Attachment Change Processes in the Early Years of Marriage

PERSONALITY PROCESSES AND INDIVIDUAL DIFFERENCES

Attachment Change Processes in the Early Years of Marriage Joanne Davila

Benjamin R. Karney

State University of New York at Buffalo

University of Florida

Thomas N. Bradbury University of California, Los Angeles

The authors examined 4 models of attachment change: a contextual model, a social-cognitive model, an individual-difference model, and a diathesis-stress model. Models were examined in a sample of ncwlyweds over the first 2 years of marriage, using growth curve analyses. Reciprocal processes, whereby attachment representations and interpersonal life circumstances affect one another over time, also were studied. On average, newlyweds became more secure over time. However, there was significant within subject variability on attachment change that was predicted by intra- and interpersonal factors. Attachment representations changed in response to contextual, social-cognitive, and individualditterencc factors. Reciprocal processes between attachment representations and marital variables emerged, suggesting that these factors influence one another in an ongoing way.

To be useful . . . working models must be kept up to date. As a rule this requires only a continuous feeding in of small modifications, usually a process so gradual that is it hardly noticeable. Occasionally, however, some major change in environment or organism occurs: we get married [italic added] . . . . At those times, radical changes of models are called for. (Bowlby. 1969. p. 82)

raises important questions. One question is whether such change occurs at all, particularly in adulthood. Adult attachment researchers generally agree that both accommodation and assimilation do occur. Working models show both stability and change over time (e.g., N. L. Collins & Read, 1994). For example, although longitudinal research has shown relatively high levels of self-reported attachment stability, some estimates indicate that approximately 30% of people change their attachment style over time. This has been shown to occur in various samples, including married couples, over time frames as long as 4 years (see Baldwin & Fehr, 1995; Davila, Burge, & Hammen, 1997; Fuller & Fincham, 1995; Kirkpatrick & Hazan, 1994). Hence, there is evidence that whereas many people appear to retain their attachment style, some people change.

A basic premise of attachment theory is that attachment style is stable and guides interpersonal functioning across the life span (Bowlby, 1969, 1973, 1980). However, attachment theory also suggests that over time attachment style may become a product of both earlier and current interpersonal circumstances. That is, working models may accommodate as well as assimilate new information and, thus, change over time. Moreover, as the quote above indicates, in the face of significant interpersonal events such as marriage, working models may change radically. This possibility

This brings us to the second, and more pressing question: Why do some people change attachment styles? Specifically, who is likely to change, and what specific interpersonal circumstances promote attachment change? At present there are three models of attachment change.

Joanne Davila. Department of Psychology. State University of New York at Buffalo; Benjamin R. Karney. Department of Psychology. University of Florida; Thomas N. Bradbury. Department of Psychology, University of California, Los Angeles. This research was funded by National Institute of Mental Health (NIMH) Grant MH48674 and by an NIMH research supplement for minority postdoctoral scholars. We thank Catherine Cohan. Matthew Johnson. Kieran Sullivan. Gregory Miller. Jeb Cozzi. Erika Lawrence. Ron Rogge, and Kathy Eldridge for their assistance in data collection and coding. Correspondence concerning this article should be addressed to Joanne Davila, State University of New York at Buffalo, Department of Psychology, Park Hall. Box 604110, Buffalo, New York 14260-4110. Electronic mail may be sent [email protected].

Models of Attachment Change

Contextual Model This model draws directly on Bowlby's (1969) notion that working models can change in the face of new information and, in a manner consistent with the opening quote, proposes that attachment style may change in response to contextual variables, particularly interpersonal events and circumstances (e.g., relationship difficulties). Yet most tests of this accommodation hypothesis have

Journal of Personality and Social Psychology. 1999. Vol. 76. No. 5. 783-802 Copyright 1999 hy the American Psychological Association. Inc. 0022-3514/99/S3.00

783

784

DAVILA, KARNEY, AND BRADBURY

failed to yield clear evidence that interpersonal life events predict changes in self-reported attachment (Baldwin & Fehr, 1995: Davila, Burge, & Hammen, 1997; Scharfc & Bartholomew, 1994). The exception is a 4-year longitudinal study in which relationship breakups were associated with some secure people subsequently becoming insecure (Kirkpatrick & Hazan. 1994). In addition, although they did not examine interpersonal events, Fuller and Fincham (1995) found, among married couples, that spouses' mental models of their partners (specifically, beliefs about trust) predicted changes in their own attachment models over 2 years. Similarly, Feeney, Noller, and Callan (1994) found that husbands' satisfaction and communication predicted changes in their attachment models over 9 months. It is important that, in these three studies, reciprocal associations between attachment status and relationship variables were found, suggesting, as attachment theorists have suspected, that an ongoing process of assimilation and accommodation is occurring.

Social-Cognitive Model A second model adopts a social-cognitive perspective on attachment change (e.g.. Baldwin, 1995: Baldwin & Fehr, 1995: Baldwin, Keelan. Fehr, Enns, & Koh-Rangarajoo, 1996). This model suggests that although people have a chronically accessible attachment model (or style) that may be stable over time, they also have a number of different attachment models, or relational schemas, that can be activated by specific contextual circumstances. Hence, attachment change is due to the accessibility of different models at different times depending on the context or frame of mind that the person is in. Baldwin et al. (1996) indeed found that priming different types of attachment experiences affected people's thinking about relationships. That is. people's thoughts about relationships were more consistent with the primed material than with their self-reported attachment style. However, Baldwin et al found that actual self-reported attachment style did not change following such priming.

Indi vidual-Difference Model In contrast to the prior models that focus on the role of current context in attachment change, one study has suggested that changes in self-reported attachment style do not represent accommodation of contextual circumstances but instead represent a manifestation of incoherent working models. Davila, Burge, and Hammen (1997) proposed that proneness to attachment change is a stable individual difference. Some people are more prone to attachment fluctuations because of stable vulnerability factors rather than contextual factors. This hypothesis was supported in a 2-year longitudinal study of young women making the transition to adulthood, which is typically a time of stress and change that might foster modification of working models. Specifically, women with a history of early personal and family dysfunction (e.g., past history of psychopathology, divorce of parents) were more prone to attachment fluctuations and had more in common with people who were stably insecure than were women without such a history. Moreover, there was little evidence of attachment change in response to stressful interpersonal experiences with family, friends, and romantic partners. Davila, Burge, and Hammen (1997) thus suggested that although attachment style change may, in some

cases, be a response to contextual factors, it is more likely to be related to particular vulnerabilities and to be an indicator of insecurity.

Conceptual and Methodological Issues in the Study of Attachment Change The models described above provide different pictures of attachment change processes. This is likely to be true because they provide information about different aspects of attachment change processes. The contextual model may be best thought of as a model proposed to address large-scale changes in attachment style in which people shift from one style to another because of the interpersonal events and circumstances they experience. However, the existing literature has told us little about whether and how this process works. For example, a limited number of appropriate interpersonal events have been studied. Prior studies have mostly examined change in romantic relationship status among dating couples or an aggregate measure of interpersonal life events over a specified time period (e.g., Baldwin & Fehr, 1995: Davila, Burge. & Hammen. 1997; Scharfc & Bartholomew, 1994). However, N. L. Collins and Read (1994) suggested that in order to effect changes in attachment style, interpersonal experiences must be relatively long in duration and emotionally significant. It is not dear whether events examined in prior studies met these criteria. In addition, change has not consistently been examined in samples that are conducive to studying change (sec Scharfe & Bartholomew, 1994). Hence, the likelihood of finding attachment style changes has been minimized. Furthermore, the extent to which change occurs immediately in response to events, versus slowly over an extended period of time, has not been examined. That is, how and when people change has not yet been examined. The social-cognitive model may best be thought of as a model that explains momentary fluctuations in attachment models or changes in people's present states of mind about attachment relationships. Whether it provides a model for changes in attachment styles, or traits, is unclear, as is whether it provides a model for long-term attachment change. Repeated activation of attachment models that differ from chronically accessible models may ultimately lead to changes in what becomes chronically accessible, but research has not addressed this issue yet. Thus, the socialcognitive model and the contextual model, although they both focus on how contextual factors affect people's attachment orientation, appear to address two different aspects of attachment change. The contextual model is focused on presumably long-term changes in traits, and the social-cognitive model is focused on short-term changes in states. Whether and how contextual factors produce short-term changes, or social-cognitive factors produce long-term changes, is unknown. Whether and how these change processes work together is also unknown. In addition, neither model speaks directly to the reciprocal processes that likely exist between attachment models and interpersonal experiences. People both actively construct their lives (e.g., select and create interpersonal environments) and respond to existing life circumstances (e.g., Caspi, 1993; Cicchetti, Rogosch, & Toth, 1994). Moreover, past research has yielded evidence of reciprocal processes (e.g.. Fuller & Fincham, 1995: Kirkpatrick & Hazan, 1994). Thus, these processes must be investigated to more fully understand stability and change over time.

785

ATTACHMENT AND MARRIAGE The individual-difference model takes a different approach than do the contextual and social-cognitive models. The individualdifference model can best be thought of as a model that explains how stable vulnerability factors (e.g., prior psychopathology, personality disturbance, parental divorce and psychopathology), rather than contextual factors, affect fluctuations in attachment orientation. The emphasis of this model, like the social-cognitive model, is on short-term attachment fluctuations. Unlike the socialcognitive model, however, the individual-difference model proposes that these fluctuations are a manifestation of insecurity. As with the other two models, of course, there are also a number of unknowns about this model. First, because only one study has tested it, replication is necessary. Second, the model was tested in a sample that may have biased the findings toward the individualdifference model and away from the contextual model. Davila, Burge, and Hammen (1997) noted that contextual variables may have a greater impact on self-reported romantic attachment style when romantic contextual variables are specifically examined in a sample for whom the romantic relationship is salient. Consistent with this, only a few studies found evidence for attachment style change in response to contextual factors, and all were conducted in samples in which the relationship was salient and in which relationship factors were used as predictors (Fuller & Fincham, 1995; Kirkpatrick & Hazan, 1994). Hence, a fairer test of whether attachment change can be a true form of accommodation to current circumstances versus a manifestation of insecurity is necessary. Third, the fact that the individual-difference model suggests that attachment changes are not due to contextual factors presents a problem in that it fails to address exactly why a person who is prone to attachment fluctuations might change his or her style at any particular moment. Are contextual factors involved in this process in any way? For example, are people who are prone to attachment fluctuations more reactive to life circumstances? If so, what are those circumstances? These are questions that must be addressed to refine our understanding of how stable vulnerability factors lead to attachment change. In addition to the conceptual issues identified above, there are a number of methodological issues to be addressed in the study of attachment change. First, the type of stability and change being studied must be clarified. Personality researchers have described two types of stability: absolute stability and relative, or differential, stability (see Adler & Scher, 1994; Caspi & Bern, 1990). Relative stability refers to the consistency of the rank order of individuals within a group on an individual-difference measure over time. This is the type of stability data that comes from analyses of the test-retest correlation over two time points—the type of data that has traditionally been reported in most studies of attachment stability and change. The problem with these data is that high correlations that appear to indicate high levels of stability may be misleading, because it is possible to have high relative stability while still having significant individual change. Individual change, which is likely to be what researchers studying attachment change are most interested in, is best addressed by examining absolute stability—change (or lack thereof) in the absolute level of the variable over time for each individual. Distinguishing these two types of change is important in order to make the most precise conclusions about attachment stability and change. A second related issue concerns the methodology used to study change. Researchers who study personality and those who study

development have struggled with the issue of how to best capture change over time (e.g., L. M. Collins & Horn, 1991; Karney & Bradbury, 1995). Many believe that current methods do not provide adequate descriptions of how and why people change. For example, most longitudinal studies, regardless of length, examine only two points in time and use multiple regression or cross-lagged correlations to predict change in one variable from initial levels of the other. This methodology is problematic in that (a) it* cannot reveal dynamic relationships between variables over time, (b) it fails to adequately describe change before trying to predict change, (c) it assumes that both variables in question change in the same way (linearly), (d) it may provide an unreliable estimate of change (e.g., Rogosa, Brandt, & Zimowski, 1982), and (e) it focuses on a between-subjects level of analysis rather than a within-subject level of analysis (see Adler & Scher, 1994; Bolger & Zuckerman, 1995; Kamey, in press; Karney & Bradbury, 1997, for a discussion of these issues). The last point warrants elaboration. Researchers interested in attachment change are likely to be interested in within-subject changes in addition to between-subjects changes. A between-subjects level of analysis addresses whether one variable, measured at one point in time, is associated with changes in another variable. This is consistent, for example, with predictions of the individual-difference model. However, a within-subject level of analysis addresses whether changes in one variable are associated with changes in another variable within the individual. This is consistent, for example, with predictions of the contextual and social-cognitive models, but prior research has not tested these within-subject associations. Hence, studies of change processes may benefit from adopting methods that allow for the analysis of multiwave data, the description of these data over time in addition to the prediction of change, and for both between- and within-subject analyses of change.

