Parent to Offspring Fear Transmission via Modeling in Early Life: A Systematic Review and Meta-Analysis

In this systematic review and meta-analysis, we followed PRISMA guidelines (Moher et al., 2009). Moreover, this meta-analysis was preregistered at OSF https://​doi.​org/​10.​17605/​OSF.​IO/​XPRUS.

Web Of Science, Pubmed, Embase, and PsycINFO databases were searched to identify relevant articles. The search was performed on the 21st of November 2022. The final search term was ((postnat* OR neonat* OR newborn OR “new-born” OR infan* OR baby OR babies OR “month old” OR “month-old” OR toddler) AND (parent* OR mother* OR father* OR caregiver* OR guardian*) AND ((“social referencing” OR acquisition OR “nonverbal transmission” OR “non-verbal transmission” OR “vicarious learning” OR “observational learning”) AND (fear* OR avoid* OR anxi* OR threat*))). For a full overview of the development of the final search term used in this study, see the Full search term list in the Supplementary Material (A). All screening steps were conducted by two independent reviewers. The interrater agreement on the inclusion of studies during the abstract screening process was high, with Cohen’s kappa of 0.85. Inconsistencies between reviewers were discussed and resolved in coding meetings. The steps of the screening process are presented in Fig. 1.

The data that were extracted are demographic information (i.e., age of the participating parents and infants, gender, ethnicity, occupation/SES, and study location) and methodological characteristics (i.e., study design, number of outcome variables, measurement tools for predictor and outcome variables, and their validity). Furthermore, we extracted means, standard deviations, effect sizes, and corresponding 95% Confidence Intervals (CI) of the variables and associations of interest. Variables of interest are infant fearful or avoidant reactions, parent anxious/fearful expressions, parent psychopathology, infant temperament, and stimulus type (i.e., social versus non-social). All effect sizes were converted to Hedges’ g. In case multiple means and standard deviations were reported in a study, for instance, due to multiple outcome measures, we averaged the outcomes to yield a single study-wide effect size. In cases where insignificant findings were reported without providing further statistical information than the sample size and non-significance, we assumed a p-value of 0.5 (one-directional) to calculate the effect size, which results in an effect size of 0 with the accompanying variance (see Dusseldorp et al., 1999). This was done as excluding the insignificant finding from analyses would inflate the effect sizes. The effect sizes for the moderators were only investigated if a subset consisted of at least four studies (k ≥ 4) (Bakermans-Kranenburg et al., 2003).

Analyses were carried out using the metafor package in R. Statistical significance of the pooled SMD was assessed using a Z test at p < 0.05. We checked for heterogeneity using the Q test. A two-tailed p significance test was used with statistical significance with p < 0.05. To enable comparisons, calculated effect sizes were transformed into standardized scores. We corrected the effect sizes to a weighted effect size (corrected for unequal n´s) and checked for publication bias with funnel plots. In case of publication bias, a trim and fill method was applied. Furthermore, to detect outliers, we checked whether the standardized residuals were between 3.29 and − 3.29.

The methodological quality of the retained articles was assessed using a checklist (presented in Table S1) based on the Cochrane Collaboration tool (ROB2) and adapted to our study design. Examples of these assessment criteria are the reliability of the outcome measures, as well as the transparency and selection of the reported results.

The study characteristics of the studies included in the systematic review and meta-analysis are summarized in Table 1. The quality ratings ranged from 67 to 100%, with a mean percentage of 93% (for the quality rating per study see Table S1 in Supplementary Material B). Most studies randomized participants into the conditions and used reliable coding systems or measures. However, some studies did not adequately describe their hypotheses and/or reported more analyses than planned a priori.

The study and sample characteristics are presented in Table 1. The studies differed in (1) design, (2) moderators, (3) child fear index, (4) parental message type, and (5) stimulus type. Below we address each of these in detail.

