Differential associations of maternal behavior to preschool boys’ and girls’ executive function

Abstract

Boys are more sensitive to environmental factors like parental behavior, an important predictor of executive function. This study examined whether the interaction between child sex and maternal behavior was associated with children’s executive function in a manner consistent with the vulnerability or differential susceptibility model. Participants were 146 36-month-old children and their mothers. Maternal responsiveness and negative reactivity were coded during structured mother-child interactions. Executive function was operationalized as latent self-control and working memory/inhibitory control (WMIC). Structural equation modelling supported a sex by responsiveness interaction for self-control but not WMIC. Consistent with a vulnerability model, less responsiveness was associated with poorer self-control for boys relative to girls. Boys’ self-control may be more vulnerable to the negative effect of unresponsive maternal behavior helping explain boys increased risk for externalizing behavior problems.

Executive function (EF) is a set of higher-order cognitive processes necessary for goal-directed behavior (Garon, Bryson, & Smith, 2008). EF undergoes rapid development during the preschool period (Clark et al., 2013; Wiebe, Sheffield, & Espy, 2012), and EF deficits during this time are a risk factor for the development of externalizing behavior problems (Oh et al., 2020; Sulik, Blair, Mills-Koonce, Berry, & Greenberg, 2015). Because of the significance of early childhood for EF development, identifying factors contributing to individual differences in EF during this period is critical. Maternal behavior is one such factor that plays an important role in supporting and compromising young children’s developing EF (Hughes & Devine, 2017; Valcan, Davis, & Pino-Pasternak, 2018). However, not all children are equally impacted by maternal behavior (Vrantsidis et al., 2021). Research on the interplay between children’s characteristics and parental behavior has primarily focused on children’s temperament, genotype, and stress physiology (Doom & Gunnar, 2013; Slagt, Dubas, Deković, & van Aken, 2016; Vrantsidis, Wuest, & Wiebe, 2022). However, boys are more sensitive to the effects of parental behavior on externalizing behavior problems suggesting that boys may be more sensitive to the impact of parental behavior on EF, as well (Daoust et al., 2021; Mohamed Ali, Kotelnikova, Kryski, Durbin, & Hayden, 2021; Tung, Li, & Lee, 2012). The goal of the present study was to examine interactions between child sex and maternal behavior on children’s EF. Examining this question has clinical implications for understanding pathways to the development of externalizing behavior problems, a set of disorders more common in boys than girls (Maughan, Rowe, Messer, Goodman, & Meltzer, 2004).

Dimensions of executive function

EF is often conceptualized as two interrelated but distinct constructs that differ in the contexts in which they are used (Zelazo & Carlson, 2020). One dimension is primarily used in situations high in emotional or motivational load (e.g., resisting eating a tempting immediate treat in favor of receiving a larger treat later), termed self-control. A second dimension is used in contexts high in cognitive load (e.g., overcoming an automatic response). In early childhood, this dimension of EF typically corresponds to the constructs of working memory (the ability to hold in mind and manipulate information) and inhibitory control (the ability to inhibit a prepotent response), referred to collectively as WMIC. In factor analytic studies of WMIC in 2- to 6-year-olds, a single-latent factor model accounts for these two processes as well as more complex, multi-factor models (Wiebe et al., 2011; Wiebe, Espy, & Charak, 2008; Willoughby, Blair, Wirth, & Greenberg, 2010; Willoughby, Blair, Wirth, & Greenberg, 2012; Willoughby, Kupersmidt, Voegler-Lee, & Bryant, 2011). Self-control and WMIC roughly map onto what some EF models term “hot” and “cool” EF (Zelazo & Carlson, 2020). In support of this conceptualization of EF, a two-factor model of EF, corresponding to self-control and WMIC, has been found to fit the data better than a one-factor model and as well as multi-factor models (Wiebe et al., 2015; Willoughby et al., 2011; but see Allan & Lonigan, 2011 for an exception). Self-control and WMIC also correspond to a neuroanatomical division of labor and are differentially associated with developmental outcomes in early childhood. Poor self-control, which involves ventral medial prefrontal cortex, is associated with externalizing problems, and WMIC, which involves dorsolateral prefrontal cortex, is uniquely predictive of academic performance (Kim, Nordling, Yoon, Boldt, & Kochanska, 2013; Willoughby et al., 2011; Zelazo & Carlson, 2020).

