Ponesse, Schachar, Logan, & Tannock, 1999). Consequently, it is .... of the box and to help the child complete the puzzle verbally but not to actually do the ...
Copyright 2001 by the American Psychological Association, Inc. 0012-1649/0I/$5.00 DO1: 10.1037//0012-1649.37.4.475
Developmental Psychology 2001, Vol. 37, No. 4, 475-490
Mothers' Emotional Expressivity and Children's Behavior Problems and Social Competence: Mediation Through Children's Regulation Nancy Eisenberg, Elizabeth Thompson Gershoff, Richard A. Fabes, Stephanie A. Shepard, Amanda J. Cumberland, Sandra H. Losoya, Ivanna K. Guthrie, and Bridget C. Murphy Arizona State University
The relations between mothers' expressed positive and negative emotion and 55-79-month-olds' (76% European American) regulation, social competence, and adjustment were examined. Structural equation modeling was used to test the plausibility of the hypothesis that the effects of maternal expression of emotion on children's adjustment and social competence are mediated through children's dispositional regulation. Mothers' expressed emotions were assessed during interactions with their children and with maternal reports of emotions expressed in the family. Children's regulation, externalizing and internalizing problems, and social competence were rated by parents and teachers, and children's persistence was surreptitiously observed. There were unique effects of positive and negative maternal expressed emotion on children's regulation, and the relations of maternal expressed emotion to children's externalizing problem behaviors and social competence were mediated through children's regulation. Alternative models of causation were tested; a child-directed model in which maternal expressivity mediated the effects of child regulation on child outcomes did not fit the data as well.
to mediate the association between children's temperament and their problem behavior (Rothbart & Bates, 1998), it is possible that child characteristics such as temperamentally based regulation mediate the effects of some aspects of parenting on the quality of children's social functioning. For example, Eisenberg, Spinrad, and Cumberland (1998) hypothesized that the effects of parenting style (e.g., general warmth) and specific emotion-related parenting practices on children's adjustment and social competence are mediated in part by children's regulatory capacities. The purpose of the present study was to examine the relations between parental emotional expressivity (toward the child and more generally at home) and children's adjustment and social competence and to test whether these relations might be indirect through children's regulatory capacities.
The goal of many contemporary developmental, clinical, and personality researchers has been the identification of factors that promote or inhibit children's adjustment and social competence. Although numerous investigators have implicated parenting (Rothbaum & Weisz, 1994) or children's temperament (e.g., regulation or emotionality; Eisenberg, Fabes, Guthrie, & Reiser, 2000; Rothbart & Bates, 1998) in the development of children's adjustment and social competence, few have examined both factors simultaneously or have assessed the nature of their joint contribution to the prediction of children's social functioning (see Bates, Pettit, Dodge, & Ridge, 1998, for an exception). Although parenting behavior has been hypothesized
Nancy Eisenberg, Elizabeth Thompson Gershoff, Stephanie A. Shepard, Amanda J. Cumberland, Sandra H. Losoya, Ivanna K. Guthrie, and Bridget C. Murphy, Department of Psychology, Arizona State University; Richard A. Fabes, Department of Family Resources and Human Development, Arizona State University. Elizabeth Thompson Gershoff is now at the National Center for Children in Poverty, Columbia University. Stephanie A. Shepard is now at the Counseling Department, University of Oregon. Bridget C. Murphy is now at the Department of Psychology, University of Oklahoma. This research was supported by Grants 1 RO1 HH55052 and 1 R01 MH60838 and Research Scientist Award KO5 M801321 from the National Institutes of Mental Health. We thank the many students who assisted in this study; the parents and children involved; the principals and teachers in the Tempe, Kyrene, Mesa, and Scottsdale School Districts; the undergraduates who assisted in data collection; and Amanda Sheffield Morris for her comments on the manuscript. Correspondence concerning this article should be addressed to Nancy Eisenberg, Department of Psychology, Arizona State University, Tempe, Arizona 85287-1104. Electronic mail may be sent to nancy.eisenberg@ asu.edu.
Relations Between Parents' Expressivity and Children's Behavior Problems and Social Competence Expressivity has been defined as "a persistent pattern or style in exhibiting nonverbal and verbal expressions that often but not always appear to be emotion related; this pattern or style is usually measured in terms of frequency of occurrence" (Halberstadt, Cassidy, Stifter, Parke, & Fox, 1995, p. 93). Parental expressivity can be measured in at least two ways (Halberstadt, Crisp, & Eaton, 1999): (a) as parents' expression of positive and negative emotion when interacting with the child and (b) as parents' general tendency to express emotion in interactions in the family. The latter is a somewhat purer measure of family (or parental) expressiveness or "the predominant style of exhibiting nonverbal and verbal expressions within a family" (Halberstadt et al., 1995, p. 93). This type of expressivity often occurs in situations that do not directly involve the child and in which the emotion is not directed at the child. Parental expressivity defined in this manner is frequently 475
476
EISENBERG ET AL.
