 |  |
| ||||
Reading Others’ Emotions: The Role of Intuitive Judgments
in Predicting Marital Satisfaction, Quality, and Stability Correspondence concerning this article should be addressed to
Robert J. Waldinger, who is now at Brigham and Women’s Hospital,
1249 Boylston Street, 3rd Floor, Boston, MA 02215. E-mail:
rwaldinger/at/partners.org  The publisher's final edited version of this article is available at J Fam Psychol. See other articles in PMC that cite the published article. | |||||||||
Abstract This study examined links between emotion expression in couple
interactions and marital quality and stability. Core aspects of emotion
expression in marital interactions were identified with the use of
naïve observational coding by multiple raters. Judges rated 47
marital discussions with 15 emotion descriptors. Coders’ pooled
ratings yielded good reliability on 4 types of emotion expression: hostility,
distress, empathy, and affection. These 4 types were linked with concurrent
marital satisfaction and interviewer ratings of marital adjustment as well as
with marital stability at a 5-year follow-up. The study also examined the extent
to which naïve judges’ ratings of emotion expression
correspond to “expert” ratings using the Specific Affect
Coding System (SPAFF). The unique advantages of naïve coding of
emotion expression in marital interaction are discussed. | |||||||||
The U.S. Census Bureau predicts that more than 90% of adults now
living in the United States will marry at some point in their lifetimes and that nearly
half of these marriages will end in divorce (Kreider
& Fields, 2001). Many of the marriages that remain intact will be
characterized by spousal dissatisfaction and poor functioning. Identifying which
marriages are likely to succeed and which are likely to fail is an essential component
of efforts to prevent marital distress and enhance relationship quality. In
observational research on couples, emotional expression has emerged as an important
predictor of both marital satisfaction and stability (Gottman, 1994; Smith, Vivian, &
O’Leary, 1990). In fact, research suggests that emotional
elements of communication may be more highly related to marital quality than actual
verbal content (Gottman, 1979; 1994; Vivian &
O’Leary, 1987), and a recent study suggests that variables
derived from observational coding of emotions can predict which couple relationships
will remain intact and which dissolve with more than 80% accuracy (Gottman, Coan, Carrere, & Swanson, 1998). In spite of these findings, researchers have often been reluctant to engage in observational coding of emotional expression in marital interactions, and for good reason. Emotions are evanescent and complicated phenomena. Gathering the kind of data on emotional expression that proves useful in predicting real-world outcomes requires a substantial commitment of resources (Fincham, 1998) and poses significant methodological challenges. These challenges include difficult decisions about which elements of emotion to code and how to maximize coders’ natural abilities to read emotion while still deriving reliable ratings. The primary aim of the research presented here was to replicate and extend previous studies in which emotional expression was found to be a powerful predictor of both marital distress and dissolution. Our approach differs from traditional studies in that we rely on the consensual judgments of unschooled raters rather than the “expert” judgments of trained coders. We also sought to add another empirical perspective to an ongoing debate in the marital literature—whether there are particular groupings of emotions beyond positivity and negativity that have implications for marital satisfaction and stability. Finally, we explored the question of how closely a commonly used set of manualized rules for identifying specific emotions corresponds to people’s intuitive identification of emotions expressed in interactions. Our alternative approach to coding using unschooled (naïve) raters is a strategy that has been effectively applied in other areas of psychology to capture elements of nonverbal behavior but has been underutilized in research on emotion in family relationships. This approach takes full advantage of human beings’ highly developed natural capacities for instantaneous recognition of emotions. | |||||||||
Research over the past 2 decades (e.g., Bradbury & Fincham, 1987; Fincham & Beach, 1999; Jacobson et al., 1994; Thomas, Fletcher, & Lange, 1997) provides clear evidence that emotion is an essential factor to consider in accounting for variability in marital quality (Bradbury, Fincham, & Beach, 2000). However, the specific nature of the association remains uncertain. Positive and negative emotion are intuitively associated with greater and lesser marital quality, respectively, and many studies have provided empirical support for these associations (e.g., Holtzworth-Munroe, Stuart, Sandin, Smutzler, & McLaughlin, 1997; Jacob, 1975). However, some research suggests that the categories of positive and negative emotion are too broad to be of maximal predictive utility and that specific negative emotions (e.g., contempt) and positive emotions (e.g., humor) are critical in predicting marital outcomes (De Koning & Weiss, 1997; Gottman, 1998). One current controversy in this area concerns the role of expressed anger in eroding or strengthening marriages. Empirical studies have provided inconsistent findings regarding the effect of anger on marital quality and change (e.g., Birchler, Clopton, & Adams, 1984; Notarius, Benson, Sloane, Vanzetti, & Hornyak, 1989). In fact, Gottman and Krokoff (1989) found that anger was related to lower concurrent marital satisfaction but also to improvement in marital satisfaction over time, suggesting the possibility that there are different short-term and long-term implications of expressing anger in marital interaction. The work of Gottman and his colleagues has been at the center of the debate about anger. In two studies, they distinguished anger from potentially related negative emotions such as criticism and contempt (Gottman, 1994; Gottman et al., 1998). They found that these latter emotions were reliable predictors of marital dissolution but that anger was not. They characterized the model explaining these findings as the “specificity of negativity hypothesis” (Gottman, 1998). As several marital researchers have emphasized (Bradbury et al., 2000; Markman & Notarius, 1987), inconsistent findings regarding the role of emotion in marriage may be related to differences in how emotions are conceptualized and operationalized in various studies. A brief look at how Gottman operationalizes anger may help clarify important issues in measuring emotion expression. Gottman’s Specific Affects Coding System (SPAFF; Gottman, McCoy, Coan, & Collier, 1996) is the most fully developed manualized coding system for observing specific emotions. The 169-page SPAFF manual sets forth rules for coding verbal and nonverbal information in order to identify 16 discrete variables. These include emotions such as anger and joy, along with a limited number of behaviors such as criticism and validation that have strong affective implications in marriage and are therefore commonly included in studies of emotions in couple interactions. The SPAFF has separate categories for 10 negative and 5 positive emotions and related behaviors. SPAFF coders rate videotaped marital interactions continuously, judging which of the 16 mutually exclusive variables (including “neutral”) is present for each instant of tape. The SPAFF system distinguishes between five specific negative aggressive emotions or behaviors: anger, contempt, disgust, belligerence, and domineering. Coders are instructed that if anger occurs in conjunction with any other negative code, the other negative code takes precedence (Gottman et al., 1996). Thus, “anger” in the SPAFF system is coded when a moment of interaction does not include contempt, belligerence, or any of several other highly negative emotions. When coders have parsed negativity into these many forms, what remains in the domain of the SPAFF anger code is likely to be of relatively low intensity.1 Because the SPAFF system does not code intensity, it is possible that raters may end up distinguishing what most people