17
Discriminant Validity of the Adult Attachment Interview Judith A. Crowell, Everett Waters, Dominique Treboux, Elizabeth O'Connor, Christina Colon-Downs, and Olga Feider State University of New York at Stony Brook Barbara Golby University of WisconsinMadison German Posada University of Denver CROWELL, JUDITH A.; WATERS, EVERETT; THEBOUX, DOMINIQUE; O'CONNOR, ELIZABETH; COLON- DOWNS, CHRISTINA; FEIDER, OLGA; COLBY, BARBARA; and POSADA, CERMAN. Discriminant Validity of the Adult Attachment Interview. CHILD DEVELOPMENT, 1996,67,2584-2599. The Adult Attach- ment Interview is a semi-structured interview developed to investigate adults' attachment repre- sentations. Subjects are asked to describe their parents as caregivers, explain these descriptions, describe how their parents typically responded to distress, and discuss their current relationships with their parents. They are also asked to describe any significant losses and/or instances of abuse during childhood. Scoring focuses on the accessibility of early experiences to memory and the coherence and plausibility of the subject's narrative. Discriminant validity is always an important issue with such measures because IQ and other cognitively loaded variables offer plausible alternative interpretations or represent important correlates fiat should be treated as covariates when the measure is used. In addition, complex, multifaceted interviews always pose the risk of assessing general social adjustment rather than a more narrowly defined construct. This study examines the discriminant validity of the AAI vis-^-vis intelligence, social desirability, discourse style, and general social adjustment in a sample of 53 native-English-speaking, married women with preschool children. They were assessed with the AAI, a written IQ test, the Social Adjustment Scale, the Employment Experience Interview (discourse style), and a measure of social desirability. There were modest but significant correlations with IQ scores and social adjustment. There was no relation between AAI classifications and discourse style or so- cial desirability. These results substantially strengthen the case for interpreting the AAI as an attachment-related measure. Attacbment theorists use the term as caregivers, explain these descriptions, de- "working model" to refer to mental repre- scribe how their parents typically responded sentations or belief systems derived from at- to distress, and discuss their own current re- tachment-related experiences. Such repre- lationships with their parents. They are also sentations are said to organize attachment asked to describe any significant losses and/ relevant feelings, experiences, and thoughts or instances of abuse during childhood, and to guide appraisal processes and behav- Scoring focuses on tbe accessibility of early ioral responses to new experiences (Rrether- experiences to memory and the coherence ton, 1985). The Adult Attachment Interview and plausibility of the subject's narrative. Al- (AAI; George, Kaplan, & Main, 1985) is a though the language and discourse style semi-structured interview developed to in- elicited in the AAI do not specify the content vestigate adults' attachment representations, of an adult's attachment working model, ac- Subjects are asked to describe their parents cessibility of early experiences, coberence. This study was funded by a grant from the National Institutes of Mental Health (ROl-MH- 4493501). We would like to tfiank our research assistants, Melissa Fleischman, Yuan Gao, and Helen Pan, and our typists, Naomi Weiner and Maryann Hannon. Special thanks go to Margaret Gonnelly for her help and support in all aspects of' the study. Requests for reprints should be addressed to the authors at Department of Psychology, SUNY-Stony Brook, NY, 11794-2500. [Child Development, 1996,67,2584-2599. © 1996 by the Society for Research in Child Development, Inc. All rights reserved. 0009-3920/96/6705-0043$01.00]

Discriminant Validity of the Adult Attachment Interview ... · 1985; Posada, Waters, & Crowell, 1995; van IJzendoorn, 1992) have proceeded far in ad-vance of detailed studies on its

  • Upload
    others

  • View
    2

  • Download
    0

Embed Size (px)

Citation preview

Discriminant Validity of the Adult AttachmentInterview

Judith A. Crowell, Everett Waters, Dominique Treboux,Elizabeth O'Connor, Christina Colon-Downs,and Olga FeiderState University of New York at Stony Brook

