Romantic Love Conceptualized as an Attachment Process Cindy Hazan and Phillip Shaver University of Denver This articleexplores the possibilitythat romantic loveisan attachment process--abiosocialprocess by which affectional bonds are formed between adult lovers,just as affectional bonds are formed earlier in life between human infants and their parents. Key components of attachment theory, developed by Bowlby, Ainsworth, and others to explain the development of affectional bonds in infancy, were translated into terms appropriateto adult romantic love.The translation centered on the three major styles of attachment in infancy--secure, avoidant, and anxious/ambivalent--and on the notion that continuity of relationshipstyle is due in part to mental models(Bowlby's "inner working models") of selfand social life. These models, and hence a person's attachment style, are seen as determined in part by childhoodrelationships with parents. Twoquestionnaire studies indicated that (a) relative prevalence of the three attachment styles is roughly the same in adulthood as in infancy, (b) the three kinds of adults differ predictablyin the waythey experience romantic love, and (c)attachment styleisrelatedin theoreticallymeaningful waysto mental modelsofselfand social relationshipsand to relationshipexperiences with parents. Implicationsfortheoriesofromanticlove are discussed, asare measurementproblemsand otherissuesrelatedto future tests ofthe attachment perspective. One of the landmarks of contemporary psychology is Bowlby's (1969, 1973, 1980) three-volume exploration of attachment, separation, and loss, the processes by which affectional bonds are forged and broken. Bowlby's major purpose was to describe and explain how infants become emotionally attached to their primary caregivers and emotionally distressed when separated from them, although he also contended that "attachment behavior [characterizes] human beings from the cradle to the grave" (1979, p. 129). In recent years, laboratory and naturalistic studies of infants and children (summarized by Bretherton, 1985, and Maccoby, 1980) have provided considerable support for attachment theory, which was proposed by Bowlby and elaborated by several other investigators. The purpose of this article is to explore the possibility that this theory, designed primarily with infants in mind, offers a valuable perspective on adult romantic love. We will suggest that romantic love is an attachment process (a process of becoming attached), We are grateful to Donna Bradshaw for sharing her expertisein the areas ofattachment theory and research, to Marry Meitus for allowing us to conduct Study 1 in the Rocky Mountain News,to Kathy Purcell for keypunching, to Rick Canfield for assistance in all phases of the project,and to MaryAinsworth, John Bowlby,Harry Gollob,Lee Kirkpatrick, Roger Kobak, Anne Peplau, Harry Reis, Judith Schwartz, Arlene Skolnick, and Robert Sternberg for helpful comments on convention presentationsand earlierdrafts ofthis article. Correspondence concerning this articleshould be addressedto Cindy Hazan or to Phillip Shaver, Department of Psychology, University of Denver, Denver, Colorado80208-0204. experienced somewhat differently by different people because of variations in their attachment histories. For our purpose, which is to create a coherent framework for understanding love, loneliness, and grief at different points in the life cycle, attachment theory has several advantages over existing approaches to love (Shaver, Hazan, & Bradshaw, in press). First, although many researchers (e.g., Rubin, 1973; Hatfield & Sprecher, 1985) have attempted to assess love with unidimensional scales, love appears to take multiple forms (e.g., Dion & Dion, 1985; Hendrick & Hendrick, 1986; Lee, 1973; Stock, Levitan, McLane, & Kelley, 1982; Sternberg, 1986; Tennov, 1979). Attachment theory explains how at least some of these forms develop and how the same underlying dynamics, common to all people, can be shaped by social experience to produce different relationship styles. Second, although various authors have portrayed certain forms of love as healthy and others as unhealthy, or at least problematic (e.g., Hindy & Schwarz, 1984; Tennov, 1979), they have not said how the healthy and unhealthy forms fit together in a single conceptual framework. Attachment theory not only provides such a framework, but it also explains how both healthy and unhealthy forms of love originate as reasonable adaptations to specific social circumstances. The portrait of love offered by attachment theory includes negative as well as positive emotions: for example, fear of intimacy (discussed by Hatlield, 1984), jealousy (e.g., Hindy & Schwarz, 1985), and emotional ups and downs (Tennov, 1979) as well as caring (Rubin, 1973), intimacy (Sternberg, 1986), and trust (Dion & Dion, 1985). Third, attachment theory deals with separation and loss and helps explain how loneliness and love are related (Shaver & Rubenstein, 1980; Parkes & Journal of