would refer to as relatively intense anger from relatively benign anger by labeling them as qualitatively different (e.g., by coding more intense anger as “belligerence” and coding less intense anger as “anger”). Ideally, a coding system for emotional expression would make meaningful qualitative distinctions among specific emotions and capture variations in the intensity of emotion expressed. Capturing variations in emotional intensity is of particular use in understanding emotion regulation. Efforts to modulate intensity are key elements of emotion regulatory processes and are increasingly the focus of emotion (Schulz & Lazarus, in press) and couple interaction research (Burman, Margolin, & John, 1993; Fincham & Beach, 1999). The challenge of developing a reliable coding manual for a system that assesses both specific emotions and a full range of emotional intensity is great, for such a manual would need to include explicit criteria for each emotion as well as specific anchors for the varying intensities of each emotion. This problem and questions about the distinctions researchers have been making among particular emotions in marital research led us to explore emotional expression using the pooled judgments of multiple naïve raters, an alternative that has the potential to add important information to our understanding of emotion expression in families. | |||||||||
Psychological researchers have used naïve coders for several decades but rarely to code emotion expression in family interactions (for a notable exception, see Smith et al., 1990). Studies indicate that untutored raters concur remarkably in their judgments of the personality and affective traits of complete strangers. When judgments are pooled, naïve raters exhibit high consensual accuracy and are able to predict important aspects of interpersonal functioning (Albright, Kenny, & Malloy, 1988; Ambady & Rosenthal, 1992, 1993; Paunonen, 1991; Rosenthal, Blanck, & Vannicelli, 1984). An approach that pools the judgments of untrained raters offers the possibility of maximizing the use of intuitive capacities to judge emotion while minimizing the bias inherent in any one individual’s impressions of another’s emotions. This method also surmounts the difficulty encountered by some researchers in training raters to code emotion reliably. Smith et al. (1990) noted that in attempting to train a group of coders to recognize emotions according to manualized criteria, “The implicit theories of affective expression possessed by the coders were too deeply ingrained for us to alter in a reliable fashion” (p. 792). Coding rules attempt to re-socialize raters to define and recognize emotion in new ways, forcing them to inhibit their own intuitive understanding of emotion in order to carry out the coding task. This need to inhibit native abilities may increase the cognitive demands on coders and may account, in part, for why it frequently takes a long time for people to learn manual-based emotion coding systems. | |||||||||
Many couple and family studies have focused on the broad dichotomy of positive versus negative emotion in analyzing family interactions (Gottman et al., 1998; Karney & Bradbury, 1997; Markman & Notarius, 1987; Mishler & Waxler, 1968). It is easier for observers to agree on whether an expressed emotion is positive or negative than to agree on finer discriminations within these two broad categories (e.g., between anger and contempt or between happiness and humor). Although research has provided some support for the predictive utility of making finer distinctions among expressed emotions in marital interactions, large numbers of specific emotions are difficult to analyze statistically. Researchers have looked for ways to group specific emotions together in some meaningful clusters beyond the positive and negative supraordinate categories, but, for the most part, these clusters have been created a priori on primarily theoretical rather than empirical grounds (e.g., Gottman, 1994; Pasch & Bradbury, 1998). Our approach to observational coding enabled us to explore an empirical basis for a middle ground between the use of a positive–negative emotion dichotomy that is easier to code and analyze but may limit predictive utility and conceptual understanding and the use of multiple emotion variables that allow for more meaningful predictions about marriage but are more difficult to code reliably and to use in data analysis. We wanted to determine the extent to which untutored raters, using specific emotion labels commonly cited in the marital literature, discriminated among emotions beyond the basic positive and negative valence. In light of the questions noted above about whether anger is distinct from other negative aggressive emotions, with different implications for marriages, we had a particular interest in examining the extent to which naïve individuals distinguished anger from other negative aggressive emotions. A review of past theory and research on couple and family interactions suggested particular dimensions that might shape the distinctions that lay coders observed in the emotions expressed in couple interactions and might have particularly strong implications for marital quality and stability. Dominance has been emphasized as an important relationship dynamic with strong implications for emotion and for quality of functioning. Citing work in child development, social learning theory, and studies of family psychopathology, Markman and Notarius noted, “There has been a clear convergence among family scholars regarding the belief that dominance is a key process in family interaction” (Markman & Notarius, 1987, p. 339). Other researchers studying gender differences in emotion in interpersonal interactions have also emphasized the importance of a dominance dimension in distinguishing among different emotions (e.g., Brody, 1999; Timmers, Fischer, & Manstead, 1998). Anger and contempt are emotions typically identified as dominant, whereas emotions such as sadness and fear are seen as more submissive. In the realm of positive emotion, studies of close relationships have highlighted a potentially important distinction between the expression of empathy or validation and the expression of affection or warmth (Linehan, 1987). Whereas affection refers to a feeling of fondness or tender attachment, empathy involves perceiving the internal frame of reference of another with accuracy (Rogers, 1975). Empathy and validation have been defined differently by different investigators, but the core aspect of both terms involves an understanding and recognition of a partner’s thoughts and feelings (Levenson & Ruef, 1992). It is possible to feel affection for a partner without understanding his or her point of view, and the distinction is potentially of great importance in couple relationships. Empathy or validation has been cited by numerous investigators as a key aspect of couples interaction and a predictor of marital satisfaction and functioning (Julien, 1989; Markman & Notarius, 1987; Schaap, 1982; Weiss & Heyman, 1990). | |||||||||
The Present Study: An Alternative Approach to Examining Links Between Emotional
Expression and Marital Satisfaction and Stability We used the pooled judgments of naïve coders to examine links between expressed emotion and three independently measured marital outcomes: interviewer-based assessments of marital quality, self-reports of marital satisfaction, and a 5-year follow-up of marital stability. As in manualized approaches to emotion coding, our coders used a set of emotion labels found in previous marital research to be relevant to marital outcomes. However, in contrast to manualized approaches, coders underwent virtually no training but instead depended on their intuitive understandings of commonly used terms (e.g., anger, sadness) to recognize emotions. By pooling the ratings of five to six coders we were able to derive reliable estimates of the type and intensity of emotional expression in 30-s epochs of interaction. To ensure a wide range of marital functioning, we chose a sample of young adults who had distinct differences in individual functioning in their adolescence and were now in committed relationships. We also report on a direct comparison of our naïve coding approach with the manualized SPAFF on a small sample of couple interactions that were generously provided by Gottman and Carrere from their studies at the University of Washington. Direct comparison of two observational coding systems across laboratories is rare. Obstacles to such comparisons include the large time commitment involved in emotion coding and the need to obtain participants’ informed consent to use their data in more than one laboratory. In this case, we were fortunate that four couples whose marital interactions were rated by expert SPAFF coders at the University of Washington had given permission for other researchers to study their videotapes. | |||||||||