assessed with parent-report measures (e.g., Bronstein, Fitzgerald, Briones, Pieniadz, & D'Ari, 1993; Halberstadt, 1986). There is mounting evidence that parents' expression of emotion is related to their children's socioemotional competence (Eisenberg, Cumberland, & Spinrad, 1998). Parents high in warmth or positive emotion and low in negativity in interactions with their children have children who display increased social competence and social understanding and decreased levels of hostility, externalizing problems, and internalizing problems (Dunn & Brown, 1991; Lindahl, 1998; Matthews, Woodall, Kenyon, & Jacob, 1996; Rothbaum & Weisz, 1994; Rubin, Hastings, Chen, Stewart, & McNichol, 1998; Scaramella, Conger, & Simons, 1999). In addition, measures of parental and family expressiveness in the home, especially positive expressivity, have been associated with children's social competence, emotional understanding, prosocial behavior, self-esteem, security of attachment, and positive emotionality and with the quality of parent-child relationships (Boyum & Parke, 1995; Bronstein et al., 1993; Cassidy, Parke, Butkovsky, & Braungart, 1992; Halberstadt et al., 1999). In contrast, negative emotional expressivity in the home tends to be inversely related to the same developmental outcomes, although the findings for negative expressivity have frequently been inconsistent (Halberstadt et al., 1999). It is likely that it is a predominance of hostile and hurtful negative emotions rather than simply any negative emotion that is associated with negative developmental outcomes. Some exposure to negative emotion, especially when it is expressed appropriately and not on an ongoing basis, is likely to be important for learning about emotions and how to regulate them (Halberstadt et al., 1999). For example, the expression of nonhostile emotion (e.g., sadness, embarrassment, distress) has been shown to be positively related to sympathy, whereas parental expression of hostile emotion has been shown to be negatively related to sympathy (Crockenberg, 1985; Eisenberg et al., 1991; Eisenberg et al., 1992). Even exposure to conflict involving parents often has no negative effects on children when the conflicts are resolved (Davies & Cummings, 1994). Nonetheless, frequent and continuing negative emotion in the home may have different effects than the expression or discussion of negative emotion in specific appropriate incidents. Dunn and Brown (1994) found that situations in which children were upset were likely to be opportunities for discourse (and thus learning about emotion) but that such situations were less likely to be opportunities for learning in families in which anger and distress were expressed frequently. The same may be true for instances in which parents experience and express negative emotion.
Processes Involved in the Relation of Parental Expressivity to Children's Social Functioning and Regulation There are a variety of conceptual reasons to expect both types of parental expressivity to be linked to children's adjustment and social competence. For example, the emotions parents express indicate to children what goals their parents value and how they are disposed to act (Dix, 1991, 1992). When parents are warm and supportive with their children, children are likely to believe that their parents are concerned with their interests, and consequently, such children are motivated by feelings of trust and reciprocity. Thus, children are likely to comply with and internalize supportive
parents' standards for appropriate behavior, including demands for self-regulation (Grusec & Goodnow, 1994; Maccoby & Martin, 1983; Parpal & Maccoby, 1985). In contrast, children whose parents display high levels of negative emotion toward them may come to feel that their parents are not concerned with their interests, and as a consequence, such children may not be motivated to comply with parents' directives or to internalize parents' standards of behavior. In addition, parents who generally are positive and supportive with their children may help them to manage their distress and to cope successfully in stressful situations (Skinner & Wellborn, 1994); such coping, in turn, might foster the development of social skills and reduce negative expectations about social interactions (Dusek & Danko, 1994; Hardy, Power, & Jaedicke, 1993). Parental expressivity may also contribute to children's abilities to interpret and understand others' emotional reactions and to their beliefs about how much and what types of emotional expressions are appropriate and effective in social interactions, and such knowledge may foster both self-regulation and social skills (Denham, Zoller, & Couchoud, 1994; Dunn & Brown, 1991; Eisenberg, Cumberland, & Spinrad, 1998). As implied in the discussion of the effects of parental expressivity on children's compliance, internalization, and understanding of emotion, there is reason to believe that children whose parents express more positive emotion around them and do not express frequent hostile or hurtful emotions (e.g., disappointment in the child) will be particularly likely to be well regulated. Children's regulatory capacities include the abilities to voluntarily focus attention, shift attention, and inhibit or initiate behavior—abilities that can be used to modulate emotional reactivity to events as well as emotionally driven behavior (Derryberry & Rothbart, 1997; Eisenberg et al., 2000; Rothbart, Ahadi, & Hershey, 1994). Although a child's regulatory abilities are thought to derive partly from heredity (Rothbart et al., 1994; Rothbart & Derryberry, 1981), experience also plays a role in shaping the expression of temperament, including regulation (Rothbart & Bates, 1998). It has been argued that children's abilities to regulate their attention, emotion, and behavior are embedded in the context of social relationships (Campos, Campos, & Barrett, 1989; Thompson, 1994; Walden & Smith, 1997) and that children's self-regulation can be facilitated by, and perhaps even partly learned from, their parents (Gottman, Katz, & Hooven, 1997; Kopp, 1982). For example, it is clear that children model adults' self-denial (Grusec, 1971; see Eisenberg & Fabes, 1998) and can be taught attentional strategies for regulating themselves (Mischel, 1981). If parents tend to be supportive with their children in stressful situations, their children may be less likely to become overaroused in those situations and better able, and more motivated, to process parents' messages and other relevant information. As a consequence, such children are likely to be relatively skilled at managing their emotion and behavior and at identifying parents' goals and expectations (Dix, 1991, 1992; Hoffman, 1983). Consequently, they are more likely than overaroused children to learn appropriate strategies for handling their emotions and emotionally driven behavior. Moreover, heightened expression of negative affect by parents may reflect their own dysregulation and can serve as a model for children's imitation or contribute in other ways to children's dysregulation (Cole, Michel, & Teti, 1994). In addition, as suggested by Cummings and Davies (1996), emotional security
477
MEDIATION THROUGH REGULATION is a goal that motivates children's regulatory processes; thus, problems of emotion regulation associated with emotional insecurity can accumulate as a function of exposure to frequent hostile and hurtful negative emotion in the family. There is very little research on the relation between parental expressivity and children's regulation. In a study of toddlers, mothers' reported positive expressivity in the family was related to higher levels of toddlers' self-soothing behavior, whereas motherreported sadness within the family was negatively related to toddlers' self-soothing (Garner, 1995). Garner and Power (1996) found that preschoolers' positive displays of emotion when they received a disappointing prize (a measure of regulation) were negatively related to negative expressivity in the family but were unrelated to observed maternal positive or negative expressiveness with the child. Children's displays of negative emotion when they received the prize were unrelated to family expressiveness and marginally linked to mothers' observed positive expressiveness with the child. In a study of college students and adults, those from negatively expressive families reported less control over feeling angry, even when intensity of anger was controlled for (Burrowes & Halberstadt, 1987). Although they did not assess family expressivity per se, Gottman and colleagues (1997) found that parents who were supportive in regard to encouraging the appropriate expression of emotion and coaching children about their emotions had children who were relatively high in regulatory abilities. However, Gottman and his colleagues did not find a relation between children's regulation and either parental derogatory behavior or parental scaffolding and praising (at least when other variables were controlled in a structural model). In a longitudinal study, Eisenberg and her colleagues (1999) found that parents' reports of punitive reactions to children's negative emotions were negatively related to children's dispositional regulation, with regulation and maternal punitive reactions apparently affecting one another over time. Although these punitive reactions did not explicitly involve the expression of negative emotion, they likely communicated negative emotion to the child. In addition, mothers' distress, avoidance, or discomfort in response to children's negative emotions at ages 6 to 8 predicted children's low levels of regulation at ages 8 to 10 (Eisenberg et al., 1999). Thus, although there is conceptual and empirical work supporting associations between parents' positive and negative expressivity and children's regulation, few researchers have examined the relation of parental expression of emotion per se, and the limited relevant findings have not been very consistent (also see Halberstadt et al., 1999).
