalizing and externalizing behavior problems (e.g. [71‐75]). When mothers suffer from stress, for instance, children consistently show an increasing risk of developing internalizing problems, like depression and anxiety, and externalizing problems, like aggressive and rule-breaking behavior [75].

The relationship between maternal stress and child behavior problems, however, is complex, as parents and children reciprocally affect each other. Whereas children are more likely to develop behavior, social or emotional problems if their parents suffer from stress, children with these problems may in turn increase parental stress, giving rise to a cycle of negative parent–child interactions with adverse outcomes for all parties involved [8, 34, 76‐79]. Longitudinal studies and methodological approaches (e.g. cross-lagged models) that assess how parental stress and child behavior problems reciprocally affect each other are therefore necessary to disentangle to what extent maternal stress contributes to child behavior problems, and/or the other way round, and/or whether other external factors can explain the link between these two variables (see e.g. [8]).

Among the most important environmental factors that affect child development are parental and family well-being (e.g., [80‐82]). On the one hand, the family environment can entail risk factors for the child development, including parental stress (see above). On the other hand, families have an essential promotive or protective function for children’s healthy development, in the face of aversive conditions. Although there is no consensus in literature, promotive factors are usually considered those factors that have direct ameliorative effects on child development, being linked to a general decrease in behavior problems (i.e. as main effect), whereas protective factors usually indicate those factors that buffer children from the occurrence of behavior problems in the face of risk (i.e. in interaction with other factors; [4]). A healthy social environment, for instance, can provide increased social support to parents, directly reducing their stress levels and increasing their psychological well-being (i.e. promotive function), or it can have an indirect function by mitigating the negative consequences of parental stress on children (i.e. protective function) [83, 84].

To date, several studies have shown that siblings can significantly contribute to family well-being. Siblings are an essential part of family systems [85], they usually have long-lasting relationships with each other and spend abundant time together [86‐88]. Therefore, siblings may learn to understand each other’s thoughts, emotions and intentions from early on, facilitating the development of social, emotional and cognitive skills [89‐91]. Moreover, siblings may serve an important promotive or protective function during child development, reducing the occurrence of child behavior problems and/or mitigating the negative consequences of stressful events [13‐16]. By generally contributing to a healthy social and cognitive development, positive sibling relationships may indeed reduce the likelihood that children show adjustment problems (e.g. [87]; for reviews, see [13, 14]). Children with siblings, for instance, are less likely to develop internalizing problems, even when controlling for maternal stress levels [75].

Moreover, siblings may exert a protective function for children, by reducing the risk that adverse conditions trigger the emergence of internalizing and externalizing problems [15, 16]. Positive sibling relationships, for example, can decrease the probability that children develop internalizing problems after experiencing stressful events [15]. However, the mechanisms beyond these processes are not clear. Siblings, especially older ones, may provide each other comfort and security, or they may more successfully distract each other when experiencing stress [15]. Either way, siblings may have a protective function in case of adverse conditions, increasing their ability to positively react to environmental stressors [15, 92].

Although longitudinal studies are especially important to further explore the promotive and protective function of siblings and temporally disentangle the complex interplay of parental stress and child behavior problems, such an approach has rarely been used [15]. However, it can be crucial to understand how parental stress and child behavior problems are linked to each other through time, and whether the presence of siblings provides direct or indirect benefits to children during their development. Moreover, siblings can have a different role depending on their age and gender. While older siblings are usually more prone to engage in caregiving and helping roles (and their protective/promotive function may thus be especially relevant), younger siblings are usually more likely to elicit help and care (see [91]). Furthermore, some studies have shown that older sisters are more likely to engage in caretaking and helping behavior toward siblings, as compared to older brothers [93, 94]. Therefore, separately assessing the role of older brother and sisters may be essential to understand the protective and/or promotive function that siblings may have during child development.

In this study, we used longitudinal data on maternal stress levels and child behavior to test (i) the effects of different social and environmental stressors on maternal stress levels, (ii) the dynamic relationship between maternal stress and child behavior problems during development, and (iii) the potential promotive/protective function of siblings during child development. In this way, we aimed to contribute novel findings on the complex interplay of socio-environmental stressors, maternal stress and child behavior problems, by especially focusing on the role of promotive/protective factors in this relationship. Moreover, we aimed to confirm previous findings linking maternal stress and child behavior problems—an important endeavor in psychology to increase reproducibility (e.g. [95]). Based on previous studies (see above), we made the following predictions. First, we predicted that maternal stress (which we assessed when children were aged 10) would be higher when mothers are less satisfied with the environment they live in, in terms of living standards (e.g. air quality, safety) and quality of social relationships with neighbors. Second, we predicted that higher levels of maternal stress would be linked to an increased probability of internalizing (e.g. anxiety, poor self-estimation) and/or externalizing (e.g. hyperactivity, aggression) behavior problems in children. In particular, we predicted that prenatal maternal stress would have a negative long-term effect on offspring behavior during childhood (which we assessed when children were 7, 8 and 10), whereas maternal stress and behavior problems (both assessed when children were 7, 8 and 10) would be linked to each other throughout development. As maternal stress and behavior problems during childhood may reciprocally affect each other, we further used longitudinal cross-lagged models to better disentangle the link between these variables. Finally, we predicted that the presence of older siblings would exert a protective or promotive function for children [15, 16], by buffering the negative effects of maternal stress on child behavior or by overall reducing the occurrence of behavior problems in children through their development (i.e. when children were 7, 8 and 10).