Participants Forty-seven heterosexual couples participated in the study. One member
of each couple was one of 146 original participants in the Adolescent and Family
Development Project (now the Across Generations Project), a longitudinal study
of psychological development begun in 1978 (see details in Hauser, 1991). On entering this longitudinal study at
age 14, participants were members of primarily Caucasian middle- and
upper-middle-class families. Approximately half were recruited from the freshman
class of a local high school (n = 76), and half
were psychiatrically hospitalized adolescents without psychosis or mental
retardation (n = 70). The predominant diagnoses
during these participants’ hospitalizations were mood or disruptive
behavior disorders.In the current study, we examined data from the first 47 original subjects who participated with their romantic partners in follow-up assessments of all participants conducted at age 32. The composition of the sample of 47 original participants was as follows: 20 men and 27 women, 29 from the high school cohort and 18 from the psychiatric cohort. The 94 individuals in these 47 couples were predominantly Caucasian (94%). The average age of participants was 32.20 years (SD = 3.6 years). Thirty-seven of the 47 couples were married (average length of relationship = 5.4 years, SD = 3.4 years), whereas the remaining 10 were living together (average length of relationship = 3.5 years, SD = 3.0 years).2 The average number of children per couple was 1.5 (range = 0–5). Among the 94 participants, the median level of education attained was some years of college without completion of a degree. The median family income was between $40,000 and $60,000 per year. Data from four additional couples were used to compare the SPAFF with our naïve coding system. These couples were participants in a longitudinal study of newlyweds conducted at the University of Washington (Gottman et al., 1998) and had given written permission for other research groups to use videotapes of their interactions for further study. The 8 individuals in these couples were mostly Caucasian, middle-class (average yearly income between $40,000 to $54,000), and college-educated. The mean global score on the Locke-Wallace Marital Adjustment Test (Locke & Wallace, 1959), a widely used measure of marital satisfaction, was 115, which was one standard deviation above the commonly cited cutoff point for marital distress (Freeston & Plechaty, 1997). Procedure Couple interaction task Participants engaged in two 10-minute laboratory-based discussions of areas of marital conflict, a task widely used in marital research (see Gottman, 1994). Independently of one another, participants were asked to identify the most important areas of disagreement in their current relationship. Participants were asked to discuss the disagreement they rated as most important in a 10-minute videotaped discussion with their partners. Each participant recorded on audiotape a one- or two-sentence statement summarizing the problem to be discussed, and this audiotape was played for the couple at the start of each discussion. In counterbalanced order, couples discussed one problem identified by the man and one identified by the woman. Among the most common discussion topics were difficulties with couple communication, disagreements over finances, and conflict over household chores. Discussions took place in a 10- ×12-foot (3.0- ×3.7-m) room in which participants sat facing each other in front of a one-way mirror. Participants were aware that they were being videotaped. Two video cameras were used to obtain clear, close images of each participant’s face and top of the torso to optimize the ability to observe facial expression and body language. The two images were recorded in a split-screen format so that partners appeared side by side. Marital discussions in the newlywed study at the University of Washington were conducted according to a similar procedure, but participants discussed one mutually agreed upon topic for 15 min (Gottman et al., 1998). Emotion coding The first of the two conflict discussions for each of the 47 couples was rated by undergraduates or recent college graduates, all of whom had completed general coursework in psychology. Coders rated participants’ emotion expression during the discussion using the emotion expression scales described in the Measures section below. Videotapes of the 10-min discussions were divided into twenty 30-s segments, and coders rated these segments in randomized order. We chose 30-s segments (as opposed to longer or shorter segments), taking into consideration the amount of time necessary to form an accurate judgment of the emotion being displayed and the practical constraints of the time required to code each segment. Clips were coded in randomized order in an effort to increase the likelihood that sequential connections between segments would reflect ongoing streams of behavior rather than artifactual connections due to common repeated-measure problems such as carryover or practice effects. Raters watched each 30-s segment twice, coding first one spouse’s emotional expression and then the other. The order was carefully counterbalanced for each segment. To minimize the influence of one partner’s behavior on the coding of the other partner, one half of the split video screen was covered by dark fabric so that only one participant was visible at a time, but no effort was made to block out the partner’s vocalizations. Two groups of coders produced the data for this study. The first group, consisting of three men and three women, coded the first 40 marital interaction videotapes. A second group of coders was assembled that included two of the original coders along with three new coders. The second group of coders (consisting of two men and three women) rated nine videotapes chosen at random from the original 40 in addition to seven new tapes of couples who participated in the ongoing Across Generations Project assessments after the original group of 40 had been coded. For the SPAFF comparison, the emotion coding procedures described above were applied to the four marital interaction videotapes provided by the Gottman laboratory. The same videotapes had been coded independently by two experienced SPAFF coders at the University of Washington. The two SPAFF coders rated the videotaped discussions continuously, classifying the participant in each moment of the discussion as expressing one of 16 categories of emotion or emotion-based behavior. Interrater reliability for all SPAFF codes as calculated using Cohen’s kappa (Cohen, 1968) was .79.3 For this study, data from only one of the two SPAFF coders were used for each participant, and the selection of which coder’s data was used was done by random alternation across the 8 participants. Measures Emotional expression The untrained coders were asked to rate participants on 18 variables for each of the 20 segments based on their own understanding of each variable. These 18 variables (see Table 1) were culled primarily from the SPAFF (Gottman, McCoy, Coan, & Collier, 1996). Five additional dyadic behavior patterns (e.g., reciprocates partner’s negativity) were coded but not analyzed in the study presented. To capture emotional intensity, coders were asked to rate the intensity of the participant’s display of each of the 18 variables during that segment of the interaction using Likert-type scales ranging from 0 (not at all) to 9 (extremely). Coders rated each of these 18 variables separately so that expression of multiple emotions during the 30-s period could be captured easily. No definitions of the individual variables or any additional instructions were given.
