Infant–Mother Attachment and Children’s Friendship Quality: Maternal Mental-State Talk as an Intervening Mechanism

Abstract

Utilizing data from the Eunice Kennedy Shriver National Institute of Child Health and Human Development’s Study of Early Child Care and Youth Development, we investigated mothers’ talk about mental states during play with their 24-month-old children as a mechanism though which infant–mother attachment was associated with children’s later friendship quality. A series of repeated measures analyses of covariance indicated that a secure versus avoidant or disorganized infant–mother attachment was associated with more maternal talk about cognitions (but not emotions or desires) at 24 months. Latent growth curve models tested within a structural equation modeling framework revealed indirect effects of infant–mother attachment on observed and mother-reported positive friendship interaction at 54 months and decreases in mother-reported negative friendship interaction from 54 months to 1st grade via maternal cognitive talk at 24 months.

A central tenet of attachment theory is that early attachment relationships set the stage for children’s later interpersonal functioning. A conservative view suggests that child–caregiver attachment relationships lay the foundation for functioning in close relationships (e.g., friendships) in particular (see Belsky & Cassidy, 1994). Accumulating evidence supports this claim. A secure infant–mother attachment, for instance, has been associated with lower levels of child–friend aggression at age 3 (McElwain, Cox, Burchinal, & Macfie, 2003), less negative interactions with a close friend at age 5 (Youngblade & Belsky, 1992), and more competence in forming close friendships at age 10 (Freitag, Belsky, Grossmann, Grossmann, & Scheurer-Englisch, 1996). Further, a meta-analysis of 63 studies examining attachment–peer associations found significantly larger effects for associations with children’s interactions with friends versus unfamiliar peers (see Schneider, Atkinson, & Tardiff, 2001). Despite accumulating evidence of attachment–friend linkages, little is known about the mechanisms accounting for such associations and, in particular, the role that the mother–child relationship, postinfancy, may play.

A critical next step in this line of inquiry, therefore, is to go beyond prediction and elucidate the processes through which early attachment relationships are associated with children’s later relationships. Through daily repeated interactions with a sensitive caregiver, the infant is expected to develop a secure attachment in which he or she constructs an internal working model of the self as worthy of care and the caregiver as available when needed (Bowlby, 1969/1982), which is then carried forward into the child’s later interactions with close others (Sroufe & Fleeson, 1986). Following from the notion of the internal working model, initial efforts to identify processes of attachment–peer linkages have largely focused on intrapersonal mechanisms, such as the child’s peer-related representations or perceptions of self-worth (see Berlin, Cassidy, & Appleyard, 2008; Booth-LaForce & Kerns, 2009). Yet, attachment security stems from ongoing transactions between child–caregiver open communication and the child’s developing representations of those experiences. In this vein, Kobak and Madsen (2008) called for greater attention to interpersonal processes— especially communicative processes— of attachment. We aimed to address this call by investigating mothers’ mental-state talk as a mechanism of attachment–friend linkages.

Our focus on maternal talk about mental states (desires, emotions, cognitions), in particular, is guided by two related but separate lines of inquiry. The first is growing emphasis by attachment theorists regarding the role of the caregiver’s attunement to and communication about mental states in fostering attachment security and, in turn, psychological discourse and understanding (e.g., Fonagy & Target, 1997; Meins, 1997). The second highlights the role of mental-state discourse and understanding for children’s peer functioning, especially friendships (e.g., Brown, Donelan-McCall, & Dunn, 1996). In bridging these literatures, we utilized data from the Eunice Kennedy Shriver National Institute of Child Health and Human Development (NICHD) Study of Early Child Care and Youth Development (SECCYD) to examine maternal mental-state talk during mother–child play at 24 months as a mechanism linking infant–mother attachment and children’s friendship quality from 54 months to first grade.

Maternal Attunement to Mental States and Child–Mother Attachment

Attachment theorists have long recognized that the caregiver’s conception of the child as a separate person with his or her own thoughts, feelings, and desires undergirds the caregiver’s sensitive responsiveness that promotes security (Ainsworth, Blehar, Waters, & Wall, 1978; Bowlby, 1969/1982). Recent reformulations of attachment theory, however, have brought the caregiver’s attunement to and communication about mental states to the forefront. In these accounts, the caregiver’s reflective function—“the mental function which organizes the experience of one’s own and others’ behaviors in terms of mental state constructs” (Fonagy & Target, 1997, p. 680)—and, similarly, the mother’s mind-mindedness— “her recognition of her child as a mental agent, and her proclivity to employ mental states in her speech” (Meins, 1997, p. 127)—are posited to be key. Importantly, the caregiver’s reflective capacity or mind-mindedness enables appropriate or accurate communication about the infant’s internal state versus communication that is inappropriate or mismatched (Fonagy et al., 1995; Meins, Fernyhough, Fradley, & Tuckey, 2001). In accordance with this theorizing, prenatal assessments of mothers’ and fathers’ propensity to discuss past attachment experiences in mentalistic terms (Fonagy, Steele, Steele, Moran, & Higgitt, 1991) and mothers’ appropriate mind-related comments (i.e., comments congruent with the infant’s behavioral cues) during play at 6 months (Meins et al., 2001) predicted infant–parent attachment security at the end of the first year. Notably, the latter association emerged above and beyond the degree to which mothers responded sensitively (Meins et al., 2001) and, thus, highlights the unique role that attunement to and talk about mental states may play in the emergence of a secure infant–caregiver attachment.

A secure attachment in infancy, in turn, is expected to provide a foundation for continued sharing of internal worlds through language during the toddler and preschool years (Bretherton, 1990; Thompson, 2000). That is, if mother–child communication and the child’s representations of those experiences are viewed as a dynamic process, it follows that a secure working model of self and other will not only be influenced by but will also influence the caregiver’s propensity to attune to and communicate about mental states. A secure working model may promote maternal mental-state talk because the child is willing and/or able to seek out the mother when needed and thereby provide opportunities for mental-state discourse (Main, Kaplan, & Cassidy, 1985), regulate behavior and affect to attend to such talk (Cassidy, 1994; Fonagy, 2004), and trust relevant information provided by the mother (Corriveau et al., 2009).

Prior research on the associations between early attachment security and later mental-state talk has focused primarily on talk about emotions. A secure infant–mother attachment, for instance, has been associated with mother–child open emotional communication at 4.5 years, but only for boys (Etzion-Carasso & Oppenheim, 2000), and with emotionally matched dialogue at 4.5 and 7.5 years (Oppenheim, Koren-Karie, & Sagi-Schwartz, 2007). In light of more recent emphases on mental-state attunement and talk (Fonagy & Target, 1997; Meins, 1997), however, studies investigating a broader array of maternal mental-state talk are needed. Although we know of no such prospective study, Mcquaid, Bigelow, McLaughlin, and MacLean (2008) reported a concurrent association between secure child–mother attachment and more appropriate mental-state talk by mothers during a coconstruction task with their preschool-age children. Yet, given the concurrent design and examination of a composite score that combined talk about emotions, cognitions, and desires, it remains unknown whether infant–mother attachment is prospectively associated with specific types of maternal mental-state talk other than emotions.

In this regard, maternal cognitive talk may prove a fruitful line of inquiry because, unlike emotions and desires, cognitions do not have corresponding behavioral cues (e.g., emotional expressions), and thus, talk about cognitions may provide stringent evidence for a mentalistic orientation. Intriguingly, infant–mother attachment security has been associated with toddlers’ use of cognitive terms at 30 months (Lemche, Kreppner, Joraschky, & Klann-Delius, 2007) and with preschoolers’ false-belief understanding (McElwain & Volling, 2004; Meins, Fernyhough, Russell, & Clark-Carter, 1998; but see Meins et al., 2002), further signifying the utility of investigating attachment-related differences in maternal cognitive talk, in particular.

Attunement to Mental States and Children’s Friendships

Mental-state discourse and understanding are also posited to be vital to children’s friendships. Young children often strive to maintain high levels of positive affect and excitement with friends and to minimize negative affect and conflict (Parker & Gottman, 1989). To do so successfully, children need to be mindful of their friends’ beliefs, desires, and feelings. Moreover, collaborative pretend play provides a natural context for mental-state talk because it requires children to distinguish reality from fantasy, negotiate pretend scripts and roles, and engage in shared meanings (Howes, Unger, & Matheson 1992; Lillard, 1998). Accordingly, preschool-age children engaged in significantly more cognitive talk with friends than mothers (Brown et al., 1996), and more cognitive talk during play with friends was associated with greater child–friend cooperation (Brown et al., 1996) and pretend play (Hughes & Dunn, 1997). Whereas these studies examined cognitive talk only, Hughes, Lecce, and Wilson (2007) investigated preschool-age children’s talk about cognitive and desire/emotion (combined) talk during separate play sessions with siblings and friends. Findings indicated that, although children talked more about desires and emotions than cognitions, children’s talk about their friends’ cognitive states was associated with better understanding of mental states (also see Hughes & Dunn, 1998). Children’s understanding of mental states (e.g., false beliefs) has been associated, in turn, with higher quality friendships (Dunn & Cutting, 1999; McElwain & Volling, 2004).

