Parenting Behavior and Child Language: A Meta-analysis

Child language in the current study refers to either receptive or expressive language assessed via parent-report questionnaires (eg, MacArthur Communicative Development Inventory³²) or standardized assessments (eg, Peabody Picture Vocabulary Test³³). In terms of observed parenting constructs, sensitive responsiveness refers to a parent’s ability to perceive and interpret the child’s signals and cues and to respond to those cues and signals promptly and appropriately.¹⁴ Warmth refers to caregiver physical affection or their positive affective quality during contact and involvement with the child.

Searches were conducted by a medical librarian in Medline, Embase, PsycINFO, Web of Science, and Dissertation Abstracts up to June 2017 (Supplemental Information). Both database-specific subject headings (when available) and text word fields were searched for the concepts of “language,” “parents,” and “children.” Synonymous terms were first combined with the Boolean “OR.” These 3 concepts were then combined with the Boolean “AND.” In all databases, truncation symbols were used in text word searches when appropriate to capture variations in spelling and phrasing. References of all included studies were also searched. No language or date limits were applied.

Inclusion criteria were as follows: (1) a typically developing sample; (2) a measure of child language, in English, including measures of receptive (eg, understanding of words) or expressive language (eg, total utterances); (3) an observational measure of sensitive responsiveness or warmth; and (4) a statistic that could be transformed into an effect size. If effect sizes could not be calculated from the statistics provided, the corresponding author was contacted for this information. Exclusionary criteria included samples of children with diagnostic language delays, intellectual disabilities, deafness (in parents or children), hearing loss or middle ear disease, autism spectrum disorders, speech anomalies, and brain injuries. Intervention studies were only included if they provided pretest (or baseline) estimates of parenting and language.

All titles and abstracts emerging from the search strategy were reviewed to determine eligibility criteria. An experienced systematic reviewer (S.M.) trained a group of coders on the title and abstract review. When titles and abstracts were insufficient to determine eligibility criteria, full texts were retrieved.

A data-extraction coding protocol was developed to extract effect sizes as well as the following moderator variables: (1) SES, ≥80% of sample considered middle to upper or low SES (mixed SES = diverse); (2) child sex (percentage of boys); (3) child age (in months at language measurement); (4) location (home versus laboratory) and duration of parenting observation (in minutes); (5) sample type (term versus preterm); and (6) study design (cross-sectional versus longitudinal). Approximately 10% of the articles meeting inclusion criteria were double coded throughout the data extraction process. Reliability on continuous measures based on Pearson correlations were r >0.93. Reliability on categorical measures based on κ estimates ranged from 0.64 to 1.0. Any disagreements were resolved with the expert coder.

An assessment of the methodologic quality and validity of each study and their risk of bias was conducted by using a 13-point quality assessment tool adapted from the National Institutes of Health Quality Assessment Tool for Observational Cohort and Cross-Sectional Studies³⁴ (Supplemental Table 3). Studies were assigned a score of 1 if the criterion was sufficiently met (Supplemental Table 4). A total score was then calculated by summing criteria with possible scores ranging from 0 to 13. Any study deemed to have low methodologic quality (scores ≤5) were removed from analyses.

We identified overlapping studies and developed a protocol for sample selection to ensure inclusion of independent effect sizes for both meta-analyses. If there were multiple studies published based on the same data set, we selected the study with the largest sample size, readily available statistics, and psychometrically sound measurement. If a single study assessed sensitive responsiveness or warmth, both effect sizes were extracted and their associations with child language were examined in separate meta-analyses. If a single study provided a measure of receptive and expressive language, these 2 effect sizes were pooled to provide the most global and representative assessment of child language. If studies assessed language at multiple time points, the latest time point of language was selected to capture the most developed language skills.^35,36 The most temporally distant effect size for parenting and child language was selected when multiple assessments of parenting behavior were provided. Finally, if studies reported multiple effect sizes for term versus preterm infants and the samples did not contain overlapping participants, they were entered into the meta-analysis separately.

Outlier detection was examined in SPSS (version 23.0; IBM SPSS Statistics, IBM Corporation) by using visual inspection of box plots. Studies with an effect size value ±3 SDs from the mean were considered outliers.