Regulation as a Mediator of the Relation Between Parental Expressivity and Children's Problem Behavior and Social Competence Given that parental emotional expressivity has been related both to children's regulatory abilities and to their behavioral and social competencies, it is reasonable to hypothesize that the effects of parental expressivity on children's adjustment and social competence are at least partly mediated by children's regulation. Children whose parents display more positive affect with and toward them and who themselves are more regulated would be expected to successfully manage negative emotions and be relatively socially competent and adjusted.
Rothbart and Bates (1998) suggested that specific temperament or personality dimensions relate in a differentiated way to internalizing and externalizing problems, with inhibition relating more to internalizing problems and unmanageability linked to later externalizing problems. Early inhibition is characterized, in part, by a high degree of behavioral inhibition in novel situations and low impulsivity, whereas unmanageability involves low ability to voluntarily inhibit behavior or modulate emotion. Similarly, Eisenberg et al. (in press) hypothesized that externalizing children are low in effortful and less voluntary modes of regulation, whereas internalizing children are low to average in modes of regulation that are voluntary and effortful (especially attentional regulation) but high in involuntary inhibition of behavior. Consistent with this view was Huey and Weisz's (1997) finding that ego undercontrol (behavioral underregulation) was related to externalizing problems in clinic-referred children, whereas ego overcontrol predicted internalizing problems. Many children with externalizing problems are characterized partly by their impulsive nature and a lack of attentional control (Newman & Wallace, 1993; Pulkkinen, 1986; Robins, John, Caspi, Moffitt, & StouthamerLoeber, 1996). Moreover, there is mounting evidence that individual differences in children's regulatory abilities predict differences in externalizing behaviors, social competence, and conscience (Caspi, Henry, McGee, Moffitt, & Silva, 1995; Eisenberg et al., 1996; Gottman et al., 1997; Kochanska, Murray, & Coy, 1997; Rothbart et al., 1994). Findings regarding the relation of internalizing problems to regulation are less consistent. Lengua, West, and Sandier (1998) found a negative relation between children's attentional focusing and depression in correlational analyses but generally not in structural equation modeling (when other variables were controlled). In their sample, regulation related more consistently with children's conduct problems than with depression. Zahn-Waxier, Schmitz, Fulker, Robinson, and Emde (1996) found that internalization of control (e.g., good impulse control, capacity to attend and focus) predicted low externalizing problems but not level of anxiety (an aspect of internalizing). On the basis of these inconsistent findings, we expected internalizing problems to be either unrelated or weakly negatively related to effortful regulation, and we were unsure whether regulation would mediate relations between parental expressivity and children's internalizing problems.
The Present Study The correlational research just reviewed is at least partly consistent with Eisenberg, Cumberland, and Spinrad's (1998) hypothesis that parents' emotional expressivity affects children's abilities to self-regulate, which in turn affect children's social competence and adjustment. To our knowledge, this mediational process has not been explicitly tested. Our goal in the present study was to test this hypothesis for both positive and negative parental expressivity and a range of child outcomes. Positive parental expressivity was expected to be associated with high regulation, and regulation was expected to mediate the relations between parental positive expressivity and both high social competence and low adjustment (especially externalizing behavior). Negative parental expressivity, especially if hostile or perhaps indicative of the child irritating, disappointing, or upsetting the parent, was expected to show the reverse pattern of relations. In addition, we tested the possibility
478
EISENBERG ET AL.
that children's regulation could cause problems in social competence and adjustment that would affect maternal expressivity. Also, because children who are socially competent with peers may have more opportunities than their less competent peers to learn how to regulate themselves, we tested a model in which maternal expressivity predicted children's adjustment and social competence, which in turn predicted children's regulation. We also examined whether sex or age moderated the mediated relations just described. The relation of family or parental expressiveness to child outcomes often varies with the sex of the child (e.g., Eisenberg et al., 1992). For example, Boyum and Parke (1995) found more relations between maternal expressed emotion and kindergarten children's sociometric status and aggression for girls than for boys, whereas Bronstein et al. (1993) found that measures of parents' expressiveness toward their children predicted quality of social behavior more for boys than for girls. Thus, we were uncertain if there would be sex differences, although mothers' expressivity more often has predicted girls' sympathy (Eisenberg et al., 1991, 1992) and therefore may be more related to social competence for girls. We also did not have any clear reason to expect the relations of interest to vary in strength from the age of 4.5 years to the age of 8 years. However, children's regulation develops rapidly during the preschool (Kopp, 1982; Reed, Pien, & Rothbart, 1984) and elementary school years (Murphy, Eisenberg, Fabes, Shepard, & Guthrie, 1999; Williams, Ponesse, Schachar, Logan, & Tannock, 1999). Consequently, it is possible that regulation becomes a more important predictor of adjustment with age because problems stemming from lack of regulation are viewed more negatively in older than in younger children.