This study was based on the prospective mother–child cohort LINA (Lifestyle and environmental factors and their Influence on Newborns Allergy risk) and included longitudinal data on maternal stress levels and child behavior from pregnancy until the age of 10 years of children life. We initially screened the data for 629 mother–child pairs recruited between March 2006 and December 2008 in Leipzig, Germany, among Caucasian mothers who did not suffer from severe infectious or immune diseases during pregnancy [96]. In the analyses presented here, we only included a LINA sub-cohort of 373 mother–child pairs for which information was available on both child behavior and maternal stress levels (see below; Fig. 1). For an overview of the main characteristics of the LINA sub-cohort analyzed in this study, including measures of socio-economic status, please refer to Table S1 in Suppl. Material. Participants were assessed longitudinally with standardized questionnaires (see below) self-administered by the parents, which measured (i) maternal stress levels when the mother was pregnant (i.e. 34^th-36^th week of gestation), and (ii) both maternal stress and child behavior in three waves (i.e. when the child was 7, 8 and 10 years of age; Fig. 1). All mothers signed an informed consent.

We assessed maternal stress using a 20-item perceived stress questionnaire (PSQ [97, 98]). The PSQ is a validated questionnaire commonly used to assess how often (on a 4-point scale) mothers experience stress (e.g. worries, loss of joy, tension). For each questionnaire, we summed all the scored answers to each item (from 0 = almost never to 3 = mostly) and then divided the sum for the maximum score that could be obtained (i.e., 60), in order to obtain individual indices of maternal stress ranging from 0 to 1 (i.e., from less to more stressed; see [98]). Maternal stress was assessed longitudinally, first during pregnancy, and then in the three waves (i.e. when the child was 7, 8 and 10 years of age; Fig. 1). Internal consistency for this questionnaire was high, with Cronbach’s alpha ranging from 0.92 to 0.94 in the three waves.

Mothers were asked to assess their children’s behavior in the previous six months, at the three waves (i.e. when the child was 7, 8 and 10 years of age), using the standardized 25-item Strength and Difficulties Questionnaire (SDQ) that assesses internalizing and externalizing problems, including emotional, mental and behavioral ones [99]. More specifically, the SDQ questionnaire contained five questions for each of the following five dimensions: (i) hyperactivity/lack of attention (e.g. how easily the child is distracted), (ii) emotional symptoms (e.g. how often the child is unhappy or tearful), (iii) conduct problems (e.g. how often the child has temper tantrums), (iv) peer relationship problems (i.e. how solitary the child is), and (v) prosocial behaviors (i.e. how helpful the child is when others are hurt or upset). Given that only the first four dimensions refer to behavior problems [99], we first aggregated the four dimensions by summing the answers to the respective items that were assessed using a response format from 0 (not true) to 2 (true), obtaining a total difficulties score that ranged between 0 and 40. For modelling purposes (i.e. to model dependent variables as proportions, varying from 0 to 1), we then divided this score by the maximum score that could be obtained (i.e., 40; see [99]), so that the individual indices ranged from 0 to 1. Internal consistency for this questionnaire was only moderate (with Cronbach’s alpha ranging from 0.53 to 0.62 in the three waves), as expected given that the questionnaire includes different dimensions [99].

We further asked mothers to assess child behavior in the three waves (i.e. when the child was 7, 8 and 10 years of age), using an adapted version of the Fremdbeurteilungsbogen für hyperkinetische Störungen (FBB-HKS), which provides a 20-item external assessment of child hyperkinetic disorders [100], and thus a more specific focus on children’s externalizing problems. Mothers were provided with 18 questions (see supplementary material for the list of questions) to be assessed on a 4-point scale (e.g. how often the child seems not to be listening when being talked to). As above, we calculated a second behavioral index ranging from 0 to 1 (i.e. from less to more problematic), by summing the scored answers (from 0 to 3, with 3 always indicating higher behavior problems) and then dividing the sum for the maximum score that could be obtained (i.e. 54). Internal consistency for this questionnaire was high, with Cronbach’s alpha ranging from 0.90 to 0.91 in the three waves. Behavior problems as assessed with the SDQ and the FBB-HKS questionnaires correlated in the three waves (Spearman exact test: p < 0.001, N = 251, rho in the first wave = 0.677, in the second wave: 0.746, and in the third wave: 0.742).

