Relational Antecedents and Social Implications of the Emotion of Empathy: Evidence from Three Studies

Abstract

Despite emotion researchers’ strong interest in empathy and its implications for prosocial functioning, surprisingly few studies have examined parent-child attachment as a context for early origins of empathy in young children. Consequently, empirical evidence on links among children’s attachment, empathy, and prosociality is thin and inconsistent. We examined such links in two longitudinal studies of community families (Family Study, N=101 mothers, fathers, and children, 14 to 80 months; Parent-Child Study, mothers and children, N=108; 15 to 45 months) and a study of low-income, diverse mothers and toddlers (Play Study, N = 186, 30 months). Children’s security was assessed in Strange Situation in infancy and rated by observers and mothers using Attachment Q-Set at toddler age. Children’s empathy was observed in scripted probes that involved parental simulated distress. Children’s prosociality was rated by parents (Family Study, Play Study). Security with mothers related to higher empathy. For mother- and father-child dyads, security moderated the path from empathy to prosociality. For insecure children, but not secure ones, variations in empathy related to prosociality. Insecure and unempathic children were particularly low in prosociality.

Positive social emotions, such as empathy, concern for others, compassion, or sympathy, and their implications for prosocial behaviors, such as helping or sharing, have all been classic topics of reflection in social psychology, developmental psychology and psychopathology, and sociology. In particular, empathy has been a strong focus of researchers studying emotion and affective processes. That research has resulted in multiple reviews, target articles, and special sections of journals (e.g., Cuff, Brown, Taylor, Howat, 2016; Eisenberg & Miller, 1987; Engelen & Röttger-Rössler, 2012; Preston & de Waal, 2002).

In developmental psychology, research on empathy has evolved considerably. In the 80’s, the predominant focus on older children, due mostly to the influence of cognitive theories and cognitive models of morality, shifted dramatically to very young children – toddlers, preschoolers, even infants, amply demonstrating rich signs of empathy, compassion, and prosociality at early ages (Emde, Biringen, Clyman, & Oppenheim, 1991; Eisenberg & Fabes, 1998; Hoffman, 1975; Radke-Yarrow et al., 1976; Radke-Yarrow, Zahn-Waxler, & Chapman, 1983; Zahn-Waxler & Radke-Yarrow, 1990; Zahn-Waxler, Radke-Yarrow, Wagner, & Chapman, 1992). That research has now matured into an influential, growing, heuristically productive field, featured prominently in special issues and handbooks (Brownell, 2013; Davidov, Zahn-Waxler, Roth-Hanania, & Knafo, 2013; Killen & Smetana, 2015; Thompson, 2006, 2013, 2015). Even very young children are now seen as capable of responding emotionally to others’ emotions and needs, evaluating others in terms of prosocial versus antisocial intentions (Hamlin, Wynn, & Bloom, 2007), and engaging in altruistic behavior (Warneken & Tomasello, 2009).

The importance of empathy has been also broadly recognized in developmental psychopathology. Compromised empathy, or low concern for others in children is a core component of callous-unemotional traits. Those traits are considered a substantial risk for future antisocial and aggressive pathways (Blair, 1995; Eisenberg & Miller, 1987; Frick & White, 2008; Miller & Eisenberg, 1988; Wakschlag, Tolan, & Leventhal, 2010). Even very early individual differences in children’s empathy to others’ distress have been shown to have profound, clinically relevant, long-term implications (Rhee et al., 2013). Consequently, understanding the origins of individual differences in the emotion of empathy and its implications for prosociality is an important goal for developmental psychology and psychopathology. Multiple dimensions of parenting, such as warmth, responsiveness and sensitivity, positive reciprocity, empathic caregiving, supportive, authoritative control, effective emotion socialization, as well as specific parenting strategies, such as induction, internal states talk, internal attributions, reinforcements, and other socialization techniques have been proposed as effective instruments of promoting emotional empathy and prosocial development. Several scholarly and thorough reviews summarize this very extensive area of research (see Eisenberg, Fabes, & Spinrad, 2006; Eisenberg, Spinrad, & Knafo-Noam, 2015; Killen & Smetana, 2015; Thompson, 2013, 2015, for most recent examples).

Surprisingly, however, few studies in the area of empathic emotions and prosociality have made contact with another dynamically growing field: attachment theory and research. Research on empathy and prosocial development has progressed largely in parallel to the flourishing research on attachment, although obvious points of intersection have been articulated (Waters, Hay, & Richters, 1986, van IJzendoorn, 1997; Waters, Kondo-Ikemura, Posada, & Richters, 1991). In a recent comprehensive review, Shaver, Mikulincer, Gross, Stern, and Cassidy (2016) argue persuasively that a secure attachment is a natural, prototypical, inherently positive ecology for the emergence of the child’s care orientation, empathy, compassion, and prosociality. They also outline several characteristics of a secure parent-child relationship that are likely antecedents and mechanisms for emerging positive forms of social behavior: The child’s responsive and caring internal working models of the caregiver and self, effective emotion regulation, necessary for the management of empathic arousal and translating it into prosocial behavior, the parent’s high mind-mindedness and reflective functioning, the child’s attunement to the caregiver’s and others’ emotions, and the child’s trust in the caregiver’s availability and protection, which frees up internal resources and allows him or her to focus on comforting others in distress. As well, the parent’s own empathy, coded from interviews about the child (Stern, Borelli, & Smiley, 2015) or measured as a degree of affective understanding and matching the child’s affect in a stressful situation (Borelli, Vazquez, Rasmussen, Teachanarong, & Smiley, 2016) has been shown to be associated with the child’s secure attachment and his or her perception of parental warmth, thus likely enhancing the effectiveness of parental modeling of children’s future empathic responses.