Personality and Social Psychology, 1987, Vol. 52, No. 3, 511-524 Colwfight 1987by the American Psychological Association, Inc. 0022-3514/87/$00.75 511 512 CINDY HAZAN AND PHILLIP SHAVER Weiss, 1983; Weiss, 1973). Finally, attachment theory links adult love with socioemotional processes evident in children and nonhuman primates; it places love within an evolutionary context (Wilson, 1981). (See Sternberg & Barnes, in press, for an anthology of recent approaches to the study of adult love.) Attachment Theory and Research Bowlby's attachment theory grew out of observations of the behavior of infants and young children who were separated from their primary caregiver (usually the mother) for various lengths of time. Bowlby noticed what primate researchers had also observed in the laboratory and the field: When a human or primate infant is separated from its mother, the infant goes through a predictable series of emotional reactions. The first is protest, which involves crying, active searching, and resistance to others' soothing efforts. The second is despair, which is a state of passivity and obvious sadness. And the third, discussed only with reference to humans, is detachment, an active, seemingly defensive disregard for and avoidance of the mother if she returns. Because of the remarkable similarities between human infants and other primate infants, Bowlby was led to consider the evolutionary significance of infant--careglver attachment and its maintenance in the face of separation. The attachment system, as Bowlby called the complex constellation of attachment feelings and behaviors, seems to have evolved to protect infants from danger by keeping them close to the mother. When very young, a human infant can do little more than cry, make eye contact, smile, and snuggle in to encourage its mother to keep it near. Once mobile, however, it can activelypursue its mother and vocalize to her. Bowlby and other observers of both human and primate behavior have noticed that when an infant is healthy, alert, unafraid, and in the presence of its mother, it seems interested in exploring and mastering the environment and in establishing aifiliativecontact with other family and community members. Researchers call this using the mother as a secure base. Attachment theory can be summarized in three propositions, phrased clearly in the second volume of Bowlby's trilogy: The first [proposition] is that when an individualis confident that an attachment figure will be available to him whenever he desires it, that person will be much less prone to either intense or chronic fear than will an individualwho for any reason has no such confidence. The second propositionconcerns the sensitiveperiodduring which such confidence develops. It postulates that confidence in the availability ofattachment figures,or lackofit, isbuilt up slowly during the years of immaturity--infancy, childhood, and adolescence-and that wliatever expectations are developed duringthose years tend to persist relatively unchanged throughout the rest of life. The third proposition concerns the role of actual experience. It postulates that the varied expectation'g of the accessibility and responsiveness of attachment figuresthat individualsdevelop during the years of immaturity are tolerably accurate reflections of the experiences those individuals haveactuallyhad. (Bowlby,1973, p. 235) The formation during early childhood of a smoothly functioning (i.e., secure) attachment relationship with a primary caregiver, although the norm in our society, is by no means guaranteed. Research by Ainsworth and others suggests that a mother's sensitivity and responsiveness to her infant's signals and needs during the first year of life are important prerequisites. Mothers who are slow or inconsistent in responding to their infant's cries or who regularly intrude on or interfere with their infant's desired activities (sometimes to force affection on the infant at a particular moment) produce infants who cry more than usual, explore less than usual (even in the mother's presence), mingle attachment behaviors with overt expressions of anger, and seem generally anxious. If, instead, the mother consistently rebuffs or rejects the infant's attempts to establish physical contact, the infant may learn to avoid her. On the basis of their observations, Ainsworth, Blehar, Waters, and Wall (1978) delineated three styles or types of attachment, often called secure, anxious/ambivalent, and avoidant. Infants in the anxious/ambivalent category frequently exhibit the behaviors Bowlby called protest, and the avoidant infants frequently exhibit the behaviors he called detachment. A major goal of this article is to apply this three-category system to the study of romantic love. In their description of the three attachment styles, Ainsworth et al. (1978) referred to infants' expectations concerning their mothers' accessibility and responsiveness. This fits with Bowlby's claim that infants and children construct inner working models of