In the three waves (i.e. when children were aged 7, 8 and 10), mothers were also asked for information on the number of siblings at home, their sex and age, and their own education level (on a 7-point scale, from pre-primary education to second stage of tertiary education; see [101]). In the third wave (i.e. when children were 10), we also used subjective measures to assess the potential impact of socio-environmental stressors, by asking mothers how satisfied they were with the environment they lived in, using 16 questions on a 4-point scale (i.e. with 0 meaning low satisfaction, and 3 high satisfaction). We included questions on mothers’ evaluation of their living standards, including (i) natural environment (e.g. air quality), (ii) safety of the area, (iii) presence of social areas and meeting places (e.g. playgrounds), (iv) shops and (v) other infrastructures (e.g. public transport). We thus calculated 5 indexes ranging from 0 to 1 (i.e. from lower to higher satisfaction), by summing the relative scored answers and dividing the sum for the maximum score that could be obtained. We further included 5 questions on the quality of social relationships with the neighbors, 2 on a 4-point scale (e.g. how good is the relationship to the neighbors) and 3 as polar questions (e.g. whether there are friends in the area, who are regularly met). As above, we calculated an index from 0 to 1 (i.e. from lower to higher satisfaction), by summing the scored answers and dividing the sum for the maximum score that could be obtained (i.e. 9). Therefore, these indexes were higher when mothers rated themselves as happier with their environment, and lower when they were less satisfied with it. Finally, mothers in the third wave (i.e. when children were 10) also provided information on the monthly family income (as a categorical variable from 0 to 9, with 0 being a lower income and 9 a higher one).

We used linear mixed models, generalized linear mixed models and longitudinal cross-lagged models to analyze the data. For this purpose, we prepared three data-sets, (i) to study the factors predicting maternal stress levels (Model 1); (ii) to assess the link between maternal stress (i.e. prenatal and postnatal) and child behavior problems, and the potential promotive or protective effect of siblings on child behavior problems (Models 2 and 3); and (iii) to study the dynamic relationship between maternal stress and behavior problems during child development (i.e. in the three waves; longitudinal cross-lagged models). In the first data-set we entered one line for each child (i.e. only including the third wave; N = 268; Fig. 1). We then entered the individual indices for maternal stress levels in the third wave, maternal satisfaction (i.e. with neighbors, natural environment, safety of the area, presence of social areas, shops and other infrastructures), family income (as low income has been associated to increased stress, e.g. [102]), maternal education level (as education may have a protective function against stress, e.g. [103]) and number of children in the household (as the presence of more children at home can increase parental stress levels, e.g. [104]). In the second data-set, we entered one line for each child and wave (i.e. up to three lines per child, as tested at 7, 8 and 10 years of age; Fig. 1). As some children were not tested at all ages, we ended up with N = 974 lines in this data-set. We then entered child identity, SDQ and FBB-HKS behavioral indexes, maternal stress as assessed (i) prenatally and (ii) in the three waves, the number of siblings living in the same household (i.e. the overall number, the number of older sisters and the number of older brothers), the child’s gender (because males were more likely be reported to show behavioral problems, e.g. [105]), the child’s age (because the occurrence of behavioral problems may vary through development), and maternal education level (as this may have a positive impact on child development, e.g. [106, 107]). Further controls (e.g. family income) could not be added in this data-set, as this information was not available for all waves. In the third data-set, we entered one line for each child (N = 373), including maternal stress levels and SDQ and FBB-HKS behavioral indexes for each of the three waves, and further specifying whether the child had older siblings. We never aggregated data of the three waves.

Data were analyzed using R (version 3.5.0 [108]). We first used the glmmTMB package (version 1.0.1 [109]) to build three linear mixed models and generalized linear mixed models (Models 1 to 3; Fig. 1). In order to test for the role of social and environmental factors on maternal stress levels, we run the first linear mixed model with the first data-set, using maternal stress in the third wave as the dependent variable (Model 1). As predictors, we included several measures of maternal satisfaction with the environment (i.e. relationship quality with neighbors, natural environment, safety of the area, presence of social areas, shops and other infrastructures). We further controlled for the number of children in the household and for family income (which is known to reliably predict parental stress [110]).