Barbara GolbyUniversity of Wisconsin—Madison

German PosadaUniversity of Denver

CROWELL, JUDITH A.; WATERS, EVERETT; THEBOUX, DOMINIQUE; O'CONNOR, ELIZABETH; COLON-DOWNS, CHRISTINA; FEIDER, OLGA; COLBY, BARBARA; and POSADA, CERMAN. Discriminant Validityof the Adult Attachment Interview. CHILD DEVELOPMENT, 1996,67,2584-2599. The Adult Attach-ment Interview is a semi-structured interview developed to investigate adults' attachment repre-sentations. Subjects are asked to describe their parents as caregivers, explain these descriptions,describe how their parents typically responded to distress, and discuss their current relationshipswith their parents. They are also asked to describe any significant losses and/or instances ofabuse during childhood. Scoring focuses on the accessibility of early experiences to memoryand the coherence and plausibility of the subject's narrative. Discriminant validity is always animportant issue with such measures because IQ and other cognitively loaded variables offerplausible alternative interpretations or represent important correlates fiat should be treated ascovariates when the measure is used. In addition, complex, multifaceted interviews always posethe risk of assessing general social adjustment rather than a more narrowly defined construct.This study examines the discriminant validity of the AAI vis- -vis intelligence, social desirability,discourse style, and general social adjustment in a sample of 53 native-English-speaking, marriedwomen with preschool children. They were assessed with the AAI, a written IQ test, the SocialAdjustment Scale, the Employment Experience Interview (discourse style), and a measure ofsocial desirability. There were modest but significant correlations with IQ scores and socialadjustment. There was no relation between AAI classifications and discourse style or so-cial desirability. These results substantially strengthen the case for interpreting the AAI as anattachment-related measure.

Attacbment theorists use the term as caregivers, explain these descriptions, de-"working model" to refer to mental repre- scribe how their parents typically respondedsentations or belief systems derived from at- to distress, and discuss their own current re-tachment-related experiences. Such repre- lationships with their parents. They are alsosentations are said to organize attachment asked to describe any significant losses and/relevant feelings, experiences, and thoughts or instances of abuse during childhood,and to guide appraisal processes and behav- Scoring focuses on tbe accessibility of earlyioral responses to new experiences (Rrether- experiences to memory and the coherenceton, 1985). The Adult Attachment Interview and plausibility of the subject's narrative. Al-(AAI; George, Kaplan, & Main, 1985) is a though the language and discourse stylesemi-structured interview developed to in- elicited in the AAI do not specify the contentvestigate adults' attachment representations, of an adult's attachment working model, ac-Subjects are asked to describe their parents cessibility of early experiences, coberence.

This study was funded by a grant from the National Institutes of Mental Health (ROl-MH-4493501). We would like to tfiank our research assistants, Melissa Fleischman, Yuan Gao, andHelen Pan, and our typists, Naomi Weiner and Maryann Hannon. Special thanks go to MargaretGonnelly for her help and support in all aspects of' the study. Requests for reprints should beaddressed to the authors at Department of Psychology, SUNY-Stony Brook, NY, 11794-2500.

[Child Development, 1996,67,2584-2599. © 1996 by the Society for Research in Child Development, Inc.All rights reserved. 0009-3920/96/6705-0043$01.00]

and plausibility are assumed to refiect theinternal consistency, extensibility, and test-ability of the underlying belief system(Bowlby, 1973, 1980, 1982; Bretherton,1985).

Discriminant validity is always an im-portant issue with such measures becauseIQ and other cognitively loaded variables of-fer plausible alternative interpretations orrepresent important correlates that shouldbe treated as covariates wben the measureis used. In addition, complex, multifacetedinterviews always pose the risk of assessinggeneral social adjustment rather than a morenarrowly defined construct. As with theStrange Situation in the mid 1970s, studiesof the AAI's reliability and stability(Bakermans-Kranenburg & van IJzendoorn,1993; Benoit & Parker, 1994; Crowell, Wa-ters, Treboux, O'Connor, & Colon-Downs,1995; Fonagy, Steele, & Steele, 1992; Sagiet al., 1994) and convergent validity (Ains-worth & Eichberg, 1991; Crowell & Feld-man, 1988, 1991; Fonagy etal., 1992; Cross-mann, Fremmer-Bombik, Rudolph, &Crossmann, 1988; Main, Kaplan, & Cassidy,1985; Posada, Waters, & Crowell, 1995; vanIJzendoorn, 1992) have proceeded far in ad-vance of detailed studies on its discriminantvalidity. To date, only Bakermans-Kianenburg and van IJzendoorn (1993) andSagi et al. (1994) have examined the discrim-inant validity of the AAI, both in non-English-speaking samples. The presentstudy examines the discriminant validity ofAJM classifications and scale scores vis-a-visintelligence, social desirability, discoursestyle, and general social adjustment in asample of native-English-speaking marriedwomen with children.