We note that conceptual and empirical research on links between attachment and early conscience or morality has been developing rapidly (Thompson, 2006, 2012, 2013; 2015; Kochanska, 1995; Kochanska, Aksan, Knaack, & Rhines, 2004). The construct of “conscience”, however, tends to encompass broadly many forms of positive social behaviors, such as moral understanding, receptive cooperation with parents’ socialization, and internalization of and regard for parental rule. Although research on conscience may include empathy and prosociality, it does not focus on those constructs specifically, making more precise understanding difficult.

The extant empirical evidence on links among attachment, empathy, and prosociality is remarkably thin and inconsistent. We have located only a handful of studies. Panfile and Laible (2012) supported a model in which security led to better emotion regulation, which in turn led to more empathy, and empathy – to more prosocial behavior in 3-year-olds. One limitation, however, was the fact that security, emotion regulation, and empathy were all rated by mothers, and only prosociality was observed as helping an adult in the context of a baby’s recorded cry. In what appeared the same sample, mother-rated attachment at 42 months predicted children’s observed empathic concern to a baby’s cry at 48 months (Panfile & Laible, 2013). Denham (1994) reported positive links between security as rated by the mother and the child’s observed empathic, concerned, prosocial response to the mother’s simulation of negative emotion in 3½-year-old children. Kestenbaum, Farber, and Sroufe (1989) observed 4-year-olds’ empathy shown to naturally occurring instances of distress expressed by their peers during interactions in preschool. Children with secure attachment histories, compared to those with avoidant histories, assessed in Strange Situation, were more empathic. Van der Mark, van IJzendoorn, and Bakermans-Kranenburg, (2002) found a modest positive relation between a history of security and child empathic concern observed in response to the experimenter’s (although not the mother’s) simulated distress in female toddlers. By contrast, Carter, Little, Briggs-Gowan, and Kogan (1999) found no associations between attachment measured in Strange Situation and mothers’ ratings of empathy and prosocial behaviors in 1-year-old children. Of note, no study has involved fathers and children (a more general gap in research on prosocial socialization, Hastings, Utendale, & Sullivan, 2007). The overall objective of the present work is to examine the role of early attachment organization with mothers and fathers in the development of children’s early empathy and positive social behavior.

One reason for the lack of consensus is the fact that conceptualizations and measures of security, empathy, and prosociality have varied broadly. As Shaver and colleagues (2016) point out, in research with young children, the definitions and operationalizations of care for others are particularly complex. They make a critical distinction between empathy, which refers to the largely emotional response to another’s distress, and prosociality, which refers to the behavioral manifestation of that emotion in social contexts. They emphasize the importance of drawing distinctions between the emotion of empathy and prosociality, and challenges of assessing each construct with sensitivity and specificity, particularly in very young children, before self-reports of internal experience are possible. We follow those recommendations in the present work, and we assess empathy and prosociality as distinct constructs, using distinct methodologies.

Obtaining robust behavioral observations of the emotion of empathy in very young children is a challenge, because empathic responses are infrequent and coding is difficult. We adapted Zahn-Waxler’s and Radke-Yarrow’s (e.g., Zahn-Waxler, Cole, & Barrett, 1991; Zahn-Waxler et al., 1992) pioneering methodology to assess young children’s empathy in scripted paradigms or probes, in which the parent or another adult simulates distress for a brief time, either pretending to have been hurt by the child or to have hurt oneself. Children’s affective, verbal, and behavioral responses are then coded. This approach is methodologically demanding, because it requires a careful implementation of the standard paradigm and sensitive behavioral coding. Prosociality was assessed by parents’ reports about the child’s functioning in the peer ecology (sharing, helping, comforting, interacting harmoniously, taking turns, being fair, etc.).

Regardless of variations in measures of security, empathy, and prosociality adopted in various studies, conceptual models of how those constructs relate to each others have also varied. Of note, as described above, Panfile and Laible (2012, 2013) proposed a causal path from security to emotion regulation to empathy to prosociality. This is consistent with the broadly believed causal association between empathy and prosociality (e.g., Decety, Bartal, Uzefovsky, & Knafo-Noam, 2016; Eisenberg et al., 2015; Hoffman, 1975, 2000). In the present work, we adopt a similar model, in which we examine the path from the child’s attachment organization with the parent to empathy to that parent’s distress to prosociality in the peer context.

This mediational path, however, captures only one way that early security may influence future prosociality. Effects of early security organization on future emotional development can be very complex. Recently, substantial evidence has accumulated to indicate that the impact of early security extends beyond simple direct effects. In particular, research has found that associations between children’s individual characteristics and developmental outcomes may vary depending on their early attachment security. In other words, early security may moderate future associations between children’s individual characteristics and developmental outcomes. In particular, children’s early negative characteristics (e.g., high anger, low concern about rules, difficult temperament) have been linked to unfavorable future outcomes, but those links were present only in parent-child dyads that were insecure in infancy. Several studies have robustly confirmed such model (Boldt, Kochanska, & Jonas, 2016; Kim, Kochanska, Boldt, Koenig Nordling, & O’Bleness, 2014; Kochanska & Kim, 2012; Kochanska, Barry, Stellern, & O’Bleness, 2009). Drawing from that work, we expected that early security may moderate the links from empathy to prosociality. More specifically, and consistent with past findings, we expected that in insecure parent-child relationships, children who are especially unempathic may be at a particular disadvantage in terms of prosocial development.