Method
Participants Families with children between the ages of 4.5 and just 8 years were recruited for participation with letters sent to parents through the schools, newspaper ads, and flyers placed at local preschools and after-school programs in a large southwestern urban area. The early school years are a time of rapid development of regulation and increased expectations for socially appropriate social behavior (due to increased exposure to peers and adults other than parents); thus, these years may be a critical time for children to consolidate their emerging social skills or problems with adjustment. Because the goal of the study was to determine the factors associated with risk for behavior problems, selective screening was conducted through phone-call administration of the Child Behavior Checklist (CBCL; Achenbach, 1991a, 1991b). Out of a pool of 315 children, all children with T scores of 60 or above on either externalizing or internalizing behavior problems (or both) were asked to participate; scores of 60-63 are viewed as indicating moderate risk for clinical problems (Achenbach, 1991b, p. 63). To equate children with and without problem behaviors on demographic variables as much as possible, we matched problem children (considering them as either externalizers or internalizers on the basis of which score was higher) with nonproblem children of the same sex and race (when possible), with similar social class (based on parents' reports of education and occupation), and of the same age. The mean and median T scores were approximately equal both for internalizing (M = 59.78, SD = 9.87. Mdn = 60.00, range = 33-90) and externalizing (M = 59.55, SD = 10.80. Mdn = 59.00, range = 35-82). Of these 214 families, 12 were excluded from the present analyses because only fathers came into the laboratory (see Eisenberg et al., in press).
We used the criterion of a T score of 60 or above as a cutoff for problem behaviors; of the 202 remaining families, 72 were nondisordered children, 34 had internalizing problems only, 30 had externalizing problems only, and 66 had comorbid problems. Of the children with scores at 60 or above for either internalizing, externalizing, or both, 16 had externalizing scores in the moderate risk range (i.e., 60-83) and 30 had internalizing scores in this range. Eighty children had T scores for externalizing behavior above 63, whereas 70 had scores for internalizing behavior above 63. The majority of the 202 mothers in the final sample reported that their children were of European American heritage (76%). The percentages of children with Hispanic, Native American, African American, Asian American, and mixed or other racial or ethnic backgrounds were 12%, 5%, 3%, 1%, and 3%, respectively. Forty-two percent of the mothers attended or completed high school but did not obtain further education, 46% attended or graduated from college, and 12% received postgraduate training. Most mothers reported working outside the home (72%), and the mean annual family income was $40,000 (range = $6,000 to $125,000; SD = $24,000). Children in the sample were an average of 6 years old (mean age = 73 months, SD = 9 months, range = 55-97 months; 46%, 73-71 months; 39%, 72-83 months; 15%, 84-97 months), and the sample included more boys (n = 113; mean age = 72 months, SD = 9 months) than girls (n = 89; mean age = 74 months, SD = 10 months).
Procedure As part of a larger study, each mother and her child came to a university laboratory and were met by an experimenter of the same sex as the child. While the mother completed a series of questionnaires, the child was videotaped unobtrusively engaging in a 5-min puzzle task. The child was shown a wooden box that contained a puzzle with geometric-shaped pieces. The box was constructed with a clear Plexiglas back (so that children's hand movements could be observed) and a cloth covering the front that had sleeves through which the children placed their arms. An experimenter instructed the child to assemble the puzzle without looking at it. However, although the cloth at the front of the box blocked the child's view of the puzzle, it was not attached at the bottom and could be lifted easily by the child so that he or she could cheat by looking at the puzzle. Children were told that if they finished the puzzle within 5 min, they would receive an attractive prize. A timer was set for 5 min and placed in front of the child so that the child would know how much time he or she had. The experimenter left the room during the puzzle task and did not return until the child called him or her back by ringing a bell or until the timer went off. The mother then joined her child for a second videotaped puzzle task. An experimenter demonstrated the puzzle box to the mother and child and placed a new puzzle of alphabet letters inside. The child and mother were instructed to complete the puzzle without the child looking at it and were told that the child would receive an attractive prize if he or she completed the puzzle within 5 min. The mother was instructed to sit on the clear side of the box and to help the child complete the puzzle verbally but not to actually do the puzzle for the child. The timer was again set, and the experimenter left the room until the bell was rung or the timer sounded. At the end of the session, the child was given a prize (regardless of his or her performance), and the mother and child were paid for their participation. The children's teachers also were asked to complete questionnaires about the children; 91 % of the teachers (ra = 183) agreed to participate and were paid for their participation. Teachers were sent questionnaires concerning the children near the end of the semester during which the children came to the laboratory.