Reliability was assessed for each emotion variable using the procedure described by Rosenthal and Rosnow (1991) for calculating the reliability of composite scores from multiple raters. Just as the composite reliability of a multi-item scale can be calculated by applying the Spearman-Brown formula to the average inter-item correlation and the number of items on the scale, the composite reliability of a score aggregated across coders can be determined by using the average interrater reliability and the number of raters. Pearson correlations were calculated between all possible pairs of coders on each variable for each 30-s segment of coded videotape. This was initially done using the ratings from the first cohort of coders. Following Rosenthal and Rosnow (1991), the mean interrater correlation for each variable was calculated, and the Spearman-Brown formula was applied to these mean correlations to derive a measure of the reliability of the composite scores for each of the 18 coded variables. Thus, for example, the mean intercorrelation among all pairs of the six coders for their ratings of the variable “critical” was .46. Using this correlation and the number of coders (six) in the Spearman-Brown formula yielded a composite reliability of .82 for the pooled ratings of all six coders for “critical.” The mean intercorrelation between all possible pairs among the first cohort of six coders for all the variables was .30. Because our final variables combined the ratings from all six coders, the mean composite reliability of the scores for all the variables was .66, with individual variables ranging from .89 (humorous) to .27 (disgust) (see Table 1). The mean correlation between all pairs of coders and the composite reliability on the 18 variables were highly similar for the second cohort of coders (mean interrater correlation = .32, composite reliability = .70).4 Although our goal was to arrive at reliably coded groupings of emotion that aggregated the individual variables into meaningful clusters (as described below), it is interesting to note that 14 of the 18 individual composite variables had reliabilities of .60 or greater. Because of poor interrater reliability, disgusted and belligerent were dropped from further analyses. Because of their conceptual importance in interpersonal interaction, fearful and tense/anxious, which were also below generally acceptable levels of reliability, were combined into one anxious/fearful variable, resulting in an effective reliability of .51. This combined variable was included with the other 14 reliable variables in further analyses, resulting in a total of 15 variables. Relationship satisfaction The Dyadic Adjustment Scale (DAS; Spanier, 1976) was used as a measure of marital satisfaction. The DAS is a widely used, 32-item measure of marital satisfaction. It has demonstrated high internal consistency and has been shown to distinguish between distressed and nondistressed couples and between abusive and discordant, nonabusive couples (Rosenbaum & O’Leary, 1981). DAS scores range from 0 to 151, with scores below 100 typically used to identify marital distress. DAS data were available on 82 of the 94 participants (41 men and 41 women).5 Husbands’ and wives’ DAS scores were highly correlated (r = .70), so scores were averaged to arrive at a couple score for use in all analyses. Marital adjustment The marital adjustment subscale of the Social Adjustment Scale (SAS) was used as a measure of marital functioning. The SAS (Weissman & Paykel, 1974) is a semistructured interview that characterizes individuals’ adaptive functioning in six domains (e.g., work, extended family) on a 7-point scale ranging from 1 (excellent) to 7 (severe impairment). Adjustment in the marital domain is assessed by considering the participant’s responses to five questions about level of conflict, conflict resolution, and the degree to which the individual’s opinions and priorities have been voiced and considered in the relationship in the preceding 2 months. The global 7-point adaptive functioning rating is based on the interviewer’s overall assessment of functioning as indicated by responses to these five questions. Lower scores indicate better adjustment. Interviewers were extensively trained using standard procedures (Weissman & Paykel, 1974). In addition, during training, all interviewers independently scored five audio-tapes that were part of a separate study (Crowell et al., 1996) and scored by an expert SAS rater. All interviews were audiotaped, and 25 of the tapes were scored by all SAS interviewers to establish interrater reliability. Pearson correlations for all pairs of interviewers were calculated. The mean correlation for ratings of global marital functioning was .86, indicating good interrater reliability. Independent interviewers’ SAS ratings of husbands and wives were highly correlated (r = .80), so scores were averaged to arrive at a couple score for use in all analyses. The marital adjustment score from the SAS was available for 91 of the 94 participants (45 men and 46 women). As expected, marital satisfaction (DAS) and marital adjustment (SAS) scores in this sample were significantly correlated, r(80) = −.56, p < .01. Marital stability Participants from the 47 couples included in our sample were contacted by telephone on average 4.7 years (SD = 1.4 years) after their laboratory visit as part of the follow-up procedures of the Across Generations Project. At this follow-up, participants were asked whether they were still married or living together with their partner. Thirty-nine of these couples were still together, and 8 had separated.6 | |||||||||
Intensity of Observed Emotional Expression The first column of Table 1 shows
the mean intensities for each of the 18 coded variables averaged over the 20
epochs for each 10-min discussion. Mean intensities for individual variables
were low, indicating minimal expression of any specific emotion or
emotion-related behavior in any given 30-s epoch. The generally low levels of
expression resulted in some variables having distributions that were positively
skewed. For this reason, we conducted a power transformation to approximate
normality for the purpose of improving the accuracy of p values
and significance levels of tests in our statistical models. All variables were
transformed using the formula 2x2/3 (Box & Cox, 1964).7Identifying Core Groupings of Emotion Using data from 94 individuals (47 couples), we conducted factor
analysis to identify meaningful clusters of the 15 variables. For the factor
analysis, participants’ mean composite scores over the entire
interaction on each of the 15 variables were subjected to principal axis
factoring with orthogonal rotation according to varimax criterion.8 According to examination of the scree plot
and the use of the criterion of eigenvalues >1, four emotion groupings
were identified that accounted for 82% of the total variance. The
lowest factor loading for any variable was .59, all above the
“good” loading level identified by Comrey and Lee (1992). One variable (warm) loaded
similarly on two factors, and for theoretical reasons this was included in the
factor interpreted as Affection. All other variables loaded strongly on only one
factor. Table 2 shows the factor loadings
for all 15 variables.