themselves and their major social-interaction partners. Because the expectations incorporated in these models are some of the most important sources of continuity between early and later feelings and behaviors, they deserve special attention. According to Bowlby, working models (which we will also call mental models) and the behavior patterns influenced by them are central components of personality. The claim of cross-situational and cross-age continuity is still controversial but is supported by a growing list of longitudinal studies from infancy through the early elementary school years (Dontas, Maratos, Fafoutis, & Karangelis, 1985; Erickson, Sroufe, & Egeland, 1985; Main, Kaplan, & Cassidy, 1985; Sroufe, 1983; Waters, Wippman, & Sroufe, 1979). This evidence for continuity adds plausibilityto the notion that a person's adult style of romantic attachment is also affected by attachment history. Continuity, according to Bowlby (1973), is due primarily to the persistence of interrelated mental models of self and social life in the context of a fairly stable family setting: Confidence that an attachment figure is, apart from being accessible, likelyto be responsive can be seen to turn on at least two variables: (a) whether or not the attachment figure isjudged to be the sort ofperson who in general responds to calls forsupport and protection; [and] (b) whether or not the selfisjudged to be the sort of person towards whom anyone, and the attachment figure in particular, is likely to respond in a helpful way.Logically these variables are independent. In practice they are apt to be confounded. As a result, the model of the attachment figure and the model of the selfare likely to develop so as to be complementary and mutually confirming. (Bowlby, 1973,p. 238) Love as Attachment So far, no one has'attempted to conceptualize the entire range of romantic love experiences in a way that parallels the typology developed by Ainsworth and her colleagues. Nor has anyone with an interest in romantic relationships pursued Bowlby's idea that continuity in relationship style is a matter of mental models of self and social life. Finally, no one has explored the LOVE CONCEPTUALIZED AS AN ATTACHMENT PROCESS 513 possibilitythat the specific characteristics of parent-child relationships identified by Ainsworth et al. as the probable causes of differences in infant attachment styles are also among the determinants of adults' romantic attachment styles. These are the major aims of this article. We derived the following hypotheses by applying Bowlby's and Ainsworth's ideas and findings as literallyas possible to the domain of adult love. Hypothesis 1 Given the descriptions of the secure, avoidant, and anxious/ ambivalent styles, we expected roughly 60% of adults to classify themselves as secure and the remainder to split fairly evenly between the two insecure types, with perhaps a few more in the avoidant than in the anxious/ambivalent category. In a summary of American studies of the three types of infants, Campos, Barrett, Lamb, Goldsmith, and Stenberg (1983) concluded that 62% are secure, 23% are avoidant, and 15% are anxious/ambivalent. Given a diverse sample of American adults, we thought it reasonable to expect approximately the same proportions. Hypothesis 2 Just as the feelings an infant presumably experiences in the relationship with his or her mother are thought to reflect the quality of attachment to her, we expected that different types of respondents--secure, avoidant, and anxious/ambivalent-- would experience their most important love relationships differently. We predicted that the most important love experience of a secure adult would be characterized by trust, friendship, and positive emotions. For avoidant adults, love was expected to be marked by fear of closeness and lack of trust. Anxious/ambivalent adults were expected to experience love as a preoccupying, almost painfully exciting struggle to merge with another person. This last style is similar to what Hindy and Schwarz (1984) called anxious romantic attachment and Tennov (1979) called limerence. Hypothes• 3 Respondents' working models of self and relationships were also expected to differ according to attachment style. Secure types should believe in enduring love, generally find others trustworthy, and have confidence that the self is likable. Avoidant types should be more doubtful of the existence or durability of romantic love and believe that they do not need a love partner in order to be happy. Anxious/ambivalent types should fall in love frequently and easily but have difficulty finding true love. They should also have more self-doubtsthan the other two types because, unlike avoidant respondents, they do not repress or attempt to hide feelings of insecurity. Hypothesis 4 Because attachment style is thought to develop in infancy and childhood, we expected respondents of the three types to report different attachment histories. According to the theory, secure respondents should remember their