Maternal behavior and executive function

Maternal behavior is thought to be one contextual factor of particular importance for the development of children’s early emerging EF (Hughes & Devine, 2017; Valcan et al., 2018). Two dimensions of maternal behavior associated with children’s EF are responsiveness and negative reactivity. Responsive behaviors include warm, sensitive, and contingent responses to children’s behavior; emotional availability; and the match between a parent’s response and child’s behavior (Clark, Massey, Wiebe, Espy, & Wakschlag, 2019; Landry, Smith, & Swank, 2006). Negative reactivity is marked by coercive and inconsistent maternal behaviors, including over-controlling behavior, hostility, negative affect, and the use of punitive discipline strategies (Hill, Maskowitz, Danis, & Wakschlag, 2008; Karreman, van Tuijl, van Aken, & Dekovíc, 2006). More responsiveness is associated with better EF in preschoolers; while more negative reactivity is associated with poorer EF in early childhood (Lengua et al., 2021; Merz, Landry, Montroy, & Williams, 2017; Valcan et al., 2018).

Sex differences in sensitivity to maternal behavior

Multiple studies suggest that boys may be more sensitive to parental behavior than girls (Daoust et al., 2021; Mohamed Ali et al., 2021; Tung et al., 2012). Boys’ increased androgen and testosterone exposure relative to girls may play a role in their heightened sensitivity to parental behavior (Del Giudice et al., 2018; Gatzke-Kopp, 2011). Prenatal and postnatal androgen and testosterone exposure are associated with alterations in the development of the hypothalamic-pituitary-adrenal axis, neurotransmitter systems (e.g., dopaminergic system), temperament, and neural plasticity; and individual differences in these domains are associated with individual differences in sensitivity to parental behavior (Del Giudice et al., 2018; Ellis & Boyce, 2011; Moore & Depue, 2016; Slagt et al., 2016). For example, more prenatal testosterone exposure is associated with increased fear reactivity in infant boys, a construct linked to increased sensitivity to parental behavior in a recent meta-analysis (Bergman, Glover, Sarkar, Abbott, & O’Connor, 2010; Slagt et al., 2016).

Boys’ increased sensitivity may be advantageous or disadvantageous depending on their environment (Bakermans-Kranenburg & van IJzendoorn, 2011; Ellis, Boyce, Belsky, Bakermans-Kranenburg, & van IJzendoorn, M. H., 2011). As such, two competing models of the form of the interaction between child sex and maternal behaviors have been proposed: the vulnerability or diathesis-stress model and the differential susceptibility model. The vulnerability model proposes that some individuals have characteristics that make them more vulnerable to negative outcomes (e.g., psychopathology) when exposed to negative environmental factors (Caspi et al., 2003; Monroe & Simons, 1991). This model predicts that boys will have poorer EF relative to girls in the presence of negative maternal behaviors, like negative reactivity. In contrast, the differential susceptibility model proposes that individuals’ characteristics confer susceptibility to both positive and negative environmental factors rather than vulnerability to negative factors (Belsky & Pluess, 2013; Belsky, Zhang, & Sayler, 2021). Increased susceptibility is hypothesized to lead to poorer outcomes in negative environments and better outcomes in positive environments. This model predicts that boys will have better EF relative to girls when exposed to more positive maternal behaviors (e.g., high responsiveness) and poorer EF relative to girls when exposed to more negative behaviors (e.g., high negative reactivity).

In support of the vulnerability model, less positive parenting at age 3 has been linked to more impulsivity at age 5 for boys relative to girls (Amicarelli, Kotelnikova, Smith, Kryski, & Hayden, 2018). Likewise, less sensitive parenting when children are 3 years old has been linked to more omission errors on a continuous performance task for 5-year-old boys but not girls (Mileva-Seitz et al., 2015). However, sensitivity was not associated with commission errors or working memory for either boys or girls. Furthermore, negative parental behaviors, like physical discipline, have been linked to boys’, and not girls’, externalizing symptomatology, conduct problems, and emotion regulation difficulties (Daoust et al., 2021; Mohamed Ali et al., 2021; Tung et al., 2012; Wang, Wang, Wang, & Xing, 2021). Together, these results suggests that boys may be more vulnerable to negative maternal behaviors, although it is important to note that the majority of these studies did not examine positive parental behaviors.

Emerging evidence from studies of positive parental behaviors and parenting interventions suggests that boys may also disproportionately benefit from positive maternal behaviors relative to girls. For example, parents use of more elaborative narrative styles is associated with better inhibitory control in 4-year-old boys than girls (Lund et al., 2021). In addition, the Incredible Years parenting intervention has been linked to a greater decline in externalizing problems among 4- to 8-year-old boys but not girls (Chhangur, Weeland, & Belsky, 2017). Similarly, positive maternal behavior and paternal involvement are associated with fewer externalizing behavior problems for 6- to 12-year-old boys but not girls (Gryczkowski, Jordan, & Mercer, 2010). Importantly, previous research using the present cohort did not find significant interactions between maternal responsiveness and child sex on WMIC or effortful control at age 5 (Clark et al., 2019). Sex differences in the association of parental behavior to child EF may be most pronounced between ages 3 and 4 years because this is a period of particularly rapid growth and inter-individual variability in EF and impulsive responding (Wiebe et al., 2012). Further research examining the form of the interaction between maternal behavior and child sex in relation to multiple dimensions of children’s EF during a period of high inter-individual variability and with a full range of negative and positive maternal behaviors is likely to help inform research on sex differences in the pathways to the development of externalizing behavior problems via EF (Belsky & Pluess, 2009; Matthys, Vanderschuren, & Schutter, 2013; Oh et al., 2020).