The Present Research The present research was designed to address a number of the existing issues in research on attachment change in order to clarify how and why people show changes in attachment representations. First, we examined attachment change processes in a context that maximizes the opportunities for change: the early marital relationship. Doing so served the dual goals of clarifying adult attachment change processes and providing information relevant to understanding the role of attachment in the course of early marriage. The early marital relationship is an optimal context for studying attachment change processes. Specifically, the marital relationship is typically the primary attachment relationship of married adults, and it is typically a time of transition both in terms of actual experiences (e.g., change in marital status, moving in with a partner, negotiating new decisions) and attachment experiences. Research suggests that over time, attachment functions are transferred from parents to peers and ultimately to romantic partners (e.g., Fraley & Davis, 1997; Hazan & Shaver, 1994). Hence, newlywed marriage may be a time during which attachment functions are being transferred to the spouse from prior caregivers or peers. This may make relationship issues particularly salient and may open working models to the possibility of revision. Furthermore, newlyweds are at a very specific life stage that tends to be experienced in a similar manner by most people. On average, newlyweds tend to be very satisfied with their marriages and then

786

DAVILA. KARNEY, AND BRADBURY

experience a gradual decline in satisfaction over time (e.g., Karney & Bradbury, 1997). This common experience facilitates examination of attachment change processes as they occur over an important phase of life and allows for the examination of whether attachment follows a related trajectory. Second, to deal with problems inherent in studying stability and change over time, we used a data analysis strategy—growth curve analysis (GCA)—that would allow us to examine change as a continuous process at both the between- and within-subject level of analysis. GCA is a procedure that transcends many of the limitations of more traditional approaches to studying change that were described earlier. Briefly, GCA proceeds in two steps. First, multiple assessments of a variable are used to estimate a trajectory that summarizes whether and how each individual is changing over time. Examining the trajectory allows researchers to describe precisely the nature of change in a sample. Second, the parameters of change estimated in the first step are treated as dependent variables and can be predicted from other between- or within-subject variables. These procedures, which are described in more detail later in this article, were used to describe the course of attachment change over time and to examine what factors predicted such change.

in their attachment representations. This is not a true test of the social-cognitive model as we did not use an experimental methodology or cognitive outcome variables of the kind used by Baldwin and colleagues (e.g., Baldwin et al., 1996). However, we believe that our test provides information about a core assumption of the social-cognitive model: that attachment representations reflect a person's present state of mind about relationships. The social-cognitive model would thus suggest that, within a specific relationship, a person's attachment representations should vary depending on whatever specific relationship information is accessible at present. Marital satisfaction is a good indicator of relationship information that is accessible at present, and attachment representations should thus change in response to it. Of course, there is evidence that attachment representations and marital satisfaction covary concurrently (e.g., Davila. Bradbury, & Fincham. 1998; Feeney et al.. 1994; Fuller & Fincham, 1995). We were interested, instead, in how ongoing changes in marital satisfaction affected ongoing changes in attachment representations over time. Hence we tested a modified social-cognitive model in order to test the alternative models within a nonexpert mental longitudinal study.

Third, we designed the study to examine hypotheses from each of the three models of attachment change. This allowed not only for a test of competing hypotheses but also for an examination of how processes from the three models might work together, as the models may not be mutually exclusive. As a test of the contextual model, we did two things. First, at the descriptive level, we examined whether attachment representations would be a reflection of marital circumstances. In this case, the marriage itself is considered to be the salient contextual factor as it has the potential to be an emotionally significant, long-lasting event—the criteria laid out by N. L. Collins and Read (1994) for the types of events that might lead to attachment change. Two patterns of change are likely. First, attachment representations might conform to general trends consistent with the course of marital satisfaction. Specifically, just as marital satisfaction among newlyweds begins very high and then wanes (e.g., Karney & Bradbury, 1997), attachment security might begin very high and wane somewhat. On the other hand, a general trend toward security might be expected as spouses receive evidence of the stability of their marriages. Spouses are likely to be transferring attachment functions to a person with whom they are in close, consistent contact, and such contact might provide increased security in the early years of marriage. We evaluated both possibilities.

As a test of the individual-difference model, we examined whether changes in attachment representations were predicted by stable vulnerability factors such as parental divorce, history of personal or family psychopathology. and personality disturbance (Davila, Burge, & Hammen, 1997). These factors may disrupt the formation of coherent models of self and others and thus result in attachment instability overtime. Thus, high levels of vulnerability should predict high levels of attachment instability (e.g.. high levels of variability in attachment ratings over time). Moreover, stable vulnerability factors should be better predictors of attachment instability than are the contextual or social-cognitive variables. We also tested a diathesis-stress model in which stable vulnerability factors may put people at increased risk for attachment change in response to marital circumstances. The diathesis-stress model represents one integration of the individual-difference model and the contextual and social-cognitive models of attachment change. Although prior research indicates that stable vulnerability factors are better predictors of attachment change than are contextual factors (Davila. Burge, & Hammen, 1997), research has not examined the proximal circumstances (e.g., contextual factors) that lead vulnerable people to change their attachment models. One hypothesis is that contextual factors are more likely to lead to attachment change among people who possess stable vulnerabilities. Stable vulnerability factors would thus render a person more reactive to contextual stimuli and, hence, more likely to report attachment changes. This hypothesis is consistent with a diathesisstress approach, which suggests that some people possess a vulnerability factor which, when coupled with a proximal stressor, results in a negative outcome. For example, in the depression literature, a sociotropic personality style (e.g., a personality style reflective of interpersonal sensitivity and dependence) is considered a diathesis that, when coupled with interpersonal life stressors, increases risk for depression (e.g., Beck, 1983). In the present conceptualization, stable vulnerability factors are seen as the diathesis that puts people at risk for attachment instability in the face of changing life circumstances. Therefore, the people who will be most likely to change their attachment representations in response

These possibilities refer to the relative stability of attachment representations in the sample as a whole. We also examined contextual factors accounting for individual differences in the stability of attachment representations at a within-subject level. In addition to marriage being an event in itself, some of the most salient contextual factors within a marriage are likely to be qualities of, behaviors of, and experiences with one's partner. As an attempt to capture such contextual factors, we examined whether spouses' attachment models would change as a result of their partners' marital satisfaction and attachment models. Hence, at the predictive level, we examined whether partners' lower levels of satisfaction and security would be associated with greater decreases in spouses' security over time. As a test of the social-cognitive model, we examined whether spouses' own levels of marital satisfaction would predict changes

ATTACHMENT AND MARRIAGE to contextual factors are those who possess the stable vulnerability factors. Finally, we examined the extent to which attachment representations and marital circumstances affect one another over time, in order to begin elucidating reciprocal processes of accommodation and assimilation. A reciprocal perspective on accommodation and assimilation suggests that there is a dynamic interplay among people's attachment representations and their ongoing life circumstances. It accounts for the process by which people construct their lives to confirm their beliefs but leaves open the possibility that people can change their beliefs to be consistent with new life circumstances. Thus, attachment representations and marital circumstances should be associated with one another over time and should predict changes in one another as well. We used a dimensional assessment of attachment in the present study (the Revised Adult Attachment Scale; N. L. Collins & Read, 1990). Thus, stability of attachment representations was assessed rather than stability of attachment style or category. Because Davila, Burge, and Hammen's (1997) research used a categorical assessment of attachment style, we chose a dimensional assessment for the present study to determine if results regarding the individual-difference model generalize across measures. We chose a dimensional approach for three additional reasons. First, many researchers believe that dimensional measures provide a more psychometrically sound assessment of attachment than do categorical measures (e.g.. Scharfe & Bartholomew. 1994). Second, a dimensional assessment provides information on specific aspects of working models that may change. For example, people may be more likely to fluctuate in their beliefs and feelings about selfworth and abandonment rather than in their beliefs about comfort with intimacy. Third, we may be likely to detect more subtle fluctuations in working models with a dimensional measure, because it is more sensitive than a categorical model. It is important to note that this choice has implications for the types of conclusions that can be drawn about attachment change. Conclusions about large-scale changes in attachment style cannot be drawn using the methods used in this study. Instead, this study focuses on ongoing changes in attachment representations, which provide information about the waxing and waning of beliefs about security in response to intrapersonal and interpersonal factors.

Method

Participants Participants were 172 newlywed couples in first marriages recruited from marriage licenses filed in Los Angeles County. To be eligible to participate, both spouses had to be over 18 years of age, have at least a tenth-grade education, speak English, have no children, and have no immediate plans to move from the area. Marriage licenses in Los Angeles County include both spouses' address, age, years of education, and number of previous marriages. Couples who were eligible on the basis of this information were sent letters inviting them to participate in a longitudinal study on newlywed marriage. Of the 3,606 letters that were sent. 637 couples (17.8%) expressed interest in participating, 41 letters were returned as undeliverable (1.1%), and 2,928 letters (81.2%) went unanswered. The 17.8% response rate is similar to that of other studies in which married couples were recruited from public records (e.g., 18% by Kurdek, 1991; 17% by Spanier, 1976). Compared to the 2,928 nonrespondents. the 637 respondents were more likely to have cohabitated premaritally (42.9% vs. 35.3%). had more years of education (for wives, 15.4 years vs. 14.5

787

years; for husbands, 15.2 years vs. 14.6 years), and the wives were slightly older (26.6 years vs. 26.2 years). As might be expected, respondents also had higher status jobs (see Karney et al., 1995). The couples who expressed interest in participating were screened further with a telephone interview to ensure that the remaining eligibility criteria were met. The first 172 couples who met the eligibility criteria and who kept their laboratory appointment formed the sample. All couples were married less than 6 months when they began participation in the study. Wives averaged 26.0 years of age (SD = 3.4) and 16.2 years of education (SD = 2.0), and their median annual income ranged from S 11,000 to $20,000. Sixty-one percent were Caucasian. 15% were Asian American/Pacific Islander, 5% were African American, 16% were Latina/ Chicana, 2% were Middle Eastern, and 1% identified themselves as "other." Husbands averaged 27.6 years of age (SD = 3.9) and 15.6 years of education (SD = 2.2). and their median annual income ranged from $21,000 to 530,000. Sixty-seven percent were Caucasian, 13% were Asian American/Pacific Islander. 4% were African American, 15% were Latino/ Chicano, and 1% were Middle Eastern.

Procedure Spouses participated in an initial in-person laboratory session at the University of California. Los Angeles (Time 1 [Tl ]) and four follow-ups (Time 2 [T2]-Time 5 [T5]). each 6 months apart. The Time 3 (T3) follow-up was also an in-person laboratory session. The T2. Time 4 (T4). and T5 follow-ups involved completing questionnaires by mail. For the laboratory sessions (Tl and T3), spouses completed measures of romantic attachment representations at home prior to the laboratory session during which the measure of marital adjustment was completed. Independent completion of home materials was stressed in the instructions to the questionnaires, in a letter that was sent to each couple, and in a telephone call to«aeh couple. Couples were paid $75 for their participation. For the T2, T4. and T5 follow-ups, spouses completed attachment and marital satisfaction questionnaires at home and mailed them back to the investigators. Independent completion of questionnaires was again stressed in a phone call and in written instructions. Couples were paid $25 for each follow-up. Questionnaires assessing stable vulnerability factors were completed at home at Tl, and interviews assessing stable vulnerabilities were completed during the Tl laboratory session.