mothers as dependably responsive and caring; avoidant respondents should report that their mothers were generally cold and rejecting; and anxious/ ambivalent respondents should remember a mixture of positive and negative experiences with their mothers. As less research has been conducted with fathers, we tentatively expected the findings related to them to be roughly similar to the findings for mothers. Hypothesis 5 Finally, because the attachment needs of insecure respondents are unlikely to be fully met, avoidant and anxious/ambivalent respondents should be especially vulnerable to loneliness. The avoidant types, however, may defend against or attempt to hide this vulnerable feeling and so report less loneliness than anxious/ambivalent respondents do. Study 1 In an initial effort to test the attachment-theory approach to romantic love, we designed a "love quiz" to be printed in a local newspaper. As explained by Shaver and Rubenstein (1983), the newspaper questionnaire method has been used in a wide variety of studies, always with results that approximate those from more expensive, more strictly representative surveys. The main difference between newspaper survey respondents and participants in representative sample surveys is that the former have slightly higher education levels. Also, depending on the topic, newspaper surveys tend to draw more female than male respondents. Neither of these biases seemed to preclude a valuable initial test of our ideas, and the gains in sample size and heterogeneity appeared to outweigh the cost of mild unrepresentative- ness. A single-item measure of the three attachment styles was designed by translating Ainsworth et al?s (1978) descriptions of infants into terms appropriate to adult love. The love-experience questionnaire, which we will describe in detail, was based on previous adult-love measures and extrapolations from the literature on infant-caregiver attachment. The measure of working models was based on the assumption that conscious beliefs about romantic love--concerning, for example, whether it lasts forever and whether it is easy or difficult to find--are colored by underlying, and perhaps not fully conscious, mental models. The measure of attachment history was a simple adjective checklist used to describe childhood relationships with parents and the parents' relationship with each other. Method Subjects. Analyses reported here are based on the first 620 of over 1,200 repliesreceived within a week followingpublication ofthe questionnaire. (The major findings were stable after the first few hundred, so additional replieswere not keypunched.) Of these 620 replies, 205 were from men and 415 were from women. The subjects ranged in age from 14 to 82, with a median age of 34 and a mean of 36. Average household income was $20,000 to $30,000; averageeducation levelwas "some college"Just over half(51%)were Protestant, 22%were Catholic, 3% were Jewish, 10% were atheist or agnostic, and 13% were "other." Ninety-onepercent were "primarilyheterosexual," 4%were "primarily homosexual," and 2% were "primarilybisexual" (3% chose not to answer). Forty-two percent were married at the time of the survey; 28% 514 CINDY HAZAN AND PHILLIP SHAVER were divorced or widowed, 9% were "livingwith a lover" and 31% were dating. (Some checked more than one category.) Measuresandprocedure. The questionnaire appeared in the July 26, 1985, issue of the Rocky Mountain News on the first and second pages of the Lifestyles section. Besides being highly visible there, it was referred to in a banner headline at the top of the paper's front page: "Tell us about the love of your life; experts ask 95 questions about your most important romance" The instructions included the following sentences: "The questionnaire is designed to look at the most important loverelationship you have everhad, why you got involvedin it, and why it turned out the way it did.... It may be a past or a current relationship, but choose only the most important one?' Given that there was only enough room to ask about one relationship, we decided to have subjects focus on the one they considered most important. The questionnaire was divided into three parts. The first contained 56 statements concerning the subject's most important relationship, for example, "I (considered/consider) one of my best friends" and "I (loved/love)~ so much that I often (felt/feel)jealous?' (The blank referred to the most important lover's name.) Responses were recorded by circlingSD, D, A, or SA to indicate points along a strongly disagreeto stronglyagreecontinuum. The 56 statements, 4 each for 14 a priori subscales, were adapted from previous love questionnaires (Dion & Dion, 1985; Hatfield & Sprecher, 1985; Hindy & Schwarz, 1984; Lasswell & Lobsenz, 1980; Rubin, 1973; Steffen, McLaney, & Hustedt, 1984) or suggested by the literature on infant-caretaker attachment (e.g., Ainsworth et al., 1978). A principal-components analysis followed by equimax rotation was performed on the 56-item