Thus, we proposed the following moderated mediation model (Figure 1). The child’s early security with the parent was (a) expected to promote the child’s empathy expressed toward the parent, and in turn, empathy was expected to promote child prosociality in a broader social ecology (an indirect effect, or mediation). But in addition, security was also expected (b) to moderate the link between empathy and prosociality, such that variations in empathy would relate to prosociality particularly for insecure children (moderated mediation). Recently developed analytic tools allow for testing precisely such a model, one that combines mediation and moderation, with the same construct (security) serving as a predictor and a moderator (Hayes, 2013).

Figure 1.

The conceptual moderated mediation model: The path from the child’s attachment organization with the parent to empathy to the parent’s distress to prosociality with peers. Attachment additionally moderates the path from empathy to prosociality.

We present three studies. Family Study followed low-risk community mothers, fathers, and children from 15 months to age 6.5 years. In Play Study, high-risk, low-income mothers and children were assessed at age 2.5 years. In Parent-Child Study, low-risk community mothers and children were followed from 14 months to 45 months. We examined the relations among children’s attachment, observed and parent-rated (all studies), their emotional empathic response to parental distress observed in laboratory paradigms (all studies), and their prosociality in peer contexts, rated by parents (Family Study, Play Study).

Overall Method

Description of the studies, designs, and participants

We present three studies. All participants were volunteers who came from the same area of the Midwest (a college town, a small city, and small rural towns and communities). Two were longitudinal studies of low-risk, two-parent community samples (Family Study and Parent-Child Study), and one was a high-risk study of exclusively low-income mothers and children (Play Study). All children were biological offspring, typically developing and free of major health problems. Recruitment involved flyers and ads posted broadly and mailed to day care providers, pediatricians, schools, etc. In Play Study, venues frequented by low-income families, such as Women, Infants, and Children offices, thrift stores, mobile homes parks, or subsidized housing projects were particularly targeted. The descriptions of the designs and participants follow.

Family Study

In this longitudinal study, mothers, fathers, and children entered in infancy; in this article, we present data from 6 assessment points to examine the relations among child attachment security, assessed at 15 months (N=101, 51 girls), the child’s empathy measures with each parent, obtained at 25 months (100, 50 girls), 38 months (N=100, 50 girls), and 52 months (N=99, 49 girls), and the child’s prosociality at 67 months (N=92, 45 girls) and 80 months (N=90, 43 girls). All measures were obtained for mother-child and father child dyads.

The families ranged in education. Among mothers, approximately 25% had a high school education (or less), 54 % had an associate or college degree, and 21% had a postgraduate education. Among fathers, the respective figures were approximately 30, 51, and 20%. The annual family incomes ranged from less than $20,000 (8 %), to $20,000–$40,000 (17%), to $40,000–$60,000 (26 %), to over $60,000 (49 %). Ninety percent of mothers were White, 3% Hispanic, 2% African American, 1% Asian, 1% Pacific Islander, and 3% other non-White; 84% of fathers were White, 8% Hispanic, 3% African American, 3% Asian, and 2% other. In 20% of families, one or both parents were non-White.

Play Study

This was an experimental study of high-risk, exclusively low income, mothers and children (N=186, 90 girls), who entered at toddler age (22–44 months, average 30 months) and then participated in a randomized parenting intervention. We present data obtained for the whole sample, at the baseline (prior to randomization), for concurrent associations among the measures of child attachment security, empathy, and prosociality.

To be eligible, the mother had to receive or qualify for aid from a federal, state, or faith-based agency, or for Earned Income Tax Credit. The average annual family income was $20,385, SD = $13,010; 55% of mothers had no more than a high school education, and 45% had an associate, bachelors, or technical degree. There were 11% Hispanic, and 88% not Hispanic mothers; 73% White, 15% African American, 2% Asian, 2% American Indian, and 8% more than one race or unreported. Fifty four percent were married, 13% cohabitated, 31% were single or divorced, and 2% in other arrangements.

Parent-Child Study

In this longitudinal study, community mothers and children entered in infancy; we present data on the relations between children’s attachment security, assessed at 14 months (N=108, 53 girls) and at 22 months (N=106, 53 girls), and their empathy at 45 months (N=101, 49 girls). No data on prosociality were available.

The sample was comparable to Family Study: two-parent volunteer community families (although only mothers and children participated). Among mothers, 26% had no education past high school, 15% had some college education, and 59% completed college or some postgraduate education. For fathers, the respective figures were 31%, 10%, and 57%. Family annual income varied: under $20,000 (7%), $40,001–$50,001 (17%), $50,001 – $60,000 (l6%), and more than $60,001 (25%). In terms of race, 97% of mothers and 92% of fathers were White.