First, concerning the design, from the 23 studies included in this systematic review four had a correlational design, whereas 19 had an experimental design. In the correlational designs, parental expressions of fear to novel stimuli were not manipulated/trained by the experimenter, but observed as it naturally unfolds during a social referencing paradigm with parents and their infants.

Second, of these studies, eight studies included a measure of parental anxiety symptoms or diagnosis. Four studies included clinical parent samples, consisting of 51% to 56% of parents with an anxiety disorder (Aktar et al., 2013, 2014, 2018; Murray et al., 2008), whereas four studies assessed anxiety (symptoms) in community samples of parents and reported no or low anxiety scores (De Rosnay et al., 2006; Dubi et al., 2008; Goodman-Wilson, 2012; Möller et al., 2014). Finally, ten studies of these 23 studies assessed infant temperament (Aktar et al., 2013, 2014, 2018; Blackford & Walden, 1998; Carpenter, 2004; De Rosnay et al., 2006; Dubi et al., 2008; Goodman-Wilson, 2012; Möller et al., 2014; Murray et al., 2008).

Third, there were also differences across studies, which child fear indices were assessed to test child acquisition of fear and avoidance (overview can be found in Table 2). Across all studies, infant fear was primarily assessed with a behavioral measure, specifically affective responses and avoidant responses toward the stimulus during the social referencing paradigm. In one study infant reactions were assessed with just a fear measure (i.e., facial, vocal, and verbal expressions of fear), in five studies only avoidance was assessed (i.e., latency touching and reaching for the toy), and in 17 studies both fear and avoidance were assessed. Looks to the caregiver were defined as an indicator of social referencing, rather than an index of infant fear. Nearly all studies reported mean interobserver reliability (ICC or Cohen’s kappa) for coding infant fear and avoidance (22 studies), which ranged from 0.56 to 1, and were all classified to be sufficient to very high interrater reliability.

Fourth, parental expressions of fear toward novel stimuli can be categorized into (1) non-verbal messages only (such as fidgeting) and (2) non-verbal and verbal messages (such as “this is scary, right?”). Out of 23 studies, five studies fall in the first category, whereas 18 studies were in the second. Furthermore, in experimental designs, the threat condition was defined as fearful/anxious non-verbal messages (based on facial, bodily, or vocal expressions), whereas the control condition could either consist of parental neutral non-verbal expressions or positive non-verbal expressions (i.e., smiling). In 21 studies, the control condition in the social referencing paradigm consisted of positively valenced non-verbal parental messages and two studies included both a positive and neutral control condition (Klinnert, 1984; Mumme et al., 1996). Most studies reported mean interobserver reliability (ICC or Cohen’s kappa) for parent variables (17 studies), which ranged from 0.39 to 1, and all except for one (Zarbatany & Lamb, 1985) were classified as sufficient to very high.

For the meta-analysis, we only included studies that reported the statistical information that is necessary for the computing of effect sizes. We contacted authors for missing statistical information (such as missing sample sizes or standard deviations). We only received sufficient statistical information to analyze effect sizes from one study (Möller et al., 2014). We received three responses that the statistical information was not available (Blackford & Walden, 1998; Walden & Baxter, 1989; Zarbatany & Lamb, 1985) and one author did not respond (Klinnert, 1984). Furthermore, three studies that were included in the systematic review (Aktar et al., 2013, 2014, 2018) contained analyses of the same infants at different developmental stages. For the meta-analysis, we chose to include only data from the first study (Aktar et al., 2013), as it contained the largest sample size. Möller et al. (2014) reported their findings separately on a mother and father sample, which participated independently with different infants. Therefore, we added them as separate samples in our analyses.