The present study

This study examined whether interactions between child sex and mothers’ responsiveness and negative reactivity were associated with children’s EF at 36 months of age, and whether the forms of the interactions were consistent with the vulnerability or differential susceptibility model. Consistent with previous research (e.g., Amicarelli et al., 2018; Mileva-Seitz et al., 2015), we expected to find an interaction between sex and maternal behavior on EF. If results supported a vulnerability model, we expected sex differences only in higher-risk contexts (i. e., low responsiveness or high negative reactivity). If results support a differential susceptibility model, we expected sex differences in both higher- and lower-risk contexts (i.e., at both extremes of responsiveness or negative reactivity). We used data from the 36-month wave of the Midwestern Infant Development Study (MIDS), a cohort prospectively recruited during pregnancy, to test these hypotheses.

Methods

Participants

Participants were 146 36-month-old children (69 girls, 77 boys; M_age = 3 years 15 days, SD = 26 days) and their mothers who participated in a follow-up of the Midwestern Infant Development Study (MIDS) when children were 36 months old. MIDS is a predominantly low socioeconomic status cohort prospectively recruited to study the effects of prenatal tobacco exposure on cognitive development, consequently, women who smoked during pregnancy were overenrolled (55% of the sample compared to 23% in the general US population) (Espy et al., 2011; Tong et al., 2013; Wiebe et al., 2015). Mothers were recruited during pregnancy at two Midwestern study sites (Carbondale, Illinois, and Lincoln, Nebraska) and provided written, informed consent prior to their participation in the study. Mothers who reported binge drinking or illegal drug use, apart from occasional marijuana use, were excluded from the cohort, as were infants born preterm (<35 weeks gestational age) or with birth complications known to affect developmental outcomes (e.g., neonatal seizures). Due to funding constraints, only children at the Lincoln, Nebraska site were included in the 36-month follow-up. Dyads included in the final sample did not differ significantly from dyads lost to attrition in terms of prenatal tobacco exposure status, maternal education, ethnicity, or child sex. Family income ranged from under $10,000 to over $100,000 (Mdn = $26, 874) with 42% percent of families living at or under the poverty line, defined as an income-toneeds ratio below 1.00 (McLoyd, 1998). Parents completed between 11 and 18 years of formal education (M = 13.95). Children’s ethnic backgrounds are European American (n = 78), African American (n = 33), Latine American (n = 28), Indigenous (n = 2), and mixed ethnicity (n = 5).

Procedure

Mother-child dyads visited a developmental laboratory at the University of Nebraska-Lincoln. A trained research assistant tested children individually over three sessions, separated by approximately one week. In the first session, children completed the Disruptive Behavior Diagnostic Observation Schedule (DB-DOS), a standardized clinical observation designed to differentiate normative misbehavior in early childhood from clinically salient patterns of behavioral dysregulation (Wakschlag et al., 2008). The DB-DOS examines children’s behavior during parent and examiner social interaction contexts with parallel demands. Each context is designed to “press” for children’s behavioral dysregulation in ecologically salient situations, including in the face of challenge, frustration, compliance demands, and social play. For this study, we focused solely on the parent context. In the remaining two sessions, children completed a battery of EF tasks. Children received a small toy after each session, and mothers received a gift card, in appreciation, after completing all three sessions. Adherence to experimental protocols was maintained by regular team meetings and reviews of session video recordings. Study procedures were approved by the University of Nebraska’s Institutional Review Board.

Measures

Maternal behavior

Mother-child dyads were video recorded during the parent context of the DB-DOS (Wakschlag et al., 2008). The parent context consisted of four 5-min mother-child interactions: child compliance during a clean-up task, completing a challenging puzzle task, mother’s withdrawal of attention, and free play (Hill et al., 2008). The examiner explained the procedure to the mother before starting the context, and flip cards with instructions were provided throughout the context. The examiner rang a bell to mark transitions between tasks. Mother-child behavior was not scripted, and mothers were encouraged to act as they normally would at home.