Measures

All studies followed the same set of methodological guidelines. Data were collected in the university laboratory. Parents and children were observed during lengthy (typically 2-–3 hours), carefully scripted sessions, which encompassed a broad range of paradigms and contexts, varying in their psychological potentials (play, snack, chores, free time, standard tasks, etc.). The sessions were videotaped through one-way mirror for later coding. Multiple teams coded behavioral data. Because this research has spanned two decades, the best practices regarding reliability statistics have varied. Consequently, we report alphas, ICC, and kappas. Reliability was typically based on 15–20% of the sample. Coders realigned periodically to prevent observer drift. To obtain robust final constructs, we systematically deployed data aggregation at multiple levels (Rushton, Brainerd, & Pressley, 1983).

Measures of Children’s Empathy to Parental Distress

The paradigms eliciting children’s empathy, behavioral coding, and data aggregation were fully parallel across Family Study, Play Study, and Parent-Child Study. Those have been published, and thus, the description is brief (Kochanska, Koenig, Barry, Kim, & Yoon, 2010).

Paradigm

In the scripted simulated distress paradigm, the parent had been first instructed by the visit coordinator, and given a detailed script. The parent and child then played with a pounding block toy; as the child was hammering down the pegs, the parent pretended that the child had hit his or her finger, and simulated distress and pain. Then, the parent said the finger was “all better”.

Coding

Because for young children, another’s distress is a very complex stimulus that elicits a complex set of emotional responses (Zahn-Waxler et al., 1992), our coding was purposely comprehensive. It encompassed microscopic codes and overall ratings. The microscopic codes were assigned to each 5-sec segment in Family Study and Parent-Child Study (3-sec segment in Play Study). Those captured the child’s various facial, behavioral, and verbal emotional expressions, and included (a) signs of concern (e.g., looking at parent, sad/concerned expression, gestures of reparation/affection, verbal statements about reparation and verbal concern about parent), (b) signs of discomfort/distress (e.g., looking away, avoidance, squirming, covering face, blaming self), and (c) signs of amusement/lack of concern (e.g., happy expression, continuing uninterrupted play).

The overall ratings were assigned to the whole paradigm, for child distress (1=none, 2= mild, 3=moderate, 4=strong) and for empathic concern (from 1=appears to enjoy distress, blames parent, hits; 2=ignores, indifferent; 3=comforts in a mild, perfunctory manner; 5=spontaneously comforts, apologizes, kisses the hurt finger). The coding started when the parent began to simulate distress and lasted, on average, in Family Study (with mother first, father second), at 25 months, 49 and 46 seconds, at 38 months, 28 and 29 seconds, and at 52 months, 36 and 35 seconds; in Play Study, 29 seconds; in Parent-Child Study, 58 seconds.

The coding reliabilities were as follows. In Family Study, at 25 months, alphas were above .70 and kappas above .66; at 38 months, alphas above .95 and kappas above .64; at 52 months, alphas above .97 and kappas above .61 (except one, which was .57). In Play Study, ICCs were all above .95, and kappas all above .76. In Parent-Child Study, kappas were all above .64.

Data reduction

Data reduction encompassed the aggregation of the microscopic, 5-sec behavioral codes and the overall ratings. For each parent and child, first, composites were created by tallying and averaging the pertinent codes, as described above, into composites of (a), concern, (b) discomfort/distress, and (c) amusement/lack of concern. Then, the final composite was created by standardizing and averaging those three composites (having reversed amusement/ lack of concern), as well as the overall (standardized) ratings of child distress and empathic concern. The comprehensive coding approach yielded internally consistent measures of child emotional empathic response. Cronbach’s alphas for the composites were: in Family Study, for the composites with mother and father, respectively: at 25 months, .71 and .78, at 38 months, .73 and .73, and at 52 months, .84 and .79; in Play Study, .79; and in Parent-Child Study, .81.

The composite scores in Family Study were subsequently standardized and averaged across 25, 38, and 52 months, for the child with each parent, as in past work (Kochanska et al., 2010). We thus created the overall child empathy composites with mother, M = −.01, SD = .63, range −1.36 – 1.56, and with father, M = −.00, SD = .68, range −1.67 – 1.70.

Measures of Children’s Attachment Security with Parents

Strange Situation Paradigm (SSP)

The classic SSP was conducted in Family Study with each parent and in Parent-Child Study with the mother, at the beginning of the session at ages 15 months and 14 months, respectively (see, for example, Kochanska and Kim, 2013, for Family Study; Kochanska et al., 2004, for Parent-Child Study). Professional coders, blind to all other information about the families, coded children’s attachment organization. Although more specific attachment categories were coded, in this article, we use the distinction between insecure (A, C, D/U) and secure (B). The numbers of insecure and secure children in each study are in Table 1. In Family Study, security status with one parent was unrelated to that with the other parent, Pearson Chi square (1) = 1.67, ns. Reliability of coding, kappas, were .78 in Family Study and .90 in Parent-Child Study. All cases coded with low confidence by one coder and all D/U cases were double coded and adjudicated.

Table 1.

Descriptive Statistics: Means and Standard Deviations for All Constructs in Three Studies.

		Family Study						Play Study	Parent-Child Study
		15 Mo.	25 Mo.	38 Mo.	52 Mo.	67 Mo.	80 Mo.	30 Mo.	14 Mo.	22 Mo.	45 Mo.
Empathya	M-C		.00 (.68)	.00 (.69)	−.00 (.78)			.00 (.74)			.00 (.75)
	F-C		.00 (.73)	−.00 (.69)	.00 (.74)
Security	M-C	45b/56c						.19d (.20)	50b/58c	.48d (.20)
	F-C	34b/66c
Prosociality with Peerse	M-rated					2.26 (.34)	2.31 (.34)	1.25 (.40)
	F-rated					2.14 (.38)	2.20 (.34)

M=Mother. F=Father. C=Child. Mo.= Months.