Comprehensive Meta-Analysis Software (version 3.0)³⁷ was used to calculate and analyze effect sizes and moderator analyses. Effect sizes were calculated directly from information provided in each study (eg, correlations, means and/or SDs, etc; see Supplemental Table 5 for details of data extraction of effect sizes). In 1 study, a nonsignificant finding was reported without accompanying statistical information, and therefore, P = .50 was entered into the Comprehensive Meta-Analysis Software.³⁸ Pooled effect sizes are represented as correlations with 95% confidence intervals (CIs). Calculations were based on a random-effects model to account for existing heterogeneity among studies.³⁸ To formally assess for heterogeneity of effect sizes, the Q and I² statistics were computed. A significant Q statistic and an I² >50% suggest moderators should be explored.²³ Heterogeneity of effect size and significance of categorical moderators were determined by Q-statistics,^38,39 and dimensional moderators were determined by mixed-effects–model meta-regressions.⁴⁰ Given the preference for studies with significant findings to be published in the literature, there is risk of meta-analyses overestimating overall effect sizes. Thus, in addition to including a dissertation database search in our search strategy, the Egger test and funnel plots were used to examine publication bias.

The number of nonduplicate abstracts identified by using this search strategy was 12 949. Figure 1 outlines the Preferred Reporting Items for Systematic Reviews and Meta-Analyses flow diagram.⁴¹ Of those identified in the search, 315 met initial screening criteria, and full-text abstracts were retrieved. Of these, 278 were excluded. No outliers were identified. In total, 37 studies (with 39 samples) met full inclusion criteria and proceeded to methodologic review for final determination of study inclusion.

A detailed description of study and sample characteristics can be found in Table 1. Sample size across studies ranged from 9 to 1026 (median = 142). The average child age at the language assessment was 33.5 months (range = 12–71 months), and the mean percentage of boys was 51.1%.

The mean score across all studies for methodologic quality was 9.24 of a possible maximum of 13.0 (range = 6–12; Supplemental Table 4). Thus, all studies met the inclusion criteria for methodologic quality (ie, a score of ≥5).

The effect size for the association between sensitive responsiveness and child language across 36 samples (7315 parent-child dyads) was significant (r = 0.27 [95% CI: 0.21 to 33]; Fig 2). There was no indication of publication bias (Supplemental Fig 3; Egger test P = .14). A separate meta-analysis was conducted to determine the pooled association between parental warmth and child language across 13 samples (1961 parent-child dyads). This association was also significant (r = 0.16 [95% CI: 0.09 to 0.21]; Fig 2). There was no indication of publication bias (Supplemental Fig 4; Egger test P = .13). CIs (95%) around the pooled effect sizes for sensitive responsiveness and language versus parental warmth and language were used to assess statistical significance between the 2 effect sizes.⁷⁸ Results indicated that the association between sensitive responsiveness and child language was statistically stronger than that between parental warmth and child language.

For the association between sensitive responsiveness and child language, the Q statistic was significant (Q = 225.11; P < .001; I² = 84.5), indicating heterogeneity of effect sizes. Moderator analyses revealed that effect sizes varied according to family SES, with stronger effect sizes seen in the low SES (k = 7; r = 0.37; 95% CI: 0.19 to 0.53) and diverse SES (k = 17; r = 0.29; 95% CI: 0.22 to 0.36) groups compared with the middle to upper SES group (k = 12; r = 0.15; 95% CI: 0.05 to 0.24). Effect sizes also varied as a function of whether the study examined associations longitudinally (k = 24; r = 0.30; 95% CI: 0.23 to 0.37) versus cross-sectionally (k = 12; r = 0.18; 95% CI: 0.09 to 0.27). No other moderators were significant (Table 2).

For the association between parental warmth and child language, between-study heterogeneity was not identified (Q = 17.5; P = .13; I² = 31.3); thus, moderator analyses were not explored.

Language development is 1 of the fundamental building blocks for school readiness, reading comprehension, academic achievement, and occupational outcomes.^2,3,79,–81 Results of the current meta-analyses indicate that children whose caregivers show higher levels of sensitive responsiveness and warmth display stronger language skills compared with children who received lower levels of such parenting behavior. The magnitude of this effect size translates into 2.8-fold increased odds that children who receive these parenting behaviors will acquire stronger language skills. This finding is in line with theoretical frameworks describing how parenting behaviors can help build the mental architecture required for children to engage with and learn from the social world.^26,82 

The meta-analytic association between sensitive responsiveness and child language was higher compared with the association between parental warmth and child language. Consistent with attachment theory,¹⁴ this finding may speak to the importance of attuned interactions between a parent and child to foster learning versus warmth alone. That is, measures of sensitive responsiveness are more likely to capture the parents’ contingent response to their children’s interests, focus of gaze, and developmental capacity. A sensitive-responsive parent can build on the moment-to-moment shifts in children’s attention, providing a finely tuned enhancement to the child’s experience.^26,82 Neural development is thought to occur through the internalization of these finely tuned, reciprocal interactions.⁸³ Warmth, on the other hand, does not involve contingency or reciprocity. Warmth can be demonstrated without solicitation if, for example, the parent is providing affection to the child, without any cues from the child. The provision of warmth can also be nonverbal, whereas sensitive responsiveness is more likely to have accompanying utterances in response to cues, which may be crucial for child language development.