Maternal behavior was coded using the Parenting Clinical Observation Schedule (P-COS), a structured observational measure designed to assess competent and problematic parental behaviors (Hill et al., 2008). Behaviors were coded to assess responsiveness (scaffolding, responsiveness to positive behaviors, warm affection, positive engagement, labelling, intensity of positive affect, and pervasiveness of positive affect), and negative reactivity (hostile behavior, verbally aggressive discipline, physical discipline, power struggles, emotional misattunement, intensity of angry/irritable affect, and pervasiveness of angry/irritable affect). Each item was coded globally (i.e., codes captured maternal behaviors across all four interactions) on a 4-point scale ranging from 0 (no evidence of the behavior) to 3 (high levels of the behavior). Coders watched videos twice before assigning final codes. Scores on the items assessing responsiveness were summed and used as the measure of responsive maternal behavior. Consistent with Hill et al. (2008), labelling was dropped from the measure because of poor loading on internal consistency analyses (corrected item-total r = 0.15). Scores on the items assessing negative reactivity were summed and used as the measure of negative reactive maternal behavior. Because scores on the intensity of angry/irritable affect and pervasiveness of angry/irritable affect items were highly correlated (r = 0.86), scores for these two items were averaged together to form one item (Hill et al., 2008). Cronbach’s alpha is not an appropriate measure of internal consistency for ordinal data, therefore, ordinal ω total was used instead (McNeish, 2018). Internal consistency was good for responsiveness (ordinal ω total = 0.80) and adequate for negative reactivity (ordinal ω total = 0.71). The responsiveness and negative reactivity measures are correlated with self-report measures of parental behavior, such as Coping with Children’s Negative Emotions (rs = −0.31–0.19, ps < 0.05; Hill et al., 2008).

Coding was completed by two coders trained to reliability (at least 80% exact agreement on each item) by a master coder involved in the development of the coding scheme. Ongoing reliability was maintained through weekly coding meetings and disagreements were resolved by consensus. Twenty percent of the videos were double coded by all three coders. Inter-rater reliability was high for these videos (responsiveness: mean κ = 0.86, exact agreement: 95%; negative reactivity: mean κ = 0.88–0.91, exact agreement: 97–98%).

Executive function

Children completed a battery of seven EF tasks (Wiebe et al., 2015). Brief descriptions of each task are presented in Table 1. Administration, psychometric properties, scoring, and validation of the EF battery are described in more detail elsewhere (Wiebe et al., 2011; Wiebe et al., 2015). Confirmatory factor analysis supported a two-factor measurement model for EF consisting of self-control and WMIC (see Wiebe et al., 2015). A two-factor model fit the data well (χ²(19) = 26.70, p = 0.12, RMSEA = 0.05, CFI = 0.96, SRMR = 0.05) and provided a better model fit than a one-factor model (Δχ²(1) = 36.84, p < 0.05) and more parsimony than a three-factor (self-control, WM, and IC) model (Δχ²(2) = 2.21, p = 0.33). Two tasks, Goody Shelf and Snack Delay, loaded on a self-control factor. Five tasks loaded on a WMIC factor, including Delayed Alternation, Nebraska Barnyard, Big-Little Stroop, Preschool Go/No-Go, and Shape School-Inhibit Condition. All factor loadings were statistically significant and standardized loadings ranged from 0.43 to 0.97 for the self-control factor and from 0.42 to 0.56 for the WMIC factor. Tests of measurement invariance supported invariance by child sex and prenatal tobacco exposure status at the configural, metric, and scalar levels (see Wiebe et al., 2015). To identify the configural model, it was necessary to fix the residual variance for the Snack Delay summary score indicator to zero for boys. If metric and scalar invariance are supported, then one can validly compare the means of latent variables across groups (Kline, 2015).

Table 1.

Executive function task descriptions.

Task	Description	Dependent Measure
Self-control
Goody Shelf	Children were instructed that they could look at, but not touch, a shelf containing appealing toys over a 5-min delay.	Summary score representing child noncompliance. Each instance of toy touching was scored between 1 (brief touches) and 3 (sustained touches where the child was resistant to examiner prompts).
Snack Delay	Children were instructed to keep their hands on a placemat marked with two handprints and stand still in front of M&M candies placed under a transparent cup during a 4-minute delay.	Summary score representing child compliance in 5-s intervals until either the child ate the snack, or the task ended. Children received up to 3 points for standing still, keeping their hands on the mat, and remaining silent. Task success: whether the child ate the snack during the delay period.
Working Memory/Inhibitory Control
Delayed Alternation	Children searched for a hidden food reward in one of two locations; the examiner hid the reward out of the child’s view, changing the location after each correct response.	Proportion of correct responses.
Nebraska Barnyard	Children listened to sequences of animal names then pressed colored buttons corresponding to the names in order on a touch screen.	Summary score calculated by summing the proportion of correct trials at each span length.
Big-Little Stroop	Children named small pictures of everyday objects, embedded within larger pictures that matched (congruent trials) or mismatched (incongruent trials) the small pictures.	Proportion of correct responses on incongruent trials.
Preschool Go/No-Go	Children pressed a button on a button box to catch fish (75% of trials), but not sharks (25% of trials).	d’ (d-prime; the standardized difference between the hit rate and false alarm rate).
Shape School Inhibit Condition	Children named the color of a cartoon shape character when the character had a happy face but remained silent when it had a sad face (inhibit trials).	Proportion of correct responses on inhibit trials.