Measures Marital satisfaction. The Marital Adjustment Test (MAT; Locke & Wallace. 1959) was used to assess marital satisfaction. The MAT is a widely used 15-item instrument that has been shown to have adequate reliability and to discriminate between nondistressed spouses and spouses with documented marital problems (Locke & Wallace, 1959). Adult romantic attachment. The I8-item Revised Adult Attachment Scale (N. L. Collins & Read, 1990) was used to assess dimensions of adult romantic attachment. This scale includes three subscales of adult attachment: Close, which measures the extent to which people feel comfortable being close to others; Depend, which measures the extent to which people are comfortable relying on others and believe that others are dependable; and Anxiety, which assesses fears about abandonment and of being unloved. The Revised Adult Attachment Scale has been shown to have adequate reliability and validity (e.g., N. L. Collins & Read. 1990). For wives and husbands, respectively, average internal consistencies (alphas) over the five time points in the present study were: Close = .82, .79; Depend = .82, .83; Anxiety = .86, .82. Stable vulnerability factors. Following Davila, Burge, and Hammen (1997), we assessed four stable vulnerability factors: personality disturbance, past history of psychopathology, family history of psychopathology, and family status (intact vs. nonintact). Personality disturbance was measured through a composite variable

788

DAVILA. KARNEY. AND BRADBURY

composed of questionnaire assessments of neuroticism, anger—hostility, impulsivity. and borderline personality disorder. Neuroticism was assessed with the Neuroticism scale of the Eysenck Personality Questionnaire (EPQN: Eysenck & Eysenck. 1978) and the Neuroticism scale of the NEO-Five Factor Inventory. Form S (NEO-N: Costa & McCrac. 1992). Both instruments arc widely used and assess the extent to which participants experience and express a range of negative affect (e.g.. tension, anxiety, sadness). The EPQN is a 23-item true-false measure that has adequate internal consistency and test-retest reliability (Eysenck & Eysenck. 1978). Coefficient alpha was .83 for wives and .86 for husbands in the present sample. The NEO-N is a 12-item scale on which participants rate the extent to which they agree with each statement on a 5-point scale ranging from strongly disagree (I) to strongly agree (5). Extensive reliability and validity data exist for the NEO-N (sec Costa & McCrae. 1992). Coefficient alpha in the present sample was .77 for wives and .75 for husbands. Tendency toward anger-hostility was assessed with a shortened version of the Multidimensional Anger Inventory (MAI: Siegel. 1986). A total subset of 25 items from the Frequency. Duration. Magnitude. Anger-In. Anger-Out. Hostile Outlook, and Range of Anger-Eliciting Situations subscales of the MAI was administered. Coefficient alpha in the present sample was .86 for wives and .86 for husbands. Impulsivity was assessed with Diekman's (1990) 12-item Dysfunctional Impulsivity Measure, which assesses nonoptimal patterns of acting without foresight. Coefficient alpha in the present sample was .82 for wives and .88 for husbands. Symptoms of borderline personality disorder were assessed with the Borderline Personality Disorder subscale of the Personality Diagnostic Questionnaire-Revised (PDQ-R: Hyler & Rieder. 1987). The subscale is a 12-item true-false self-report questionnaire with items that are keyed to the diagnostic criteria tor borderline personality disorder in the Diagnostic and Statistical Manual of Mental Disorders {\DSM-II1-R\\ American Psychiatric Association. 1987). (Note that relatively few changes have been made to the diagnostic criteria for personality disorders from DSM-III-R through DSM-IV [American Psychiatric Association. 1994].) Participants were instructed to complete the items on the basis of how they had felt "in the past several years." Examples of items include: "I feel empty and bored much of the time." "People that I have worshiped often ended up disappointing me." and "I've had more than my share of identity crises." Diagnoses derived from scores on this scale show moderate correspondence with those arrived at through semistructured interviews (Hyler. Skodol. KeUman. Oldham. & Rosnick. 1990). Scores on the scale are also correlated with scores on other self-report measures of borderline personality disorder in a nonclinical sample (Trull. 1995). Continuous scores, not diagnoses, were used in the present study. We converted the scores on the five personality disturbance measures to r scores and summed them to form a composite score of personality disturbance. Higher scores are consistent with higher levels of personality disturbance. Coefficient alpha in the present study was .72 for husbands and .82 for wives. Past history of psychopathology was assessed with the Structured Clinical Interview for DSM-IV (SCID: Spit/er, Williams. Gibbon, & First. 1994). The SCID has acceptable interrater and test-retest reliability that are comparable to those of other widely used structured clinical interviews (Spit/.er et al., 1994). Information about past history of psychopathology was gathered during the initial interview. For every disorder, the diagnostic data were converted into a 3-point continuous scale: 0 = no symptoms, 1 = some symptoms, 2 = DSM-IV diagnosable disorder. All ratings for all past disorders were summed. For example, if a participant had diagnosable depression and some symptoms of substance abuse in his or her past, he or she would receive a rating of 3 (2 + 1 = 3). Family history of psychopathology was assessed with the Family History—Research Diagnostic Criteria (FH-RDC: Endicott. Andreasen. & Spit/.er. 1975). The FH-RDC is a semistructured interview that assesses

symptomatology and level of impairment in family members. It has been shown to have adequate reliability (Andreasen. Endicott, Spitzer, & Winokur. 1977). and it is considered a conservative assessment of family psychopathology (Endicott et al.. 1975). Interviewers indicated the presence of significant symptomatology and whether the family member met criteria for a diagnosis. A total score of family psychopathology was computed by summing the number of family members who had at least symptoms of one disorder. We assessed family status by asking spouses whether their families were intact (i.e.. their parents were married) or nonintact (i.e., their parents were separated or divorced). Twenty-four percent of wives and 22% of husbands came from nonintact families. To simplify data analysis, we converted the scores on the four stable vulnerability factors to ; scores (as necessary) and summed them to form an overall stable vulnerability factor.

Data Analysis We conducted the primary tests of the contextual, social-cognitive, and diathesis-stress models of attachment change as GCAs using hierarchical linear modeling (HLM; Bryk & Raudenbush, 1992) and the HLM/2L computer program (Bryk. Raudenbush, & Congdon. 1994). HLM is advantageous in that all available data for each participant are used to estimate the trajectory. Hence, participants who do not have data at every time point can be included in the analyses. In the present study. 163 of the 172 couples had data at at least three time points and could be included in the analyses.' In addition, HLM computes effects on each parameter of the trajectory through simultaneous equations. As such, effects on one parameter of change are estimated controlling for effects on other parameters of change. In all of the HLM analyses, we estimated husbands' and wives' trajectories simultaneously in a couple-level model, according to procedures-outlined by Raudenbush. Brennan. and Barnett (1995). We conducted the main test of the individual-difference model using multiple regression procedures and will present it after the growth curve analyses. We did this because GCAs do not provide information about whether people with stable vulnerabilities are likely to be more variable on attachment representations over time, which is the central prediction of the individual-difference model. For example. GCA provides information about whether people differ on rate of change over time according to level of vulnerability (e.g.. whether less vulnerable people show more rapid systematic changes in anxiety about abandonment). This analysis, because it can be conducted only at the between-subjects level, does not tell us. as the within-subject analyses do, whether vulnerability makes people more likely to deviate from their trajectories—a question more central to the individual-difference model. Therefore, we conducted an additional set of analyses to examine whether vulnerability was associated with variability on attachment representations over time.

Results Descriptive Statistics and Preliminary Analyses The means and standard deviations for all variables, at all time points, for husbands and wives, are shown in Table 1. On average, husbands became significantly less anxious about abandonment 1 Of the 172 couples participating at Tl. 135 completed all follow-ups through T5. Of the 37 who did not complete all follow-ups. 9 divorced, 5 dropped out, and 23 were missing data from two or more assessments. The 9 couples not included in the present analyses are part of the group of 23 with missing data. Couples who completed all follow-ups (n = 135) did not differ from the other 37 couples on attachment, marital satisfaction, or stable vulnerability variables at Tl.

789

ATTACHMENT AND MARRIAGE over time, r(l 15) = 2.89, p = .005, as did wives. r(121) = 3.03. p = .003. Consistent with prior research (e.g., Karney & Bradbury. 1997), husbands. r(12I) = 3.77, p < .001. and wives. r(122) = 4.56, p < .001, also became significantly less maritally satisfied over time, on average. Husbands and wives differed significantly at Tl through T4 on the Close. Depend, and Anxiety subscales of the Revised Adult Attachment Scale, and on marital satisfaction, with wives scoring higher on all variables at all time points (p < .05). However, at T5, husbands and wives differed only on comfort with closeness, with wives scoring higher (p < .05). Husbands and wives differed significantly on four of the Tl stable vulnerability factors. Wives scored higher on the two measures of neuroticism and on past symptomatology. Husbands scored higher on impulsivity (p < .001). The correlations between all variables in the analyses are shown in Tables 2. 3. and 4. Table 2 shows that attachment

Table 1 Means and Standard Deviations of All Variables in the Analyses Husbands

Wives

Variable

M

SI)

M

SI)

Time 1 (Tl) Close Time 2 (T2) Close Time 3 (T3) Close Time 4 (T4) Close Time 5 (T5) Close Tl Depend T2 Depend T3 Depend T4 Depend T5 Depend Tl Anxiety T2 Anxiety T3 Anxiety T4 Anxiety T5 Anxiety Tl Satisfaction T2 Satisfaction T3 Satisfaction T4 Satisfaction T5 Satisfaction Tl EPQN Tl NEO-N Tl Anger Tl Impulsivity Tl BPD Tl Family history Tl Past symptoms

21.86 21.96 21.89 22.00 21.75 19.21 19.48 19.68 20.21 19.79 13.36 12.88 12.47 12.21 12.26 126.22 122.29 122.45 121.56 121.29 6.63 29.18 62.40 14.85 21.60 2.55 1.73

4.59 4.69 4.35 4.14 4.62 4.50 4.70 4.69 5.20 5.27 4.79 4.17 4.60 4.30 4.86 17.56 20.48 22.61 19.91 20.74 4.93 6.04 13.05 7.34 2.68 2.61 2.31

23.10 23.00 23.69 23.47 23.30 20.35 20.75 20.95 21.23 20.85 13.45 13.03 12.67 12.48 12.43 130.04 126.16 125.97 126.22 124.02 9.28 31.69 62.40 12.56 21.59 2.84 2 33

4.69 4.69 4.42 4.83 5.03 4.64 4.87 5.07 5.33 5.24 5.17 4.96 5.39 5.07 5.27 16.24 17.96 19.18 17.31 18.55 4.84 6.56 13.98 5.90 2.85 2.75 2.55

Note. N ranges from 172 for Tl to 117 for T5. Close = comfort-withcloseness dimension of the Revised Adult Attachment Scale: Depend = comfort-depending-on-others dimension of the Revised Adult Attachment Scale; Anxiety = anxiety-about-abandonment dimension of the Revised Adult Attachment Scale; Satisfaction = marital satisfaction on the Marital Adjustment Test; EPQN = Eysenck Personality Questionnaire. Neuroticism subscale: NEO-N — NEO Five-Factor Inventory. Neuroticism subscale; Anger = Multidimensional Anger Inventory; Impulsivity = Dysfunctional Impulsivity Measure; BPD = Borderline Personality Disorder subscale of the Personality Diagnostic Questionnaire—Revised; Family history = family psychopathology assessed with the Family History— Research Diagnostic Criteria interview; Past symptoms = past history of psychopathology assessed with the Structured Clinical Interview for DSM-IV (Diagnostic and Statistical Manual of Mental Disorders. 4th ed.).

Table 2 Attachment Stability Correlations for Husbands and Wives Time point

Time 1 (Tl)

Time 2 (T2)

.72 .71 .72 .66

.62 — .77 .69 .72

Time 4 (T4)

Time 5 (T5)

.69 .68 — .76 .70

.66 .63 .77

.63 .69

.71 .66

.58 .68 .79

Time 3 (T3) Close

Tl

T2 T3 T4

T5

.75 .81

.80

Depend Tl T2 T3 T4 T5

.67 .63 .66 .63

.68

.69 .71 .67

.72 .71

.58 .70 .70 .74

.72

Anxiety Tl T2 T3 T4 T5

.74 .69

.72 .63 .75

.70 .62 .66

.73 .64

.70 .68

.50 .60 .64

.61 .67 .68 .73

.69

Note. N ranges from 172 for Tl to 105 for T5. Husbands' data are above the diagonal. Wives' data are below the diagonal. All correlations are significant (/) < .001. one-tailed). Close = comfort-with-closeness dimension of the Revised Adult Attachment Scale; Depend = comfortdepending-on-others dimension of the Revised Adult Attachment Scale; Anxiety = anxiety-about-abandonment dimension of the Revised Adult Attachment Scale.

representations had high levels of relative stability over time, with all correlations in the range of .58-.81. This is consistent with past research from samples of married couples (e.g.. Feeney et a!.. 1994; Fuller & Fincham, 1995). Table 3 shows that marital satisfaction also had high levels of relative stability over time, with correlations ranging from .43 to .87. Table 4 shows that, consistent with prior research (e.g.. Davila et al., 1998; Feeney et al.. 1994; Fuller & Fincham. 1995), attachment representations were consistently correlated with marital satisfaction cross-sectionally. Table 4 also shows that for husbands and wives, depend and anxiety were correlated with the composite stable vulnerability factor at all time points. Close was correlated with the composite stable vulnerability factor at T2, T4, and T5 for husbands and Tl, T2, and T3 for wives. Table 3 shows that husbands' marital satisfaction was significantly associated with the composite stable vulnerability factor at Tl and T2. Wives' marital satisfaction was significantly associated with the composite stable vulnerability factor at all time points. In summary, all expected associations were present, rendering subsequent analyses plausible.

Growth Curve Analyses: Do Attachment Representations Change Systematically Over the Early Years of Marriage? Before predicting attachment change, we first sought to describe the course of attachment over time to determine whether attach-

790

DAVILA, KARNEY, AND BRADBURY Table 3

Correlations Between Marital Satisfaction Variables and the Stable Vulnerability Composite Variable for Husbands and Wives 1

2

3

4

5

6

_ .59 .55 .43 .57 -.28

.66 — .63 .47 .62 — 22

.55 .55 — .65 .64 -.35

.59 .66 .66 — .72 -.19

.60 .67 .60 .87 — -.28

-.16 -.20 -.09" -.13a -.13" —

Variable 1. 2. 3. 4. 5. 6.