Of importance, the child’s ability to understand and talk about mental states, which appear central to friendship competence, is fostered by mental-state talk in the family. When family members— especially mothers—talk more about desires, emotions, and/or cognitions, children tend to use more mental-state language (Dunn, Bretherton, & Munn, 1987; Jenkins, Turrell, Kogushi, Lollis, & Ross, 2003), demonstrate advanced social understanding (Dunn, Brown, Slomkowski, Telsa, & Youngblade, 1991; Ruffman, Slade, & Crowe, 2002; Symons, Fossum, & Collins, 2006), and show greater conscience development (Laible & Thompson, 2000).

Maternal Mental-State Talk as an Intervening Mechanism

Together, these findings suggest that attachment security is associated with greater maternal attunement to and talk about mental states, that maternal talk about mental states predicts children’s mental-state talk and understanding, and that these abilities, in turn, have implications for friendship competence. Thus, in the context of a secure attachment, mothers may scaffold a psychological understanding of behavior through talk about beliefs, desires, and feelings, and adopting such a mentalistic orientation has ramifications for the quality of children’s friendships. Moreover, toddlerhood may be a critical period during which to assess maternal mental-state talk. As children’s autonomy seeking and negative emotional displays increase during this developmental phase (see Brownell & Kopp, 2007), mothers’ talk about beliefs, desires, and feelings also increases, and individual differences in maternal mental-state talk may become more pronounced (Brown & Dunn, 1991; Dunn et al., 1987). Also, toddler-age children demonstrate great strides in internal state language (e.g., Shatz, Wellman, & Silber, 1983) and, therefore, may benefit especially from maternal input.

No prior study has tested maternal mental-state talk as a mechanism of attachment–friend linkages, although findings from two studies are pertinent. Freitag et al. (1996) found that connected communication between mothers and their 10-year-old children partially mediated a longitudinal association between a composite of child–mother attachment security and competence in forming close friendships at age 10. Additionally, McElwain, Booth-LaForce, Lansford, Wu, and Dyer (2008) reported that open emotional communication between mothers and children at 54 months was part of an indirect pathway through which child–mother attachment security at 36 months was related to greater friendship quality in third grade. Neither study examined maternal mental-state talk specifically, yet the findings provide initial evidence for related aspects of mother–child communication as intervening mechanisms.

The Current Study

Utilizing data from the NICHD SECCYD, our primary objective was to test maternal mental-state talk during a mother–child play session at 24 months as an intervening mechanism between infant–mother attachment at 15 months and children’s subsequent friendship quality. The proposed process model is shown in Figure 1, and the hypothesized indirect pathway from infant–mother attachment to later friendship quality through maternal mental-state talk is in bold. Notably, controlling for potential third variables increases confidence in tests of indirect effects in nonexperimental studies (see Frazier, Tix, & Barron, 2004). Because maternal sensitivity has shown positive, although modest, associations with maternal mental-state talk (e.g., Meins et al., 2001) and children’s friendship quality (e.g., Gazelle & Spangler, 2007), we included maternal sensitivity during the mother–child play session at 24 months as a covariate to rule out the alternative explanation that hypothesized associations were accounted for by more sensitive mothering. Moreover, mothers who talk more about mental states may simply talk more in general or do so because their children talk more about mental states; thus, we also controlled for total maternal utterances during the play session and child mental-state talk at 24 months.

Figure 1.

Conceptual model depicting maternal mental-state talk at 24 months as an intervening mechanism between infant–mother attachment and subsequent friendship quality. The hypothesized indirect pathway is in bold. Maternal sensitivity, total maternal utterances, and child mental-state talk are included as covariates.

Though not shown in Figure 1, the process model was tested for two sets of friendship outcomes: (a) mother-reported friendship quality at 54 months, kindergarten, and first grade and (b) observed child–friend interaction at 54 months. Further, because mental-state talk has been related to positive but not negative aspects of friendship during the preschool period (Brown et al., 1996; Hughes & Dunn, 1997), we tested the process model separately for positive and negative child–friend interaction. Children’s social reasoning abilities undergo marked change between 4 and 7 years of age, which may account for general declines in negative peer interactions during this period (see Fabes, Martin, & Hanish, 2009). Thus, the benefits of maternal mental-state talk for managing negative friendship interactions (e.g., competition, aggression) may not emerge as early as 54 months but may be associated with more rapid declines in negative interaction between 54 months and first grade. Repeated assessments of mother-reported friendship quality enabled an investigation of this possibility as maternal mental-state talk was examined as a predictor of initial levels of, as well as change in, positive and negative friendship interaction.

In view of theory and evidence underscoring the differential antecedents and sequelae of the avoidant, resistant, and disorganized attachment groups (see Ainsworth et al., 1978; Bowlby, 1969/1982; Cassidy, 1994; Main & Solomon, 1990), we differentiated among the insecure groups. In particular, the defensive processes characterizing an insecure–avoidant attachment, which stem from interactions with a consistently rejecting caregiver and a corresponding working model of the caregiver as unavailable for support, may impede mental-state discourse. For instance, infants classified as avoidant (vs. secure or resistant) tended to have mothers who exhibited lower levels of reflective capacity (Fonagy et al., 1991) and made fewer appropriate mind-related comments (Meins et al., 2001). Relatedly, Main et al. (1985) reported that an avoidant attachment history was associated with more restricted parent–child discourse at 6 years (e.g., talk about impersonal topics and inanimate objects). A disorganized (vs. secure) attachment history—in which the infant lacks a coherent, organized strategy for coping with attachment-related distress because of experiences with a caregiver who is frightened and/or frightening— has also been associated with more mismatched or dysfluent parent–child discourse (Etzion-Carasso & Oppenheim, 2000; Main et al., 1985; Oppenheim et al., 2007), whereas evidence of impaired mental-state talk among children in the resistant group has been equivocal (e.g., Etzion-Carasso & Oppenheim, 2000). We expected, therefore, that the hypothesized process depicted in Figure 1 would be pronounced for comparisons of the secure versus avoidant and disorganized attachment groups.

Because a significant association between the predictor (attachment) and putative mechanism (mental-state talk) is a first and necessary criterion for detecting indirect effects (see Kenny, Kashy, & Bolger, 1998), and because so few studies have examined infant–mother attachment and maternal mental-state talk prospectively, a secondary objective was to conduct a fine-grained assessment of such associations. To this end, mothers’ mental-state talk was coded for type (desires, emotions, cognitions) and referent (child, self, other), and child references were additionally coded as appropriate or inappropriate. On the basis of prior theory and evidence, we expected that a secure infant–mother attachment would be associated with more frequent talk about emotions and cognitions. Our examination of desire talk was exploratory yet important given the prevalence of desire talk among mothers and their toddler-age children (Bartsch & Wellman, 1995; Brown & Dunn, 1991). We also expected that for a given type of mental-state talk, a secure attachment would relate to more appropriate and fewer inappropriate references to the child’s mental states. Prior studies have typically focused on mothers’ talk about the child’s mental states, and we explored whether attachment-related differences extended to mothers’ references to her own and others’ mental states. We also explored child gender as a moderator of hypothesized associations in light of prior findings that child–mother attachment was associated with emotional communication (Etzion-Carasso & Oppenheim, 2000) and emotional communication was associated with peer competence (McElwain et al., 2008) for boys only.

Method

Participants

The NICHD SECCYD is a prospective study of the links between early environments and children’s development (see NICHD Early Child Care Research Network [ECCRN], 2005). The study initially enrolled 1,364 one-month-old infants and their families located in or near 10 sites across the United States. Because 115 participants were missing data on all study measures examined in this article, we utilized data from a subsample of 1,249 participants. Mothers averaged 28.3 years of age (SD = 5.61) and 14.3 years of education (SD = 2.49), as reported by mothers when study children were 1 month old. The average family income-to-needs ratio (combined across data collected at 6 and 15 months of age) was 3.7 (SD = 3.03). For 82% of the families, the mother’s partner or spouse lived in the home (combined across data collected at 1, 6, and 15 months). Of the study children, 48% (n = 604) were female. With respect to child ethnicity, 77% were European American non-Hispanic, 12% were African American non-Hispanic, 6% were Hispanic, and 5% were another race or more than one race. When compared with the 115 families excluded because of missing data, families included were characterized, on average, by higher maternal age, maternal education, and family income-to-needs ratio and were more likely to have a partner living in the home. Children included versus excluded were more likely to be European American. No significant difference emerged for child gender.