Overall, of the 23 studies included in the systematic review, 19 studies entailing 20 samples were also included in the meta-analysis. Two studies had a correlational design (Aktar et al., 2013; Möller et al., 2014) and the remaining seventeen studies had an experimental design. Fourteen studies entailed non-social stimuli, two had only social stimuli, and three studies included both social and non-social stimuli. When a study measured infant acquired fear to both social and non-social stimuli, we combined the effect sizes (if relevant statistical information was available). Thirteen studies assessed infant fear/ anxiety with behavioral indices of fear and avoidance separately, three studies just assessed avoidance based on infant behavior, and one assessed only infant fear with a behavioral measure. Two studies assessed infant avoidance additionally with a cognitive measure. When we had multiple outcomes of fear or avoidance, we combined the effect sizes. If we could not combine indices, we chose the statistics in the following order: (1) behavioral measure of infant fear or avoidance and (2) cognitive measure of infant avoidance (such as frequency of looks). No study assessed physiological indices of fear.

Six studies that were included in the meta-analysis assessed parental anxiety, of which the sample of two studies consisted of 51% to 54% of parents with an anxiety disorder (Aktar et al., 2013; Murray et al., 2008). Four studies assessed the absence of an anxiety disorder/symptoms (De Rosnay et al., 2006; Dubi et al., 2008; Goodman-Wilson, 2012; Möller et al., 2014). However, only three studies that were included in the meta-analysis of main effects reported findings on parental anxiety as a moderator, and therefore we could not perform analyses on its effect size. Eight studies that were included in the meta-analysis assessed infant temperament and reported relevant statistical information (Aktar et al., 2013; Carpenter, 2004; De Rosnay et al., 2006; Dubi et al., 2008; Goodman-Wilson, 2012; Möller et al., 2014; Murray et al., 2008).

A summary of the main findings can be found in Table 3. Based on social fear learning theories (Olsson et al., 2007; Rachman, 1977), we expected that infants express more fear and anxiety toward novel stimuli when these stimuli are paired with parents’ fear/anxiety expressions than non-anxious parental expressions. Of the 23 studies reviewed, infant fear was assessed in 17 studies, infant avoidance was assessed in 21 studies, and a combined fear and avoidance measure was utilized in two studies. We found that six out of 17 (44%) studies found an effect of parental expressions of fear/anxiety on infant fear (measured as infant fearful/negative affect/distress) and the other eleven did not. Furthermore, 11 out of 21 (52%) studies found an effect on infant avoidance (Aktar et al., 2013; De Rosnay et al., 2006; Dubi et al., 2008; Gerull & Rapee, 2002; Hirshberg & Svedja, 1990; Kim et al., 2010; Mumme et al., 1996; Sorce et al., 1985, Walden & Ogan, 1988; Walden & Baxter, 1989; Walden et al., 1991), one study found an effect on avoidance of toys but not of the stranger (Goodman-Wilson, 2012), and another only on part of the stranger task (pick up but not approach phase) (Murray et al., 2008). Lastly, one of the two studies that included a combined fear and avoidance measure found no relationship between parental non-verbal signals of threat and infant reaction to novel stimuli (Aktar et al., 2014), whereas the other one did (Rosen et al., 1992).

Multiple studies investigated additional moderating effects of for example parental gender, infant gender, and/or age in the link between parental-expressed fear and infant fear/avoidance to novel stimuli. In one study, authors found an effect of parental threat on infant fear and avoidance, but only when the father conveyed the fearful signals and not the mother (Möller et al., 2014), whereas another study found only maternal, but not paternal expression being related to subsequent infant fear when the infant was one year old (Aktar et al., 2018). Carpenter (2004) found an effect of parental threat on infant fear and avoidance only in specific age ranges. Nine-month old’s looked less to stimuli in fear vs happy condition and 18-month old’s showed overall less approach in fear versus happy condition, whereas Walden and Baxter (1989) found an effect in the 13- to 23-month-olds but not in 6–12- or 24–40-year old’s. Another study found mothers’ messages to only affect female infants, who stayed less close to the toy in the fearful versus happy condition (Rosen et al., 1992).