measure. Thirteen factors had eigenvalues greater than 1.0, and 12 corresponded to a priori scales. Items loading above .40 on 1 of the 12 predicted factors were analyzed for reliability, and items that reduced coefficientalpha were deleted. Table 1provides the names of the 12 scales and a sample item, the number of items retained, and coefficient alpha for each. Alpha ranged from .64 to .84 with a mean of.76, which seemed adequate for preliminary tests of the hypotheses. Part 2 of the questionnaire asked whether the described relationship was current or past (61% were current, 39% were past), what the subject's relationship to that person was at the time of the survey, how long the relationship had lasted, how many times the subject had been in love,and whether he or she had experienced crushes before age 10. This part of the questionnaire also contained demographic questions• Part 3 dealt with attachment style and attachment history. It included sections dealing with the subject's childhood relationships with his or her mother and father and the parents' relationship with each other (the specific items will be discussed more fully in the Results andDiscussion section). Also included were questions concerning how the subject typically felt in relationships (the exact wording appears in Table 2) and what he or she believed concerning the typical course of romantic love• The questionnaire concluded with the open-ended question "Can you add anything that might help us understand romantic love?" and a request for the subject's name and phone number if he or she was willing to be interviewed. (Over 60% ofthe subjects provided this information•) Subjects were asked to mail their reply forms to the RockyMountain Newswithin a week. Results and Discussion Frequencies ofthe three attachment styles. Hypothesis 1 concerned whether newspaper readers could meaningfully classify themselves as avoidant, anxious/ambivalent, or secure in their most important romantic relationship, given fairly simple descriptions of the three attachment styles, and in particular whether the frequencies of the types would be similar to those found in studies of infants and young children. Table 2 shows Table 1 Information on Love-Experience Scales No. of Scale name Sample item items a Happiness My relationship with ~ 4 .84 (made/makes) me very happy. Friendship I (considered/consider) 4 .78 one of my best friends. Trust I (felt/feel)complete trust in 4 .83 Fear of closeness I sometimes (felt/feel)that 3 .64 getting too close to could mean trouble. Acceptance I (was/am) wellaware of 2 •67 ' s imperfections but it (did/does) not lessen my love. Emotional I (felt/feel)almost as much 3 .81 extremes pain asjoy in my relationship with _ _ Jealousy 1(loved/love)__ so 4 .82 much that I often (felt/ feel)jealous. Obsessive Sometimes my thoughts 3 .70 preoccupation (were/are) uncontrollably on Sexual attraction I (was/am) very physically 4 .80 attracted to Desire for union Sometimes I (wished/wish) 3 .79 that and I were a single unit, a "we" without clear boundaries. Desire for More than anything, I 3 .70 reciprocation (wanted/want) ~ to return my feelings. Love at first sight Once I noticed __, I was 4 .70 hooked. how the alternatives were worded and provides the percentage of subjects endorsing each description. Just over half (56%) classified themselves as secure, whereas the other half split fairly evenly between the avoidant and anxious/ambivalent categories (25% and 19%, respectively). These figures are similar to proportions reported in American studies of infant-mother attachment (Campos et al., 1983, summarized the proportions obtained in these studies as 62% secure, 23% avoidant, and 15% anxious/ambivalent). Our results suggest, but of course do not prove, that subjects' choices among the alternatives were nonrandom and may have been determined by some of the same kinds of forces that affect the attachment styles of infants and children. The remainder of the results argue for the validity of subjects' self-classifications. Differences in love experiences. The second hypothesis predicted that subjects with different self-designated attachment styles would differ in the way they characterized their most important love relationship. Table 3 presents the mean subscale scores (each with a possible range of I to 4) for each attachment type, along with the F ratio from a one-way analysis of variance (ANOVA) on scores for each subscale. In line with the hypothesis, secure lovers described their most LOVE CONCEPTUALIZED AS AN ATTACHMENT PROCESS 515 Table 2 Adult Attachment Types and Their Frequencies (Newspaper Sample) Question: Which ofthe followingbest describesyour feelings? Answers andpercentages." Secure (N = 319, 56%): I find it relativelyeasyto get closeto others and am comfortabledepending on them and havingthem depend on me. I don't often worry about beingabandoned or about someone gettingtoo closeto me. Avoidant (N = 145, 25% ): I am somewhatuncomfortable beingclose to others;I