Overview of Procedures

Infant–mother attachment was assessed during a laboratory visit at 15 months. A laboratory mother–child play session at 24 months was originally intended as a developmentally appropriate context in which to assess key aspects of mother–child interaction (e.g., maternal sensitivity and intrusiveness) across multiple ages during early childhood (see NICHD ECCRN, 1999, for further details). For the purposes of this article, the mother–child play session at 24 months was recoded for mothers’ mental-state talk. Children’s friendship quality was assessed with (a) maternal reports at 54 months, kindergarten, and first grade and (b) observations of child–friend interaction at 54 months. To rule out potential third variables, maternal sensitivity and total maternal utterances were each assessed from the mother–child play session at 24 months, and children’s mental-state talk was measured by means of maternal report at 24 months. All videotapes for a given assessment (i.e., attachment, sensitivity, mental-state talk, child–friend interaction) were coded at one site, and coders were independent across assessments and unaware of other child and maternal characteristics. Additional details about study procedures and measures can be found on the NICHD SECCYD website (http://secc.rti.org). Because maternal mental-state talk was not coded as part of the NICHD SECCYD, information about the mental-state talk codes is not available on the study website, and we describe this coding scheme in detail below.

Infant–Mother Attachment

Mothers and their 15-month-old infants visited a laboratory playroom and were videotaped in the Strange Situation procedure. Following Ainsworth et al. (1978) and Main and Solomon (1990), infants were classified as insecure–avoidant (A), secure (B), insecure–resistant (C), disorganized (D), or unclassifiable (U) by a team of trained coders. All Strange Situation assessments were double coded, and disagreements were resolved by consensus. Interobserver agreement (before conferencing) was 83% (κ = .69) for the five-category classification system (ABCDU). Infants who received an unclassifiable code (n = 41, 3.4%) were not included in the analyses. Of the 1,149 cases with attachment data, 160 (13.9%) were classified as insecure–avoidant, 710 (61.8%) as secure, 102 (8.9%) as insecure–resistant, and 177 (15.4%) as disorganized.

Maternal Mental-State Talk

During a laboratory visit at 24 months, mother–child dyads were videotaped in a 15-min semistructured play session. The mother was presented with three boxes, each of which contained an age-appropriate toy, and instructed to have her child play with the toys in the following order: (a) storybook (Barnyard Tracks by Dee Dee Duffy; this book did not contain any mental-state terms), (b) toy stove, and (c) toy house. From video recordings of the mother–child play session, maternal utterances and mental-state talk were assessed. We chose to assess mental-state talk in this context because the child-centered nature of the play session provided ample opportunity for mothers to focus on the thoughts, desires, and feelings of their children, as the interaction was not constrained by competing goals or caregiving tasks (also see Meins et al., 2001). Moreover, maternal mental-state talk during play may be especially relevant to developing skills children need to effectively manage interactions with friends.

In line with Shatz and Gelman (1973), an utterance was defined as a complete clause (i.e., a group of words forming a statement or a question), which could include a complete sentence (e.g., “That’s all that’s in there”) or a sentence fragment (e.g., “Down the chimney”). A single utterance could include more than one mental-state term. Any phrase or sentence in which the mother was reading the actual words of the storybook was flagged and not coded as a maternal utterance. False starts, nonwords (fillers, play noises), and exact, immediate repetitions of an utterance were not counted as utterances and, therefore, were not coded for mental-state terms. In coding mothers’ mental-state talk, utterances containing one or more mental-state terms were transcribed, and each mental-state term was coded for type and referent. Mothers’ references to the child’s mental states also were coded as appropriate or inappropriate. Because accurate coding often required consideration of contextual cues (see below), coders viewed transcripts and videotapes together when assessing maternal mental-state talk.

Type of mental state.

Coding for type of mental state was based on previous work (Bartsch & Wellman, 1995; Jenkins et al., 2003; Ruffman et al., 2002; Shatz et al., 1983) and included mothers’ use of desire, emotion, and cognition terms. Desire terms refer to wishes or desires (e.g., want, hope, wish, prefer). Emotion terms refer to emotional states (e.g., happy, sad, unhappy, angry, scared, afraid, disappointed). Cognitive terms refer to a mental activity or process, as well as beliefs, memories, or knowledge (e.g., think, know, remember, believe, understand, expect). Because we were interested in mothers’ references to internal mental states, words that denoted an observable behavior (e.g., cry, laugh, frown), physical state (e.g., sick, tired, hungry), or evaluation (e.g., neat, cool, nice, special) were not included as emotion terms in our coding scheme (also see Dunn et al., 1987; Ruffman et al., 2002). One exception was mothers’ use of the word fun, which was coded as an emotion term given its strong connotation of internal experience (e.g., “This is fun!”). Like and love were coded as desire terms when denoting a want (“I’d like to play with the kitchen”) and as emotion terms when conveying a positive feeling about someone or something (“I like these toys”).

Following Bartsch and Wellman (1995), coders differentiated between uses of mental-state terms that were genuine versus those that were conversational. Examples of the latter included cognitive terms used to indicate a yes or no response (e.g., “I think so”), engage in turn taking (e.g., “Let’s go to the park. What do you think?”), elicit attention (“You know what?”), or act as a pause filler (“You know, this toy is fun!”). The phrase “I don’t know” was also considered a conversational usage if it was not clear what was unknown. Conversational or social convention forms of desire terms were also coded (“I don’t care” without referring to object of care and the objectless statement of desire) but were rarely observed. Only mothers’ genuine uses of mental-state terms were examined in this study.

Referent of mental-state term.

The referent of each mental-state term was coded as self (i.e., mother refers to her own mental state), child (i.e., mother refers to the child’s mental state), or other (i.e., mother refers to the mental state of a toy figure or individual not present). If the mother used a nonspecific pronoun (“That is interesting.” “This is frustrating, isn’t it?”), the coder used contextual clues to decide whether the mother was talking about her own experience (“I find this interesting”) or the child’s experience (“You’re frustrated, aren’t you?”).

Appropriate versus inappropriate references.

For each mental-state term referring to the child, coders also assessed whether the term appropriately reflected the child’s desire, emotion, or cognition. On the basis of prior work by Meins et al. (2002) and Symons et al. (2006), a mental-state term was coded as appropriate if the term (a) reflected the child’s desire, emotion, or cognition as evidenced by his or her speech or behavior (child puts down book and heads to new toy, and mother says, “Do you want to play with another toy now?”; child is actively exploring the kitchen, and mother says, “Do you like the kitchen?”) or (b) connected the child’s current activity to a relevant past or future event, object, or activity (e.g., child points to a picture of a dog, and mother says, “Does that remind you of our doggy at home?”). A mental-state term was coded as inappropriate if (a) the term did not match the child’s speech or behavior (e.g., child pushes the book away, and mother says, “Do you want to look at the book first?”), (b) the mother referred to a past or future event that was not relevant to the child’s current activity (e.g., child points to a picture of a dog, and mother says, “Do you want to stop for ice cream on the way home?”), or (c) the focus of the mental-state term was nonspecific or vague.

Interobserver agreement and scoring.

Coders were trained on a subsample of tapes until they reached 80% agreement. Interobserver reliability checks were conducted throughout the coding process, and 30% (n = 339) of the protocols were double coded. Interobserver agreement on the total number of maternal utterances, assessed with intraclass correlation, was .94. Kappas were utilized to assess interobserver agreement for the mental-state talk codes and were .87 (type of mental-state term), .87 (referent), and .84 (appropriate/inappropriate usage). Composites of maternal mental-state talk were created by summing each type of mental-state talk (desire, emotion, cognition) within referent (self, child, other). For terms referring to the child, composites were computed separately for appropriate and inappropriate references. Additionally, because self and other references were relatively infrequent, we summed across these references within each type of term (desire, emotion, cognition). These self/other composites enabled more robust comparisons with mothers’ references to the child’s mental states.

Friendship Quality

Maternal reports.