In order to test for the link between maternal stress and child behavior problems, as well as the potential protective effect of siblings, we ran two generalized linear mixed models with the second data-set, using the SDQ (Model 2) and FBB-HKS (Model 3) behavioral indexes in the three waves, respectively, as dependent variables (Fig. 1). In both models, we included prenatal maternal stress and maternal stress in the wave as test predictors. To assess the protective role of siblings as social buffers against maternal stress, we further included as predictor the three 2-way interactions of maternal stress in the three waves with i) the overall number of siblings at home, ii) the number of older brothers and iii) the number of older sisters, respectively, and their main terms. If the 2-way interactions were non-significant, the interaction was removed and the model was re-run by including only the single predictors, thus allowing us to test for the promotive role of siblings on child behavior problems. We also included children’s sex as predictor, as sex may also predict behavior problems in children [111]. Finally, we controlled for children’s age (as our data were collected in three different waves), and included child identity as random factor to control for lack of independence, as the same children were entered multiple times in the data-set.

As the dependent variables in Models 1 to 3 were modelled as proportions including 0 and 1, we used a beta distribution after transforming them [112]. We also z-transformed continuous test predictors (i.e. age) to facilitate convergence. We compared full models to null models (i.e. only containing control predictors and random factors) using likelihood ratio tests [113]. When full and null models significantly differed, we obtained the p values for each test predictor using the R function summary. We detected no over-dispersion, convergence or stability issues in our models. There was also no problem of collinearity in our models (maximum VIFs across all models = 2.08).

In order to test for the dynamic relationship between changes in maternal stress and changes in child behavior problems during development, we run a further set of models, using the third dataset and the lavaan package (version 4.0 [114]) to build longitudinal cross-lagged models [115] (Fig. 1). These longitudinal models allow to disentangle the interplay of two variables across waves, and thus better infer possible causal pathways linking the two variables. In particular, it is possible to assess both cross domain-relationships (i.e. the predictive effect of one variable on the other one as tested in the following wave) and autocorrelations (i.e. the predictive effect of one variable on the same variable as tested in the following wave). In these longitudinal models, we always used full information maximum likelihood for missing data to reduce bias and increase statistical power [115]. We compared models of increasing complexity (see below) using likelihood ratio tests and also reported fit indices for the individual models, including robust Root Mean Square Error of Approximation (RMSEA), robust Comparative Fit Index (CFI), robust Tucker-Lewis Index (TLI) and Standardized Root Mean Square Residual (SRMR).

To run our longitudinal cross-lagged models, we first constructed child behavior problems as a latent variable, measured by a set of two observed measurements (i.e. SDQ and FBB-HKS behavioral indexes). We established measurement invariance through time by using a multiple indicator univariate model. In particular, we established configural, metric and scalar invariance by sequentially constraining factor loadings, error terms and intercepts through time [115]. All models fit the data well (configural model with no parameter constraints: χ² = 19.28, df = 3, RMSEA = 0.130, CFI = 0.985, TLI = 0.925, SRMR = 0.024; metric model with factor loading and error term invariance: χ² = 20.09, df = 11, RMSEA = 0.066, CFI = 0.986, TLI = 0.980, SRMR = 0.037; scalar model with intercept invariance: χ² = 24.00, df = 13, RMSEA = 0.064, CFI = 0.984, TLI = 0.982, SRMR = 0.038). Crucially, there were no significant differences between models with increasing constraints (configural-metric model comparison: χ² = 7.61, df = 8, p = 0.473; metric-scalar model comparison: χ² = 4.23, df = 2, p = 0.121), suggesting scalar measurement invariance for child behavior problems through time.

After establishing measurement invariance, we tested for possible relationships between changes of maternal stress and changes in child behavior problems during the three waves (i.e. at 7, 8 and 10 years of age), using a longitudinal cross-lagged model. We decomposed each of the two variables into its score in the preceding wave and its change from the preceding to the current wave [115]. We then tested all possible relationships between maternal stress and child behavior problems: (i) stress-behavior covariance in the first wave, (ii) stress to behavior coupling, to assess whether stress scores in one wave predicted rates of changes in behavioral scores, (iii) behavior to stress coupling, to test whether behavioral scores in one wave predicted rates of changes in stress scores, and (iv) estimates of correlated changes, to assess whether stress-behavior changes co-occur after accounting for the coupling pathways (see [115] for details). In the model, we also specified an auto-regression parameter for maternal stress and another one for child behavior problems, to measure how change in each variable depended on the scores in the preceding wave. To obtain finer-graded measures of these relationships, and given the low number of waves in our model, we only assessed changes between adjacent waves [116]. Finally, we tested whether the relationships between maternal stress and child behavior problems differed for children with or without older siblings, and assessed whether this more complex model provided an increase of fit as compared to the preceding one [115].