Factor 1, which we labeled Hostility, included the following variables: defensive, critical, angry, irritable, contemptuous, and domineering. Factor 2, labeled Empathy, included the following variables: acknowledges partner’s perspective, interested in understanding partner, and tuned in to partner’s feelings. Factor 3 included affectionate, humorous, and warm; this factor was labeled Affection. Factor 4 included the following variables—sad, withdrawn, and anxious/fearful—and was labeled Distress. Individual scale scores for each participant were derived by taking the mean of all items on that factor (Tabachnick & Fidell, 1996). Analyses below use these scale scores aggregated over the whole discussion for each participant.9 The effective inter-rater reliabilities for these scale scores were all at or above .80 except for Distress, which was .74 (see Table 3). Correlations among the four scales ranged from .11 (Hostility and Distress) to .56 (Affection and Empathy), as shown in Table 3.10
Anger and Other Negative Aggressive Emotions Given the interest in the field in the relations between anger and other
negative aggressive emotions, we examined these linkages more closely.
Specifically, we looked at the correlations between anger and three of the
negative emotions that Gottman distinguishes from anger in his specificity of
negativity hypothesis: criticism, contempt, and defensiveness. Correlations
between anger and these negative aggressive emotions were all large in magnitude
for both men and women, sharing as much as 74% of their variance,
r(47) = .86, p < .001.
The only correlation below r = .74 was the link
between defensiveness and anger for women, r(47) =
.49, p < .001. These results suggest that unschooled
observers do not make clear distinctions between anger and these other negative
emotions.Links between emotional expression and marital satisfaction and functioning Men in more maritally satisfied couples were seen by coders as expressing greater Empathy, r(44) = .40, p < .01; greater Distress, r(44) = .30, p < .05; and less Hostility, r(44) = −.53, p < .001, in their marital interactions. There was a marginally significant positive association between couple satisfaction and men’s expression of Affection, r(44) = .28, p < .07. Women in more maritally satisfied couples expressed greater Empathy, r(44) = .35, p < .05, in their marital interactions. There was a marginally significant positive association between couple satisfaction and women’s expression of Affection, r(44) = .25, p < .10. Correlations between couple satisfaction and women’s expression of Hostility, r(44) = − .19, p = .22, and Distress, r(44) = − .09, p = .59, did not reach the level of statistical significance. Men who were in couples rated by SAS interviewers as having poorer marital adjustment were observed to express more Hostility, r(47) = .55, p < .001, and less Empathy, r(47) = − .43, p < .01, during marital interactions. Women in couples rated by SAS interviewers as having poorer marital adjustment were observed to express more Distress, r(47) = .50, p < .001, and less Empathy, r(47) = − .32, p < .05, during marital interactions. Women in less well-adjusted couples, as rated by our interviewers, were marginally less likely to express Affection, r(47) = − .26, p < .10. The correlation between poorer marital adjustment and the expression of Hostility for women, r(47) = − .14, p = .34, was not statistically significant nor was the correlation between poorer marital adjustment and the expression of Distress for men, r(47) =.05, p = 73. Predicting marital dissolution Couple breakup at follow-up was used as a dichotomous variable and correlated with each of the four emotion groupings using point biserial correlations. For men, expression of Empathy during the marital interaction was negatively correlated with subsequent break-up, r(47)= − .29, p < .05, and expression of Affection was negatively correlated with marital dissolution at a trend level, r(47) = −.27, p < .10. Expression of Hostility, r(47) = −.18, p = .22, and Distress, r(47) = −.14, p = .37, were not significantly correlated with break-up for men. For women, expression of Affection during the marital interaction was significantly negatively correlated with subsequent marital dissolution, r(47) = −.29, p = .05, but expression of Hostility, r(47) = −.01, p = .93, Distress, r(47) = .012, p = .93, and Empathy, r(47) = −.16, p = .26 were not significantly linked with breakup. An additional logistic regression analysis was conducted to examine the overall predictive power of all four emotion variables in combination.11 Separate models were estimated for women and for men using the four emotion scale scores indexing their emotion expression. In both the men’s and the women’s models, we correctly identified 85% of the couples whose relationships remained intact. Of the couples whose relationships dissolved, 75% were correctly identified using the women’s emotion expression variables, whereas 63% were correctly identified from the men’s emotion expression variables. Overall, these statistics indicate that we correctly identified whether couples would remain together or break up by the 5-year follow-up with 83% accuracy using women’s emotional expression and with 81% accuracy using men’s emotional expression. These accuracy rates would be even higher if we combined men’s and women’s data on emotional expression into one model. Links Between the Naïve Coding System and the SPAFF Having provided evidence for the validity of this coding method by using
indices of marital functioning, we now turn to the question of how the SPAFF and
the naïve coding method compare in their abilities to detect a range
of emotional expression. We were particularly interested in whether the
naïve coding system captured low levels of emotion that were not
captured by the SPAFF.We compared the frequency of observed emotions as coded by the SPAFF to the intensity of observed emotions as captured by our naïve coding system using two strategies. First we examined the mean frequencies and intensities of the emotions observed by coders using the two systems. Then we correlated variables derived from both systems to investigate the degree to which the two systems capture similar constructs. In order to compare SPAFF data with naïve emotion coding data, the 16 SPAFF variables were grouped into scales by summing across items that matched the four factors identified by the principal axis factoring of naïve coding data reported above. Hostility was composed of the aggregate frequency of anger, disgust, contempt, belligerence, domineering, and defensiveness. Distress included sadness, tension, whining, and stonewalling. Empathy incorporated interest and validation, whereas Affection included affection, humor, and joy. SPAFF scores for the frequency of expression of Hostility, Distress, Empathy, and Affection for each participant were calculated for each 30-s epoch that constituted the coding units for naïve coders. These SPAFF frequency scores were compared with naïve coding intensity scores. Table 4 contains descriptive information on the SPAFF-derived frequencies of emotional expression and the intensities of emotional expression derived from naïve coding. In the SPAFF-derived data, hostile and distressing emotions or emotion behaviors were the most frequently coded in each 30-s epoch, although they were coded at low frequency. On average, 2.32 seconds of each 30-s epoch were coded as Hostile, and 1.42 s were coded as Distress. Empathy and, especially, Affection were coded much less frequently. The generally low mean frequencies for all four categories of emotional expression indicate that much of the interaction did not contain sufficient emotional expression to trigger a specific affect code by the SPAFF and therefore was coded as neutral. Of the 240 30-s epochs, 77% were coded as having some kind of emotional expression for at least one second. However, 39% of the epochs in which emotion was identified had frequencies of 1 (22%) or 2 (17%), indicating that emotion had been coded for only 1 or 2 s of that 30-s interval and that the rest of the epoch was coded as neutral. When our four specific categories of emotion are examined, the low frequency of emotion coded becomes even more evident. For example, 98% (all but 6) of the 240 thirty-second epochs were seen by SPAFF coders as displaying no Affection, and 81% were seen as devoid of Empathy.