On a behavioral level, responsive parenting may encourage children into social interactions that enhance learning. For instance, children of responsive mothers have been found to have an eager and willing stance toward others⁸⁴ and are more motivated, exploratory, and enthusiastic to seek out new information. This, in turn, enhances the likelihood that they will learn from their caregivers and others in their environment.⁸⁵ This willing stance positions children to benefit considerably from stimulating interactions. That said, a consequence of children’s limited language and/or limited interest or engagement in reciprocal exchanges with a parent is that parents may find fewer opportunities to engage in sensitive-responsive parenting. Future research should examine this possibility through cross-lag models that examine the directionality of associations between parenting behavior and child language over time.⁴³ 

Findings indicated that the associations between parental sensitive responsiveness and child language were comparatively larger in samples with low and diverse SES groups, as compared to middle to upper SES groups. A possible interpretation of this finding is that maternal sensitive responsiveness is particularly advantageous to children’s language when they are raised in socially disadvantaged families. These findings are in line with previous evidence from observational research documenting the protective effect of high-quality parent-child interactions in the context of adversity.^8,86 These findings also lend additional support to economic analyses suggesting that early investments in preventive interventions in disadvantaged groups may have the best return on investment.^87,88 Although these interventions can be expensive to undertake, initial investment costs are often recovered because of cost-saving reductions in social services and health professional use. Specifically, it has been shown that early investments in disadvantaged children aged 0 to 5 result in a 13% return on investments.⁸⁸ 

The association between sensitive responsiveness and child language was larger in studies using longitudinal versus cross-sectional study designs. The sheer number of longitudinal studies is notable (k = 24), especially given the arduous nature of collecting long-term research and the potential they provide in terms of distinguishing predictions from outcomes and addressing directionality of associations.⁸⁹ Methodologically speaking, longitudinal designs are more stringent and less susceptible to bias and are therefore considered to yield a truer estimate of effect sizes. Longitudinal designs are also more likely to capture the enduring effect of parenting on child development.⁹⁰ 

Several limitations should be noted. First and foremost, meta-analyses of observational studies are correlational in nature and thus do not permit conclusions about causality. Conclusions about causality can only be made in the context of experimental studies,^58,91,–93 and approximations of causality may be derived from longitudinal designs that address directionality and temporal precedence.⁹⁴ Second, these meta-analysis focused on typically developing samples, and thus, findings are not generalizable to children with language delay, intellectual disability, autism, and/or hearing or vision difficulties. Third, the generalizability of the current findings is also limited to samples of mother-child dyads. Although fathers engage in sensitive and responsive parenting,⁹⁵ the large majority of studies retrieved in this review reported on maternal, as opposed to paternal, measures of parenting behavior. Given the unique role that fathers’ language plays in children’s language development,^96,97 there is a need for more published research in this area. Finally, because of insufficient studies stratifying results on the basis of birth order,^98,–100 we were unable to examine this variable as a potential moderator of associations.

Early language development is an essential developmental skill that fosters academic, social, and behavioral success and well-being.^29,101 As such, understanding the antecedents to individual differences in children’s language abilities is critical to informing the policy and practice guidelines that aim to lay the foundation for healthy developmental trajectories. The findings indicate a moderate association between sensitive-responsive parenting and children’s language skills.^102,103 Sensitive responsiveness is a modifiable risk factor that has been successfully trained in parents in randomized controlled trials and shown to improve the language development of children.^{25,58,91,–93} Thus, the demonstration of a significant association between sensitive responsiveness and children’s language coupled with the evidence from randomized controlled trials on improving parental responsiveness suggests the importance of addressing this skill in parents, particularly in economically disadvantaged families.

Cheri Nickel, MLIS (University of Calgary), conducted the literature search.

CI

confidence interval

SES

socioeconomic status