Measures Mothers' Emotion and Behavior With Their Children Mothers' expressions of positive and negative emotion with their children were assessed with both observational measures and self-reports of
MEDIATION THROUGH REGULATION emotional expressivity. Because of computer problems, all mother-report data were lost for 5 mothers. Some data were lost for 4 additional mothers because of missing responses. Observed positive and negative affect. During the mother-child puzzle task, each mother's positive emotion was coded every 30 s on a scale from 1 (low positive affect) to 5 (high positive affect) on the basis of her facial expressions (e.g., smiling) and laughing. Both intensity and duration of positive affect were considered in making these ratings. We averaged the 10 positive affect codes across the 5-min interaction to compute a mean positive affect score for each mother (a = .83). All coders were blind to the hypotheses. In addition to her positive affect, the mother's expressed warmth and positive affect toward her child during the task was coded once by observers on a scale from 1 (low) to 5 (high). Examples of warmth included smiling at the child and displaying physical affection. This index was viewed as primarily tapping expressions of positive emotion toward the child; it overlapped with the coding of positive emotion but included physical displays of positive emotion (affection) toward the child. Interrater reliabilities on these continuous scales for 25% of the interactions (n = 52) were .83 for mothers' positive affect and .65 for mothers' global warmth. Because these two measures were strongly positively related, rs(199) = .66, p < .0001, and because the two constructs tapped the mother's expressions of positive feeling toward the child in the same interactional context, we averaged them to form a positive expressivity composite score. Each mother's negative affect, as displayed through her facial expressions and tone of voice, was also coded every 30 s during the mother-child puzzle task on a 5-point scale (1 = no negative affect; 5 = negative affect high in intensity, duration, or both). Cues such as frowning, biting the lips, irritated looks, and negative voice tone were used as indicators of negative affect, which could include sadness, anger, worry, and anxiety; 54% of the mothers exhibited some negative emotion. Because many of the negative looks were subtle and relatively infrequent, anger and irritation were not coded separately from other negative emotions. Moreover, mothers often were not looking directly at the camera, which made it difficult to code negative emotion more precisely. Even if mothers were anxious because they were in an experiment or trying to perform well, given the context of children trying to succeed at a task, the children might have taken their mothers' negative emotion—sadness or anxiety as well as irritation or anger—personally. We averaged the 10 negative affect codes across the 5-min interaction to compute an observed negative affect score, a = .82. The interrater reliability, based on 52 families, was .78. Mothers' self-reported expressivity. Each mother reported her own emotional expressiveness with her family by completing the SelfExpressiveness in the Family Questionnaire (Halberstadt et al., 1995). Mothers rated items on a 9-point scale (1 = / rarely express these feelings; 9 = 1 frequently express these feelings) to indicate the frequency with which they expressed a range of emotions in a variety of settings typical to most families. The Positive Expressivity subscale was used as an index of mothers' self-reported positivity (e.g., "Telling family members how happy you are" and "Expressing gratitude for a favor"; 14 items; a = .85). The Negative Dominant subscale was used as an index of mothers' selfreported negative expressivity (e.g., "Expressing anger at someone else's carelessness" and "Showing dislike for someone"; 10 items; a = .80). Halberstadt et al. (1995) reported high test-retest reliability, internal consistency, construct validity, and discriminant validity for this measure with several samples.
479
persisted on the puzzle task rather than being off task or cheating (i.e., lifting up the cloth to look at the puzzle) when alone were coded by two observers. (Different people coded children's persistence and maternal emotion.) The time a child spent persisting was divided by the total time he or she spent with the puzzle task (usually 5 min). Interrater reliability for stopwatch timings of the number of seconds children persisted on the task was .97. Mothers' and teachers' reports. Children's regulation was assessed with adults' reports on a variety of aspects of children's regulation and with a behavioral measure. Mothers and teachers rated children's regulation using subscales from the Child Behavior Questionnaire (Goldsmith & Rothbart, 1991; Rothbart & Ahadi, 1992). The teacher version was adapted slightly to reflect the school rather than the home context. Three subscales were used to compute a composite of children's regulation (alphas for mothers and teachers are in parentheses): (a) Attention Focusing (e.g., "When drawing or coloring in a book, shows strong concentration"; 10 items; as = .74 and .85), (b) Attention Shifting (e.g., "Can easily shift from one activity to another"; 11 items; as = .83 and .88), and (c) Inhibitory Control (e.g., "Can lower his/her voice when asked to do so"; 13 items; as = .85 and .88). The intercorrelations among these subscales were significant both for mothers and teachers. For Attention Focusing with Attention Shifting, Attention Focusing with Inhibition Control, and Attention Shifting with Inhibition Control, respectively, rs(196) = .38, .61, and .71, ps < .001, for mothers; for teachers, rs(183) = .61, .74, and .78, ps < .001. The three subscales were standardized and averaged, and the reliabilities were .91 and .95 for mothers and teachers, respectively.
Children's Externalizing Behavior Children's externalizing behavior was assessed with teachers' and mothers' reports. Mothers and teachers completed Achenbach's (1991a, 1991b) CBCL and TRF (Teacher Report Form), respectively. Children's feelings and behaviors over the past 6 months were rated on a 3-point scale ranging from not true to very true. The Aggressive Behavior (as = .91 and .96 for mothers and teachers) and Delinquent Behavior subscales (as = .65 and .70 for mothers and teachers) were highly correlated: rs (202 and 182) = .68 and .72, ps < .0001, for mothers and teachers, respectively. The total externalizing behavior score was used in the analyses (externalizing on the CBCL and the TRF is based on these two subscales; as = .91, .95, and .91 for mothers, teachers, and fathers, respectively). Consistent with the recommendations of Achenbach (1991b), the continuous scores were used in correlational and structural equation analyses rather than the T scores.
Children's Internalizing Behavior The Withdrawn, Anxious/Depressed, and Somatic Complaints subscales of the CBCL and the TRF (which include all of the internalizing items) were internally consistent for mothers and teachers: as = .73 and .81 for mothers and teachers, respectively, for Withdrawn; as = .81 and .84 for Anxious/Depressed, and as = .59 and .77 for Somatic Complaints. Correlations among the subscales for mothers and teachers were as follows: Withdrawn with Anxious/Depressed, rs(202 and 182) = .66 and .63 for mothers and teachers, respectively, ps < .001; Withdrawn with Somatic Complaints, rs(202 and 182) = .33 and .13, ps < .001 and .08; Anxious/ Depressed with Somatic Complaints, r(202 and 182) = .43 and .31, ps < .001. The total internalizing scale, which includes items from these three subscales, was used in the major analyses (as = .86, .89, and .90 for mothers, teachers, and fathers, respectively).