Covariates

Because mothers who smoked during pregnancy were overenrolled in the MIDS cohort, prenatal tobacco exposure status was included as a covariate. Prenatal tobacco exposure status was dummy coded (exposed = 1; not exposed = 0). Mothers completed timeline-follow-back interviews about daily smoking at two points during pregnancy and shortly after their child’s birth. Exposure status was verified by assaying maternal urine and infant meconium for cotinine, a metabolite of nicotine (Espy et al., 2011). Household socioeconomic status was included as a covariate as it is a robust and important predictor of child EF (Lawson, Hook, & Farah, 2018). Highest household degree was used as a measure of socioeconomic status. Mothers reported on each parent or caregiver’s highest degree. For two-parent households, the highest level of education was used as the measure of parental education. For single-parent households, mother’s highest degree was used. Maternal mental health was also included as a covariate because it is sometimes related to child EF (Vrantsidis, Clark, Chevalier, Espy, & Wiebe, 2020). Mothers completed the Brief Symptom Inventory (BSI; Derogatis, 1993), a self-report measure assessing psychological distress along nine dimensions, such as depression and hostility. Scores on each question were summed and divided by the number of questions answered to create the Global Severity Index, which was used as the measure of psychological distress.

Analytic strategy

Univariate distributions for all dependent variables were examined for non-normality and outliers. In total, 5% of the data were missing, ranging from <1% (maternal psychological distress) to 34% (Shape School); see Table 2 for more details. To minimize the number of participants lost due to missingness on exogenous variables, missing BSI scores (n = 1) were replaced with BSI scores from when children were 6 months old. Results were unchanged when the participant with missing data was dropped from the analyses; therefore, the imputed BSI score was retained. Missing data on endogenous variables were dealt with using full information maximum likelihood estimation using an expectation maximization algorithm. Because FIML assumes that data are missing at random a series of logistic regression models were computed to test whether missingness on endogenous measures was related to demographic characteristics. Missingness was unrelated to prenatal tobacco exposure status, maternal education, child ethnicity, and child sex (ps > 0.05).

Table 2.

Descriptive statistics for the measures of maternal behavior, executive function, and covariates.

Construct	N	M	SD	Range
Responsiveness (composite score)	146	13.13	2.82	0.00–18.00
Negative reactivity (composite score)	146	0.88	1.29	0.00–7.00
Child sex (% male)	146	53%
Goody Shelf (rule-breaking)	140	3.54	7.26	0.00–33.00
Snack Delay (movement score)	135	50.58	32.24	3.00–117.00
Snack Delay (ate treat)	135	0.33	0.47	0.00–1.00
Delayed Alternation (accuracy)	140	0.50	0.18	0.00–0.94
Nebraska Barnyard (composite score)	134	3.31	1.74	0.58–8.06
Big-Little Stroop (conflict trial accuracy)	133	0.25	0.25	0.00–1.00
Go/No-Go (d’)	140	0.54	1.00	−1.37–3.12
Shape School inhibit (accuracy)	97	0.36	0.27	0.00–1.00
Prenatal tobacco exposure status (% exposed)	146	55%
Parental education (years)	146	13.95	1.57	11.00–18.00
Psychological distress (composite score)	146	0.51	0.53	0.00–3.64

Structural equation modelling (SEM) was conducted using MPlus 7.3 (Muthén & Muthén, 2017). Model fit was assessed using the chi-square (χ²) statistic, root mean square error of approximation (RMSEA), comparative fit index (CFI), and standardized root mean square residual (SRMR). Values indicating good fit were <0.06 for the RMSEA, 0.95–1.00 for the CFI, and <0.08 for the SRMR (Hu & Bentler, 1999). Values indicating adequate model fit were a RMSEA between 0.06 and 0.08, 0.90–0.94 for the CFI, and <0.10 for the SRMR (Hu & Bentler, 1999; Kline, 2015). The chi-square difference (Δχ²) test was used to compare nested models (Kline, 2015). When the test was significant (p < 0.05), the less constrained model was retained; otherwise, the more parsimonious model was favored.

The model that best fit the data was established using a backward trimming approach (see Kline, 2015; Little, 2013). First, responsiveness and negative reactivity were mean-centered, and product terms representing the interactions between maternal behavior and sex were calculated. Child sex was dummy coded as 0 (girls) and 1 (boys). Second, a structural regression model was tested by regressing EF on maternal behavior, child sex, covariates, and the above-mentioned interaction terms. Third, non-significant paths (p > 0.10) were trimmed sequentially, beginning with higher-order interaction terms and starting with the highest p-values (Kline, 2015).