^aAll empathy scores represent composites of standardized constituent variables, behaviorally coded.

^bSecurity in Strange Situation: Number of insecure children.

^cSecurity in Strange Situation: Number of secure children

^dSecurity in AQS (performed by observers in Play Study, and by mothers in Parent-Child Study).

^eHBQ Prosocial Behavior Scale in Family Study, ITSEA Prosocial Peer Relations Scale in Play Study.

Attachment Q-Set (AQS, Version 3.0; Waters, 1987; Waters & Deane, 1985)

In Play Study, highly trained coders observed the child and parent during the entire mother-child laboratory session (2 hr 45 min) in psychologically diverse contexts, and then performed the sort (Boldt, Kochanska, Yoon, & Koenig Nordling, 2014). Inter-observer reliability, ICCs, across several teams of coders ranged from .78 to .90. In Parent-Child Study, the mothers performed the sort at age 22 months. Data reduction in all studies was conducted according to standard guidelines (after 90 cards had been sorted into nine 10-card piles ranging from 1 = “most uncharacteristic” to 9 = “most characteristic’, the sort was correlated with the prototypical “secure child” criterion sort).

Measures of Children’s Prosociality in Peer Contexts

In Family Study, at ages 67 months and 80 months, mothers and fathers completed the Prosocial Behavior scale as part of Health Behavior Questionnaire (HBQ; Essex et al., 2002). It has 20 items (e.g., offers to share, helps, tries to be fair, considerate of others, comforts others, rated from 1 = rarely applies to 3 = certainly applies). Cronbach’s alphas were as follows: at 67 months, for mothers and fathers, .88 and .91; at 80 months, .89 and .88. The scores correlated across the two assessments: for mothers, r(88) = .69, p < .001, for fathers, r(85) =.63, p < .001, and were combined into one score for each parent; mothers, M = 2.28, SD = .32, range 1.58 – 2.28, fathers, M = 2.18, SD = .33, range 1.44 – 2.90.

In Play Study, mothers completed Infant–Toddler Social and Emotional Assessment, (ITSEA; Briggs-Gowan, Carter, Bosson-Heenan, Guyer, & Horwitz, 2006; Carter, Briggs-Gowan, Jones, & Little, 2003), an established instrument that has shown good psychometric qualities in large studies. We used the scale of Prosocial Peer Relations (5 items, e.g., takes turns when playing, asks for things nicely, rated 0= not true/rarely, 1= somewhat true/sometimes, or 2= very true/often); Cronbach’s alpha was .57. Table 1 presents the overview of all three studies and descriptive data for all measures.

Overall Results

Overview of Analytic Approach

We first compared (in Family Study and Parent-Child Study) children who returned vs. did not return for the final assessment. There were no significant differences for any construct. Next, as a preliminary step in all studies, we examined correlations among the variables.

In the main analyses, we tested the proposed model (Figure 1) for Family Study (separately for mother-child dyads and father-child dyads) and for Play Study (recall that in those two studies, data on children’s attachment security, empathy, and prosociality were available; but no data on prosociality were obtained in Parent-Child Study). Toward that end, we implemented PROCESS (Hayes, 2013; Model 74), a versatile computational tool to investigate the concurrent combination of mediation and moderation in various models. We tested the indirect effects from child security to empathy to prosociality, and the moderation of the empathy-prosociality link by security. Once the proposed significant moderation effect was supported (child security found to moderate the link between child empathy and prosociality), we conducted the follow-up analyses of that interaction using simple slopes and regions-of-significance (RoS) analyses (Aiken & West, 1991; Hayes & Matthes, 2009; Preacher, Curran, & Bauer, 2006).

In Parent-Child Study, data were available for child security (assessed in both Strange Situation and through AQS) and child empathy. We tested the associations between security and empathy in a straightforward hierarchical multiple regression.

Given that the rates of missing data were low – Family Study, 10%, Play Study, 6% for mothers’ ITSEA ratings, and Parent-Child Study, 6% – we employed listwise deletion, the default option in PROCESS. Child age was covaried in Play Study (recall that age ranged from 22 to 44 months). Children’s security in SSP was coded as 0=insecure and 1=secure (Family Study, Parent-Child Study). In Play Study and Parent-Child Study, AQS scores were continuous (in Figure 5, it was dichotomized exclusively for the ease of graphing and conveying RoS).

Figure 5.

Play Study: The child’s attachment with the mother (observer’s AQS, 30 months) moderates the effect of empathy to mother’s distress (observed, 30 months) on prosociality with peers (mother-rated, 30 months). The solid line represents a significant simple slope; the dashed line represents a nonsignificant simple slope. The shaded area is the region of significance.

Preliminary Analyses

We examined the correlations for all constructs (only significant correlations are reported below). In Family Study, for mother-child dyads, children’s security in SSP modestly, positively related to their empathy, r (99) = .20, p < .05 (point-biserial), and empathy related to prosociality, r (90) = .26, p < .025. There were no significant correlations for father-child dyads. Additionally, the child’s empathy to the mother and father was positively correlated, r (99) = .51, p < .01, and so were the parents’ ratings of child prosociality, r (88) =.39, p < .01.