Issues in the Interpretation of AAIClassifications

Intelligence.—The first discriminantquestion addressed is whether the AAI issignificantly infiuenced by intelligence orverbal ability. A subject's ability to speak inao organized, coherent, thoughtful, free-fiowing manner about early experienceswith parents is a key factor in determiningattachment classification (Main & Goldwyn,1994). Hauser (1976, 1992) and Borman, Al-len, Cole, and Hauser (1996) have reportedthat AAI discourse coherency is related toeducational level and also to ego develop-ment, a construct known to correlate signifi-cantly with intelligence (Hauser, 1976). Incontrast, several studies have found that AAIclassifications and scale scores are unrelatedto intelligence tests (Bakermans-Kranenburg

Crowell et al. 2585

& van IJzendoorn, 1993; Rosenstein & Horo-witz, 1993; Sagi et al., 1994; Ward, Botyan-ski, Plunket, & Carson, 1991).

Tbese findings suggest that the relationbetween tbe AAI and intelligence needs fur-ther investigation. For example, the coher-ency score and the classification system, al-though highly related, may have differentcorrelates. There are also circumstances inwhich intelligence or education may play arole in attachment security, for example, inthose individuals termed "earned secure"who have been able to overcome the effectsof negative childhood experiences with par-ents and are scored as secure based upontheir high coherence scores (Liu, Colon-Downs, Lord, Wang, & Crowell, 1995; Mas-ten, Best, & Garmezy, 1990; Pearson, Cohn,Cowan, & Cowan, in press).

This is an important issue because thecomplexity of the AAI and the emphasis onlanguage and cognition in the scoring sys-tem lend themselves to intrusion of intelli-gence-related variance. In addition, many ofthe stability results and competence relatedcorrelates ofthe AAI are consistent with this.A strong relation between AAI scoring andIQ would raise serious questions about theattachment-relatedness of the AAI. Modestcorrelations would suggest measuring IQ asa covariate in research with the AAI.

Trait-like discourse style.—The classi-fications of the AAI are determined by care-ful examination ofthe adult's discourse styleon the topic of childhood attachment experi-ences. It is important to determine if the dis-course style or manner of speech scored inAAI represents a broad cognitive style. Spe-cifically, does inability to speak coherently,clearly, and consistently in an AAI interviewreflect problems in an underlying attacb-ment working model or simply an inabilityto speak well about any topic? Our strategyfor assessing this is to conduct an interviewabout employment experience during ado-lescence and early adulthood and score thetranscripts using coherence, clarity, and con-sistency criteria that parallel standard AAIscoring.

Social desirability.—It has long beenknown that there are reliable trait-like indi-vidual differences in tbe extent to whichsubjects present themselves in a positivelight iri self-report questionnaires and inter-views. The AAI addresses a wide range ofself-related issues that have both personaland social significance to the subject and thesubject's appearance to the interviewer.

2586 Child Development

Thus intrusion of social desirability varianceinto the interview seems likely. The issue,of course, is whether such variance infiu-ences the AAI scoring system.

It is possible that this issue has receivedlittle attention in previous research (seeBakermans-Kranenburg & van IJzendoorn,1993) because attachment theorists considerworking models to exist outside of con-sciousness (Bowlby, 1988) and because theAAI is specifically intended to "surprise theunconscious" (George et al., 1985). Re-sponses to the interview are taken to bespontaneous and refiect cognitive constructsoperating out of conscious awareness andnot calculated to make a specific impact onthe interviewer. Nonetheless, because socialdesirability responding is both reliable andstable and has a wide range of competence-related correlates, many of the stability re-sults and competence-related correlates ofthe AAI lend themselves to a social desir-ability interpretation. We address this issueby including a standard self-report measureof" social desirability response set in our as-sessment.