Time 1 Satisfaction Time 2 Satisfaction Time 3 Satisfaction Time 4 Satisfaction Time 5 Satisfaction Stable vulnerability

Note. N ranges from 172 for Time 1 to 103 for Time 5. Satisfaction = marital satisfaction on the Marital Adjustment Test. Husbands' data are above the diagonal. Wives' data are below the diagonal. Correlations > . 15. p < .05, one-tailed; correlations > .20. p < .01. one-tailed; correlations > .28. p < .001. one-tailed. '' ns.

merit representations changed systematically during the early years of marriage. One variant of the contextual mode! of attachment change suggests a linear model of change in that attachment security either decreases as marital satisfaction decreases, or increases as a function of consistent contact with partners and as spouses gain evidence of marital stability. Hence, this view predicts systematic changes in attachment representations over the period of time following marriage. Empirically, the preliminary analyses suggested that, at least for anxiety about abandonment, a linear decline might capture the appropriate trajectory. Hence, we tested a linear model of change over time. This model summarizes the trajectory of change with two parameters: the intercept, which captures the initial level at which the trajectory begins, and the slope, which captures the subsequent rate of change in the trajectory over time. An alternative perspective predicts no systematic changes in attachment representations over the early years of

marriage. To test this perspective, we estimated a mean-andvariance model. This model describes the trajectory of change in terms of the mean level across time and deviations from the mean level at each assessment. The model suggests that multiple assessments of attachment should fluctuate around an individual's mean level. That is, everyone has a baseline level around which they fluctuate. Hence, this model is potentially consistent with the contextual, social-cognitive, individual-difference, or diathesisstress models. The linear model and the mean-and-variance model were tested for each of the three attachment representations, resulting in six analyses. There are two steps in evaluating the relative fit of the two models (linear vs. mean and variance). First, the parameter estimates are examined to determine whether each model provides meaningful estimates and whether any parameters can be dropped. For example, when testing a linear model, if a slope coefficient is

Table 4 Correlations Between Attachment Variables, Marital Satisfaction, and the Stable Vulnerability Composite Variable for Husbands and Wives Husbands Variable Tl T2 T3 T4 T5 Tl T2 T3 T4 T5 Tl T2 T3 T4 T5

Close Close Close Close Close Depend Depend Depend Depend Depend Anxiety Anxiety Anxiety Anxiety Anxiety

Vuln. -.08" -.17 -.10" -.16 -.22 -.32 -.41 -.34 -.43 -.37

.38 .37 .38 .32 .32

Tl Sat. .21 .22 .21 .27 .25 .25 .27 .25

.28 .19 -.27 -.30 -.33 -.46 -.24

T2 Sat. .10"

.26 .21 22 .24

.29 .32 .30 .33 .25 -.18 -.30 -.24 -.25 -.20

Wives

T3

T4

Sat.

Sat.

.17 .10" .14 .17 .09" .19 .22 .26 .22 .11" -.17 -.21 -.20 -.25 -.08"

.10" .12" .14" .16 .14" .16 .27 .18 .23 .13" -.14" -.14" -.16 -.26 -.09"

T5 Sat. .13" .13" .16" .15" .18 .20 .29 .21 .26 .19 -.21 -.19 -.21 -.31

Tl

Vuln. -.17 -.16 -.16

-.08" -.05"

-.31 -.26 -.32 -.30 -.24 .31 .27 .39 .30 .23

Sat.

.26

.19 .25 .28 .14"

.34 .22 .22 .23

.16 -.29 -.19 -.29 -.38 -.26

T2 Sat. .22 .16

.18 .25 .16 .34 .25 .26 .21 .22 -.31 -.27

-.29 -.41 -.26

T3 Sat. .20 .11" .22 .26 .18 .41 .34 .34 .30 .33 -.33 -.23 -.37

-.39 -.29

T4 Sat. .15

.17 .11" .18 .21 .31 .27 .17 .24 .27 -.23

-.19 -.18 -.32 -.19

T5 Sat.

.06" .06" .07" .15 .07" .25 .21 .22 .18 .26 -.25 -.23 -.24 -.33 -.30

Note. N ranges from 172 for Time 1 (Tl) to 110 for Time 5 (T5). Correlations > .14, p < .05, one-tailed; correlations > .20, p < .01, one-tailed; correlations > .30, p < .001. one-tailed. Vuln. = stable vulnerability composite variable; Sat. = marital satisfaction on the Marital Adjustment Test; T2 = Time 2; T3 = Time 3; T4 = Time 4; Close = comfort-with-closeness dimension of the Revised Adult Attachment Scale; Depend = comfort-dependingon-others dimension of the Revised Adult Attachment Scale; Anxiety = anxiety-about-abandonment dimension of the Revised Adult Attachment Scale. " ns.

791

ATTACHMENT AND MARRIAGE not different from zero, then that parameter could be dropped because it suggests that systematic linear changes are not occurring. Second, the deviance statistics of the two models are compared to determine whether the models provide an equally adequate fit to the data. Because the linear models and mean-andvariance models are nested, their fit can be compared to determine whether they differ significantly from one another by subtracting the deviance statistics of the nested model (the mean-and-variance model) from that of the larger model (the linear model). If the resulting difference, distributed as chi-square, is significant, the model with the fewest parameter constraints can be said to provide a better description of the data. If, for example, a linear model resulted in slope estimates significantly different from zero, and if the deviance statistics from the linear and mean-and-variance models were significantly different, then the linear model could be considered the better fitting model. If a linear model resulted in slope estimates that were not significantly different from zero, and if the deviance statistics from the linear and mean-and-variance models were not significantly different, then the mean-andvariance model could be considered the best-fitting model. Models for comfort with closeness. Results for this and all models are shown in Table 5. The linear model for comfort with closeness did not fit the data well. First, although the model produced reliable estimates of the intercept (reliability = .68 for husbands and .69 for wives), the slope estimates captured less than one third of the true variance in spouse's change scores (reliability = .31 for husbands and .29 for wives). Second, neither the husbands' nor wives' mean slope estimates were significantly different from zero, indicating that, on average, comfort with closeness was not changing linearly over time. However, the individual variability in husbands' and wives' slope estimates was significant, indicating that some spouses were experiencing changes in their comfort with closeness over time. This suggests that a mean-and-variance model may better approximate the data, and comparisons of the deviance values across the two models supported this. The resulting difference value (12.48, df= 7) was not significant. Hence, a mean-and-variance model was selected as the best model to describe change in representations of comfort with closeness over time. The mean-and-variance model also indicated that there was significant between-subjects variability in mean levels of comfort with closeness: for husbands, M = 22.01, SD = 3.60, x2 test of variance (171, N = 172) = 1672.19, p < .000; for wives, M = 23.33, SD = 4.07, )f test of variance (171, N = 172) = 2006.06, p < .000. Models for comfort depending on others. The linear model for comfort depending on others provided an adequate fit to the data. Reliable estimates of the intercept (reliability = .57 for husbands and .58 for wives) resulted, although the slope estimates captured less than one third of the true variance in spouse's change scores (reliability = .27 for husbands and .19 for wives). However, slope coefficients were positive and significantly different from zero for husbands and marginally significant for wives, suggesting that comfort depending on others shows at least a tendency to increase linearly over time. On average, spouses became more comfortable depending on others as their marriage progressed. There was also significant between-subjects variability in slopes over time for both husbands and wives. The mean-and-variance model also indicated that there was significant between-subjects variability in mean levels. However, the difference in the deviance statistics

Table 5 Mean Trajectories of Attachment Representations and Marital Satisfaction Across 2 Years Variable

M

SE

f

X1 test of variance

Intercepts Close Husbands Wives Depend Husbands Wives Anxiety Husbands Wives Satisfaction Husbands Wives

21.95 23.07

0.37 0.37

— —

569.17*** 515.16***

19.18 20.42

0.35 0.36

—

379.88*** 408.61***

13.52 13.65

0.39 0.41

—

482.02*** 545.74***

127.45 131.66

1.46 1.31

— —

450.31*** 363.53***

Slopes Close Husbands Wives Depend Husbands Wives Anxiety Husbands Wives Satisfaction Husbands Wives

0.02 0.10

0.08 0.08

0.31 1.25

246.55*** 235.48***

0.18 0.14

0.09 0.08

2.04* I.69t

213.35** 212.01**

-0.31 -0.33

0.09 0.08

-3.65*** -4.17***

214.78** 188.87t

-1.74 -2.11

0.38 0.40

-4.53*** -5.29***

281.76*** 325.12***

Note. N = 163 couples; df = 162. Close = comfort-with-closeness dimension of the Revised Adult Attachment Scale; Depend = comfortdepending-on-others dimension of the Revised Adult Attachment Scale; Anxiety = anxiety-about-abandonment dimension of the Revised Adult Attachment Scale; Satisfaction = marital satisfaction on the Marital Adjustment Test. a The / test addresses the hypothesis that a given parameter differs significantly from zero. Because the lowest possible scores on the measures is greater than zero, these tests are not meaningful and hence are not reported. ip < .10 (marginally significant). * p < .05. ** p < .01. *** p < .001.

(25.98, df=l) for the two models was significant, suggesting that the models do not describe the data equally well and that the model with the fewest parameter constraints (the linear model) provides a better fit to the data. Because of this and the linear trends in the data, we estimated linear changes in comfort depending on others in all subsequent models. Models for anxiety about abandonment. The linear model for anxiety about abandonment provided an adequate fit to the data. Reliable estimates of the intercept (reliability = .65 for husbands and .69 for wives) resulted, although the slope estimates captured less than one third of the true variance in spouse's change scores (reliability = .27 for husbands and .14 for wives). However, slope coefficients were negative and significantly different from zero for husbands and wives, suggesting that anxiety about abandonment shows a tendency to decrease linearly over time. On average, spouses became less anxious about abandonment as their marriage progressed. There was also significant between-subject variability in slopes over time for husbands and marginal variability for

792

DAVILA. KARNEY. AND BRADBURY

wives. The mean-and-variance model also indicated that there was significant between-subjects variability in mean levels. However, the difference in the deviance statistics (42.00, df = 1) for the two models was significant, suggesting that the models do not describe the data equally well and that the model with the fewest parameter constraints (the linear model) provides a better fit to the data. Because of this and the linear trends in the data, we estimated linear changes in anxiety about abandonment in all subsequent models. Models for marital satisfaction. Although the focus of this study was not on change in marital satisfaction, we estimated its trajectory in order to use this information in subsequent analyses in which changes in attachment are predicted from changes in marital satisfaction and in subsequent analyses exploring reciprocal processes. There is growing evidence to suggest that marital satisfaction follows a linear decline over time (e.g., Karney & Bradbury, 1997). Hence, we estimated a linear model, which resulted in a good fit to the data. Reliable estimates of the intercept (reliability = .63 for husbands and .55 for wives) and slope (reliability = .41 for husbands and .46 for wives) resulted. In addition, slope coefficients were negative and significantly different from zero for husbands and wives, suggesting that marital satisfaction declined linearly over time. On average, spouses became less satisfied as their marriage progressed. There was also significant betweensubjects variability in slopes over time for husbands and wives. The mean-and-variance model also indicated that there was significant between-subjects variability in mean levels. However, the difference in the deviance statistics for the two models (132.45, df = 7) was significant, suggesting that the models do not describe the data equally well and that the model with the fewest parameter constraints (the linear model) provides a better fit to the data. Because of this and the linear trends in the data, we estimated linear changes in marital satisfaction in all subsequent models.

Growth Curve Analxses: What Accounts for Attachment Change Over Time? We tested the four theoretical models (contextual, socialcognitive, individual-difference, and diathesis-stress) by estimating separate models for each of the three attachment representations. In these analyses we examined the effect of both within- and between-subjects variables on attachment trajectories. At the within-subject level, the model examines whether changes in one variable are associated with deviations from the trajectory of another variable. Specifically, we examined whether changes in own marital satisfaction, partner marital satisfaction, and partner attachment representations (which all change over time) were associated with deviations from the trajectory of own attachment representations. These analyses involve entering the within-subject variables into the within-subject models as time-varying covariates (see Karney. in press). The resulting coefficient of the timevarying covariate estimates the extent to which changes in, for example, marital satisfaction between assessments affects attachment representations at each assessment, controlling for the overall trajectory of attachment representations over time. Thus, this analysis tells us whether changes in marital satisfaction make people deviate from their trajectories of attachment representations. We tested the contextual and social-cognitive models in this manner.