In Play Study, children security in AQS modestly, positively related to their empathy, r (183) = .24, p < .01, and to prosociality, r (173) = .22, p < .01. In Parent-Child Study, children’s security in SSP was positively related to their empathy, r (97) = .26, p < .01 (point-biserial), but security in AQS was unrelated to empathy (recall that no prosociality data were available). We entered both security scores in a multiple regression; security in SSP predicted empathy, B = .39, SE = .15, 95% CI [.10, .69]; security in AQS was not a significant predictor, B = .20, SE = .38, 95% CI [−.55, .95]. For the overall equation, R² = .07, F (2, 95) = 3.58, p < .05.

Testing the Model in Comprehensive Moderated Mediation Analyses

We conducted these analyses in two steps. (a) We used PROCESS (Hayes, 2013, Model 74) to test our proposed conceptual model (Figure 1). PROCESS allows for examination of indirect effects of a predictor on an outcome through a mediator, depending on varying values of the predictor. The indirect effects were estimated by the bootstrapping method. This method is particularly advantageous with small- to moderate-size samples, due to the absence of assumptions on the sampling distribution of the conditional indirect effects and a specific formula for the standard error (MacKinnon, Lockwood, & Williams, 2004; Shrout & Bolger, 2002). (b) We followed up the significant moderation effects, using simple slopes and RoS.

Family Study: Mothers and children

PROCESS results are in Figure 2A. The child’s attachment organization with the mother in SSP at 15 months (insecure versus secure) was modeled as the predictor, the child’s empathy to the mother’s distress at 25–52 months as the mediator, and the mother-rated prosociality at 67–80 months as the outcome. The predictor, attachment, was also assumed to moderate the effect of the mediator (empathy) on the outcome (prosociality).

Figure 2.

Family Study: The moderated mediation model of the path from the child’s attachment with the parent (SSP, 15 months) to empathy to parent’s distress (observed, 25–52 months) to prosociality with peers (parent-rated, 67–80 months). The solid line represents a significant effect; the dashed line represents a nonsignificant effect. A: mother-child dyads, B: father-child dyads. M=Mother, F=Father, C=Child, Mo=Months. + p < .10. *p < .05. ** p < .01.

The child’s security with the mother predicted greater empathy to the mother’s distress; and an increase in empathy, in turn, led to an increase in prosociality. Further, the moderation effect of attachment security on the link between empathy and prosociality was also significant. The significant main effect of attachment security on empathy and the significant moderation effect suggest that the indirect effects from security to empathy to prosociality vary depending on the value of security. In insecure mother-child dyads, the indirect effect was present: B = .07, SE = .04, and its bias-corrected bootstrap confidence interval, CI, [.01, .19], did not include zero at .05 alpha level. By contrast, in secure mother-child dyads, the indirect effect was absent: B = .01, SE = .02, and its bias-corrected bootstrap CI, [−.04, .06], did include zero at .05 alpha level.

To probe the significant moderation effects in Figure 2A, we estimated the simple slopes for children who had been insecure and those who had been secure with their mothers in Figure 3A (Aiken &West, 1991). Furthermore, we examined the RoS where the levels of prosociality for insecure and secure children were significantly different (Hayes & Matthes, 2009; Preacher et al., 2006). For the graphing purpose for lower and upper bounds of the RoS, the levels of children’s empathy were expanded from −2 SD (very low) to +2 SD (very high) beyond the conventional −1 SD (low) and +1 SD (high) range around the mean.

Figure 3.

Family Study: The child’s attachment with the parent (SSP, 15 months) moderates the effect of empathy to parent’s distress (observed, 25–52 months) on prosociality with peers (parent-rated, 67–80 months). The solid line represents a significant simple slope; the dashed line represents a nonsignificant simple slope. The shaded area is the region of significance. A: mother-child dyads, B: father-child dyads.

In Figure 3A, the simple slope of children’s empathy to their mothers on prosociality was significant for children who had been insecure with their mothers, B = .23, SE = .07, p < .01, but not for those who had been secure, B = .02, SE = .07, ns. Thus, the variation in children’s empathy to mothers was associated with future prosociality only for children who had been insecure. There was no such link in secure mother-child dyads. The lower and upper bounds of RoS were −.90 and 4.89 respectively (note that although calculable, the upper bound is well beyond the observed range from −1.36 to 1.56, and is thus of no practical utility). Therefore, we can infer that the predicted scores of the two regression lines are significantly different only below the lower bound (approximately −1.5 SD; recall that M=−.01, SD = .63). In other words, among children with low empathy (below approximately 1.5 SD below the mean), insecure children were significantly less prosocial than their secure peers. The shaded area in Figure 3A represents the RoS.

Family Study: Fathers and children

In Figure 2B, the moderated mediation model was tested for father-child dyads. Children’s attachment with their fathers did not have a significant effect on child empathy to the father’s distress. An increase in child empathy, however, led to an increase in prosociality. Although the moderation effect of attachment security on the link between empathy and prosociality was significant, the indirect effects from security to empathy to prosociality were absent in both insecure, B = −.02, SE = .04, bias-corrected bootstrap CI, [−.11, .04], and secure, B = −.01, SE = .01, bias-corrected bootstrap CI, [−.04, .002] father-child dyads.