The average intensities of emotion expressed (see Table 4), as coded by our naïve coding system, were comparable to those obtained for the group of 47 couples. Empathy was the scale rated as being expressed most intensely during the 30-s epochs. Hostility, Distress, and Affection were rated as being expressed at significantly lower intensities. The naïve coding system was designed to capture a range of emotion intensities, and it appeared to do this well. For example, all epochs but one were coded as having displayed some Empathy, with 50% of the epochs being coded with an intensity of 2.67 or higher. All but 10 of the epochs (96%) were rated as showing evidence of at least some Hostility, and 50% received a score above 0.54. Of the 240 coded epochs, 210 were coded as displaying some Distress and some Affection. Although the distributions yielded by the naïve ratings are still somewhat skewed, there is a substantial range of intensities across epochs, suggesting that differences in intensities between epochs can be meaningfully investigated. Of the 56 segments coded as neutral on the SPAFF (23% of the 240 segments), all were coded as displaying some degree of intensity of positive and negative emotional expression by our naïve coders. The mean intensities during these SPAFF neutral segments were as follows: Hostility = 0.69, Distress = 0.51, Affection = 0.80, and Empathy = 3.21. This suggests that naïve coding of emotional intensity may meaningfully differentiate low levels of affective expression that are marked as neutral by the SPAFF system because they fail to exceed the threshold required by the SPAFF system for coding the presence of emotion. It is also possible that this naïve coding system overidentifies emotion, but the meaningful associations with marital quality reported above would argue against this possibility. The first column of Table 5 contains correlations between the SPAFF and naïve coding emotion expression scores for data at the 30-s epoch level. In these analyses, we correlated emotion scale scores derived from the SPAFF and from the naïve coding system for the 240 epochs coded from the four videotapes of 8 participants. There was a moderate degree of consistency between the frequency with which Hostility was coded using the SPAFF and the intensity of Hostility observed by naïve coders. Similarly, the SPAFF-derived frequency of Empathy also correlated at a moderate level with the intensity of Empathy derived from naïve coding. The connection between the two systems was weak for Distress and absent for Affection. The relatively high number of epochs in the SPAFF data in which no emotion was coded limits the degree of association that can be found at the epoch-level between the SPAFF and naïve coding methods.
As shown in the second column of Table 5, we then examined correlations between the overall frequency and average intensity of Hostility, Distress, Empathy, and Affection expressed by each participant over the entire course of the marital discussion. For this analysis, the emotion codes were aggregated across epochs for each individual. Hostility, Distress, and Empathy correlated at very high levels, suggesting that the SPAFF and naïve coding methods rank individuals in highly similar orders in terms of their degree of expression on these three categories of emotion. Ratings of Affection from both systems were moderately correlated even though Affection was coded infrequently using the SPAFF system. | |||||||||
In this study we addressed three primary questions: (a) Can we confirm previous findings that emotion expression predicts marital quality and stability? (b) Is it possible to identify groupings of emotions beyond positivity and negativity that are theoretically meaningful and have implications for marital quality and stability? (c) Given the multiplicity of perspectives on emotion, how do the ratings of unschooled coders compare with the ratings of expert coders using manualized rules? In order to address these questions, we developed a method for using the pooled judgments of multiple untrained raters to assess both the intensity and type of emotion expressed in couple interactions. The Predictive Power of Emotion Expression Consistent with previous research, we found that emotion expressed in
marital interactions related in meaningful ways to (a) self-reported marital
satisfaction, (b) interview-based assessments of marital quality, and (c)
marital dissolution. We found that current marital quality (as measured by the
Locke-Wallace Short Marital Adjustment Test and the SAS) was linked to the
intensity of expression of four types of emotion. Eight of the 16 correlational
links examined (4 emotion scales × 2 indices of marital quality, all
calculated separately for men and women) were significant, and 3 additional
correlations were marginally significant. All but one correlation were in the
expected direction, and there was impressive convergence in the pattern of
findings across the interview-based and self-report measures.The finding that men’s but not women’s Hostility and women’s but not men’s Distress were significantly correlated with interviewer ratings of poorer marital adjustment may reflect differences in the kinds of emotions that men and women express in distressed marriages.12 Prior research suggests that at least in some contexts, women may be more likely to express sadness and vulnerability and men may be more likely to express hostile emotions (Brody, 1999). The unexpected correlation between men’s Distress and greater couple satisfaction may be an indicator of men’s greater willingness to express vulnerability in more satisfying relationships. In this case, the distinction is between distressing and hostile emotions. By contrast, men’s hostile emotions, including anger, were associated with poorer concurrent marital functioning. In fact, men’s Hostility accounted for 28% of the variance in the couples’ reports of marital satisfaction. These findings lend support to the argument that it is useful to capture distinctions among negative emotions, because not all types of negative emotions function in the same way in marriage. With regard to our positive emotion groupings, there were connections between both Empathy and Affection and our positive marital outcomes in the expected directions. In addition to the expression of warmth, the expression of the desire to understand one’s partner may be particularly helpful to marriages (Gottman, 1994; Julien, 1989; Markman & Notarius, 1987; Schaap, 1982; Weiss & Heyman, 1990). However, the similar pattern of connections and the high degree of correlation between these two groupings suggests some caution in assessing their utility as separate groupings. Expressed emotions also had implications for marital stability. In fact, the four types of emotion in combination predicted marital dissolution five years hence with more than 80% accuracy. This finding is particularly impressive when one considers that the coders were untrained college-age young adults. Our correlational analyses indicated that the expression of positive emotions was significantly linked with marital stability, whereas the expression of negative emotions was not. For men the expression of greater Empathy (and at the trend level, greater Affection), and for women the expression of greater Affection predicted that the couple would remain together in the 5 years following the observed interaction. This finding is consistent with those of the Gottman et al. (1998) newlywed study and the Pasch and Bradbury (1998) study of social support in marriage, both of which found that positive emotions and behaviors predicted marital stability. Increasing attention to the role of positive and supportive behaviors in marriage is clearly warranted (Cutrona, 1996). Finding a Middle Ground: Hostility, Distress, Affection, and Empathy Researchers continue to search for an appropriate compromise between the
convenience of parsing expressed emotion into two global groupings of positivity
and negativity, and the promise of greater understanding that might come with
considering multiple specific emotions. The findings reported above indicate why
such a compromise may be useful. The ratings of our untrained judges clustered
into four emotion groupings that were differentially linked with key indices of
marital functioning. Moreover, the four groupings make sense in light of prior
theory and research on dimensions of emotion in interpersonal interactions. Our
raters clearly distinguished between two types of negative emotion: Hostility
and Distress. It may be argued that Hostility and Distress differ most markedly
along the dominance dimension observed by Russell and Mehrabian (1974) and discussed extensively in the
literature on gender differences in interaction (Brody, 1999; Timmers et al.,
1998). Our naïve raters did not clearly distinguish anger
from other negative aggressive emotions such as criticism and contempt. This
finding suggests caution in differentiating anger from related negative
aggressive emotions. Recent findings using the SPAFF have suggested that anger
is not toxic to marriages but that criticism, contempt, defensiveness, and
withdrawal are. It is possible that anger as defined by the SPAFF is not
qualitatively different from these other negative emotions but that it simply
represents a less intense form of negative aggressive emotionality. Because of
the potentially important clinical implications of this issue, more research is
warranted.Our untrained raters also distinguished between two types of positive emotions or emotion-relevant behaviors: Affection and Empathy. Although caution is warranted because these groupings of emotion correlated within this sample at a level of .56 and other research has suggested the presence of only one positive factor (Smith et al., 1990), this distinction is supported by other empirical investigations (Gottman, 1994; Julien, 1989; Markman & Notarius, 1987; Schaap, 1982; Weiss & Heyman, 1990) in which empathy (or validation) and affection (or warmth) differ in the extent to which they predict marital stability and satisfaction. Further research is needed in which additional emotion descriptors conceptually linked with empathy or affection are incorporated into the rating system, so that the degree of overlap or independence can be clarified. Comparing the Perspectives of Naïve and Expert Coders Our naïve coding approach also had predicted links with the
SPAFF despite the fact that the SPAFF system codes for the presence or absence
of an emotion and our naïve coding approach rates intensity.