In line with social fear learning models (Olsson et al., 2007; Rachman, 1977), there was a significant effect of parental fear expressions on infant fear and avoidance, supporting the idea of parental modeling as a social fear learning pathway with small to medium effect sizes (Hedges’ g = 0.39 and 0.46, respectively). When investigating the causal effect of parental modeling on infant fear and avoidance using exclusively experimental studies, the effect sizes were also small to medium (Hedges’ g = 0.44 and 0.44, respectively).

While experimental designs allow us to make stronger inferences on causality, their findings might be less generalizable to real-life interactions/daily life (Kazdin, 2021). A previous review that summarized findings on child fear acquisition via verbal threat information has argued for assessing child fear acquisition with this social fear learning paradigm in more ecologically valid contexts (Muris & Field, 2010). This reasoning also applies to modeling pathways in early life, as both experimental as well as prospective, more naturalistic designs are necessary to gain further insight on the impact of parental fear/anxiety expressions on infant fear and avoidance acquisition toward novel stimuli.

Furthermore, there seem to be inconsistent findings regarding infant fear and avoidance. Some studies that assessed both fear and avoidance toward novel stimuli as separate constructs, found different results for fear and avoidance outcome measures in the same children (i.e., support for an effect of condition on avoidance but not infant affect) (for example, Walden & Ogan, 1988). The mixed findings suggest that infant affect and avoidance do not have to co-occur, and different behavioral indices of fear might become relevant at different developmental stages. Infants’ ability to understand and judge how threatening a novel stimulus is, as well as to what extent they can regulate emotions accordingly can influence how infants express fear (LoBue & Adolph, 2019). During childhood, over time the inclination to avoid novel stimuli increases more strongly than to show distress (Rapee & Spence, 2004; Sumter, et al., 2009). As infants get older, infants may also show fewer fearful and more avoidant strategies in response to parental expression of fear (Aktar et al., 2018). Longitudinal studies covering the periods including and extending from the toddlerhood years are needed to shed light on these developmental differences in fearful and avoidant responses.

Susceptibility models suggest that infants with fear-sensitive temperamental dispositions are more susceptible to exposure to environmental stressors, including parental anxiety signals (Belsky & Pluess, 2009; Ingram & Luxton, 2005; Nigg, 2006). In line with this model, our meta-analysis suggests that there is a small moderating effect of BI on infant avoidance (Hedges’ g = 0.25). Also the systematic review lends some support to the idea that behaviorally inhibited infants display an increased avoidant response after exposure to parental fear expressions, but only in three out of eight studies. However, based on the experimental studies we did not find a moderation by temperament in the causal link between parental fearful expressions to novel stimuli and child avoidance (Hedges’ g = 0.18). Possibly, infants are more likely to display their usual/learned responses in response to more naturalistic parental expressions of fear (in correlational studies), than manipulated fear expressions (in experimental studies). Given that temperamentally fearful infants have the trait tendency to withdraw and avoid novel stimuli (Stifter & Augustine, 2019), this effect might be more visible in studies with correlational designs. Possibly, a third variable that is related to both parental fear expressions and infant avoidance may have inflated the effect size of this link. The reported correlations could, for example, be influenced by genetic similarity or the learning history of parental behaviors. Hence, we do not know to what extent the link between parental fear expressions to novel stimuli and child fear or avoidance is due to genes or to what extent it represents the habitual reaction of infants that has been reinforced over time. Specifically, parents may have supported children’s avoidant behaviors in anxiety-inducing or novel situations or removed the child from these situations (Reinforcement pathway described in Fisak & Grills-Tacquechel, 2007). In experimental studies, these confounds are controlled for by manipulating parental expressions/reaction. However, in experimental studies, the findings might also not be representative of the effect that parental fear expressions have on child fear and avoidance in real life, because manipulated parental expressions/reaction to novel stimuli might be different to infants’ previous experiences and expectations. A previous study suggests that expectancy violations in infants might influence social learning processes (Colomer & Woodward, 2023). It is up to future research to elucidate the role of BI in the parent-to-infant transmission of fear to novel stimuli by assessing the additional effects of genetic traits or assessing the influence of BI on the repeated exposure to parental fearful expressions.