At 54 months, kindergarten, and first grade, mothers completed the 19-item Quality of Child’s Friendship questionnaire (adapted from the Quality of Classroom Friends questionnaire; see Clark & Ladd, 2000), which assesses the child’s relationship with a best friend. Items were identical across measurement times and were rated on a 4-point scale, ranging from 1 (strongly disagree) to 4 (strongly agree). Composite scores of children’s positive interaction (11 items; e.g., play happily together) and negative interaction (8 items; e.g., compete for toys) were created by averaging across items. Maternal ratings showed good internal consistency for both positive interaction (α = .80, .79, .and 81 at 54 months, kindergarten, and first grade, respectively) and negative interaction (α = .83, .80, and .82, respectively) subscales. Prior research using this questionnaire measure has shown that greater parent–child connectedness was associated with more harmonious child–friend relationships as rated by kindergarten teachers (Clark & Ladd, 2000).

Observations.

At 54 months, children were observed interacting with a close friend during a semistructured play session. Friend participants, who were within 18 months of age of the study child, were identified by the study child’s parent or child-care provider. Child–friend observations took place in the child-care setting or in the home of the study child or friend. To maintain standardization across settings, child–friend dyads were observed in a portable playroom constructed of cardboard (3 feet high × 5 feet in diameter and open at the top). After a brief introduction to the playroom, the experimenter presented the children with three toys in the following fixed order: (a) a Mickey Mouse pop-up game, (b) a Viewmaster with one slide, and (c) a Fisher-Price doctor kit and doll. Each toy session lasted approximately 5 minutes, and child–friend interaction was videotaped through a curtained opening in the playroom.

Trained observers used a coding scheme developed for the NICHD SECCYD to code the videotaped play sessions for child and friend behaviors. In this study, we examined the study child’s prosocial behavior (e.g., turn taking, turn offering, sharing), contribution to positive interaction (e.g., study child initiates new topics of joint play, elaborates on an ongoing theme, attends to the friend’s activity, or responds enthusiastically to the friend), aggressive behavior (e.g., physical or verbal acts that serve to disrupt the friend’s play, demean the friend, or forcefully gain access to a toy in the friend’s possession), and contribution to negative interaction (e.g., study child controls the play session through whining, making demands, insisting on having his or her own way; shows annoyance at the friend; competes in a way that puts down the friend). Behaviors were rated on a 5-point scale, ranging from 1 (low) to 5 (very high). Because very few children received ratings of 4 or 5 on the aggressive behavior or contribution to negative interaction codes, these ratings were transformed to 3-point scales (i.e., scores of 4 or 5 were rescored to a value of 3). Ratings were made at the end of each toy session and averaged across the three sessions. Interobserver reliability was assessed through intraclass correlations (Winer, 1971) for 20% of the cases, and estimates were .72 (prosocial behavior), .83 (positive contribution), .80 (aggressive behavior, 3-point scale), and .81 (negative contribution, 3-point scale). The quality of child–friend interaction assessed with this coding scheme has been associated with child characteristics (e.g., anxious solitude, school readiness) and parenting behavior (e.g., sensitivity) in expected ways (Gazelle & Spangler, 2007; NICHD ECCRN, 2006).

Covariates

Maternal sensitivity, total maternal utterances, and child mental-state talk were included as covariates. Maternal sensitivity was assessed by means of observations of mother–child interaction during the play session at 24 months. A composite of maternal sensitivity (α= .74) was created by summing sensitivity to non-distress, positive regard for the child, and intrusiveness (reversed), which were rated on global 4-point scales, ranging from 1 (not at all characteristic) to 4 (highly characteristic). Of the tapes, 20% were randomly assigned to two coders, and interobserver reliability (Winer, 1971) for the sensitivity composite was .84. A score of mothers’ total utterances (see coding criteria described earlier) was created by summing across maternal utterances during the 24-month mother–child play session. Notably, maternal mental-state talk, maternal sensitivity, and total maternal utterances were all assessed from the same mother–child play session; independent coding teams assessed mental-state talk/utterances and sensitivity.

Child mental-state talk was assessed with the MacArthur Communicative Development Inventories (Fenson, Dale, Reznick, Thal, & Reilly, 1991). During a home visit at 24 months, mothers completed the toddler version of this measure, which was designed for children between 16 and 30 months of age and captures vocabulary production, sentence complexity, and use of irregular nouns and verbs. In assessing vocabulary, mothers indicate their child’s use of words on a vocabulary checklist containing 680 words, organized into 22 semantic categories (e.g., animals, vehicles, body parts). From this list, 13 items tapping child mental-state talk were identified (e.g., happy, hate, like, love, mad, think, want), and responses were summed across items (α = .89). Because preliminary coding of child mental-state talk from the mother–child play session at 24 months indicated very low frequency of child mental-state talk during this brief play session, we chose to utilize items from the MacArthur Communicative Development Inventories to obtain a more representative assessment of this construct. This measure is well established with strong test–retest reliability and concurrent and predictive validity (Fenson et al., 1994).

Data Analytic Strategy

Latent growth curve models were tested within a structural equation modeling framework to assess maternal mental-state talk as an intervening mechanism linking infant–mother attachment and mother-reported friendship quality across 54 months, kindergarten, and first grade. Compared with other statistical methods used to assess change over time, latent growth curve modeling permits tests of indirect effects in a single analysis model (see Burchinal, Nelson, & Poe, 2006). Structural equation models also were tested to examine the process model predicting the 54-month observed friendship outcomes. Mplus 5.0 (Muthén & Muthén, 1998–2007) was utilized to test the process models. Because the chi-square test of model fit is sensitive to sample size and is typically significant when samples are large (>200), we examined two additional fit indices that have been shown to be robust to fluctuations in sample size (Fan, Thompson, & Wang, 1999). Specifically, the comparative fit index (CFI) and root-mean-square error of approximation (RMSEA) were examined to assess model fit. A model is considered to provide a good fit to the data when CFI values are .95 and above (Hu & Bentler, 1995) and when RMSEA values are less than .05 (Browne & Cudeck, 1993). Full-information maximum likelihood was used to handle the missing data. Compared with other methods, such as listwise deletion, full-information maximum likelihood offers less biased estimates (Schafer & Graham, 2002).

In examining indirect effects, recent simulation studies have indicated that traditional z tests, even when conducted with large samples, may be biased because of nonnormal distribution of the indirect effect when the null hypothesis is false (MacKinnon, Lockwood, & Williams, 2004). Resampling techniques provide more unbiased estimates and result in greater power to detect indirect effects (Shrout & Bolger, 2002). Comparisons across six resampling techniques indicated that the bias-corrected bootstrap method, which corrects for bias in the central tendency of the estimate, is the method of choice when assessing indirect effects (MacKinnon et al., 2004). Thus, by means of the bootstrap procedure in Mplus, we utilized bias-corrected confidence intervals (CI_bc) to assess indirect effects, and we specified 5,000 replications for this bootstrap procedure. An indirect effect was considered significant if the confidence interval did not include 0. We estimated intervals at 95% ( p < .05) and 99% ( p < .01) confidence levels.

Results

Preliminary Analyses

Descriptive data for all study measures are shown in Table 1. Because frequency of maternal inappropriate usage when referring to the child’s emotions and cognitions was very low (i.e., ≤1% of all mental-state terms), mothers’ talk about child-inappropriate emotions or cognitions was not considered further. Because of positive skewness, log transformations were conducted for all other measures of mental-state talk and were examined in subsequent analyses.

Table 1.

Descriptive Statistics for the Study Measures

Study measure (measurement time)	M	SD	Range
Maternal mental-state talk (24 months)
Desire, child appropriate	6.07	4.31	0–28
Desire, child inappropriate	2.51	2.98	0–38
Desire, self/other	1.02	1.66	0–17
Emotion, child appropriate	1.34	1.71	0–14
Emotion, child inappropriate	0.08	0.42	0–7
Emotion, self/other	0.94	1.54	0–14
Cognition, child appropriate	2.28	3.08	0–24
Cognition, child inappropriate	0.19	0.62	0–7
Cognition, self/other	3.07	3.31	0–23
Friendship interaction
Mother-reported positive (54 months)	3.02	0.28	1.82–4.00
Mother-reported negative (54 months)	2.38	0.46	1.00–3.75
Mother-reported positive (kindergarten)	3.17	0.34	1.64–4.00
Mother-reported negative (kindergarten)	2.25	0.53	1.00–3.88
Mother-reported positive (first grade)	3.21	0.36	2.09–4.00
Mother-reported negative (first grade)	2.16	0.53	1.00–3.75
Observed prosocial (54 months)	2.83	0.53	1.00–5.00
Observed positive contribution (54 months)	2.87	0.68	1.00–5.00
Observed aggressive (54 months)	1.10	0.25	1.00–2.67
Observed negative contribution (54 months)	1.44	0.51	1.00–3.00
Covariates
Maternal sensitivity (24 months)	9.36	1.76	3–12
Maternal total utterances (24 months)	170.06	57.86	3–398
Child mental-state talk (24 months)	5.01	3.79	0–13

Note. Of the 1,249 cases included in this study, data were available on infant–mother attachment at 15 months (n = 1,149), on maternal sensitivity at 24 months (n = 1,172), on maternal mental-state talk at 24 months (n = 1,114), on child mental-state talk at 24 months (n = 1,073), on mother-reported friendship quality at one or more measurement times (ns = 1,077 and 1,070 for the positive and negative interaction subscales, respectively), and on observed friendship quality at 54 months (n = 751).