At the between-subjects level, we examined whether a variable measured at Tl predicts the trajectory of another variable that changes over time (i.e., whether different groups of people show different trajectories over time). Specifically, we examined whether people who have different levels of stable vulnerabilities also have different trajectories of attachment representations. Furthermore, GCA allows for examination of between-subjects differences in the strength of the within-subject associations measured by the time-varying covariates. The analysis provides information about whether a between-subjects variable moderates the within-subject association between the other two variables (see Karney, in press). This allowed us to examine whether stable vulnerabilities measured at Tl moderated the association between the trajectories of attachment representations and the other withinsubject variables (own marital satisfaction, partner marital satisfaction, partner attachment representations). Hence, predictions about the individual-difference and diathesis-stress models were tested in this manner. Finally. HLM provides final parameter estimates for each variable controlling for the other variables in the analysis. Hence, the unique contributions of each variable can be examined- This is useful in the present analyses in order to test the competing hypotheses. Tests of the contextual model of attachment change. To test the contextual model, we examined whether spouses' attachment representations would change as a result of the partners' marital satisfaction and attachment representations. To do so we examined the within-subject associations between these variables as described above. Results are shown in the first two columns of Table "6. Across the three attachment representations, contextual effects were evident, but they were largely in an unexpected direction. Specifically, as spouses' attachment security increased (e.g., high levels of comfort with closeness, high levels of comfort depending on others, low levels of anxiety about abandonment). their partners' attachment security decreased. Changes in husbands' comfort with closeness predicted deviations in wives' levels of comfort with closeness from their mean, controlling for their mean and controlling for Tl level of stable vulnerability. Changes in either spouse's comfort depending on others predicted deviations in their partners' levels of comfort depending on others from their linear trajectory, controlling for their trajectory and controlling for stable vulnerabilities. Changes in either spouse's anxiety about abandonment predicted deviations in their partners' levels of anxiety about abandonment from their linear trajectory, controlling for their trajectory and controlling for stable vulnerabilities. These findings are inconsistent with zero-order correlations between spouses' attachment representations, in which higher levels of security in partners were associated with higher levels of security in spouses cross-sectionally and over time. Assuming these counterintuitive findings do not represent some kind of methodological artifact, they could be taken as evidence for conceptualizing attachment in marriage as a dyadic phenomenon. For example, attachment in couples may follow a homeostatic process in which couples strive (not necessarily consciously) to maintain a couplebased level of security. Increases in one spouse's security might thus result in decreases in the partner's security. One additional contextual effect was evident. Changes in husbands' marital satisfaction predicted deviations in their wives' levels of anxiety about abandonment from their linear trajectory,

793

ATTACHMENT AND MARRIAGE Table 6

Growth Curve Analyses Testing the Contextual Model: Within-Subject Associations Between Marital Satisfaction and Attachment Representations Partner satisfaction --> Spouse attachment Attachment dimension

Coefficient

Partner attachment -H> Spouse attachment

Effect size

7(161)

Close Depend Anxiety

-.01 -.02 -.02

-0.59 -1.22 -1.37

.05 .09

Close Depend Anxiety

-.01 -.01 - .02

-1.36 - 1.22

.10 .09 .17

'(161)

Coefficient

satisfaction

Effect size

Coefficient

.12 .67

.35

.26 .67 .34

Spouse = Husband -.07 -1.59 -.41 -11.47*** -.17 -4.66***

.1 1

Spouse attachment —• • Partner

Spouse -- Wife - .14 -3.53** -.39 -11.30*** -.18 -4.52***

'(161)

Effect size

.40

2.58*

-.12 .37

-0.88 2.46*

.19 .07 .19

.19 .01 .10

1.04 0.06 0.72

.08 .00 .06

Note. N — 163 couples. Satisfaction — marital satisfaction on the Marital Adjustment Test; Close = comfort-with-closeness dimension oi the Revised Adult Attachment Scale; Depend = comfort-depending-on-others dimension of the Revised Adult Attachment Scale; Anxiety = anxiety-aboutabandonment dimension of the Revised Adult Attachment Scale. */> < .05. **/) < .01. ***/; < .001.

satisfaction predicted deviations in their own levels of anxiety about abandonment from their linear trajectory, controlling for their trajectory and controlling for stable vulnerabilities. Tests of the individual-difference model. As one test of the individual-difference model we examined whether spouses with different levels of stable vulnerabilities at Tl showed different rates of change in attachment representations over time, by examining trie between-subjects associations between these variables. Hence, we examined whether stable vulnerabilities at Tl predicted the slope of comfort depending on others and anxiety about abandonment (comfort with closeness did not change linearly). Stable vulnerabilities were not associated with rates of linear change, suggesting that stable vulnerabilities do not contribute to linear changes in attachment over time (for husbands' comfort depending on others, coefficient = —.06. '[16IJ = —1.58. ns: for wives'

controlling for their trajectory and for stable vulnerabilities. As husbands' marital satisfaction decreased, wives' anxiety about abandonment increased. Tests of the social—cognitive model. To test the socialcognitive model, we examined whether spouses' attachment representations would change as a result of their own marital satisfaction, by examining the within-subject associations between these variables. Results are presented in the first column of Table 7. This model was supported largely with respect to changes in anxiety about abandonment and comfort with closeness. As spouses became less satisfied, they also became less secure. Specifically, changes in wives' marital satisfaction predicted deviations in their levels of comfort with closeness from their mean, controlling for their mean and controlling for Tl level of stable vulnerability. For husbands and wives, changes in own marital

Table 7 Growth Curve Analyses Testing the Social—Cognitive Model: Within-Subject Associations Between Marital Satisfaction and Attachment Representations Attachment dimension

Spouse satisfaction —> Spouse attachment

Spouse attac hment —> Spouse satisfaction

Coefficient

Effect size

Coefficient

Spouse = Husband .12 .09 .27

.28 .36 -.44

2.36* -2.90**

.14 .18 .22

.23 .37 -.15

1.53 2.78** -1.12

.12 .21 .09

'(161)

Close Depend Anxiety

.02 .01 -.04

1.60 1.13 -3.63***

Close Depend Anxiety

.02 .01 -.03

2.05* 0.99 -2.96**

'(161)

1.791

Effect size

Spouse = Wife .15 .08 .23

Note. N = 163 couples. Satisfaction = marital satisfaction on the Marital Adjustment Test; Close = comfort-with-closeness dimension of the Revised Adult Attachment Scale; Depend = comfort-depending-onothers dimension of the Revised Adult Attachment Scale; Anxiety = anxiety-about-abandonment dimension of the Revised Adult Attachment Scale. t/> < .10 (marginally significant). */> < .05. * * / ? < . 0 1 . ***/? Spouse attachment Attachment dimension

Coefficient

'(161)

Effect size

Close Depend Anxiety

.00 -.01 .00

0.76 -1.39 0.71

.06 .11 .05

Close Depend Anxiety

.00 .00 -.00

0.12 1.01 -0.76

.01 .08 .06

Partner attachment —J• Spouse attachment Coefficient

'(161)

Spouse attachment —s• Partner satisfaction

Effect size

Coefficient

.04 .02 .07

-.03 -.08 .11

-0.55 2.26*

.04 .14 .17

.18 .03 .01

-.04 .12 .02

-0.57 2.25* 0.41

.04 .17 .03

'(161)

Effect size

Spouse = Husband -.01 .00 .01

-0.49 0.31 0.85

-1.78t

Spouse == Wife .03 -.00 .00

2.35* -0.37 0.19

Note. N = 163 couples. Satisfaction = marital satisfaction on the Marital Adjustment Test; Close = comfort-with-closeness dimension of the Revised Adult Attachment Scale; Depend = comfort-depending-on-others dimension of the Revised Adult Attachment Scale; Anxiety = anxiety-aboutabandonment dimension of the Revised Adult Attachment Scale, t p < .10 (marginally significant). *p < .05.

795

ATTACHMENT AND MARRIAGE

Table 9 Growth Curve Analyses Testing the Diathesis-Stress Model for Social-Cognitive Factors: Effects of Stable Vulnerabilities on the Within-Subject Associations Between Marital Satisfaction and Attachment Representations Spouse satisfaction —» Spouse attachment

Spouse attachment —> Spouse satisfaction

Coefficient

4161)

Effect size

Coefficient

'(161)

Effect size

Close Depend Anxiety

.00 .01 -.01

0.05 2.41* -1.49

Spouse = Husband .00 .19 .12

.06 .10 -.08

1.01 1.83| -1.42

.08 .14 .11

Close Depend Anxiety

.00 -.00 .00

-1.52 -0.32 0.57

Spouse = Wife .12 .03 .04

-.02 -.00 -.00

-0.43 -0.10 -0.09

.03 .01 .01

Attachment dimension

Note. N = 163 couples. Satisfaction = marital satisfaction on the Marital Adjustment Test; Close = comfort-with-closeness dimension of the Revised Adult Attachment Scale; Depend = comfort-depending-onothers dimension of the Revised Adult Attachment Scale; Anxiety = anxiety-about-abandonment dimension of the Revised Adult Attachment Scale. t p < .10 (marginally significant). * p < .05.

wives' satisfaction predicted changes in their own comfort and their anxiety about abandonment, this suggests that husbands' comfort with closeness predicts changes in wives' satisfaction, which in turn predicts changes in wives' comfort with closeness and anxiety about abandonment. Hence, the contextual effects of husbands' comfort with closeness on wives' marital satisfaction may then lead to a social-cognitive process of change in wives' attachment security. Regarding reciprocal associations relevant to the contextual model for comfort depending on others, two associations emerged, and they were moderated by stable vulnerabilities (Table 8, last column). First, husbands' stable vulnerabilities moderated the association between wives' comfort depending on others and husbands' satisfaction. As vulnerability levels increased, the association between comfort depending on others and satisfaction also

increased (became more positive). This suggests that among vulnerable husbands, their marital satisfaction changes in response to their wives' comfort levels, such that as wives become less comfortable depending on others, vulnerable husbands become less satisfied. Coupled with findings from the diathesis-stress analyses of social-cognitive factors predicting comfort depending on others (Table 9, first column), this suggests that, particularly among vulnerable husbands, wives' comfort depending on others predicts changes in husbands' satisfaction, which in turn predicts changes in husbands' comfort depending on others. Also, coupled with findings from the social-cognitive analyses (see Table 7), in which changes in husbands' satisfaction were associated with changes in their anxiety, the results suggest that wives' comfort depending on others predicts changes in husbands' satisfaction, which in turn predicts changes in husbands' anxiety about abandonment. Similar

H Close

H Satisfaction W Satisfaction

H Depend, H Anxiety, W Anxiety W Close, W Anxiety

H Depend

H Satisfaction* W Satisfaction*

H Depend*, H Anxiety, W Anxiety W Close, W Anxiety

H Anxiety

H Satisfaction W Satisfaction*

H Anxiety, H Depend, W Anxiety W Anxiety, W Close

W Depend

W Satisfaction H Satisfaction

W Close, W Anxiety H Depend*, H Anxiety, W Anxiety

Figure 1. Summary of the results of the reciprocal analyses. Results are not presented in the order described in the text. H = husband; W = wife; Close = comfort-with-closeness dimension of the Revised Adult Attachment Scale; Depend = comfort-depending-on-others dimension of the Revised Adult Attachment Scale; Anxiety = anxiety-about-abandonment dimension of the Revised Adult Attachment Scale; Satisfaction = marital satisfaction on the Marital Adjustment Test. * The association between this variable and the preceding variable is moderated by stable vulnerabilities.