In Figure 3B, we graphed the significant interaction. The simple slope of children’s empathy to their fathers on their prosociality was significant only for children who had been insecure with their fathers (B = .23, SE = .09, p < .05), but not for those who had been secure (B = .02, SE = .06, ns). Thus, the variation in children’s empathy to their fathers was associated with future prosociality only for children who have been insecure. There was no such link in secure father-child dyads. The lower and upper bounds of RoS were −.01 and 75.63, respectively (again, although calculable, the upper bound is of no practical utility, given that the observed range was from −1.67 to 1.70). Therefore, we can infer that the predicted scores of the two regression lines are significantly different only below the lower bound (which was close to the mean; recall M=−.00, SD=.68): Among children with lower-than-average empathy (approximately below the mean), insecure children were significantly less prosocial than their secure peers. The shaded area in Figure 3B represents the RoS.

Play Study

In Figure 4, the child’s attachment security with the mother (AQS) was modeled as the predictor, the child’s empathy to the mother’s distress as the mediator, and the mother-rated child prosociality as the outcome (note that the predictor, security, is continuous). Security was also assumed to moderate the effect of the mediator on the outcome.

Figure 4.

Play Study: The moderated mediation model of the path from the child’s attachment with the mother (observer’s AQS, 30 months) to empathy to mother’s distress (observed, 30 months) to prosociality with peers (mother-rated, 30 months). The solid line represents a significant effect; the dashed line represents a nonsignificant effect. Child age was covaried. M=Mother, C=Child, Mo = Months. *p< .05. **p < .01.

An increase in AQS score was associated with the increase in empathy, but empathy had no significant effect on prosociality. The moderation effect of attachment security on the link between empathy and prosociality was significant.

With regard to the indirect effects, when the level of security was high (+1 SD), the indirect effect was absent: B = −.01, SE = .01, and its bias-corrected bootstrap CI, [−.04, .002] did include zero at .05 alpha level. In contrast, when attachment security was low (−1 SD), the indirect effect was present: B = .02, SE = .01, and its bias-corrected bootstrap CI, [.002, .05] did not include zero at .05 alpha level. Consequently, we infer the presence of an indirect effect from child security to empathy to prosociality only for children with low AQS scores.

Prior to conducting the simple slope analysis, we dichotomized the AQS security scores at the median exclusively for the ease and consistency of graphing and conveying the RoS. In Figure 5, the simple slope of children’s empathy to their mothers on prosociality was significant for the children with the relatively lower security, B = .14, SE = .06, p < .05, but not for those with relatively higher security, B = −.08, SE = .05, ns. Thus, the variation in children’s empathy to their mothers was associated with future prosociality only for the children who were lower in security. There was no such effect for the higher-security children. The lower and upper bounds of RoS were .17 and 3.19 respectively (again, although calculable, the upper bound is well beyond the observed range, −1.32 to 1.93, and is thus of no practical utility). Therefore, we can infer that the predicted scores of the two regression lines are significantly different only below the lower bound (approximately 1/4^th SD above the mean; recall M= .00, SD=.74): Among children with low empathy (whose empathy scores were lower than approximately 1/4^th SD above the mean), insecure children were significantly less prosocial than their more secure peers. The shaded area in Figure 5 represents the RoS.

Overall Discussion

It is broadly believed that a secure attachment relationship constitutes a natural ecology for the development of the child’s emotion of empathy (Shaver et al., 2016); and yet, empirical evidence regarding associations between attachment and empathy is fragmented, thin, and inconsistent. The main objective of this research was to address associations among young children’s attachment organization, their empathy toward the parents’ distress, and their prosociality in peer relations. In so doing, we heeded recommendations articulated by Shaver and colleagues (2016): We defined security, empathy, and prosociality as separate constructs and assessed them using distinct paradigms and measures.

Our approach embraces the recently growing strong emphasis on the key importance of replicability in behavioral sciences. We drew from three large studies: two longitudinal studies of community families and a study of a high-risk, low-income, diverse group of mothers and children. The availability of relatively large and quite distinct samples enhances robustness and generalizability of our findings. We utilized a multi-method approach that included observations in scripted, standardized paradigms (SSP, simulated parental distress) and lengthy naturalistic interactions (observer-rated AQS), as well as established parents’ ratings (mother-rated AQS and ITSEA, mother- and father-rated HBQ). Most of the methodologies were fully parallel across the studies, which is also a substantial strength.

Several findings were impressively replicated across the studies. In all three studies, in mother-child relationships, early security, observed in SSP or in naturalistic mother-child interactions by trained observers who performed AQS, was positively associated with children’s empathy to maternal distress. Although the effect size of the associations was in small-to-medium range, this nevertheless provides robust empirical support for the widespread conceptual belief in such links (Shaver et al., 2016). It is unclear why we failed to find this effect for father-child relationships. It has been suggested that mother–child and father–child attachment may have different implications (Berlin, Cassidy, & Appleyard, 2008; Bretherton, 2010; Thompson, 2006). Complex and nuanced differences in how mothers and fathers emotionally communicate with their infants have been reported (Moller, Majdandzic, & Bogels, 2014). Perhaps in the context of early care, mothers engage more in comforting and reassurance in response to child distress than do fathers (recall that Shaver et al., 2016, posited that it may be one of the main mechanisms of early empathy). Some studies of older children have found fathers to be less supportive of children’s negative emotions than mothers. For example, Eisenberg, Fabes, and Murphy (1996) found that fathers reported more punitive and minimizing response to children’s negative emotions, whereas mothers reported more encouragement of expression, and emotion- and problem-focused strategies. McElwain, Halberstadt, and Volling (2007) obtained a similar pattern of findings with regard to parents’ reports of their response to children’s negative emotions. On the other hand, fathers may be more supportive of children’s emotions in play contexts (MacDonald & Parke, 1984). More research is certainly needed.