Although comparison of the two systems at the level of 30-s segments of a
couple’s interaction yielded relatively low correlations, we found
high correlations between the two sets of ratings for participants over the
entire course of a discussion. This suggests that the two coding methods are
consistent in their ratings of the degree to which individuals express
particular emotions in a marital interaction. SPAFF data, collected
second-by-second, are particularly useful for fine-grained sequential analysis
of emotion patterns within dyads. Data from the naïve coding system
appear well suited to examining the intensity of emotional expression, an aspect
of emotion that is especially relevant in the study of emotion regulation. We
recognize of course that the sample was quite small on account of the practical
constraints noted above, and we must consider these results with caution.
However, the magnitude of links between the two systems at the participant level
was very strong.The magnitude and consistency of the links between our four emotion groupings and (a) the SPAFF and (b) both current indices of relationship functioning and long-term marital stability provide compelling evidence for the validity of these groupings. Pooling the ratings of untrained coders yields reliable estimates of the intensity with which 14 different specific emotions or emotion-relevant behaviors were expressed in marital interactions. It is noteworthy that reliability was calculated at the level of 30-s segments of the discussion and would be higher for ratings pooled for the entire marital discussion. Good reliability of these individual variables at the 30-s epoch level allows for the possibility of using these data to examine sequences of emotional expression in couples—for example, patterns of reciprocation of negative emotions that have been found in other studies to predict marital dissatisfaction and divorce (Gottman, 1994). We believe that naïve coding has both theoretical and practical advantages for researchers studying emotional expression in couples. The ecological validity of a cultural informants’ approach is a particular strength of this method. By not demanding adherence to a prescribed set of rules, naïve coding takes advantage of human beings’ well-honed and highly adaptive abilities to read others’ expressions of emotion. The emotion groupings that naïve coders consistently identified in this research may represent fundamental typologies of emotional expression that guide people’s evaluations of interactions in close relationships. Not coincidentally, the groupings identified—Hostility, Distress, Affection, and Empathy—are directly linked to the descriptors that individuals commonly use to characterize the interpersonal style of others. Research on marital interaction is likely to benefit from consideration of these readily identifiable types of emotional expression. We see several practical strengths that our naïve coding system brings to the study of emotional expression in the particular context of marital interactions. Coders require little or no training. Good to excellent effective interrater reliability can be achieved using composite scores obtained from multiple raters, especially for the larger emotion clusters we have identified. The system allows for incorporation of multiple perspectives within the group of coders, which is particularly important given gender and cultural differences in assessing emotional expression in marriage. The system also is sensitive to low intensities of emotional expression. These low-intensity emotional displays may be quite meaningful for the spouses in the interaction and therefore important for researchers to identify. By capturing low intensity and variations in intensity across the range of emotional expression, the naïve coding approach also permits researchers to more carefully examine processes of emotion regulation in couples’ relationships. Moreover, in contrast to the SPAFF, our naïve rating system also allows coders to rate the intensity of multiple emotions at the same time, allowing us to gather data about emotions that are expressed virtually simultaneously (“emotion blends”). Manualized coding systems that employ a binary decision-making framework (i.e., coding for the presence or absence of emotion) must include clear guidelines about the level of emotion required to trigger an emotion code. In such systems, there is likely to be subtle pressure to establish a relatively high threshold for triggering an emotion code. Instructing raters to code an emotion only when there is overwhelming evidence for its presence (e.g., telling raters to code an emotion only when it “hits you over the head”) may increase interrater reliability. This added reliability, however, comes at the cost of losing information about low-intensity emotion displays. The naïve coding approach presented in this report, which involves assessments of varying intensities rather than the presence or absence of emotion, appears to yield good levels of inter-rater reliability and to capture meaningful but low levels of emotion expression. Perhaps the greatest disadvantage of the naïve coding approach is the task of coordinating the efforts of multiple coders. This effort must be weighed against the extensive training and reliability testing required by manualized coding systems. Judging from our experience, two alterations in the coding method presented here may be warranted. Although coding segments of a videotaped discussion in random order may help minimize carryover effects from one segment of tape to another, this advantage may be outweighed by the importance of seeing each segment of a discussion in the context of what has come before it. Similarly, covering half of the video screen so that only one participant is visible for coding may be unnecessary and may hinder assessment of the full context of an individual’s emotional expression. Both of these issues will be tested empirically in future investigations. It is important to keep in mind several limitations of this work. The sample of 47 couples, while representing a range of psychological functioning, is small for the use of factor-analytic and logistic regression techniques. Future replication will be important. We also recognize that the choice of individual variables that we offered to naïve coders undoubtedly affected the factor structure that emerged. These variables had been shown in prior couples research to be relevant to marital functioning, but it is possible that the inclusion of other emotion labels would result in different emotion groupings. Ideally, we would have compared the SPAFF-derived emotion variables with our naïve coding approach using a larger sample. However, the hurdles to comparing two microanalytic observational coding systems are so great and the direct comparison of such systems is so rare that we believed these analyses warranted presentation. It is possible that the strength of some of the concurrent and predictive validity findings is due in part to our use of two cohorts that originally differed in their levels of functioning in important ways. However, it is important to recognize that these differences were observed during adolescence, at least 15 years prior to the timing of the assessments used