Social adjustment.—The AAI delvesdeeply into family relationships in order toassess the subjects' mental representationsand expectations of primary attachment fig-ures. Both common sense and a wide rangeof research attest to the negative effects ofpsychopathology on family relationships.This raises the possibility that some of theinformation elicited in the AAI refiectsbroadly defined problems in social adjust-ment rather than specifically attachment-related problems.

A primary function of early attachmentrelationship is to support social explorationsand acquisition of social skills and experi-ence (Allen & Crowell, 1995; Bretherton,1985; Waters & Sroufe, 1983). A wide rangeof studies has demonstrated that early at-tachment security is empirically related tosocial competence and social adjustment inchildhood and adolescence (Elicker, En-glund, & Sroufe, 1992; Kerns, 1995; Waters,Wippman, & Sroufe, 1979). Thus a small tomoderate relation between attachment secu-rity and social adjustment in a nonclinicsample with ordinary life circumstances canbe anticipated on theoretical and empiricalgrounds. Nonetheless, neither skill deficitsand adjustment problems in nonclinic sam-ples nor the social behavior correlates ofclinical problems in childhood are viewedas primarily disorders of attachment. Thus, a

strong relation between the AAI and generalsocial adjustment would raise a serious issueof discriminant validity. As with IQ and so-cial desirability, individual differences insocial adjustment are reliable and stableand, by definition, have a wide range of com-petence-related correlates. Thus they, too,might offer an alternative interpretation orcontribute significantly to important AAI re-sults. We address this issue by including inour assessments an interview measure of so-cial adjustment.

In summary, the goal of this study is todetermine whether the AAI is related to in-telligence, discourse style, social desirabil-ity, and/or social adjustment in ways thatwould affect our understanding and inter-pretation of past studies or the use of theAAI in the future.

MethodSubjects

Participants were 53 mothers of pre-school children. The children did not partic-ipate in this study. One woman was Asian-American, the remainder were White. Theirmean age was 34.5 years (range 25.4—43.5years). Using Hollingshead's (1975) ratings,they were from middle- to upper-middle in-come families, and their median occupa-tional status (based upon tbe women's mostrecent employment outside of the home)was a score of 6 (e.g., technician, semi-professional, or small business owner), aswas that of their spouses. Forty-nine of thewomen were married with their husbandsliving at home, and on average they had twochildren (range = 1-5). The average educa-tion level was 14 years; 26 of the participantsreported working outside the home at leastpart-time at the time of the study (n = 11part-time, n = 15 full time). The sample iscomparable to other samples of nonclinicalmothers in which the AAI has been used(van IJzendoorn & Bakermans-Kranenburg,1996).

MeasuresAssessment of adult attachment repre-

sentations: Adult Attachment Interview(AAI; George et al, 1985).—The AAI is a45-100-min interview. The subject is askedto give an overview of the childhood rela-tionship with her parents and to provide setsof five adjectives which describe her child-hood relationship with each parent. She isthen asked to cite incidents or experiencesfrom childhood that illustrate or explain thechoice of each adjective. The subject is

Crowell et al. 2587

asked about feelings of rejection, experi-ences of being upset, ill, and hurt, about sep-arations, losses, and abuse. The subject isalso asked to discuss cbanges in the relation-ships with her parents since childhood, todescribe her current relationship with herparents, and to explain her understanding oftheir behavior. Finally, the subject is askedabout the effects of early childhood experi-ences on her adult personality and parentingas well as her concerns and hopes for herown children. In scoring the interviews, spe-cial attention is paid to the coherence andconsistency between the general descrip-tions and explanations of the relationshipsand specific examples of behavior and expe-riences.

The interviews were scored from typedtranscripts from which all identifying namesand places were deleted (Main & Goldwyn,1991). Transcripts are scored on a set ofnine-point scales which summarize the scor-ers' evaluations of the subject's childhoodexperiences with each parent (loving behav-ior, rejection, involving/role reversing be-havior, neglect, and pressure to achieve) andthe subject's present state of mind witb re-spect to attachment (discourse coherence oftranscript, believability, amount of informa-tion, relevance to topic, manner, idealizationof eacb parent, stated lack of recall, currentpreoccupying anger toward each parent, der-ogation of each parent and of attachment,passivity of speech, and fear of loss of childthrough death). The state of mind variablescontribute to an overall discourse coherencerating which is closely related to the sub-ject's classification as secure or insecurewith respect to attachment (Ffyfe & Waters,1996). For this reason, the discriminant va-lidity ofthe discourse coherence scale is ex-amined along with the attachment classifi-cations in the present analyses.