For descriptive purposes, means and standard deviations for the friendship interaction measures by the 15-month attachment categories are shown in Table 2. Univariate analyses of variance with planned comparisons (A vs. B, C vs. B, D vs. B) indicated that the secure group differed from the avoidant group on mother-reported positive interaction at 54 months, F(1, 909) = 6.59, p < .01, and at first grade, F(1, 768) = 4.00, p < .05, on mother-reported negative interaction at kindergarten, F(1, 724) = 6.01, p < .05, and on observed child contribution to positive interaction at 54 months, F(1, 704) = 2.98, p < .10 (see Table 2 for means).

Table 2.

Descriptive Statistics for the Friendship Interaction Measures by 15-Month Attachment Categories

Friendship interaction measures	Avoidant (A)		Secure (B)		Resistant (C)		Disorganized (D)		Planned comparisons
	M	SD	M	SD	M	SD	M	SD
Maternal reports
Positive interaction (54 months)	2.96	0.26	3.03	0.28	2.99	0.25	3.02	0.28	B > A**
Negative interaction (54 months)	2.44	0.45	2.39	0.47	2.45	0.39	2.34	0.46
Positive interaction (kindergarten)	3.12	0.35	3.18	0.33	3.15	0.30	3.16	0.35
Negative interaction (kindergarten)	2.38	0.53	2.23	0.53	2.28	0.50	2.25	0.53	A > B*
Positive interaction (first grade)	3.15	0.37	3.23	0.36	3.20	0.30	3.19	0.33	B > A*
Negative interaction (first grade)	2.21	0.57	2.14	0.52	2.20	0.50	2.19	0.54
Observed interaction (54 months)
Prosocial behavior	2.78	0.54	2.86	0.53	2.77	0.51	2.80	0.52
Contribution to positive interaction	2.77	0.75	2.91	0.67	2.91	0.71	2.80	0.65	B > A†
Aggressive behavior	1.11	0.26	1.10	0.26	1.05	0.13	1.11	0.26
Contribution to negative interaction	1.48	0.56	1.43	0.49	1.39	0.47	1.50	0.58

Note. Because of missing data, Ns ranged from 708 to 913 for these analyses.

^†p < .10.

^*p < .05.

^**p < .01.

Univariate analyses of variance were also conducted to assess whether the covariates differed by the infant–mother attachment groups, and planned comparisons (A vs. B, C vs. B, D vs. B) indicated that (a) maternal sensitivity was lower for the avoidant group (M = 8.80) and the disorganized group (M = 9.24) versus the secure group (M = 9.54), Fs(1, 1084) = 22.77 and 3.95, ps < .001 and .05, respectively; (b) maternal total utterances were lower for the avoidant group (M = 161.68) versus the secure group (M = 172.40), F(1, 1064) = 4.14, p < .05; and (c) child mental-state talk was lower for the avoidant group (M = 4.38) versus the secure group (M = 5.14), F(1, 989) = 4.52, p < .05.

As noted earlier, maternal sensitivity, child mental-state talk, and mothers’ total utterances were included as covariates because they are proximal to maternal mental-state talk and may potentially act as third variables. We also examined whether maternal education and family income-to-needs ratio were associated with the main study variables, controlling for our more proximal covariates. Associations were largely nonsignificant, and neither demographic measure showed a significant association with 15-month attachment. Thus, we did not include these variables as additional covariates in our main analyses.

Maternal Mental-State Talk as a Function of Infant–Mother Attachment Groups

Because assessing the predictor-mechanism association is a first and necessary step in testing indirect effects (see Kenny et al., 1998), we first examined the extent to which maternal mental-state talk at 24 months differed by the infant–mother attachment categories and whether attachment-related differences were moderated by referent and/or child gender. To this end, a repeated measures analysis of covariance (ANCOVA; controlling for maternal sensitivity, child mental-state talk, and maternal total utterances) was conducted for each type of mental-state talk (desire, emotion, cognition), with referent as the repeated variable and child gender and attachment groups as the between-subjects variables. Attachment-related differences were of primary interest; therefore, we focus on the main effects and interactions involving the attachment groups. F tests for the attachment group main effect for each type of mental-state talk (across referent) are shown in Table 3. Additionally, adjusted means for the attachment groups and standardized betas for the covariates are shown for each type of mental-state talk (by referent) in Table 3.

Table 3.

Maternal Mental-State Talk at 24 Months as a Function of Infant–Mother Attachment Categories at 15 Months

Predictor variable	Desire talk by referent			Emotion talk by referent		Cognitive talk by referent
	Child-APP	Child-INAPP	Self/other	Child-APP	Self/other	Child-APP	Self/other
	β
Covariate
Child gender (male = 0)	.06	−.04	.00	−.04	.03	.00	−.00
Total utterances (24 months)	.39***	.30***	.31***	.28***	.19***	.34***	.33***
Sensitivity (24 months)	.06*	−.16***	−.04	.03	.13**	.21***	.20***
Child MST (24 months)	.01	−.07*	.08*	−.02	.10**	.04	.08**
	F(df)
Attachment main effect	F(3, 956) = 0.44, ns			F(3, 956) = 1.41, ns		F(3, 956) = 3.84** B > A*, C**
	Madj (SE)
A (n = 131)	6.64 (0.35)	2.73 (0.26)	0.99 (0.14)	1.32 (0.15)	0.87 (0.13)	1.85 (0.25)	2.83 (0.27)
B (n = 595)	6.18 (0.16)	2.45 (0.12)	1.00 (0.06)	1.37 (0.07)	1.00 (0.06)	2.56 (0.12)	3.22 (0.13)
C (n = 86)	6.43 (0.43)	2.67 (0.32)	1.09 (0.17)	1.27 (0.18)	0.81 (0.16)	2.41 (0.31)	3.55 (0.33)
D (n = 149)	5.50 (0.33)	2.63 (0.24)	0.93 (0.13)	1.31 (0.14)	0.91 (0.12)	1.86 (0.24)	2.95 (0.25)

Note. Attachment group means are adjusted for covariates in the model. Because of missing data, n = 964 for this set of analyses. Child-APP = child-appropriate references; child-INAPP = child-inappropriate references; MST = child’s mental-state talk as reported by mothers; A = avoidant, B = secure; C = resistant; D = disorganized. Child-inappropriate scores for maternal talk about emotions and cognitions were dropped because of very low frequency. Log transformations of the maternal mental-state talk variables were examined in the analyses. Yet, for ease of interpretation, adjusted means are based on raw scores.

^*p < .05.

^**p < .01.

^***p < .001.

A 3 (referent: child appropriate, child inappropriate, self/ other) × 2 (child gender) × 4 (attachment: ABCD) repeated measures ANCOVA was conducted for mothers’ desire talk. The attachment main effect and interactions were nonsignificant. Next, 2 (referent: child appropriate, self/other) × 2 (gender) × 4 (attachment) repeated measures ANCOVAs were conducted for maternal talk about emotions and cognitions, respectively.¹ The attachment main effect and interactions were nonsignificant for emotion talk. The main effect of attachment, however, was significant for cognitive talk, F(3, 956) = 3.84, p < .01. Planned comparisons (A vs. B, C vs. B, D vs. B) revealed that children with a secure attachment experienced more maternal talk about cognitions than did children with an avoidant, F(1, 956) = 5.82, p < .05, or disorganized, F(1, 956) = 7.00, p < .01, attachment (see Table 3 for adjusted means). This main effect emerged above and beyond effects of mothers’ total utterances, F(1, 956) = 178.98, p < .001, maternal sensitivity, F(1, 956) = 68.00, p < .001, and child mental-state talk, F(1, 956) = 6.15, p < .05 (see Table 3 for parameter estimates). All interactions were nonsignificant.