796

DAVILA. KARNEY. AND BRADBURY

to the prior analysis, the contextual effects of wives' comfort depending on others on husbands' marital satisfaction may then lead to a social-cognitive process of change in husbands' attachment security, particularly among vulnerable husbands. Finally, coupled with findings from the contextual analyses (see Table 6), in which changes in husbands" satisfaction were associated with changes in wives' anxiety about abandonment, these results suggest that wives' comfort depending on others predicts changes in husbands' satisfaction, which in turn predicts changes in wives' anxiety about abandonment. Second, there was a trend for wives" stable vulnerabilities to moderate the association between husbands' comfort depending on others and wives" marital satisfaction (Table 8, last column). As wives" vulnerability levels increased, the nonsignificant negative association between changes in husbands' comfort levels and changes in wives" satisfaction levels (shown in Table 6, last column) increased (became more negative). Although difficult to interpret, this suggests a trend that among vulnerable wives, changes in their husbands' comfort depending on others are more likely to be negatively correlated with changes in wives' marital satisfaction. Coupled with findings from the social—cognitive analyses (see Table 7), in which changes in wives' satisfaction were associated with changes in their comfort with closeness and anxiety about abandonment, the results suggest that husbands' comfort depending on others predicts changes in wives' satisfaction, which in turn predicts changes in wives' comfort with closeness and anxiety about abandonment. Similar to the prior analysis, the contextual effects of husbands' comfort depending on others on wives' marital satisfaction may then lead to a social-cognitive process of change in wives' attachment security, particularly among vulnerable wives. Regarding reciprocal associations relevant to the contextual model for anxiety about abandonment (Table 6, last column), contrary to expectations, when husbands' anxiety increased, wives' satisfaction increased, controlling for their trajectories of marital satisfaction and their stable vulnerabilities. Moreover, this association was moderated by wives' stable vulnerabilities (Table 8, last column). As vulnerability levels increased, the association between anxiety about abandonment and satisfaction also tended to increase (became more positive). This suggests that among vulnerable wives, their marital satisfaction tends to change in response to their husbands" anxiety levels, such that as husbands become more anxious, vulnerable wives become more satisfied. Coupled with findings from the analyses of the social-cognitive model, in which changes in wives' marital satisfaction predicted changes in wives" comfort with closeness and anxiety about abandonment (Table 7, first column), this suggests that, particularly among vulnerable wives, husbands' anxiety about abandonment predicts changes in wives' satisfaction, which in turn predict changes in wives' comfort with closeness and anxiety about abandonment. Again, the contextual effects of husbands' attachment security on wives' marital satisfaction may then lead to a socialcognitive process of change in wives' attachment security, particularly among vulnerable wives. In this case, wives are happier and, thus, more secure when their husbands exhibit more anxiety about abandonment. Regarding reciprocal associations relevant to the socialcognitive model (i.e., associations between own attachment and own marital satisfaction), across all three attachment dimensions,

as husbands' attachment security decreased (e.g., lower scores on Close and Depend, higher scores on Anxiety), their marital satisfaction decreased, controlling for their trajectories of marital satisfaction and their stable vulnerabilities (this association was marginally significant for comfort with closeness; see Table 7, last column). Moreover, for comfort depending on others, this association was marginally moderated by stable vulnerabilities (Table 9, last column). As vulnerability levels increased, the association between comfort depending on others and satisfaction also tended to increase (became more positive). This suggests that among vulnerable husbands, their marital satisfaction tends to change in response to their own comfort depending on others, such that as they become less comfortable, they also become less satisfied. The findings paralleled those from (a) the diathesis-stress analyses of social cognitive factors (Table 9, first column), in which husbands' vulnerability increased the strength of the association between changes in husbands" satisfaction and changes in their comfort depending on others; (b) the social-cognitive analyses (Table 7, first column), in which changes in husbands' satisfaction were associated with changes in husbands" anxiety about abandonment; and (c) the contextual analyses (Table 6. first column), in which changes in husbands' satisfaction were associated with changes in wives' anxiety about abandonment. Husbands, particularly those who are vulnerable, thus evidence reciprocal associations between changes in their marital satisfaction and changes in their own and their wives' attachment security. In only one circumstance was wives" attachment associated with their own satisfaction (Table 7, last column). As wives' comfort depending on others decreased, their marital satisfaction decreased, controlling for their trajectories of marital satisfaction and their stable vulnerabilities. Coupled with findings from the analyses of the social-cognitive model, in which changes in wives' marital satisfaction predicted changes in wives' comfort with closeness and anxiety about abandonment (Table 7, first column), this suggests that changes in wives' comfort depending on others predict changes in wives' satisfaction, which in turn predict changes in wives' comfort with closeness and anxiety about abandonment. To summarize, numerous reciprocal processes for husbands and wives were evident. A number of these fell within a socialcognitive framework, suggesting that people's current state of mind about marriage and their current state of mind about attachment security are closely linked and predict changes in one another. A number fell within a contextual framework, suggesting that people's experiences of their spouses and themselves are linked and predict changes in one another. The other reciprocal associations suggested an integration of the contextual and socialcognitive frameworks. Contextual factors (partners' attachment security) predicted changes in the spouses' state of mind about marital satisfaction, which then predicted changes in spouses' attachment security. Finally, a number of the reciprocal associations occurred most strongly among vulnerable spouses, supporting a diathesis-stress model of attachment change. However, the nature of the diathesis-stress process appeared different for husbands and wives. Vulnerable husbands became more dissatisfied in response to changes in attachment insecurity, whether the changes be their own or their partners'. However, vulnerable wives became more satisfied in response to their partners' attachment insecurity. It should be noted, however, that gender differences were not

ATTACHMENT AND MARRIAGE

tested statistically; neither were conclusions about reciprocal processes that imply mediation.

A Further Test of the Individual-Difference Model: Attachment Variability Analyses The GCAs provide information about whether stable vulnerabilities are associated with the slopes of attachment representations and whether stable vulnerabilities moderate the association between attachment trajectories and other within-subject variables. The GCAs do not address whether people with stable vulnerabilities are likely to be more variable on attachment representations over time, which is the central prediction of the individualdifference model. For example, an association between TI vulnerability and the slope of attachment representations provides information about whether people differ on rate of linear change according to level of vulnerability (e.g., whether less vulnerable people show more rapid declines in anxiety about abandonment). This between-subjects analysis does not tell us, as the withinsubject analyses do, whether vulnerability makes people more likely to deviate from their trajectories—a question central to the individual-difference model. Thus, we conducted an additional set of analyses to examine whether vulnerability was associated with variability on attachment representations over time. We calculated variability in attachment scores using a procedure similar to that used by researchers to assess the stability of selfesteem (e.g., Kernis, Cornell, Sun, Berry, & Harlow, 1993). We computed the standard deviation of each attachment dimension across the five assessment points for each spouse. Higher scores (i.e., higher standard deviations) indicated more attachment fluctuation (i.e., greater variability) over time. First, we examined the zero-order correlations between the stable vulnerability factor and each attachment variability score. There was one significant correlation: Higher vulnerability was significantly associated with wives' higher variability on anxiety about abandonment (r = .25, p < .01). Second, we conducted a simultaneous regression analysis predicting variability on anxiety about abandonment, controlling for: (a) the intercept and slope values for anxiety about abandonment to control for individual trajectory of anxiety about abandonment and (b) the intercept and slope values for own marital satisfaction, husband marital satisfaction, and husband anxiety about abandonment. We did this to control for factors found to affect anxiety about abandonment in the GCAs. The overall model was significant, R2 = .25, F(9, 154) = 5.54, p < .001, and vulnerability was a significant predictor of variability on anxiety about abandonment, (3 = .16, r(154) = 2.08. p = .04. Wives' and husbands' anxiety about abandonment intercepts and slopes also were significant predictors (p < .05), but wives' and husbands' marital satisfaction slopes and intercepts were not significant predictors. Hence, there was support for the individual-difference model for wives: Wives with higher levels of stable vulnerabilities fluctuated more on anxiety about abandonment.2'3

Discussion We designed the present study to examine attachment change processes in the early years of marriage using advanced methods that allowed for both the description and prediction of change over time. We examined four models of change, along with reciprocal

797

processes among attachment representations and marital variables. Results indicated a number of consistent patterns of change, although variables from no one model emerged as the sole predictors of change. Instead, attachment change followed a process that is more complex than any of the models suggest. These processes are described below.

Summary of Findings Describing systematic changes in attachment representations: Marriage as a contextual change factor. First we addressed whether, on average, attachment representations followed a predictable course of change over the early years of marriage. Given that marriage is likely to be an emotionally significant, long-lasting event, did being newly married serve as a contextual factor leading to systematic attachment changes? Two possibilities were examined. First, as marital satisfaction declines, which, on average, it does (see Karney & Bradbury, 1997), do attachment representations follow? That is, are people's attachment representations in a "honeymoon phase," just like their ratings of marital satisfaction, only to become less ideal over time? Or do people become more secure over time, as some have suggested (see Ha/an & Shaver, 1994)? The present data support the latter hypothesis: Spouses tend to become more secure over time—specifically, more comfortable depending on others and less anxious about abandonment. This may be due to spouses becoming more comfortable in their marriages as their relationships develop, and gaining more evidence that their marriage will last, or to spouses transferring attachment functions to a person with whom they are in close, consistent contact. In either case, spouses may be accommodating these indicators of a safe, stable marriage and therefore becoming more secure over time. Klohnen and John (1998) also found evidence of increased security over time in a 25-year longitudinal study of women—however, the women were not all married.

2 We conducted an additional simultaneous regression analysis predicting wives' variability on anxiety about abandonment from vulnerability, controlling for variability on wives" satisfaction, husbands' satisfaction, and husbands' anxiety about abandonment, to assess whether people who were variable on anxiety about abandonment were also variable on marital satisfaction and had spouses who were variable on satisfaction and anxiety about abandonment. The only significant predictor in this analysis was vulnerability. /3 = .26. /(157) = 3.31./) = .001. 3

We reconducted the entire set of GCAs and variability analyses twice to determine if there were differential effects of the different components of the stable vulnerability factor. In the first reanalysis, stable vulnerabilities were defined as personality disturbance. In the second reanalysis stable vulnerabilities were defined as past history of personal and family psychopathology and family status. We did this because, although the latter variables truly are stable, personality disturbance may not be and. as such, our measure of stable vulnerabilities may be assessing two different things. The reanalyses were largely consistent with the initial analysis. No systematic differences were observed except that personality disturbance was a significant predictor of the intercept values of attachment representations, whereas the other stable vulnerabilities were not. In the attachment variability analyses, neither the past history variables on their own, nor personality disturbance on its own, was a significant predictor of variability on anxiety about abandonment, suggesting that it is the compounded level of vulnerability resulting from the composite of both types of variables that confers the greatest proneness to attachment change.

798

DAVILA, KARNEY, AND BRADBURY

Hence, the extent to which the trend toward security is actually reflective of a marital process rather than a more global maturational process is not clear. Processes predicting within-subject changes in attachment representations. Second, we addressed the question of why attachment representations change. What factors predict change over time? We examined three sets of factors—contextual, socialcognitive, and individual-difference—and found some support for each. Respite the fact that, on average, attachment representations showed a trend toward security over the early years of marriage, significant individual fluctuations occurred over time. This was true for husbands and wives, and, consistent with a socialcognitive model (e.g., Baldwin et al., 1996), was largely in response to how maritally satisfied they were. In support of the individual-difference model (Davila, Burge, & Hammen, 1997), wives' attachment fluctuations were also directly affected by their levels of stable vulnerabilities: Higher levels of vulnerabilities were associated with greater fluctuations. Furthermore, in support of the diathesis-stress model, which integrates the individualdifference model and the social-cognitive and contextual models, vulnerability moderated a number of associations. The clearest example was that as vulnerable husbands became less maritally satisfied, they also became less comfortable depending on others. This association was weaker for less vulnerable husbands. Hence, vulnerability creates a strong link between marital satisfaction and attachment security for husbands. This suggests that vulnerable husbands may be very reactive to even minor indicators of marital distress, or they may even overinterpret benign situations as evidence of their partners' unavailability. Although these results may be disappointing to advocates of one specific model, we believe they reflect the complex factors associated with, and likely necessary for, attachment change to occur. The process of changing any core set of attitudes or beliefs is a difficult, multifaceted endeavor, and the literatures on attitude change and self-concept change attest to this (e.g., Pelham, 1991; Sears, 1983; see Petty & Wegener, 1998, for a review). Our findings indicate that changes in attachment representations are a product of both intrapersonal and interpersonal factors. People's past experiences, their current states of mind about relationships, and their experiences with partners all affect how secure they feel in relationships. The clearest example of this in the present study was changes in wives' anxiety about abandonment. There was evidence that wives' anxiety about abandonment fluctuated over time in response to their partners' marital satisfaction, their own marital satisfaction, and their levels of stable vulnerabilities. Hence, there are numerous factors that determine the extent to which a wife will worry about rejection and abandonment. It is not surprising that multiple factors would be involved in the process of attachment change, particularly the factors studied here. For example, husbands' and wives' marital satisfaction tend to be highly correlated, and marital satisfaction also tends to be correlated with the sorts of stable vulnerability factors studied here (e.g., Beach & O'Leary, 1993; Kelly & Conley, 1987; McLanahan & Bumpass, 1988). Hence, it is easy to imagine how vulnerabilities might be associated with one partner's dissatisfaction, which would also be associated with the other partner's dissatisfaction and therefore result in lowered security. In fact, we suggest that research on continuity and change in attachment must take a multivariate approach in order to best understand how people

experience themselves and their social worlds over time. In addition, although the present research did not focus on large-scale changes in attachment styles, a topic that certainly requires attention, this study provides important information about the ups and downs of people's self- and social perception or, stated another way, their ongoing fluctuations in working models of self and others. Reciprocal associations between attachment representations and marital circumstances. The analyses of reciprocal processes shed even more light on the fluctuations of people's wofldng models and perhaps provide the clearest description of continuity and change in attachment representations. The notion behind testing reciprocal processes is that there is a dynamic interplay among people's attachment representations and their ongoing life circumstances. Reciprocal processes can thus provide information about how people maintain their levels of attachment security (i.e., assimilate information into their working models) and about how people change (i.e., accommodate new information). Numerous reciprocal processes were evident in the present study, and they indicated associations between attachment and marital satisfaction at both the within-spouse level (consistent with the social-cognitive model) and the cross-spouse level (consistent with the contextual model). At the within-spouse level, for both husbands and wives, attachment representations changed in response to spouses' states of mind about their marital satisfaction and created changes in their states of mind about marital satisfaction. At the cross-spouse level, spouses' attachment representations both changed in response to partners' levels of marital satisfaction and created changes in their partners' marital satisfaction. In both cases, satisfaction bred security, and security bred satisfaction. One of the most consistent sets of reciprocal patterns indicated that both within-spouse and cross-spouse reciprocal associations were occurring together. In many cases, spouses' changes in attachment representations were associated with their partners' changes in marital satisfaction (consistent with a contextual model), and partners' changes in marital satisfaction were associated with their own changes in attachment security (consistent with the social-cognitive model). In other cases, changes in spouses' attachment security were associated with changes in their own marital satisfaction (a social-cognitive process), and changes in their marital satisfaction were associated with changes in partners' attachment security (a contextual process). Hence, although mediation was not tested, these results suggest that changes in spouses' attachment security may be related to changes in their partners' attachment security through the effect security has on changes in marital satisfaction. We also found that stable vulnerabilities moderated a number of the reciprocal associations and that the nature of the reciprocal associations appeared different for husbands and wives. In associations involving changes in husbands' marital satisfaction, it was always the case that higher levels of security of either spouse were associated with higher levels of husbands' marital satisfaction. This was particularly true for more vulnerable husbands who showed stronger associations between their own marital satisfaction and their own comfort depending on others than did less vulnerable husbands. However, wives' patterns appeared to be different. In two of the three cases in which wives' marital satisfaction changed in re-