Moreover, our meta-analysis did not find a moderating effect of BI on infant fear (Hedges’ g = 0.07). This aligns with findings from our systematic review where no such effect on infant fear was observed for the majority of the studies (7 out of 8 found no significant effect). Although this finding is not in line with the susceptibility theory (Belsky & Pluess, 2009), it might be explained by the fact that BI is most relevant in fear acquisition regarding social stimuli. BI is a more prominent risk factor for social anxiety (Clauss & Blackford, 2012) than for specific phobias (Pérez-Edgar & Fox, 2018), and in our systematic review most studies finding an effect included social stimuli in their social referencing design. Given that the majority of the included studies assessed fear toward non-social stimuli, no firm conclusions can be drawn on the moderating role of BI in the context of fear acquisition regarding social stimuli. Future experimental studies that incorporate social stimuli in their design will help to clarify this.

Next, we investigated the moderating role of parental anxiety. In the absence of a sufficient amount of studies with statistical information regarding parental anxiety, we could only assess its influence on infant fear and avoidance of novel stimuli by means of a systematic review. While parental anxiety is one of the biggest risk factors for child anxiety, we did not find support of infants of anxious parents showing stronger fear acquisition via modeling parents’ anxious expressions. This suggests that infants of anxious parents might not be more sensitive toward novel stimuli in the context of a single exposure to parental fear expressions. However, two studies, which did not find stronger fear or avoidance in the infants of anxious parents immediately after being exposed to parent’s expression of fear, found these expressions to be predictive of later avoidance toward that stimulus (Aktar et al., 2014; Murray et al., 2008). Specifically, Murray et al. (2008) found that infants of mothers with social phobia did become more avoidant of strangers picking them up between 10 and 14 months. The study by Aktar et al. (2014) found that the link between parental expressions of threat at 12 months with infant fear/avoidance at 30 months was stronger for infants of parents with lifetime comorbid social and other anxiety diagnoses. Therefore, it could be that over time, the repetitive nature of infant’s modeling of parental expressions of fear in families with anxious parents, entailing a higher frequency of parental anxious expressions to novel stimuli, could explain the familial aggregation of anxiety. Furthermore, anxious parents may be more likely to support infants’ avoidant behaviors in anxiety-inducing or novel situations or remove the infant from these situations/stimuli (Fisak & Grills-Tacquechel, 2007). Anxious parents might also provide adaptive emotion regulation strategies less frequently, such as offering a security object, or displaying alternative problem-solving behaviors (Stifter & Augustine, 2019). Moreover, they might also react unsupportive, for example, dismissing or ignoring infants’ emotional reactions. This in turn could decrease infants’ feelings of self-efficacy for self-regulation and increase the distress or fear response (Stifter & Augustine, 2019). Given the limited number of studies investigating the moderating role of parental anxiety in parent–infant fear transmission, we need more studies investigating its role.

Heightened offspring fear or fear learning in response to (potentially) threatening stimuli represents an evolutionary-adaptive and normative process (Kiel & Kalomiris, 2019). Nevertheless, by understanding how social fear learning processes unfold and differ between healthy and at-risk families, we might eventually shed more light on the specific processes and factors to target in prevention and treatment efforts. In our study, we found a small to medium effect of parents displaying fearful reactions to novel stimuli on infant fear and avoidance toward these stimuli, independent of parental anxiety levels. While this fear acquisition pathway in itself can be seen as an adaptive response to threatening/novel situation, it does not exclude the possibility that in at-risk families, where exposure to parental anxious expressions in daily life can be more frequent or intense, the impact of this fear acquisition pathway can be amplified. Incorporating psychoeducation targeting the potential pathways of social fear transmission in parents and children might be helpful in the prevention of anxiety risk in the offspring. Given that the effect of parental fearful reactions to novel stimuli on infant avoidance was stronger for children with more fearful temperaments—psychoeducation might be valuable for parents with children who are behaviorally inhibited.