To test for significant interactions, paths from each interaction term retained in the best-fitting model were fixed to zero one at a time. Interactions were deemed significant if the fixed model resulted in significantly poorer model fit based on a significant chi-square difference test. Significant interactions were probed based on Roisman et al.’s (2012) recommendations. First, we graphed significant interactions. Values of maternal behavior did not range two standard deviations above and below the sample mean (Roisman et al., 2012); therefore, we plotted the observed data and bound the data at +/− two standard deviations when the observed data extended beyond this range. Next, we conducted a regions of significance analysis, using the Johnson-Neyman technique (Johnson & Fay, 1950), to test whether potential interactions were consistent with the vulnerability or differential susceptibility model. A regions of significance analysis identifies the values of maternal behavior at which sex differences are significant (Hayes, 2018). Finally, using tools developed by Fraley (2012), we examined the Proportion of Interaction (POI). The POI index compares the area between the two regression lines above and below the crossover point on the interaction plot. A vulnerability model is supported if the area is largely on the side of the crossover point associated with risk (a POI closer to 0). A differential susceptibility model is supported if the areas are roughly equal (a POI closer to 50).

Results

Descriptive statistics

Descriptive statistics for the main variables used in the analyses are presented in Table 2, and correlations among these variables and latent variables are presented in Table 3. Boys and girls did not differ in their exposure to responsiveness (t(144) = 0.59, p = 0.56) or negative reactivity (t(144) = −1.38, p = 0.17). Latent self-control (Δχ²(1) = 1.86, p = 0.16) and WMIC means (Δχ²(1) = 1.68, p = 0.20) did not differ significantly between boys and girls.

Table 3.

Correlations between maternal behavior, covariates, executive function, and latent executive function factors.

Measure	1.	2.	3.	4.	5.	6.	7.	8.	9.	10.	11.	12.	13.	14.
1. Responsiveness	–	−0.35**	−0.05	−0.02	0.23**	−0.21**	0.07	0.09	0.14	−0.02	0.03	0.05	0.00	0.17+
2. Negative reactivity		–	0.11	0.21*	−0.16+	0.01	−0.31**	−0.05	−0.10	−0.09	−0.05	−0.07	−0.15+	−0.09
3. Child sex			–	−0.05	0.15+	−0.04	−0.15+	−0.11	−0.06	0.06	0.03	0.11	0.10	0.07
4. Prenatal tobacco exposure				–	−0.15+	0.11	−0.18*	−0.26**	−0.22*	0.00	−0.06	−0.13	0.01	−0.12
5. Parental education					–	−0.17*	0.09	0.05	0.09	0.15+	0.25**	−0.01	0.17*	0.14
6. Psychological distress						–	−0.24**	−0.26**	−0.24**	−0.20*	−0.20*	−0.09	−0.14+	−0.12
7. Good Shelf (reversed)							–	0.41**	0.36**	0.14	0.13	0.03	0.19*	0.09
8. Snack Delay (movement score)									0.75**	0.13	0.25**	0.30**	0.11	0.22*
9. Snack Delay (ate treat; reversed)									–	0.06	0.22*	0.12	0.09	0.12
10. Delayed Alternation										–	0.30**	0.23**	0.20*	0.14
11. Nebraska Barnyard											–	0.20*	0.35**	0.06
12. Big-Little Stroop												–	0.24**	0.30**
13. Go/No-Go													–	0.29**
14. Shape School														–
15. Latent self-control	0.07	−0.07	−0.12	−0.28**	0.07	−0.27**
16. Latent WMIC	0.06	−0.19+	0.14	−0.10	0.32**	−0.31**

Note. WMIC = working memory/inhibitory control.

⁺p < 0.10;

^*p < 0.05;

^**p < 0.01.

Latent WMIC and self-control are significantly correlated (r = 0.39, p < 0.001).

Regression results

Self-control and WMIC were regressed on responsiveness, negative reactivity, sex, the maternal behavior × sex interactions, and covariates. Non-significant paths from the interaction terms were trimmed sequentially. Of the interaction terms, only the responsiveness × sex effect for self-control was retained in the final model (Δχ²(1) = 7.26, p = 0.01).

The final model is presented in Fig. 1. Model fit to the data was good (χ²(62) = 73.79, p = 0.15, RMSEA = 0.04, CFI = 0.95, SRMR = 0.06). The final model accounted for 20% of the variability in self-control. For self-control, the main effects of responsiveness and sex were marginally significant and qualified by a significant two-way interaction. The main effect of negative reactivity was not significant. Of the covariates, prenatal tobacco exposure status and maternal psychological distress were associated with poorer self-control. The effect of parental education was not significant.

Fig. 1.

Path diagram illustrating the main and interaction effects of maternal behavior and child sex on self-control and working memory/inhibitory control (WMIC). Both unstandardized and standardized (in parentheses) parameters are presented; error variances are not shown. ⁺p < 0.10; *p < 0.05; **p < 0.01.