In Family Study and Play Study, we replicated, across mother-child and father-child dyads, the link between empathy and prosociality; however, in both studies, that link was present only for insecure children. Further, the entire cascade from security to empathy to prosociality was present only for insecure children, in mother-child relationships in Family Study and Play Study. Specifically, unempathic children who were also insecure were significantly less prosocial than their equally unempathic – but secure – peers. Those findings add to the accumulated evidence of differential socialization dynamics that unfold in insecure and secure relationships, now documented across several longitudinal studies, different populations, various children’s ages, and a range of measures. The emerging picture is consistent. For example, children who had difficult, angry temperaments (Kochanska & Kim, 2012), highly negative (Boldt et al., 2016), or showed relatively little remorse following their transgressions (Kim et al., 2014) had more maladaptive outcomes, but only if they were also insecurely attached to their parents. That risk appears defused in secure parent-child dyads.

To explain the current and past findings, we have offered a general hypothesis why, in secure dyads, children may still achieve positive outcomes despite the presence of a risk factor, such as negativity, low remorse, or, as in the current study – low empathy. We argue that perhaps in those dyads, risks are defused because multiple alternative mechanisms, triggered by security, may be involved in promoting positive developmental outcomes. In the present work, in secure parent-child relationships, several mechanisms other than empathy may lead to prosociality with peers. Examples include modeling of prosocial behavior by parents, children’s internalization of parents’ prosocial values, or effective emotion regulation that facilitates social competence in interactions with peers (Shaver et al., 2016).

Despite strengths, this work has several limitations. Most importantly, the measures of security, empathy, and prosociality in Play Study were concurrent. Consequently, our ability to draw inferences about directions of effects is severely limited. Note that even in Family Study and Parent-Child Study, in which the measures were not concurrent, causal inferences should be made with great caution, because no experimental designs were employed. It is, of course, more likely (and, in fact, has been shown, Panfile & Laible, 2013) that security with the caregiver contributes to the child’s empathy than vice versa. Nevertheless, our conclusions essentially refer to associations among constructs.

Our measures of children’s empathy focused on their responses to parental simulation of distress; children were led to believe that they had caused the parent’s pain and distress. Although observational methodology, parallel across all studies, was robust, it captured only one type of eliciting event. Future research should employ a broader range of stimuli to elicit empathy, for example, distress that is witnessed as well as caused by the child, and a broader range of negative emotions expressed by another person (e.g., fear or sadness). Perhaps most importantly, we assessed the child’s security and empathy within the same relationship (mother-child and father-child). Given the early stage of this research, it seemed the most clear-cut approach. It will be critical, however, to expand empathy measures to include distress expressed by adults and children other than parents. We would expect that child empathy would likely generalize to other recipients. As well, including biological measures of empathic responding would be advantageous (Van Hulle et al., 2013).

Further, we note that our coding of child empathy targeted several components: child expressions of concern, distress, and lack of concern or disregard for the parent’s pain. Although our composites, integrating those measures, evinced robust coherence (Cronbach alphas .71--.84 across studies and assessments), there is evidence that those components may have different implications for future development (Rhee et al., 2013). Consequently, we conducted exploratory analyses using the individual components instead of the composite. Those measures, however, did not perform as well as the overall empathy composite, and by and large, those exploratory analyses failed to replicate consistently the effects obtained using the composite. We therefore concluded that the integrative and broad approach to young children’s observed emotional of empathy was useful and appropriate in the context of our model.

Of course, future studies using more precise empathy measures will be important. More nuanced measures of empathy – for example, distinct assessments of personal distress versus other-oriented distress – may provide new insights with regard to our findings that showed that at high levels of empathy, both insecure and secure children were seen as prosocial. It is theoretically possible that insecure and secure children’s empathy differs in its quality – perhaps reflecting personal distress and other-oriented distress, respectively – but given sufficient quantity, it will be associated with prosociality.

Additionally, our measures of children’s prosociality were all based on parental reports of children’s functioning in peer relations. Although such measures are valuable, employing behavioral observations of children’s peer interactions would be much desired in future work. Additionally, internal consistency of the measure in Play Study was relatively low, perhaps due to the small number of items, or to children’s limited peer contacts at the young age.

Although child gender was not covaried in the analyses presented, we conducted all analyses controlling for gender, and found all results virtually unchanged. There was, however, one exception: The interaction between child security and empathy predicting prosociality in father-child dyads in Family Study became marginally significant, p = .10. Of note, the graphed slopes and RoS remained unchanged and consistent with the hypotheses. Nevertheless, that interaction effect should be interpreted with caution and replicated in future research.

Overall, the present work contributes to an integration of research on parent-child attachment and children’s emerging emotion of empathy and more generally, positive, prosocial developmental trajectories. Such integration has been long urged and conceptually anticipated by scholars studying emotional development, but empirical research has lagged behind. It remains an important goal for developmental science.