In real life, infants might not only get exposed to a fearful reaction of one (anxious) parent in isolation, but another parent or significant other may display the same or conflicting emotional responses. As fear modeling seems to lead to an infant’s fear acquisition toward novel stimuli, modeling of parents’ positive emotions or confident reactions may reduce or prevent fear acquisition, even when one parent displays anxious responses. This was recently summarized in a systematic review investigating whether infants and children’s positive modeling (of parents, strangers, and peers) in experimental, non-clinical contexts can reduce/prevent acquired fears (Krause & Askew, 2022). Although their conclusions rely on a limited amount of studies, positive modeling seems to be a promising technique to prevent fear acquisition and reduce fear responses in infants and children. Understanding how fears are acquired in developmentally sensitive designs can inform us of potential strategies to reduce or prevent parent-to-child fear transmission in at-risk families.

This is the first systematic review and meta-analysis on the effect of parental fear expressions on infant fear and avoidance of novel stimuli. Although this work provides a relatively less biased synthesis of available evidence on parent–infant fear transmission via modeling, this study is not without shortcomings and echoes the limitations of the singular empirical work. First, in our systematic review and meta-analysis we heavily relied on studies with WEIRD (Western, educated, industrialized, rich, and democratic) samples. It is important to acknowledge the role of cultural differences in the emotional development of infants since parents’ emotional expressions during daily interactions are part of commonly shared socialization practices (Halberstadt & Lozada, 2011). For example, infant’s attention to (parental) emotional expressions in daily life can vary depending on their socio-economic status (SES) (Clearfield & Jedd, 2013). Regarding our meta-analysis, this means we cannot generalize our findings to non-WEIRD samples. Future research that replicates previous studies in new or more heterogeneous samples, or compare the fear acquisition pathway across different cultural environments, can give us more insight on the generalizability of our findings (Nielsen et al., 2017).

Second, studying fear modeling in strict experimental lab designs allows stronger conclusions, but it restricts the ecological validity of the findings. In daily life, infants are usually exposed to ambiguous stimuli, such as novel toys in their own home or daycare, surrounded by familiar people, instead of in a new and ambiguous place with strangers (i.e., a lab, which does not characterize their common experience). Also training the parents to show specific emotional expressions in lab settings may not capture the intensity that the parent in real-life displays to novel stimuli. Furthermore, infants might not only get exposed to a fearful reaction of one parent in isolation, but often the two parents or significant others display similar or conflicting emotional responses, either simultaneously or successively. More research is needed to investigate fear modeling in multiple contexts, as well as naturalistic observations in clinical samples. Future studies might also investigate repeated exposure to parental fearful expressions (either via experimental manipulation or by inclusion of anxious parents) to examine whether repeated exposure predicts fear or avoidance to novelty over time and whether the relationship becomes stronger. This might also represent real life more accurately, as infants most likely will not only get exposed to parental expression to a novel stimulus just once.

Another limitation concerns the fact that multiple studies measured fear as a behavioral response to novel stimuli by exclusively focusing on either fear or avoidance. Studies focusing on singular indices of fear may not be sufficient to capture the entirety of infant fear reactions and do not allow us to investigate relationships between different fear indices. Measuring fear with singular indices can increase the likelihood of falsely identifying an infant’s reaction as fear (LoBue & Adolph, 2019). Therefore, to decrease the chances of misattribution, measuring fear in infants should contain multiple complementary methods, such as multiple behavioral (infant distress and avoidance) and physiological indices of fear (LoBue & Adolph, 2019). This would also allow us to investigate whether parental fear expressions influence various behavioral but also physiological reactions in infants. Moreover, in a longitudinal design, one could also assess which indices of infant fear can predict later development of fear or anxiety to novel stimuli.