Children's Regulation Children's regulation was assessed with both a behavioral index and adults' reports. Observed behavioral measure. Persistence during the puzzle task was used as a behavioral index of regulation. The amounts of time children
Children's Social Competence Children's social competence was reported by mothers and teachers using a scale developed by Eisenberg et al. (1997; Eisenberg et al., 2000). Two subscales from this measure were used as indices of social compe-
480
EISENBERG ET AL.
tence (on a scale from 1 to 4): Socially Appropriate Behavior (e.g., "This child is usually well-behaved"; four items; as = .79 and .89 for mothers and teachers, respectively) and Popularity (e.g., "This child has a lot of friends"; three items; as = .84 and .93 for mothers and teachers, respectively). In addition, teachers rated each child's popularity with classmates using three items developed by K. A. Dodge (personal communication, July 1993; Lemerise & Dodge, 1988). Using a 5-point scale ranging from in the top 15% of the class to in the bottom 15% of the class, teachers rated the child on the following questions: "Overall, how much is this child liked by classmates?" "How many peer nominations would this child receive for 'Liked the most'?" and "How many peer nominations would this child receive for 'Liked the least'?" (reversed). We averaged responses to these three items to obtain a measure of popularity. Mothers' ratings of children's socially appropriate behavior and popularity were positively correlated, /-(199) = .46, p < .001, and were standardized and averaged to form a composite measure of social competence (a = .83). Teachers' ratings of children's socially appropriate behavior, popularity, and sociometrics also were strongly intercorrelated, rs(180) = .57 to .79, all ps < .001. After being standardized, the two teacher ratings of popularity and sociometrics were averaged (r = .79), and then this score was averaged with teachers' ratings of socially appropriate behavior.
Results We used structural equation modeling (SEM) to test our hypothesis that children's regulatory abilities would mediate the association between mothers' affective expressivity and children's social competence and adjustment. Our use of the EQS software (Bentler, 1993) necessitated the deletion of any families for whom there were missing data for any of the main variables. A total of 96 boys and 80 girls had complete data for the variables of interest and were included in the SEM analyses.1 Table 1 presents the means and standard deviations for the measured variables for the total sample and for the subsample used in the models. All but the three regulation variables were not normally distributed (skew/5£ > 2 or < —2) and thus were subjected to either logarithmic or square root transformations (Tabachnick & Fidell, 1989). Table 2 contains the correlations among the transformed measured variables used as indicators of the hypothesized latent variables; the clear majority of the correlations were in the expected directions.2
Descriptive Analyses The relations of age and sex with the major study variables were examined for the total sample. There were only two significant correlations with age: Older children were higher in mother-rated internalizing, r(202) = .16, p < .03, and in mother-rated regulation, r(197) = .15, p < .04. In contrast, there were numerous sex differences. According to t tests, girls were rated by mothers and teachers as more regulated, rs(195 and 181) = 3.07 and 3.33, ps < .002 and .001, respectively, and by teachers as more socially competent, f(179) — 3.15, p < .002. Boys were rated as higher in externalizing behavior by mothers, r(200) = — 2.14, p < .04, and teachers, f( 180) = -2.42, p < .02. In addition, girls persisted during the puzzle task more than did boys, ?(199) = 2.98,/? < .01. To determine if families' socioeconomic status (SES) was associated with their scores on any of the measured variables, we created a composite of standardized measures of mother education, father education, and family income and correlated it with the main variables. Five significant relations were found: SES was associated with higher ratings of regulation by teachers, r( 181) = .22,
Table 1 Means and Standard Deviations for All Measured Study Variables All 202 families Variable Observed maternal affect Positive affect" Negative affect Mother self-reported expressivity Positive Negative Regulation Persistence Mother report0 Teacher reporta Externalizing behaviors Mother report Teacher report Internalizing behaviors Mother report Teacher report Social competence Mother report ab Teacher reporta
M
176 families in the models
SD
M
SD
.02 1.22
.92 .32
.03 1.21
.93 .32
7.31 3.99
.99 1.23
7.33 3.91
.99 1.21
.55 -.01 .01
.30 .86 .89
.55 .01 .02
.30 .82 .89
16.83 9.49
9.88 11.86
16.39 9.34
9.83 11.77
11.65 5.96
7.53 6.48
11.60 5.84
7.32 6.30
.01 .01
.86 .90
.05 .01
.78 .90
a
These variables were composites of two or more standardized measures. b Means for mother report of social competence are significantly different at p < .05; all other means are not significantly different.
p < .01, more observed maternal positive expressivity, r(199) = .24, p < .001, less observed maternal negative expressivity, r(198) = -.18,/) < .01, less mother-rated externalizing, r(200) = -.15,/? < .04, and less mother-rated internalizing, r(200) = —.16, p < .03. When child sex, child age, and family SES variables were partialed from the correlations in Table 2, the correlations were highly similar. For the total sample (N = 202), 4 correlations that were significant decreased to the level of a trend; 12 correlations remained significant but decreased in level of significance (e.g., from p < .001 to p < .01); and the other 62 correlations retained their significance levels. In the case of the SEM sample (n = 176), 11 correlations decreased their level of significance but remained significant; 3 correlations dropped to the level of a trend; 1
This group of 176 families was very similar demographically to the 26 families excluded for missing data. Only one significant difference was found between the two groups of participants: Mothers of families not used in the SEM analyses reported more negative expressivity with their families than did mothers included in the SEM analyses. 2 Although we were most interested in parents' expressions of negative dominant emotions (e.g., anger), we explored the possibility that parents' negative submissive emotions [i.e., softer negative emotions such as loss, pouting, or sadness; a = .70; rs(196) with positive emotion and with negative dominant emotion = .22 and .48, ps < .01 and .001, respectively] might also function in the hypothesized ways. However, parents' selfreported expressions of negative submissive emotions were not significantly correlated either with child regulation or with child behavior problems and competence. This lack of associations eliminated the possibility that the same mediational processes hypothesized for negative dominant emotion would hold for negative submissive emotion.