ATTACHMENT AND MARRIAGE sponse to changes in husbands' attachment security, the associations were stronger among vulnerable wives, and they were in a counterintuitive direction. Specifically, among vulnerable wives, when husbands became more secure, wives became less satisfied and in turn became less secure. There are a number of possible explanations for this finding that rest on the evidence that wives' vulnerability is associated with their own insecurity. For example, increases in husbands' security may signal to a vulnerable wife that she and her husband are no longer similar. Given that similarity is associated with satisfaction (e.g., Bentler & Newcomb, 1978), this signal may thus reduce her satisfaction. Vulnerable wives may also be engaging in a social comparison process in which they evaluate themselves negatively in comparison to their secure husbands and thus become dissatisfied and insecure (e.g., Festinger, 1954; Salovey & Rodin, 1984). Vulnerable wives might also perceive their husbands' increased security as a sign that their husbands no longer need them or even that they no longer love them or want to be with a wife who has such vulnerabilities. This hypothesis is consistent with the finding that vulnerable wives react negatively only to husbands' increases in comfort depending on others and decreases in anxiety about abandonment. Wives become more satisfied when their husbands become more comfortable with closeness. Therefore, when husbands provide wives with evidence of being closer, which wives may interpret as husbands being emotionally open and interested in the relationship, wives are happy and secure. However, when husbands provide vulnerable wives with evidence that they are not afraid of being abandoned and feel certain of the availability of others, vulnerable wives may interpret this as a sign that their husbands do not want or need to be emotionally close to or open with them, or are less interested in or dependent on the relationship. Women tend to be more sensitive to certain aspects of the marital relationship than are men (e.g., Floyd & Markman, 1983). For example, emotional and supportive aspects of the relationship matter more to wives' marital satisfaction and psychological health than to husbands' (e.g., Acitelli & Antonucci, 1994; Julien & Markman, 1991). Hence, vulnerable wives may be particularly sensitive. Of course, these hypotheses are speculative. Future research is needed to replicate the findings and determine their causes. Given the evidence of reciprocal associations, we must ask what the existence of an ongoing reciprocal process means. Is it evidence for accommodation, assimilation, or both? This is a difficult question to answer. Reciprocal associations certainly imply that people accommodate new information. People change their levels of attachment security in response to their marital circumstances. However, the present study could not determine whether these were lasting changes, which are the type of changes that would provide the clearest evidence of true accommodation. Reciprocal associations also imply that people assimilate information into their existing attachment representations. People change their views about their marriage on the basis of how secure they feel. Hence, the marriage gets assimilated into existing attachment structures. It is thus very possible that these reciprocal associations suggest that some people experience a process of long-term assimilation. For instance, suppose a wife begins her marriage with relatively insecure attachment representations and, because attachment and marital satisfaction are correlated, she and her spouse are also relatively less satisfied than others. Then suppose her and her spouse's lowered satisfaction leads her to feel even less secure,

799

which in turn leads her and her spouse to feel even less satisfied. This scenario exemplifies long-term assimilation. The wife's attachment representations may be changing (i.e., she is getting more insecure), but she is not actually changing her style (i.e., she is still insecure). Hence accommodation may occur, but attachment style change may not. If long-term assimilation is occurring for some people, what might bring it about? One possibility is that people, particularly vulnerable or insecure people, may be contributing to their negative marital circumstances and ultimately to the worsening or maintenance of marital dissatisfaction and attachment insecurity. This may occur through a process consistent with that of stress generation, by which people with various psychological and personality problems contribute to interpersonal stress in their lives (e.g., Daley, Hammen, Davila, & Burge, 1998; Davila, Bradbury, Cohan, & Tochluk, 1997; Hammen, 1991), or through a selfverification process, by which vulnerable people look for and actually receive feedback that confirms their negative sense of self (e.g., Swann, 1987, 1992). Insecure people might thus seek out or contribute to negative marital circumstances that result in dissatisfaction and further insecurity. Of course, secure people would maintain their security by seeking out and contributing to positive circumstances that result in satisfaction and further security.

Implications and Conclusions The present study has a number of implications for adult attachment theory and research and for understanding the course of marriage. First, the results of the present study provide further evidence that the consciously held beliefs that people hold about attachment security (a) are reliably associated with important interpersonal circumstances over time and (b) show changes over time that are psychologically meaningful. Together, this suggests that the beliefs assessed with self-report adult attachment measures are useful, reliable indicators of people's consciously accessible ideas about relationship security. Second, there is evidence that people, on average, show positive changes in attachment beliefs over the early years of marriage. This suggests that there may be various interpersonally salient life phases that may lead to more or less stability in attachment security. Identification of such phases is an important step in understanding interpersonal continuity and change over time. Third, despite the overall trend toward security, spouses' attachment representations are changing in response to marital circumstances, and they are influencing changes in marital satisfaction. However, these accommodation and assimilation processes may be relatively complex in that different processes may be operating for different spouses at different times in marriage. The challenge is to continue to identify who will change, when, and for what reasons. Regarding reasons, in the present study we assessed changes in attachment only in response to marital satisfaction and crossspouse attachment representations. These variables are only one indicator of marital circumstances. The role of specific marital events or objectively assessed marital behaviors in attachment change may provide further information about the circumstances that lead to attachment change. Furthermore, in the present study we assessed only fluctuations in attachment representations, not changes in attachment styles. Whether these different types of change result from similar processes is unknown and is an impor-

800

DAVILA, KARNEY. AND BRADBURY

tant area for future research. For example, it may be that the processes identified in the present study are precursors to lasting attachment style change. Many authors have suggested that lasting change may result from the consistent experience of a new (i.e., different) attachment relationship (e.g.. Dozier & Tyrell. 1998; Wachtel, 1991). The extent to which spouses experience a new attachment relationship with their partners, and the resulting gradual changes in their attachment representations, may thus have important implications for long-term changes in their attachment style. Fourth, the present results raise a number of important issues for each of the models of attachment change. Essentially, the results suggest that each model may benefit from integration with the other models. Clearly, people's attachment representations do fluctuate in response to contextual circumstances, be they people's present states of mind about relationships, as the social-cognitive model suggests, or other salient interpersonal experiences, as the contextual model suggests. However, people with stable vulnerabilities may. in some cases, be more likely to fluctuate in response to these circumstances. This informs the contextual and socialcognitive models by illuminating a specific group of people who may react to contextual circumstances. This finding also informs the individual-difference model by illuminating proximal circumstances to which vulnerable people react. Fifth, the present results suggest that ihe course of newlywed marriage may be tied to the course of spouses' attachment representations. A number of cross-sectional and short-term longitudinal studies have shown that attachment security is associated with greater marital satisfaction and better communication (Cohn. Silver. Cowan, Cowan. & Pearson. 1992; Feeney et al.. 1994; Fuller & Fincham. 1995: Kobak & Hazan. 1991: Scnchak & Leonard. 1992). The present study confirms this association and further suggests that attachment representations can have an ongoing influence on marital satisfaction. The results also highlight a process that informs understanding of the course of marriage for vulnerable spouses in particular. The results from the reciprocal analyses suggest that vulnerable people may be more likely to change their feelings about their marriage in response to changes in their levels of attachment security. Hence, vulnerable people are reactive to their internal experiences of security, which then affect how they view their relationships. This has important implications for understanding the course of marriage, particularly for understanding who will be at high risk for ongoing marital dissatisfaction and possibly dissolution. Prior research has shown that stable vulnerabilities, such as parental divorce, psychopathology, and personality dysfunction, are associated with a poorer course of marriage (e.g.. Beach & O'Leary, 1993; Kelly & Conley. 1987; McLanahan & Bumpass, 1988: see Davila & Bradbury, 1998, for a review of the association between psychopathology and marriage). The present work suggests a possible mechanism for this association. Spouses with stable vulnerabilities tend to be insecure. As described above, their levels of marital satisfaction tend to be highly correlated with how secure they feel. They may thus tend to assimilate marital information into their insecure representations, thereby increasing the likelihood that they will continue to feel insecure and dissatisfied in their marriages. Finally, the present study focused solely on changes in selfreported attachment security in adulthood. Other findings may

result if attachment is assessed using alternate methods. Alternate methods may tap different aspects of the attachment system, which may change for different reasons. For example, attachment assessed by means of interview may tap attachment style and its associated defensive strategies, and, as such, provide information about large-scale changes in attachment style. As described earlier, large-scale changes may result only from emotionally significant events or continued new interpersonal experiences. Research on attachment change assessed by means of interview is in its early stages, because results from long-term longitudinal studies are just emerging, but they have implications for understanding large-scale attachment change. For example, data from longitudinal studies in which attachment style was measured by means of the Strange Situation (Ainsworth, Blehar, Waters, & Wall, 1978) in childhood and by means of the Adult Attachment Interview (George, Kaplan. & Main, 1984) in late adolescence and young adulthood show that stably insecure people were more likely to have been maltreated during childhood and had worse family functioning during early adolescence than people who became secure over time (Weinfeld, Sroufe, & Egeland. 1998). In addition, compared with stably secure people, people who became insecure had mothers who experienced more depression. Hence, stable vulnerability factors clearly put people at risk for insecurity, and the extent to which young people experienced significant interpersonal events such as family problems or a depressed mother was associated with changes in attachment styles. Another example of when different findings may result if attachment is assessed with methods other than self-report comes from experimental tests of the social-cognitive model of attachment change (e.g.. Baldwin et al., 1996). These methods assess changes in attachment representations by tapping into less dominant attachment schemas that people hold. These methods show that peoples' attachment schemas change frequently, depending on the circumstances the person is in. However, as discussed earlier, these methods do not provide information about dispositional changes in attachment. When self-reported attachment is assessed, as in the present study, the results highlight how consciously held experiences of security fluctuate depending on people's intra- and interpersonal experiences. Moreover, when attachment in a specific relationship is assessed, as in the present study, the results highlight how security in a specific relationship can change in response to experiences in that relationship. As we have shown, self-reported attachment security fluctuates regularly in response to marital circumstances. Whether these changes correspond to changes in general or dispositional attachment representations or whether the changes are truly specific to the marital relationship is unclear. Whether these changes result because people hold multiple attachment schemas also is unclear. Continued examination of the stability of self-reported and other measures of attachment will help illuminate how people's cognitive, emotional, and behavioral indicators of security change and how changes in each domain are related. To conclude, let us revisit the quote with which we began. Is marriage a time for radical revision of working models, as Bowlby (1969) suggested? On average, probably not. Change does happen. However, instead of the straightforward association suggested in Bowlby's quote, the process of attachment change, during marriage or any time, is likely a complex process that may occur on

ATTACHMENT AND MARRIAGE

many levels, in response to numerous intra- and interpersonal experiences. More exact specification of this complex process will be an important task for future research.