findit difficultto trust them completely,difficultto allow myselfto depend on them. I am nervous when anyone getstoo close, and often, lovepartners want meto be more intimatethan I feel comfortable being. Anxious/Ambivalent (N = 110, 19%): I findthat others are reluctant to get ascloseas I wouldlike.I often worry that my partnerdoesn't reallyloveme or won'twant to staywith me. I want to merge completelywith another person,and this desiresometimesscares peopleaway. Note. Twenty-one subjects failed to answer this question, and 25 checkedmore than one answeralternative. important love experience as especially happy, friendly, and trusting. They emphasized being able to accept and support their partner despite the partner's faults. Moreover, their relationships tended to endure longer: 10.02 years, on the average, compared with 4.86 years for the anxious/ambivalentsubjects and 5.97 years for the avoidant subjects,/7(2, 568) = 15.89, p < .001. This was the case even though members of all three groups were 36 years old on the average. Only 6% of the secure group had been divorced, compared with 10% of the anxious/ ambivalent group and 12% of the avoidant group, F(2, 573) = 3.36, p < .05. The avoidant lovers were characterized by fear of intimacy, emotional highs and lows, and jealousy. They never produced the highest mean on a positive love-experiencedimension. The anxious/ambivalentsubjects experienced love as involvingobsession, desire for reciprocation and union, emotional highs and lows, and extreme sexual attraction and jealousy. They provided a close fit to Tennov's(1979) description of limerence and Hindy and Schwarz's (1984) conception of anxious romantic attachment, suggestingthat the differencebetween what Tennov called love and limerence is the difference between secure and anxious/ambivalentattachment. Although the average love experiences of people in the three different attachment categories differed significantly, for most of the subscales all three types scored on the same side of the midpoint (2.50), emotional extremes and jealousy being the only exceptions. Thus, there appears to be a core experience of romantic love shared by all three types, with differences in emphasis and patterning between the types. The results also support the ideas that love is a multidimensionalphenomenon and that individuals differ in more ways than the intensity of their love experiences. Especially noteworthy was the fact that the ordering of means for the different attachment styles differed for different dimensions. For the dimensions of happiness, friendship, trust, and fear of closeness, secure subjects differed significantly from avoidant and anxious/ambivalent subjects but these two insecure groups did not differ from each other. On the dimensionsof obsessive preoccupation, sexual attraction, desire for union, desire for reciprocation, and love at first sight, anxious/ambivalent subjects differed significantly from avoidant and secure subjects, who did not differ from each other. On the acceptance dimension,avoidant subjects (the least accepting) differed from anxious/ambivalent and secure subjects, and on emotional extremes and jealousy, all three groups were statistically distinct. This variety of patterns supports the claim that there are three different love styles, not simply three points along a love continuum. Differences in mental models. We attempted to assess what Bowlby (1969) called working models of relationships by using the items shown in Table 4. Each was either checked or not checked as describing how the subject generally "view[s] the course of romantic love over time" These dichotomous answers were analyzed by attachment style, using a one-way ANOVA. (Because the answers were scored as either 0 or 1, the means can be read as proportions.) In line with the third hypothesis, secure lovers said that romantic feelings wax and wane but at times reach the intensity experienced at the start of the relationship and that in some relationships romantic love never fades. The avoidant lovers said the kind of head-over-heels romantic love depicted in novels and movies does not exist in real life, romantic love seldom lasts, and it is rare to find a person one can really fall in love with. The anxious/ambivalent subjects claimed that it is easy to fall in love and that they frequently feel themselves beginningto fall, although (like the avoidant subjects) they rarely find what they would call real love. Like the secure subjects, the anxious/ambivalentsubjects said they believethat romantic feelingswax and wane over the course of a relationship. Differences in attachment history. Attachment history with parents was assessed in two ways. Subjects were asked whether Table 3 Love-Subscale Means for the Three Attachment Types (Newspaper Sample) Anxious/ Scalename Avoidant ambivalent Secure F(2, 571) Happiness 3.19a 3.31. 3.51b 14.21"** Friendship 3.18, 3.19, 3.50b 22.96*** Trust 3.1 la 3.13, 3.43b 16.21"** Fear ofcloseness 2.30. 