481
MEDIATION THROUGH REGULATION
Table 2 Intercorrelations for All Measured Variables for Total Sample (N = 202) and SEM Sample (n = 176) 1
Variable
2
3
4
5
6
.01
-.13t
.13+ .16*
.08 .24**
.12+ —
-.17* -.07
7
8
9
10
11
12
13
.04 .20**
.20** .15*
-.04 -.21**
-.20** -.13+
Positive mother expressivity 1. Observed 2. Self-reported Negative mother expressivity 3. Observed 4. Self-reported Regulation 5. Persistence 6. Mother report 7. Teacher report Externalizing 8. Mother report 9. Teacher report Internalizing 10. Mother report 11. Teacher report Social competence 12. Mother report 13. Teacher report
.20**
.19** -.48** — .01
__47*** .01 -.11 -.01 .10 .10 .23**
.16* — .13t - . 1 0 .22** -.22** -.26*** .19* — .21** -.14+
-.08 -.20**
-.16* -.06
.19* -.00
-.13* -.04
.04 .20**
— .14t
.17* .14t
.15* .26*** -.12 .17* -.10 -.20**
.24*** -.09 .20** -.17*
-.19** - . 2 1 * * -.24*** -.15+
— .28*** .24**
.27*** — .48***
.15* .27***
-.22** -.07
.17* -.14+
.26*** -.10 .17* .18*
-.02 -.06
.18* .08
.24*** -.23*** -.10 -.02 -.02 .50*** -.64*** _ - 4 2 *** „ 27*** -.18* — __41*** -.68*** -.15* -.27***
22** .68*** .41***
.25*** 45*** .71***
.25*** -.22** -.10 .16*
-.63*** - . 4 1 * * * -.40*** -.66***
— .42***
.41*** —
.45*** .10
.10 -.54*** _ 34*** .38*** -.34*** -.68***
.15+ .07
-.24*** -.13+ -.16* -.25***
.43*** .09
.08 .36***
— .11
.13+ —
-.18* -.12
-.03 -.03 .27*** 24***
.69*** .43***
.39*** -.54*** -.32*** - . 3 1 * * * - . 2 1 * * .71*** -.67*** -.17* -.41***
-.36*** -.18* -.23** - . 4 1 * * * — .43**
.45*** —
Note. Correlations above the diagonal are for the total sample (N = 202); correlations for the sample used in structural equation modeling analyses are below the diagonal (n = 176). t p < . 1 0 . * p < . 0 5 . * * p < . 0 1 . *** p < .001.
and the remaining 67 correlations retained their significance levels. Thus, controlling for children's sex, age, and family SES did not attenuate the overall pattern of relations.3
The Structural Equation Models Because the sample size was relatively small for SEM analyses (i.e., N < 200), we took two steps to ensure that the analyses were appropriate for the data. First, we searched the data for Heywood cases (negative error variances) and out-of-range covariances (standardized estimates greater than 1; Hoyle & Panter, 1995). Because neither of these problems was found, all of the parameter estimates were deemed plausible. Second, standard fit statistics for the models were compared with fit statistics calculated with robust parameters. The use of robust parameters has been found to be more trustworthy when the distribution of the data does not conform to the assumed distribution (Bentler, 1993; Chou & Bentler, 1995; West, Finch, & Curran, 1995). Added confidence in the results is achieved when the standard and robust statistics agree (Bentler, 1993). As is indicated below, this was the case with our models. Our strategy of using multiple raters strengthens the viability of the patterns found in the data. However, it is possible that estimates of relations among constructs measured by reports from a single reporter may be upwardly biased. To deal with this shared measurement error, we allowed error terms within a rater to covary (Kenny & Kashy, 1992; Thomson & Williams, 1984). Thus, error terms of all variables were covaried with one another for mothers' ratings (e.g., self-reported mother negative emotion with motherreported child social competence), for teachers' ratings (e.g., teacher-rated child regulation with teacher-rated externalizing problem behaviors), and for observer ratings of mothers' emo-
tions.4 To simplify the presentation of results (and because they are of little conceptual interest), we do not display these measurementerror covariances in the model figures. As the first step in our analyses, we constructed and tested a measurement model of six latent factors with 13 measured indicator variables. Mothers' observed positive affect and self-reported positive affect served as indicators of the construct of Mothers'
3 Although not used in the present study, children's negative and positive emotions [interrater reliabilities: rs(60) = .89 and .78, respectively] during the puzzle task also were coded in 30-s intervals in the same manner as was the mothers' emotion (as = .95 and .87). Children's positive and negative emotion were not significantly related to mothers' self-reported positive or negative expressivity. Child positive and negative emotion correlated with mothers' observed negative expressivity, rs(199) = —.23 and .18. ps < .001 and .01, and child observed positive emotion also correlated positively with mothers' observed positive affect, r(200) = .15, p < .05. Child positive affect was unrelated to any other variable; child negative affect was significantly negatively related to mother-rated regulation, r(196) = —.17,/? < .05, teacher-rated social competence, r(180) = —. 18, p < .02, and observed persistence in the puzzle task, r(200) = —.15, p < .05, and positively related to mother-rated externalizing problems, r(201) = .20, p < .01. When the two child affect variables were partialed from the correlations in Table 2, the correlations were highly similar. Seventy-three of the 78 correlations (94%) remained at the same significance level or improved slightly, 3 decreased fromp < .001 top < .01, 1 decreased from p < .01 to p < .05, and 2 decreased from p < .05 to p < . 10. Thus, controlling for children's emotion during the puzzle task did not substantially diminish the overall pattern of relations. 4 The observers who coded mothers' affect during the puzzle task and children's persistence were not the same coders, and thus the corresponding error terms were not covaried.
482
EISENBERG ET AL.