Effect sizes (Cohen’s d) were also computed for comparisons between the secure group and each of the insecure groups by calculating the difference in the adjusted means (i.e., adjusted for covariates in the ANCOVA) and dividing by the root-mean-square error from the ANCOVA (see Cortina & Nouri, 2000; NICHD ECCRN, 2006). Cohen (1992) indicated d values of .20, .50, and .80 as representing small, medium, and large effects, respectively. Because no interactions of the repeated variable (referent) were found earlier, we computed effects sizes for each type of maternal mental-state talk averaged across referent. Cohen’s d was .07 (desires), .17 (emotions), and .23 (cognitions) for the avoidant–secure comparison; .08 (desires), .15 (emotions), and .03 (cognitions) for the resistant–secure comparison; and .04 (desires), .07 (emotions), and .24 (cognitions) for the disorganized–secure comparison. Thus, for the significant differences in maternal cognitive talk that emerged, effects were small (d = .23 and .24).

Maternal Cognitive Talk as an Intervening Mechanism

Because infant–mother attachment predicted maternal cognitive talk only, we proceeded to test maternal cognitive talk as an intervening mechanism linking infant–mother attachment status and children’s later friendship quality. A separate process model was tested for each friendship outcome: (a) mother-reported positive interaction, (b) mother-reported negative interaction, (c) observed positive interaction, and (d) observed negative interaction. For each model, maternal talk about child-appropriate and self/ other cognitive states were indicators of a latent variable of cognitive talk, and attachment groups were entered as three dummy variables (with the secure group coded as the comparison group). As depicted in Figure 1, we controlled for maternal sensitivity, maternal total utterances, and child mental-state talk in each model. For ease of presentation, we do not present the covariates, error terms, or within-time covariances in the figures.² Significant paths between the covariates and the friendship outcomes, however, are reported in the text. Note that associations between the covariates and maternal cognitive talk were similar in magnitude to those reported in Table 3 and, thus, are not presented in the text. For descriptive purposes, intercorrelations among all continuous measures examined in the process models are shown in Table 4.

Table 4.

Correlations Among Study Measures in the Latent Growth Curve/Structural Equation Models

Study measures (measurement time)	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15
1. Total utterances (24 months)	—	.40***	.37***	.11***	.25***	.09**	−.03	.10**	−.09*	.09*	−.10**	.04	.01	−.06	−.06
2. COG appropriate (24 months)		—	.47***	.12***	.30***	.11***	−.04	.14***	−.06	.11**	−.11**	.10**	.06	−.08*	−.11**
3. COG self/other (24 months)			—	.15***	.30***	.12***	−.04	.14***	−.06	.14***	−.14***	.10**	.08*	−.02	−.06
4. Child MST (24 months)				—	.13***	.13***	−.05	.06	−.03	.10**	−.04	.08*	.06	.01	−.04
5. Sensitivity (24 months)					—	.13***	−.10**	.15***	−.14***	.15***	−.14***	.14***	.09*	−.13***	−.17***
6. FQ POS (54 months)						—	−.50***	.43***	−.28***	.35***	−.24***	.07	.06	−.10**	−.07
7. FQ NEG (54 months)							—	−.28***	.43***	−.23***	.37***	−.07	−.02*	.12**	−.11**
8. FQ POS (kindergarten)								—	−.44***	.42***	−.29***	.13**	.05	−.04	−.07
9. FQ NEG (kindergarten)									—	−.29***	.51***	−.11**	−.08	.06	−.13**
10. FQ POS (first grade)										—	−.45***	.05	.06	−.09*	−.04
11. FQ NEG (first grade)											—	.01	.02	.07	.05
12. OBS prosocial (54 months)												—	.55***	−.25***	−.36***
13. OBS positive (54 months)													—	−.15***	−.06
14. OBS aggressive (54 months)														—	−.52***
15. OBS negative (54 months)															—

Note. COG = maternal cognitive talk; Child MST = child’s mental-state talk as reported by mothers; FQ POS = mother-reported positive friendship interaction; FQ NEG = mother-reported negative friendship interaction; OBS = observed child–friend interaction.

^*p < .05.

^**p < .01.

^***p < .001.

In assessing the process model, we distinguished between mediated and indirect effects. A mediated effect requires a significant association between the predictor and outcome and may be considered a special case of indirect effects (Holmbeck, 1997). An indirect effect, in contrast, does not require a significant predictor–outcome path but implies the predictor has an effect on the outcome through the intervening variable (Kenny et al., 1998). With this distinction in mind, we conducted preliminary tests of direct effects of the 15-month attachment categories on each set of friendship outcomes, controlling for 24-month maternal sensitivity, maternal verbosity, and child mental-state talk. For each model described, direct effects of attachment categories on friendship interaction were nonsignificant. Moreover, inclusion of the direct effect pathways in the full model examining maternal cognitive talk as an intervening mechanism did not significantly improve model fit in any instance: ${\chi }_{\text{diff}}^2\left(6\right)=5.04$ and 5.47, ns, for mother-reported positive and negative interaction, respectively; ${\chi }_{\text{diff}}^2\left(3\right)=1.97$ and 1.57, ns, for observed positive and negative interaction, respectively. In addition, significance levels of indirect effects remained identical to those reported later. In the analyses that follow, therefore, we refer to indirect rather than mediated effects. As discussed by Shrout and Bolger (2002), consideration of indirect (vs. mediated) effects may be especially important when examining processes that are distal in time.

Prior to testing each model, we also explored child gender as a moderator. Multigroup analyses in which paths composing the indirect effect of interest were constrained to be equal across gender indicated that the unconstrained model did not provide an improved fit for models predicting mother-reported positive or negative interaction, respectively, ${\chi }_{\text{diff}}^2\left(5\right)=6.44$ and 6.15, ns, or observed positive or negative interaction, respectively, ${\chi }_{\text{diff}}^2\left(4\right)=4.23$ and 3.33, ns. Thus, the models presented next were tested for all cases combined.

Mother-reported friendship interaction.

For the latent growth curve models predicting mother-reported positive and negative friendship interaction across time (54 months, kindergarten, first grade), factor loadings for the intercept were constrained to 1 at each measurement time point, and the factor loadings for the linear slope were constrained to 0, 1, and 2, respectively. A significant path predicting the intercept indicated an association with friendship interaction at 54 months, whereas a significant path predicting the slope indicated an association with rate of change in friendship interaction from 54 months to first grade. An unconditional latent growth curve model (i.e., predictors are not included) was first tested for each outcome to assess whether there was significant interindividual variability in the intercept and slope to merit testing the full model. Tests of the unconditional model indicated that, on average, positive interaction increased (B = .10, SE = .01, p < .001) and negative interaction decreased (B = —.11, SE = .01, p < .001) over the three measurement time points. Moreover, for each model, the association between the intercept and slope was nonsignificant (B = —.01 and —.01, SE = .004 and .01, for positive and negative interaction, respectively), suggesting that initial level of friendship interaction was not related to rate of change in interaction over time. Of importance, significant interindividual variation was found for initial level of friendship interaction (B = .05 and .12, SE = .01 and .02, p < .001, for positive and negative interaction, respectively) and rate of change (B = .01 and .03, SE = .004 and .01, p < .01 and .001, for positive and negative interaction, respectively). Thus, because the unconditional models indicated significant variability across individuals in initial levels of friendship interaction and rate of change, we proceeded with tests of the full process model. Results for the model predicting mother-reported positive and negative friendship interaction outcomes are presented in turn.

For the model predicting positive interaction, the chi-square test was significant (which would be expected given the large sample size), χ²(31, N = 1249) = 63.06, p < .001, and fit indices showed good fit (RMSEA = .029, CFI = .965). As shown in Figure 2, children with avoidant or disorganized (vs. secure) attachment experienced less maternal cognitive talk at 24 months. Maternal cognitive talk, in turn, was related to more positive friendship interaction at 54 months (intercept), and this association emerged above and beyond a significant contribution made by child mental-state talk (β= .13, p < .01). The following indirect effects were significant: (a) avoidant versus secure → cognitive talk → positive friendship intercept (95% CI_bc [—.028, —.002]) and (b) disorganized versus secure → cognitive talk → positive friendship intercept (99% CI_bc [—.033, —.001]). In summary, an avoidant or disorganized (vs. secure) infant–mother attachment was related to lower levels of mother-reported positive friendship interaction at 54 months via lower levels of maternal cognitive talk. Predictors of rate of change in positive interaction (i.e., slope) were all nonsignificant.

Figure 2.

Maternal cognitive talk as an intervening mechanism between infant–mother attachment and mother-reported positive friendship interaction at 54 months, kindergarten, and first grade. Standardized path estimates and R² estimates for endogenous latent variables are shown. Paths represented by dotted lines were nonsignificant. Maternal sensitivity, total maternal utterances, and child mental-state talk were included as covariates. Yet, for ease of presentation, covariates and error terms are not shown. COG child-APP = appropriate maternal references to the child’s cognitive states; COG self/other = maternal references to her own and others’ cognitive states. ** p < .01. *** p < .001.