References

801

psychopathology perspective on depression in children and adolescents. In W. M. Reynolds & H. F. Johnson (Eds.). Handbook of depression in children and adolescents (pp. 123-141). New York: Plenum. Cohn. D. A.. Silver. D. H.. Cowan. C. P.. Cowan, P. A.. & Pearson, J. (1992). Working models of childhood attachment and couple relationships. Journal of Family Issues. 13. 432-449. Collins, L. M.. & Horn. J. L. (1991). Best methods for the analysis

Acitclli. L. K... & Antonucci. T. C. (1994). Gender differences in the link between marital support and satisfaction in older couples. Journal of

Personality and Social Psychology. 67. 688-698. Adler. A. G., & Scher, S. J. (1994). Using growth curve analyses to assess personality change and stability in adulthood. In T. F. Heatherton & J. L. Weinberger (Eds.), Can personality change'.' (pp. 149-173). Washington. DC: American Psychological Association. Ainsworth. M. D. S., Blehar. M. Waters. E.. & Wall. S. (1978). Patterns of attachment: A psychological study of the Strange Situation. Hillsdaic. NJ: Erlbaum. American Psychiatric Association. (1987). Diagnostic and statistical manual of mental disorders (3rd ed.. rev.). Washington, DC: Author. American Psychiatric Association. (1994). Diagnostic and statistical manual of menial disorders (4th ed.). Washington. DC: Author. Andreasen, N.. Endicott. J.. Spit/.er. R.. & Winokur. G. (1977). The family history method using diagnostic criteria. Archives of General Psychiatry. 43. 266-271. Baldwin, M. W. (1995). Relational schemas and cognition in close relationships. Journal of Social and Persona! Relationships. J2. 547-552. Baldwin, M. W.. & Fehr. B. (1995). On the instability of attachment style ratings. Personal Relationships, 2. 247-261. Baldwin. M. W.. Keelan. J. P. R.. Fehr. B.. Enns, V.. & Koh-Rangarajoo. E. (1996). Social-cognitive conceptualization of attachment working models: Availability and accessibility effect. Journal of Personality and Social Psychology. 71. 94-109. Beach, S. R. H.. & O'Leary. K. D. (1993). Dysphoria and marital discord: Arc dysphoric individuals at risk for marital maladjustment1' Journal of Marital and Family Therapy. 19. 355-368. Beck. A. T. (198.3). Cognitive therapy of depression: New perspectives. In P. J. Clayton & J. E. Barrett (Eds). Treatment of depression: Old controversies and new approaches (pp. 265-290). New York: Raven Press. Bentler. P. M., & Newcomb. M. D. (1978). Longitudinal study of marital success and failure. Journal of Consulting and Clinical Psychology. 46. 1053-1070. Bolger. N., & Zuckerman. A. (1995). A framework for studying personality in the stress process. Journal of Personality and Social Psychology. 69. 890-902. Bowlby, J. (1969). Attachment and loss: Vol. 1. Attachment. New York: Basic Books. Bowlby. J. (1973). Attachment and loss: Vol. 2. Separation: Anxiety and anger. New York: Basic Books. Bowlby. J. (1980). Attachment and loss: Vol. 3. Loss. New York: Basic Books. Bryk, A. S.. & Raudenbush, S. W. (1992). Hierarchical linear models: Application and data analysis methods. Newbury Park. CA: Sage. Bryk. A. S., Raudenbush. S. W.. & Congdon. R. T. (1994). HLM: Hierarchical linear modeling with the HLM/2L and HLM/3L programs. Chicago: Scientific Software International. Caspi. A. (1993). Why maladaptive behaviors persist: Sources of continuity and change across the life course. In D. C. Funder. R. D. Parkc, C. Tomlinson-Keasey. & K. Widaman (Eds.). Studying lives through lime: Personality and development (pp. 343-376). Washington, DC: American Psychological Association. Caspi, A., & Bern. D. J. (1990). Personality continuity and change across the life course. In L. A. Pervin (Ed.), Handbook of personality theory and research (pp. 549-575). New York: Guilford Press. Cicchetti, D.. Rogosch. F. A., & Toth, S. L. (1994). A developmental

of

change: Recent advances, unanswered questions, future direction. Washington. DC: American Psychological Association. Collins. N. L., & Read. S. J. (1990). Adult attachment, working models, and relationship quality in dating couples. Journal of Personality and Social Psychology. 58. 644-663. Collins. N. L.. & Read, S. J. (1994). Cognitive representations of attachment: The structure and function of working models. Advances in Personal Relationships. 5. 53-90. Costa. P. T.. & McCrae. R. R. (1992). The NEO Personality InventoryRevised (professional manual). Odessa. FL: Psychological Assessment Resources. Daley. S. K.. Hammen. C . Davila. I . & Burge, D. (1998). Axis II symptomatology, depression, and life stress during the transition from adolescence to adulthood. Journal of Consulting and Clinical Psychology. 66. 595-603. Davila. .[., & Bradbury. T. N. (1998). Psychopathology and the marital dyad. In L. L'Abate (Ed.). Handbook of family psychopathology (pp. 127-157). New York: Guilford Press. Davila. J., Bradbury. T. N.. Cohan. C . & Tochluk. S. (1997). Marital functioning and depressive symptoms: Evidence for a stress generation model. Journal of Personality and Social Psychology. 73. 849-861. Davila. J.. Bradbury. T. N.. & Fincham. F. D. (1998). Negative affectivity as a mediator of the association between attachment and marital satisfaction. Personal Relationships. 5. 467-484. Davila. .).. Burge. D., & Hammen. C. (1997). Why does attachment style change? Journal of Personality and Social Psychology, 73, 826-838. Dickman. S. J. (1990). Functional and dysfunctional impulsivity. Journal of Personality and Social Psychology. 58. 95-102. Do/icr. M.. & Tyrell. C. (1998). The role of attachment in therapeutic relationships. In J. A. Simpson & W. S. Rholes (Eds.), Attachment theory and close relationships (pp. 221-248). New York: Guilford Press. Endicott. J.. Andreasen. N.. & Spit/.er. R. (1975). Family history—Research diagnostic criteria. New York: Biometrics Research. New York State Psychiatric Institute. Eysenck. H. J.. & Eysenck, S. B. G. (1978). Manual for the Eysenck Personality Questionnaire. Kent, England: Hodder and Stoughton. Feency. J. A.. Noller. P.. & Callan. V. J. (1994). Attachment style. communication, and satisfaction in the early years of marriage. Advances in Personal Relationships. 5. 269—308. Festinger, L. (1954). A theory of social comparison processes. Human Relations. 7. 117-140. Floyd, F.. & Markman, H. (1983). Observational biases in spouse observation: Toward a cognitive/behavioral model of marriage. Journal of Consulting and Clinical Psychology. 51. 450-457. Fraley, R. C & Davis, K. E. (1997). Attachment formation and transfer in young adults' close friendships and romantic relationships. Personal Relationships. 4. 131-144. Fuller, T. L., & Fincham. F. D. (1995). Attachment style in married couples: Relation to current marital functioning, stability over time, and method of assessment. Personal Relationships. 2. 17-34. George, C . Kaplan. N.. & Main. M. (1984). Attachment interview for adults. Unpublished manuscript. University of California. Berkeley. Hammen, C. (1991). The generation of stress in the course of unipolar depression. Journal of Abnormal Psychology. 100. 555-561. Ha/.an. C , & Shaver. P. R. (1994). Attachment as an organizational framework for research on close relationships. Psychological Inquiry. 5. 1-22.

802

DAVILA, KARNEY, AND BRADBURY

Hyler, S. E., & Rieder, R. O. (1987). Personality Diagnostic Questionnaire-Revised. New York: New York State Psychiatric Institute. Hyler, S. E., Skodol, A. E., Kellman, H. D., Oldham, J. M., & Rosnick, L. (1990). Validity of the Personality Diagnostic Questionnaire-Revised: Comparison with two structured interviews. American Journal of Psychiatry, 147, 1043-1048. Julien, D., & Markman. H. J. (1991). Social support and social networks as determinants of individual and marital outcomes. Journal of Social and Personal Relationships, 8, 549-568. Karney, B. R. (in press). Depressive symptoms and marital satisfaction in the early years of marriage: The implications of a growth curve analysis. In S. R. H. Beach (Ed.), Marital and family processes in depression. Washington, DC: American Psychological Association. Karney, B. R., & Bradbury, T. N. (1995). Assessing longitudinal change in marriage: An introduction to the analysis of growth curves. Journal of Marriage and the Family. 54, 582-594. Karney, B. R., & Bradbury. T. N. (1997). Neuroticism, marital interaction, and the trajectory of marital satisfaction. Journal of Personality and Social Psychology, 72, 1075-1092. Karney, B. R.. Davila, J., Cohan, C. L.. Sullivan. K. T., Johnson, M. D., & Bradbury, T. N. (1995). An empirical investigation of sampling strategies in longitudinal research on marriage. Journal of Marriage and the Family, 57. 909-920. Kelly, E. L., & Conley, J. J. (1987). Personality and compatibility: A prospective analysis of marital stability and marital satisfaction. Journal of Personality and Social Psychology, 52, 27-40. Kernis, M. H., Cornell. D. P., Sun, C , Berry, A., & Harlow, T. (1993). There's more to self-esteem than whether it is high or low: The importance of stability of self-esteem. Journal of Personality and Social Psychology. 65, 1190-1204. Kirkpatrick, L. A.. & Hazan, C. (1994). Attachment styles and close relationships: A four-year prospective study. Personal Relationships, I, 123-142. Klohnen, E. C , & John. O. P. (1998). Working models of attachment: A theory-based prototype approach. In J. A. Simpson & W. S. Rholes (Eds.), Attachment theory and close relationships (pp. 115-142). New York: Guilford Press. Kobak, R. R., & Hazan, C. (1991). Attachment in marriage: Effects of security and accuracy of working models. Journal of Personality and Social Psychology, 60, 861-869. Kurdek, L. A. (1991). Predictors of increases in marital distress in newlywed couples: A 3-year prospective longitudinal study. Developmental Psychology, 27, 627-636. Locke, H., & Wallace, K. (1959). Short marital adjustment and prediction tests: Their reliability and validity. Marriage and Family Living, 21, 251-255. McLanahan, S., & Bumpass, L. (1988). Intergenerational consequences of family disruption. American Journal of Sociology, 94, 130-152. Pelham, B. W. (1991). On confidence and consequence: The certainty and

importance of self-knowledge. Journal of Personality and Social Psychology, 60, 518-530. Petty, R. E., & Wegener. D. T. (1998). Attitude change: Multiple roles for persuasion variables. In D. T. Gilbert, S. T. Fiske, & G. Lindzey (Eds.), Handbook of social psychology (4th ed.. Vol. 1. pp. 323-390). New York: McGraw-Hill. Raudenbush, S. W., Brennan. R. T., & Barnett, R. C. (1995). A multivariate hierarchical model for studying psychological change within married couples. Journal of Family Psychology, 9, 161-174. Rogosa, D., Brandt, D., & Zimowski, M. (1982). A growth curve approach to the measurement of change. Psychological Bulletin, 92, 726-748. Salovey. P.. & Rodin, J. (1984). Some antecedents and consequences of social-comparison jealousy. Journal of Personality and Social Psychology, 47, 780-792. Scharfe, E., & Bartholomew, K. (1994). Reliability and stability of adult attachment patterns. Personal Relationships, 1, 23-43. Sears, D. O. (1983). The persistence of early political predispositions: The roles of attitude object and life stage. In L. Wheeler & P. Shaver (Eds.), Review of personality and social psychology (Vol. 4, pp. 79-116). Beverly Hills. CA: Sage. Senchak, M., & Leonard, K. E. (1992). Attachment styles and marital adjustment among newlywed couples. Journal of Social and Personal Relationships, 9, 51-64. Siegel, J. M. (1986). The Multidimensional Anger Inventory. Journal of Personality and Social Psychology, 51, 191-200. Spanier, G. (1976). Measuring dyadic adjustment: New scales for assessing the quality of marriage and similar dyads. Journal of Marriage and the Family, 42, 15-27. Spitzer, R. L., Williams, J. B. W., Gibbon. M , & First, M. (1994). Structured Clinical Interview for DSM-IV (SC1D). New York: New York State Psychiatric Institute, Biometrics Research. Swannv W. B., Jr. (1987). Identity negotiation: Where two roads meet. Journal of Personality and Social Psychology, 53, 1038-1051. Swann, W. B., Jr. (1992). Seeking "truth," finding despair: Some unhappy consequences of a negative self-concept. Current Directions in Psychological Science. I. 15-18. Trull, T. (1995). Borderline personality disorder features in nonclinical young adults: I. Identification and validation. Psychological Assessment. 7, 33-41. Wachtel. P. L. (1991). The role of "accomplices" in preventing and facilitating change. In R. C. Curtis & G. Strieker (Eds.), How people change: Inside and outside therapy (pp. 21-28). New York: Plenum. Weinfeld, N. S.. Sroufe, L. A., & Egeland, B. (1998). Attachment from infancy to early adulthood in a high risk sample: Continuity, discontinuity, and their correlates. Manuscript submitted for publication.

Received July 16, 1998 Revision received November 27, 1998 Accepted November 30, 1998 •