Table 3 Summary of Fit Indices for All Models Model
CFI
AIC
x2
df
1. Measurement model 2. Hypothesized model 3. (a) Moderation by sex, initial (b) Moderation by sex, final 4. Moderation by age 5. (a) Moderation by SES, initial (b) Moderation by SES, final 6. Alternative A: child directed 7. Alternative B: behavior mediated 8. Alternative C: observed and selfreported mother expressivity
.999 .990 .931 .984 .998 .940 .996 .992 .992
-27.52 -31.03 -80.81 -110.60 -127.61 -87.49 -117.19 -28.68 -28.71
28.48 44.97 177.20 133.40 130.39 170.51 122.81 39.32 39.29
28 38 129 122 129 129 120 34 34
.44 .20 .003 .23 .45 .008 .41 .24 .24
.992 .992
39.42 39.31
.24 .24
.011 .033 .047 .025 .011 .044 .027 .030 .030
.970
-19.23
61.80
40
.02
.968
61.80
.02
.014
RCFI
.999 .989 — —
Robust
28.50 45.78
RMSEA
.44 .18
—
Note. Robust tests are not available for multigroup models. AICs should be compared only for similar levels of models (not for moderational vs. nonmoderational models). CFI = comparative fit index; AIC = Akaike information criteria; RCFI = robust comparative fit index; RMSEA = root mean square error of approximation; SES = socioeconomic status.
Positive Expressivity, whereas mothers' observed and selfreported negative affect were indicators of the construct of Mothers' Negative Expressivity. Children's Externalizing Behavior Problems, Internalizing Behavior Problems, and Social Competence each were measured with separate mother and teacher reports. The latent construct of children's Regulation was indicated by three measured variables: observed persistence during the puzzle task, mothers' report, and teachers' report. In testing this measurement model, we allowed all latent constructs to covary freely (a covariance path was included for every possible combination of latent constructs; Kline, 1998) and included covariances between all within-rater error terms as described above. The measurement model provided a strong fit to the data and provided the basis for the structural models: comparative fit index (CFI) = .999; ^(28, N = 176) = 28.48, p = .44; robust comparative fit index (RCFI) = .999; Satorra-Bentler ^ ( 2 8 , N = 176) = 28.50, p = .44; root mean square error of approximation (RMSEA) = .011; Akaike information criteria (AIC) = —27.52. Acceptable models should have RMSEAs of less than .05 (the lower the better), and smaller AICs indicate a better model fit. The AIC evaluates the likelihood function (between the true and expected likelihood distribution) and adds a penalty that depends on the number of parameters in the model, with the lowest value representing the most preferable model. Fit statistics from this and all subsequent models are presented in Table 3.
SEM Analysis of the Hypothesized Model The overall hypothesized model was found to have a strong fit to the data as indicated by both a CFI of .990 (RCFI = .989; RMSEA = .033, AIC = —31.03) and a nonsignificant chi-square, ^(38, N = 176) = 44.97, p = .20, Satorra-Bentler ^ ( 3 8 , N = 176) = 45.78, p = .18 (see Figure 1). Mothers' reports of internalizing behavior did not load significantly on the latent factor of Internalizing Behavior Problems (the value for teachers' report of internalizing behavior had been set to 1.0, per standard procedures). Children's internalizing behavior was retained because the source measure of the measured variable—namely, Achenbach's CBCL—has been shown to have strong validity and reliability (Achenbach, 1991a, 1991b). No direct paths from the exogenous
maternal emotionality latent constructs to any of the endogenous child adjustment/social competence latent variables were suggested by the LaGrange multiplier test. We attempted to add a covariance between the constructs of Mothers' Positive and Negative Expressivity. It was nonsignificant (covariance = —.10, ns) and did not improve the model, ACFI = .001, A ^ l , TV = 176) = .15, ns. Thus, it was dropped in the interest of parsimony. Because some children were high in both internalizing and externalizing problems, we also added a covariance between the disturbances of the two endogenous factors of Internalizing and Externalizing Behavior Problems. This covariance was not significant, (covariance = - . 0 3 , ns), and did not improve the model, ACFI = .003, A ^ ( l , N = 176) = 2.80, ns. The original hypothesized model without these potential covariances was preserved. The nonstandardized parameter estimates and significance levels for the structural paths among the latent constructs are presented in Figure I 5 ; standardized coefficients are in the text. As predicted (and consistent with the correlations in Table 2), Mothers' Positive Expressivity was significantly positively related to children's Regulation, )3 = .54, p < .01, whereas Mothers' Negative Expressivity was negatively related to Regulation, j3 = - .69, p < .01. Child Regulation, in turn, related to low levels of children's Externalizing and Internalizing Behavior Problems, J3s = - . 8 4 and -.60.ps < .001 and .066, and high levels of Social Competence, /3 = .92, p < .001. Confirming the presence of mediation, the indirect effects of the mother expressivity constructs on the child competence constructs were significant for Mothers' Positive Expressivity through Regulation to child Externalizing Behavior Problems and Social Competence, /3s = -.45 and .50, ps < .01, and were marginally significant for child Internalizing Behavior Problems, )3 = —.32, p < .10. Similarly, Mothers' Negative Expressivity had significant indirect effects 5
According to Bentler (1993) and Kline (1998), in structural equation models, one can either set an observed variable on an exogenous (not endogenous) latent factor to 1 or set the variance of one of the indicators to 1 (then one obtains estimates on all variables making up the latent variable). This is true only for the exogenous variables with nothing leading into them.
483
MEDIATION THROUGH REGULATION
ChilcF Externalizing Behavior tobleim
Mother s Positive Expressivity
Mother Report
R2 = .36 Child's Internalizing Behavior Problems
Child's Regulation
Mother s Negative Expressivity
Mother Report
Mother Report
RCFI = .989, Satorra -Bentler x 2 (4f= 38) 45.78,/>= .18, AIC = -31.03, RMSEA = .033. + p