For the model predicting negative interaction, the chi-square test was significant, χ²(31, N = 1249) = 48.46, p < .05, and fit indices showed good fit (RMSEA = .021, CFI = .982). As shown in Figure 3, children with an avoidant or disorganized (vs. secure) attachment experienced less maternal cognitive talk at 24 months. Maternal cognitive talk, in turn, was related to more rapid decreases in negative friendship interaction from 54 months to first grade (slope). The following indirect effects were significant: (a) avoidant versus secure → cognitive talk → negative friendship slope (95% CI_bc [.002, .030]) and (b) disorganized versus secure → cognitive talk → negative friendship slope (95% CI_bc [.002, .031]). In summary, an avoidant or disorganized (vs. secure) infant–mother attachment was related to less rapid decreases in mother-reported negative friendship interaction from 54 months to first grade via lower levels of maternal cognitive talk at 24 months. Additionally, maternal sensitivity (but not cognitive talk) predicted lower levels of mother-reported negative friendship interaction at 54 months (i.e., intercept; β= —.13, p = .01).

Figure 3.

Maternal cognitive talk as an intervening mechanism between infant–mother attachment and mother-reported negative friendship interaction at 54 months, kindergarten, and first grade. Standardized path estimates and R² estimates for endogenous latent variables are shown. Paths represented by dotted lines were nonsignificant. Maternal sensitivity, total maternal utterances, and child mental-state talk were included as covariates. Yet, for ease of presentation, covariates and error terms are not shown. COG child-APP = appropriate maternal references to the child’s cognitive states; COG self/other = maternal references to her own and others’ cognitive states. * p < .05. ** p < .01. *** p < .001.

Observed friendship interaction.

For the structural equation model predicting observed positive interaction at 54 months, study children’s prosocial behavior and contribution to positive interaction were indicators of a latent variable of positive interaction. The chi-square test for this model approached significance, χ²(25, N = 1249) = 36.78, p < .10, and model fit was good (RMSEA = .019, CFI = .986). As shown in Figure 4, an avoidant or disorganized (vs. secure) attachment at 15 months was associated with less maternal cognitive talk at 24 months. Maternal cognitive talk, in turn, was associated with more positive friendship interaction at 54 months, and this association emerged above and beyond a significant contribution made by maternal sensitivity (β= .11, p < .05). The following indirect effects were significant: (a) avoidant versus secure → cognitive talk → positive interaction (95% CI_bc [—.052, —.001]) and (b) disorganized versus secure → cognitive talk → positive interaction (95% CI_bc [—.052, —.002]). In summary, an avoidant or disorganized (vs. secure) infant–mother attachment was related to lower levels of observed positive friendship interaction at 54 months via lower levels of maternal cognitive talk.

Figure 4.

Maternal cognitive talk as an intervening mechanism between infant–mother attachment and observed positive friendship interaction at 54 months. Standardized path estimates and R² estimates for endogenous latent variables are shown. Paths represented by dotted lines were nonsignificant. Maternal sensitivity, total maternal utterances, and child mental-state talk were included as covariates. Yet, for ease of presentation, covariates and error terms are not shown. COG child-APP = appropriate maternal references to the child’s cognitive states; COG self/other = maternal references to her own and others’ cognitive states. * p < .05. ** p < .01. *** p < .001.

For the model predicting observed negative interaction, study children’s aggression and contribution to negative interaction were indicators of a latent variable of negative interaction. The chi-square test for this model was significant, χ²(25, N = 1249) = 38.66, p < .05, and the model showed good fit (RMSEA = .021, CFI = .983). An avoidant and disorganized (vs. secure) attachment at 15 months was associated with less maternal cognitive talk at 24 months, yet maternal cognitive talk was not significantly related to negative interaction (β= —.09, ns); thus, tests of indirect effects were nonsignificant. Maternal sensitivity, however, was associated with less negative friendship interaction observed at 54 months (β= —.18, p < .01).

Discussion

Theory and research suggest that the infant–mother attachment relationship sets the stage for more positive friendships in childhood. Yet, the mechanisms that account for attachment–friend linkages have been less well understood, and studies of interpersonal processes are especially lacking. Guided by attachment theorists’ emphasis on communicative processes, as well as growing evidence for the centrality of mental-state attunement and talk to the child–mother attachment relationship and children’s friendships, our primary objective was to test maternal mental-state talk as an intervening mechanism linking infant–mother attachment and children’s later friendship quality.

As a first and necessary step toward our main objective, we assessed the degree to which infant–mother attachment was associated with mental-state talk by mothers with their 24-month-old children. Few studies have investigated such associations prospectively, and the current study provided a fine-grained analysis in that mothers’ desire, emotion, and cognitive talk were assessed separately, and the moderating effects of referent of maternal mental-state talk (child vs. self/other) and child gender were considered. Results indicated that maternal talk about cognitive states, combined across child-appropriate (e.g., “How did you know that opened?” “What do you think this book is going to be about?”) and self/other (e.g., “I think it goes like this”; “I don’t know how to get it [the toy] out”) references, was greater among children with a secure versus insecure–avoidant or disorganized attachment history. Notably, these associations were not accounted for by maternal sensitivity, maternal verbosity, or the child’s mental-state talk and were not moderated by referent or child gender. That attachment-related differences emerged for mothers’ cognitive talk, in particular, is consistent with recent emphases on maternal mind-mindedness (Meins, 1997) and reflective function (Fonagy & Target, 1997) as central to early child–mother attachment security. Cognitive states, versus desire or emotion states, are less likely to have corresponding behavioral or facial cues and are viewed as strictly representational (Bartsch & Wellman, 1995). Consequently, talk about cognitions may be especially indicative of the caregivers’ propensity to consider and reflect on mental states. It is also noteworthy that, on average, mothers’ cognitive talk with children increases from 24 to 36 months, although children’s cognitive talk is not prevalent until the later preschool years (Brown & Dunn, 1991; Taumoepeau & Ruffman, 2008). The current findings, therefore, suggest that mothers’ cognitive talk may burgeon relatively early in the context of a secure attachment.

Infant–mother attachment, however, was not related to mothers’ talk about desires or emotions at 24 months. In the child-centered play session examined here, mothers frequently referred to their child’s desires, and such talk was typically appropriate. Attachment-related differences in desire talk may emerge in contexts in which the child’s and mother’s desires are more likely to conflict and/or the child’s desires are more likely to be person oriented versus object oriented (e.g., desiring the mother’s attention vs. desiring a certain toy). We recall, however, that our examination of maternal talk about desires was exploratory, and no prior theoretical or empirical work has pointed toward associations between infant–mother attachment and desire talk, specifically. In contrast, the nonsignificant result for emotion talk was surprising given the central role that emotional communication is posited to serve in the development and maintenance of secure attachment relationships. Again, we consider the play context in interpreting this null finding. That is, mothers may be most likely to talk about emotions when their child is exhibiting moderate upset (Dunn & Brown, 1994), and it is at such times that maternal talk about emotions may correspond most closely with attachment security. Moreover, prior studies reporting positive associations between child–mother attachment security and emotion talk have used tasks specifically designed to elicit talk about emotions (e.g., Laible & Thompson, 2000; Oppenheim et al., 2007). It is likely that the mother–child play situation examined here, in which mothers showed relatively little talk about emotions—probably because children did not display much negative affect and/or because the toys were not chosen to elicit emotion talk—was not optimal for eliciting meaningful individual differences in emotion talk. Alternatively, by controlling for maternal sensitivity, maternal verbosity, and child mental-state talk, the current study provided a rigorous test of attachment–emotion talk associations, and such associations may be accounted for by one or more of these third variables.³

Turning to our primary objective, we note that because infant–mother attachment predicted mothers’ cognitive talk only, we focused on this type of mental-state talk as an intervening mechanism linking infant–mother attachment and children’s later friendship quality. The latent growth curve/structural equation models showed that the latent variable of maternal cognitive talk at 24 months, for which mothers’ child-appropriate and self/other cognitive terms were indicators, was related to more mother-reported and observed positive friendship interaction at 54 months and more rapid decreases in mother-reported negative friendship interaction from 54 months to first grade. Moreover, a secure versus insecure–avoidant or disorganized attachment at 15 months was associated with these friendship outcomes via maternal cognitive talk. The process model provided a stringent test of these associations, as the indirect paths controlled for maternal sensitivity, maternal total utterances, and child mental-state talk. Further, multiple measures of friendship were available at 54 months, and the consistent pattern of results found across mother-reported and observed friendship interaction at this time point increases confidence in the results.