To determine whether the sex by responsiveness interaction was consistent with a vulnerability model or differential susceptibility model, we tested for regions of significance and calculated a POI index. Consistent with a vulnerability model, boys’ and girls’ self-control significantly differed at values of responsiveness below 12.57 (0.20 standard deviations below the mean; see Fig. 2). Sex differences in self-control did not emerge when responsiveness was high. The POI index was calculated using values of responsiveness ranging from two standard deviations above the mean to two standard deviations below the mean. The POI index further supported a vulnerability model: 78% of the area between the regression lines was below the crossover point, indicating that the interaction effect was largely attributable to sex differences in self-control for children with less responsive mothers.

Fig. 2.

Relation between responsive maternal behavior and self-control by child sex. The region of significance indicates the value of responsiveness (12.57, 0.20 standard deviations below the mean) below which boys’ and girls’ self-control significantly differs.

The final model accounted for 22% of the variability in WMIC. There was no support for significant interactions between sex and maternal behavior on WMIC, as neither interaction term was retained in the final model. More negative reactivity was marginally associated with poorer WMIC. Neither responsiveness nor sex were significantly associated with WMIC. Of the covariates, maternal psychological distress was associated with poorer WMIC, and higher parental education was associated with better WMIC. Prenatal tobacco exposure was not associated with WMIC.

Discussion

This study examined whether interactions between child sex and maternal behavior on child EF were consistent with the vulnerability or differential susceptibility model. We adopted a multi-faceted approach to examine these pathways, assessing multiple dimensions of maternal behavior (responsiveness and negative reactivity) and EF (self-control and WMIC) using state-of-the-art, direct assessment methods. Results were partially consistent with a vulnerability model. Less responsiveness was associated with poorer self-control for boys relative to girls. Responsiveness was not associated with WMIC for either boys or girls. The interactions between sex and negative reactivity on EF were not significant. Further, the main effects of responsiveness, negative reactivity, or sex on children’s EF were not significant.

In partial support of the predictions of the vulnerability model, boys had poorer self-control relative to girls when responsiveness was low. Results were not consistent with the differential susceptibility model as sex differences in self-control did not emerge when responsiveness was high. This finding is consistent with two previous longitudinal studies that found that boys were more impulsive than girls at age 5 when positive parenting was low at age 3, providing additional support for the notion that boys’ self-control is more vulnerable in higher-risk contexts than girls and extends these results to a younger age – 3 years – a period of particularly rapid growth and inter-individual variability in EF (Amicarelli et al., 2018; Mileva-Seitz et al., 2015; Wiebe et al., 2012). Nonetheless, it is important to note that this finding is not always consistent as previous work with the MIDS cohort at age 5 did not find a significant interaction between child sex and responsiveness on effortful control (Clark et al., 2019). Support for an effect of maternal behavior at age 3, but not age 5, on children’s self-control is consistent with the suggestion that boys may be particularly vulnerable to adverse environmental experiences that occur at younger ages given sex differences in rates of neurophysiological and psychosocial maturation (Etchell et al., 2018; Schore, 2017). Further research is needed examining whether sex differences in vulnerability to maternal behavior on EF changes across the transition to middle childhood. It is also possible that these results reflect differences in the measurement of EF as Clark et al. (2019) used questionnaires to assess effortful control. Behavioral tasks may be more sensitive to fine-grain differences in performance compared to questionnaires which are thought to measure complex, real world-behaviors (Friedman et al., 2020).

The results of the present study are also consistent with a large body of research suggesting that boys are more vulnerable than girls to the detrimental impact of less positive and more negative parental behaviors on the development of externalizing behavior problems (Daoust et al., 2021; Mohamed Ali et al., 2021; Tung et al., 2012). Low responsiveness in early childhood is a well-established risk factor for the development of externalizing behavior problems (Wakschlag & Hans, 2002). If boys are more vulnerable relative to girls to poorer quality maternal behavior, low responsiveness could serve as a risk factor for the development of externalizing problems in boys via its’ impact on self-control. This is a potentially important developmental pathway since self-control is a powerful predictor of health and psychosocial functioning across the lifespan (Moffitt, Poulton, & Caspi, 2013). Further, interventions promoting positive parenting are associated with decreases in children’s behavior problems (Tully & Hunt, 2016). Increasing parental responsiveness in early childhood could serve as a protective factor for the development of boys’ self-control, potentially buffering against the development of behavior problems.