in this study. Despite these limitations, the results of this study are noteworthy in several respects. Similar to other studies, we found that emotional expression was linked to concurrent marital quality and to relationship stability over a nearly 5-year period. In contrast to other studies, our findings are based on the judgments of untrained college-age young adults. That is, the judgments of unmarried college-age individuals, when pooled appropriately, tell us a great deal about how couples are doing and about the likelihood of them remaining together. These pooled judgments also shed light on four types of emotional expression that may be particularly salient in marital interactions and can predict long-term marital stability with considerable accuracy. The relevance of these four groupings of emotion to marital functioning and satisfaction is consistent with prior research on couples and families. In emphasizing these four emotion groupings, we do not deny the importance of studying discrete emotions. However, most couple and family researchers have, of necessity, reduced data on discrete emotions to larger categories. Many researchers have hesitated to engage in observational coding of emotions because of the myriad emotion variables that could be coded. This study provides empirical support for focusing on four categories of emotional expression in couples’ interactions that are characterized by theoretically meaningful distinctions among discrete emotions. Implications for Application and Public Policy Interventions to help couples modify emotion expression and regulation
processes are key elements of many approaches to marital therapy (Christensen & Jacobson, 2000; Gottman, 1999; Stanley, Blumberg, & Markman, 1999). The
findings of this study point to central elements of emotional expression in
marital interaction that clinicians may want to consider in their work with
distressed couples. This study suggests that it is important to move beyond a
simple focus on positive versus negative emotions to a more differentiated
perspective that distinguishes between emotions associated with hostility and
distress on the negative side and between affection and empathy on the positive
side. Just as the results of this study can guide the efforts of future
observational coding of marital interactions, these findings can also help
clinicians focus on fundamental aspects of emotion expression that may be
important to marital functioning and stability. | |||||||||
Footnotes Robert J. Waldinger and Stuart T. Hauser, Judge Baker Children’s
Center and Harvard Medical School; Marc S. Schulz, Department of Psychology,
Bryn Mawr College; Joseph P. Allen, Department of Psychology, University of
Virginia; Judith A. Crowell, Department of Psychiatry, State University of New
York at Stony Brook. 1Gottman et al. (1998) appeared to arrive
at a similar conclusion when they experimented with giving numerical weights to
specific SPAFF variables based on their empirical correlations with marital
satisfaction in previous research. They defined high-intensity negativity as
contempt, defensiveness, and belligerence whereas anger was classified as
low-intensity negativity. 2Because the majority of couples were married, we refer to the partners in all
relationships as husbands and wives to facilitate fluency of writing. 3More information on the reliability of this coding can be found in previous
reports by Gottman and colleagues (Gottman,
Coan, Carrere, & Swanson, 1998; Gottman, Swanson, & Murray, 1999). 4The average pairwise correlation between old and new coders on the 18 overlapping
participants was .30, which is of a similar magnitude as the average pairwise
correlation among the original coders of the first 80 participants, suggesting
adequate reliability. However, paired t tests comparing the
original coder averages to the new coder averages on the 15 emotion variables
yielded differences on the majority of the variables; the new raters
systematically assigned higher scores, suggesting a cohort effect. To equalize
the metrics used by old and new coders we calculated the ratio of the mean score
for the old raters to the mean score of the new raters on each of the 15
variables. We then transformed all new coders’ data by multiplying
their scores by the appropriate ratio for each variable. As with any Likert-type
scale, it is important to be cautious about reifying the meaning of the absolute
score. We do not, therefore, attach particular significance to the difference in
absolute scores given by either cohort. Rather, our goal was to gauge reliably
the variability in emotion expression across 30-s epochs and across individual
participants. What is important is that this variability be measured
consistently across coding groups, and our analyses indicate that it was. 5Comparison (by t test) of the emotion scale scores of the 12
participants for whom DAS data were missing with those of the other 82
participants revealed no significant differences. 6An ANOVA revealed no significant link between breaking up and the time between
the marital interaction and telephone follow-up. 7Data transformation was based on inspection of the data and was carried out
according to procedures recommended in Box and
Cox (1964) and Tabachnick and Fidell
(1996). 8Principal axis factor analysis was used because of our interest in extracting all
meaningful theoretical factors from the data. Additional principal-components
analyses produced a substantially similar solution. 9Inspection of the distribution of scores on the emotion expression scales
revealed two significant outliers. For men, a single outlying score on Distress
was four standard deviations above the mean; for women, a single outlying score
on Hostility was four and a half standard deviations above the mean. In
accordance with procedures outlined in Tabachnick and Fidell (1996), we transformed each score so that it
was two standard deviations above the mean, and these were used in subsequent
analyses. 10Consistent with the factor-analytic results, the mean intercor-relation among
variables within each scale (median r = .66) was
noticeably higher than the mean intercorrelation among variables from different
scales (median r = .22). (Median
rs were calculated with the absolute values of
r.) 11Our final models included all significant interactions among the four
variables. 12The correlations of men’s and women’s Hostility with
marital adjustment were significantly different from each other,
t(43) = 7.24, p < .01, as
were the correlations of men’s and women’s Distress with
marital adjustment, t(43) = − 7.37,
p < .01. The study was supported by grants from the National Institute of Mental Health
(K08 MH 01555 and MH 44934-11). We gratefully acknowledge the contributions of
those whose advice and assistance contributed significantly to this project:
Sybil Carrere, John Gottman, Heidi Gralinski-Bakker, Carl Morris, Robert
Rosenthal, Katie Swanson, and Erica Woodin. | |||||||||
All author affiliations Robert J. Waldinger, Judge Baker Children’s Center and Harvard Medical School. Stuart T. Hauser, Judge Baker Children’s Center. Marc S. Schulz, Bryn Mawr College. Joseph P. Allen, University of Virginia. Judith A. Crowell, State University of New York at Stony Brook. | |||||||||
References
| |||||||||