2.15, 1.88b 22.65*** Acceptance 2.86, 3.03b 3.01b 4.66** Emotional extremes 2.75. 3.05b 2.36¢ 27.54*** Jealousy 2.57a 2.88b 2.17c 43.91"** Obsessive preoccupation 3.01a 3.29b 3.01, 9.47*** Sexualattraction 3.27, 3.43b 3.27, 4.08* Desire for union 2.81~ 3.25b 2.69, 22.67*** Desire for reciprocation 3.24a 3.55b 3.22, 14.90"** Loveat firstsight 2.91~ 3.17b 2.97, 6.00** Note. Withineach row,means with differentsubscriptsdifferat the .05 levelofsignificanceaccordingto a Scheff~test. * p < .05. **p < .01. ***p < .001. 516 CINDY HAZAN AND PHILLIP SHAVER Table 4 Proportion ofRespondents Who Endorsed Each Mental-Model Statement About Love (NewspaperSample) Anxious/ Statement Avoidant ambivalent Secure F(2,571) I. The kind ofhead-overheelsromantic love depicted in novelsand movies doesn'texistin real fife. .25~ .28a .13b 8.81"** 2. Intense romantic love iscommon at the beginningofa relationship,but it rarely lastsforever. .41, .34~ .28b 3.83* 3. Romantic feelingswax and waneoverthe courseofa relationship,but at times they can be as intense asthey wereat the start. .60, .75b .79b 9.86*** 4. In somerelationships, romantic lovereally lasts;it doesn'tfade with time. .41, .46a .59b 7.48*** 5. MostofUScould love manydifferentpeople equallywell;there is no "one true love" which is "meant to be" .39 .36 .40 ns 6. It'seasyto fallin love. I feelmyselfbeginning to fallin loveoften. .04, .20b .09, 9.33*** 7. It's rare to find someone you can reallyfallin lovewith. .66, .56, .43b 11.61"** Note.Withineach row,means with differentsubscriptsdifferat the .05 levelofsignificanceaccordingto a Scheff6test. * p < .05. **p < .01. ***p < .001. they had ever been separated from either parent for "what seemed like a longtime" and whether the parents ever separated or divorced. They were also asked to describe how each parent had generallybehaved toward them during childhood (using 37 adjectives, such as responsive,caring, critical, and intrusive, derived from a pilot study in which subjects answered open-ended questions about their childhood relationshipswith parents) and the parents' relationshipwith each other (using 12 similarlyderived adjectives such as affectionate, unhappy, and argumenta- tive). There were no significantdifferences among the three attachment types in likelihood or duration of separation from parents during childhood, even when analyzedby reason for separation. In addition, parental divorce seemed unrelated to attachment type, even though quality of relationships with parents was associated with type. The best predictors of adult attachment type were respondents' perceptions of the quality of their relationship with each parent and the parents' relationship with each other. A one-way ANOVA,with attachment style as the independent variable, on each of the 86 child-parent and parent-parent relationship variables yielded 51 Fs that were sitmificant at the .05 level, clearly more than expected by chance. (Thirty-seven of these were significantat the .01 level; 15 were significant at the .001 level.) Because many of the variables were correlated, which meant that many of the ANOVAresults were redundant, a hierarchical discriminant-functionanalysiswas performed to assess predictability of membership in the three attachment categories from a combinationof attachment-historyvariables. Subjects with no missing data on the variables involved (N -- 506) were included in the analysis. The 22 attachment-history variables shown in Table 5 (plus one with a correlation below .20) were retained as significantpredictors of attachment type. Both discriminantfunctions(two being the maximum possible number given three target groups) were statistically significant, with a combined x2(46, N= 506) = 131.16, p < .001. After removal of the first function, x2(22, N = 506) was 40.94 (p < .01). The two functions accounted for 69.87% and 30.13%, respectively, of the between-groupsvariability. As shown in Figure 1, the first discriminant function separated secure subjects from the two kinds of insecure subjects. The second function separated avoidant from anxious/ambivalent subjects. Together, the two functions correctly classified 56% of the avoidant subjects, 51% of the anxious/ambivalent subjects, and 58% of the secure subjects. (The incorrectly classified subjects were distributed fairly evenlyacross the remaining categories.) Correlations of the 22 predictor variables with the two disTable 5 Significant Correlations Between Attachment-History Variables and Discriminant Functions (NewspaperSample) Function Function Variable 1 2 Affectionateparental relationship .44* Respectfulmother .43* .22 Intrusive mother -.42* Caring father .41" Demanding mother -.40" Lovingfather .40* .25 Humorous father .40* Confidentmother .35" Unhappy parental relationship -.34* .24 Acceptingmother .33* Caring parental relationship .32* Responsiblemother .31" Affectionatefather .30* .26 Sympathetic father .28* Strongmother .28* Disinterestedmother -.28* Unresponsivefather -.24* Unfair father -.20 .47* Humorous mother .43* Likablemother .38* Respectedmother .30 .37* Rejectingmother -.27 -.30" Note. Correlations marked with an asterisk in the first column correlated more highlywith Function 1than with Function2; the reverseis true in the secondcolumn. LOVE CONCEPTUALIZED AS AN ATI'ACHMENT PROCESS 517 t= _o 3 I: O E= E k. U "O e- o I * AnxlouslAmbivalent * Avoldant Secure * -.6 -.4 -.2 0 .2 .4 .6 First discriminant function Figure 1. Plot of three group centroids on two discriminant functions derived from attachment-historyvariables(newspaper sanaple). criminant functions are shown in Table 5. The best discriminators between secure and insecure subjects included (a) a relationship between parents that was affectionate (r = .44), caring (.32), and not unhappy (-.34); (b) a mother who was respectful of the subject (.43), confident (.35), accepting (.33), responsible (.31), not intrusive (-.42), and not demanding (-.40), among other qualities; and (c) a father who was, among other things, earing (.41), loving (.40), humorous (.40), and affectionate (.30). The top discriminators between avoidant and anxious/ambivalent groups, with positively correlated variables being those named more frequently by anxious/ambivalent subjects, ineluded (a) no parental relationship variables; (b) a mother who was relatively humorous (.43), likable (.38), respected (.37), and not rejecting (-.30); and (c) a father who was relatively unfair (.47). These results can be summarized by saying that secure subjects, in comparison with insecure subjects, reported warmer relationships with both parents and between their two parents. Avoidant subjects, in comparison with anxious/ambivalent subjects, described their mothers as cold and rejecting. Anxious/ambivalent subjects saw their fathers as unfair. Both sets of correlations are compatible with expectations based on Ainsworth et al.'s (1978) studies ofinfant-caregiver attachment. Sex differences and similarities. There were a few significant sex differences on individual items. Most notably, respondents tended to describe their opposite-sex parent more favorably than their same-sex parent. For example, 62% of the women (vs. 44% of the men) described their fathers as loving, t(563) = 4.16, p < .00 l, and 78% of the men (vs. 69% of the women) described their mothers as loving, t(614) = 2.36, p < .05. This same pattern was found for the adjectives affectionate and understanding. Moreover, on negative trait dimensions, respondents tended to judge their same-sex parent more harshly. For instance, 39% of the women, but only 27% of the men, described their mothers as critical, t(614) = 2.91, p < .01. When reporting about their fathers, on the other hand, 53% of the men chose critical compared with 39% of the women, t(563) = 3.06, p < .01. The same was true for demanding. There were no significant sex differences in prevalence of the three attachment styles and only small differences on two of the love dimensions: Men agreed slightly more than women did with the sexual-attraction items (3.35 vs. 3.26), t(618) = 1.99, p < .05, and also reported greater desire for union (2.94 vs. 2.78), t(616) -- 2.45, p < .05. Overall, what stood out was the marked similarity of the results for men and women. Study 2 Method Study 1 suffered from several limitations that made it desirable to conduct a conceptual replication. First, the newspaper sample might have been biased because of self-selection. This could have affected our estimate of the prevalence of each of the three attachment types and distorted other results in unanticipated and undetectable ways. It seemed wise, therefore, to test a non-self-selected college-student group in our second study, students beingthe usual subjects in social psychological research. Second, Study I examined onlylimited aspects of subjects' mental models. An interesting part of Bowlby's (1969) analysis was the claim that these models involve complementary portrayals of selfand relationships. In Study 1,because of space limitationsimposed by newspaper editors, we neglected the self side of subjects' mental models; in Study 2 we focused on them. Third, because previous research on loneliness(e.g., Rubenstein& Shaver, 1982) has linkedloneliness to attachment history without using the attachment-classification item designed for our research on romantic love, we decided to include in Study 2 briefmeasures of state and trait loneliness(Shave~ Furman, & Buhrmester, 1985). The hypotheses were the same as in Study 1, but Hypotheses 4 and 5 were especially important in Study 2 because new self-model items and measures oflonelinesswere included. Subjects. One hundredeight undergraduates(38 men and 70 women) who were enrolled in a course entitled UnderstandingHuman Conflict completed the questionnaire as a class exercise. Approximately three fourths of the students were first-quarter freshmen; the mean age was 18 years. Measures and procedure. As in Study 1, subjects were asked to describe their most important love relationship in terms of 56 agree-