For WMIC, the interaction between responsiveness and child sex was not significant. While this finding was unexpected because WMIC and self-control are correlated, it is consistent with the work of Mileva-Seitz et al. (2015) who found a significant interaction between child sex and parental behavior on impulsivity but not working memory. A differential association between parental behavior and EF dimensions in boys and girls may be due to biological factors. For example, there is some evidence that males have lower basal dopamine levels than females (Laakso, Vilkman, Haaparanta, & Solin, 2002). Low basal dopamine levels are theorized to be associated with reduced reward sensitivity, which results in increased reward seeking and poorer self-control (Matthys et al., 2013). These sex differences in dopaminergic functioning suggest that boys may be vulnerable to impairments in self-control, specifically, rather than EF, more generally. Nonetheless, it is important to note that in this study, boys and girls did not significantly differ in mean latent self-control, although differences could emerge with age.

Contrary to our hypotheses and previous research (Chang, Olson, Sameroff, & Sexton, 2011; Valcan et al., 2018), the interaction between negative reactivity and sex on EF was not significant, nor was there a significant main effect of negative reactivity. Several factors may have contributed to the present study’s unexpected findings. First, an association between negative reactivity and EF may emerge as children grow older. In a longitudinal study of the effects of negative reactivity on children’s WMIC, negative reactivity was not associated with WMIC before age 5 (Cuevas et al., 2014). Similarly, Talwar et al. (2011) found a positive association between teachers’ use of physical discipline and children’s self-control in kindergarten and a negative association in Grade 1. Longitudinal research can help identify the origins of these pathways and developmental changes in the association between negative reactivity and EF. Second, the present study only examined the association between mother’s negative reactivity and children’s EF. Fathers’ use of physical discipline is more strongly correlated with behavior problems for boys than girls; while the lowest correlations between physical discipline and behavior problems are seen in motherson dyads (Deater-Deckard & Dodge, 1997). It is possible that boys are more sensitive to their father’s negative reactivity than their mother’s. This suggests a need to consider the role of fathers in children’s EF development in larger samples that can account for the multi-faceted nature of these pathways.

Findings from the current study should be interpreted within the context of the study’s strengths and limitations. The key strengths of this study were the use of developmentally sensitive, multi-dimensional, direct assessment methods of both maternal behavior and EF. The use of a latent variable approach to EF measurement also improved construct reliability, combating the task impurity problem. Individual EF tasks are often unreliable EF measures because task performance reflects variations in both EF abilities and the basic abilities (e.g., motor abilities) required to complete the task (Miyake et al., 2000). The use of a latent variable approach resulted in a model with good fit to the data, likely improving our ability to detect any potential associations between maternal behavior and child sex on EF. The use of a latent variable approach, compared to a composite score, may introduce bias when the amount of shared variance among indicators is low (Camerota, Willoughby, & Blair, 2020). The maximal reliability of the self-control and WMIC factors were .93 and .63, which is, respectively, higher than and consistent with the adult EF literature (Willoughby, Holochwost, Blanton, & Blair, 2014).

The present study also had several limitations. First, the sample size of the current study was relatively small for structural equation modelling (Kline, 2015). A larger sample size would increase the precision of estimates. Replication of the present findings using longitudinal designs is warranted before strong conclusions about sex differences in the association between maternal behavior and children’s EF can be drawn. Second, the pattern of results was modest with only one of eight pathways from parental behavior reaching statistical significance. Possible reasons for not finding a more robust effect of parental behavior include low statistical power, the restricted range of the maternal behavior measures, and differences in the measurement of EF across studies. Third, maternal behavior and children’s EF were measured concurrently precluding strong conclusions about the directions of the relations between maternal behavior, child sex, and EF. Although we did not find significant sex differences in EF, two meta-analyses found that boys have significantly poorer inhibition and self-control than girls (Else-Quest, Hyde, Goldsmith, & van Hulle, 2006; Silverman, 2003). Children with poorer EF may elicit harsher, less responsive maternal behavior (Eisenberg, Taylor, Widaman, & Spinrad, 2015). The cross-sectional design of the current study meant we could not test for bidirectional associations between maternal behavior and children’s EF. Fourth, our sample represented a slightly more clinical population as 55% of the children were exposed to tobacco prenatally. This may limit the generalizability of our findings to children at lower risk for poorer EF as maternal behavior may have a greater impact on EF in children from higher risk households (Rochette & Bernier, 2014).

Not all children are equally impacted by maternal behavior (Clark et al., 2019; Vrantsidis et al., 2021). Boys are thought to be more sensitive to parental and family factors than girls (Del Giudice et al., 2018), suggesting that child sex and maternal behavior may interact to shape children’s early EF. In support of this argument, less responsive maternal behavior was associated with poorer self-control for boys compared to girls. The present study’s findings suggest that adverse maternal behavior, as characterized by low responsiveness, is an important risk factor for the development of poorer self-control in boys. Given that boys are at increased risk for developing disorders characterized by EF deficits (Matthys et al., 2013; Maughan et al., 2004), identifying the mechanisms driving the association between low responsiveness and poorer self-control in boys may help with the development of interventions aimed at improving EF and decreasing their risk for developing disorders, such as behavior problems.

Data availability

The authors do not have permission to share data.