We interpret our finding for maternal cognitive talk at 24 months as an intervening mechanism of attachment–friend linkages in light of prior research highlighting cognitive talk during this developmental period as central to the child’s developing understanding of mental states and use of mental-state language. Namely, whereas maternal talk about desires (vs. cognitions or emotions) at 15 months predicted children’s mental-state language and emotion understanding at 24 months (Taumoepeau & Ruffman, 2006), maternal talk about cognitions (vs. desires or emotions) at 24 months was a more consistent predictor of children’s mental-state language and emotion understanding at 33 months (Taumoepeau & Ruffman, 2008; but see Symons et al., 2006). Likewise, Jenkins et al. (2003) reported that 2-year-old children who experienced cognitive talk by family members showed greater cognitive talk at 4 years of age, controlling for the child’s general language ability and use of cognitive terms at age 2. It is interesting that such associations did not emerge between 4 and 6 years of age for the children’s older siblings (Jenkins et al., 2003). Taken together, these findings suggest that maternal cognitive talk may scaffold the child’s own budding mind-mindedness and that this maternal input may be especially critical during the third year of life—a time when children are rapidly acquiring internal state language and demonstrating nascent understanding of others’ minds (e.g., Shatz et al., 1983). The child’s mental-state understanding (Dunn & Cutting, 1999; McElwain & Volling, 2004) and cognitive talk with friends (Brown et al., 1996; Hughes & Dunn, 1997), in turn, have been shown to foster more harmonious friendship interactions during the later preschool years. Of course, this interpretation is tentative as the current study did not include measures of children’s mental-state understanding, nor did we assess children’s mental-state language during interactions with friends. These are limitations of the current study, and future studies that include the child’s attunement to mental states, particularly in the context of close peer relationships, are needed to further illuminate the ways in which maternal talk about cognitions fosters harmonious child–friend interactions.

Of importance, we differentiated among the insecure attachment groups, and as hypothesized, the proposed processes emerged for secure versus avoidant and secure versus disorganized comparisons. Consistent with past studies (e.g., Etzion-Carasso & Oppenheim, 2000; Fonagy et al., 1991; Main et al., 1985), our findings suggest that children’s repeated experiences with a caregiver who is unavailable and rejecting (avoidant) or frightened/frightening (disorganized) may result in restricted patterns of psychological discourse and, in turn, more problematic interactions with friends. In contrast, maternal cognitive talk did not differ for children with a secure versus insecure–resistant attachment history (also see Etzion-Carasso & Oppenheim, 2000; Fonagy et al., 1991 for null findings). It may be that dyads in which the child has a resistant attachment history— emerging from experiences with an inconsistently responsive caregiver and resulting in heightened vigilance toward the caregiver—were able to function adequately in the brief, low-stress play session examined here. Future research using extended, naturalistic observations of mother–toddler dyads across a range of situations is needed to examine whether secure versus resistant group differences in maternal mental-state talk emerge in more high-stress contexts.

Although not a main aim of the study, it is noteworthy that maternal sensitivity showed a modest, positive association with maternal cognitive talk, indicating that the two are related but distinct constructs (also see Meins et al., 2001). We also note that, in certain instances, maternal sensitivity uniquely contributed to children’s friendship quality. Specifically, maternal sensitivity (but not cognitive talk) during mother–child play at 24 months predicted less negative child–friend interaction as reported by mothers at 54 months and as rated by independent observers at 54 months. In contrast, maternal cognitive talk (but not sensitivity) predicted rate of change in mother-reported negative friendship interaction from 54 months to first grade. During early childhood, peer disputes are predominantly about objects (see Shantz, 1987), and children may minimize such negative interactions with friends through behavioral means (e.g., sharing, turn taking) that they learn during interactions with a behaviorally sensitive mother. During the early school years, conflicts among friends come to involve more abstract issues (e.g., play suggestions, expectations for behavior, social comparison). Resolving such disputes likely requires sophisticated psychological understanding; thus, mothers’ cognitive talk, rather than behavioral sensitivity, may promote children’s ability to effectively manage negative interactions with friends during this developmental shift (Fabes et al., 2009).

The longitudinal design, measurement of key constructs at developmentally salient time points, and inclusion of potential third variables are strengths of the current study and increase confidence in tests of the hypothesized indirect effects (see Frazier et al., 2004; Shrout & Bolger, 2002). Nonetheless, we caution that tests of alternative models are needed to further strengthen conclusions about maternal mental-state talk as an intervening mechanism. In this vein, we note that (a) equivalent models (i.e., models in which paths among the exact same measures are reversed or changed to test alternative directions of effect) can largely be rejected given the temporal ordering of assessments in this study and (b) our ability to test other alternative models was restricted because maternal mental-state talk was assessed at only one time point. Because parental attunement to mental states has been shown to precede infant–parent attachment security in prior research (Fonagy et al., 1991; Meins et al., 2001), it is especially important to rule out the possibility that maternal mental-state attunement underlies a secure infant–mother attachment and later high-quality friendships. That is, it is plausible that the child–mother attachment relationship plays no causal role in promoting maternal mental-state talk or children’s friendship quality but is simply a by-product of the mother’s mental-state attunement (see Fonagy & Target, 1997; Harris, 1999). Yet, neither attachment nor mental-state talk are static entities, and we suspect a more dynamic process in which mentally minded caregivers promote attachment security, and attachment security, in turn, promotes caregiver–child talk about mental states, especially during the third year when mental-state talk is on the rise. The latter causal pathway may emerge because children with a history of secure attachment create opportunities for caregivers’ mental-state talk through their cooperative stance toward (e.g., Carpendale & Lewis, 2006) and trust in (Corriveau et al., 2009) the caregiver. Longitudinal studies assessing attachment and mental-state talk together over multiple time points are needed to illuminate these potential directions of effect. Furthermore, studies utilizing moment-to-moment sequential analyses would provide a window onto how mother–child talk about mental states unfolds in real time, thus enabling assessment of child-versus mother-directed talk about mental states that would complement the examination of direction of effects in longitudinal studies.

Additional limitations should be noted. Our measure of child mental-state talk was based on a maternal checklist that included a limited number of mental-state terms; a more comprehensive assessment of child mental-state language is needed in future research of this kind. Further, families examined in this study, although diverse with respect to geographic location and socioeconomic level, were characterized by higher average income, maternal age, and education compared with participants excluded because of missing data. Moreover, the NICHD SECCYD sample was not intended to be nationally representative, and families with certain characteristics (e.g., mother was under 18; child was hospitalized for more than 7 days following birth; see NICHD ECCRN, 1999, for further details) were not enrolled. Caution is needed, therefore, in generalizing results to families who may experience high levels of economic hardship and/or stress. Likewise, we acknowledge that the findings that emerged for our predominantly European American sample may not generalize to other cultural groups. Although studies indicate strong support for Ainsworth et al.’s (1978) patterns of child–caregiver attachment across a wide range of Western and non-Western cultures, cross-cultural evidence for the caregiving antecedents (e.g., sensitivity) and child outcomes (e.g., positive peer relationships) of child–mother attachment security is less well established (see Van IJzendoorn & Sagi-Schwartz, 2008). Moreover, the degree to which individuals construe others’ behavior in terms of mental states varies widely across cultures, with individuals from some collectivist societies engaging in little mental-state talk (see Vinden & Astington, 2000). Future investigations are needed, therefore, that assess the cultural specificity of maternal mental-state talk as an intervening mechanism.

Last, effect sizes for attachment-group differences in maternal mental-state talk and amounts of variance explained for the friendship outcomes were small. Because associations were examined over time and because the constructs examined here are multidetermined, modest effects would be expected. In this regard, the majority of children included in this study were from two-parent families; yet, only the mother–child relationship was considered. Inclusion of fathers in this area of study would likely increase the amount of variance explained and is warranted given the unique role fathers may play in fostering children’s peer relationships (Booth-LaForce et al., 2006), attachment security (Grossmann et al., 2002), and understanding of mental states (LaBounty, Wellman, Olson, Lagattuta, & Liu 2008).

Despite these limitations, this study provides one of the largest and most comprehensive analyses, to date, of the associations between infant–mother attachment and maternal talk about mental states. To the best of our knowledge, it is also the first to examine maternal mental-state talk as an intervening mechanism of attachment–friend linkages. Such an investigation is an important first step to better understanding the interpersonal processes through which a secure infant–mother attachment relationship may promote the child’s subsequent abilities to engage in more positive and less negative interactions with friends. Our results underscore mothers’ talk about cognitive states with their 2-year-old children as one such mechanism.