Based on the overall pattern ofthe inter-view and guided by scale ratings, tbe subjectis assigned to one of tbree classifications: (1)Secure/autonomous witb respect to attach-ment, (2) Dismissing of attachment, or (3)Preoccupied with attachment. In addition tothe primary attachment classification, sub-jects scoring 6 or more on a nine-point Unre-

solved Loss scale or Unresolved Traumascale are designated Unresolved. Table 1summarizes the criteria for each attachmentclassification.

Scoring was conducted by coderstrained by Mary Main and Eric Hesse (JCand DT), and interrater agreement was as-sessed using 23 of 53 transcripts (43%).Agreement for this sample on the three ma-jor attachment classifications was 81%, k =.64, p < .001, and 78% for four classifica-tions, k = .67, p < .001. Rater agreement forthe discourse coherence scale was r — .86,p < .001.

Disagreements between coders were re-solved by conference. The distribution ofthree classifications was secure, 58% (n =31); dismissing, 30% (n = 16); preoccupied,11% (n = 6), and for four classifications wassecure, 51% (n = 27); dismissing, 26% (n =14); preoccupied, 8% (n = 4), and unre-solved, 15% (n = 8). This distribution is sim-ilar to that reported in a meta-analysis of pre-vious studies using the AAI (van IJzendoorn& Bakermans-Kranenburg, 1996).

Intelligence: The Henmon-Nelson Testof Mental Ability (Lamke 6- Nelson,1973).—This measure is a timed, paper-and-pencil measure of general intellectual abil-ity. It is an established measure, most re-cently standardized in 1973, which yields asingle score (for reviews, see Buros, 1965;Jensen, 1980; Thorndike, Cunningham,Thomdike, & Hagen, 1991). This 90-itemmultiple-choice test includes vocabulary,patterns of number sequences, and analo-gies and can be administered in less than 20min. Alpha reliabilities range between r =.85 and r = .95 (Thorndike et al., 1991).Henmon-Nelson scores correlate well withother group test assessments of intelligence,with grades and achievement tests, and theIQ score can be used as a basis for estimatingWAIS Full Scale IQ scores (Kling, Davis, &Knost, 1978; Sternberg, Conway, Ketron, &Bernstein, 1981; Tborndike et al., 1991).

Subjects were allowed 15 min to workon the test, and raw scores were used in theanalyses.^ The range of items completed

^ Speed tests are administered to prevent the achievement of perfect scores; hence, reducingthe time allowed for the test makes it more difficult to achieve a perfect score (Anastasi, 1982).Reducing the time of a test may lower the scores of certain individuals who require more time(personal speed is independent of intelligence, Jensen, 1980). However, timing is not a highlycritical factor (subject's rank is largely unaffected) in tests where "the items are evenly gradedin difficulty, have plenty of 'top' items, . . . 'fast' subjects can finish although they reach theirdifficulty ceiling before the end of the test"; furthermore, the test should not be so short that

o

Io

"o a .1 "3 .S "S I i

I OH

-0

O

ft

S 5);53

ft-

S

g .-S

0)13 O 0^ '

1a ID .S

a

IZ;Id

a

aQ

oII

• i i2 -3

s ift

<H-i

o

ao

o>ft

a „II

-s -a

I

i(Zl

u

o

OH

Crowell et al. 2589

within the 15-min time frame was 30—90items, with a median number of 62 itemscompleted and a mean of 63.6. The range ofitems correctly answered (raw scores) was20-85, with a median of 50 items correct anda mean of 52.8.

Discourse style: The Employment Expe-rience Interview (EEI).—The EmploymentExperience Interview (Crowell, Treboux, &Waters, 1992) is a semi-structured interviewdeveloped for this study. It covers the sub-jects' opinions and conceptualizations oftheir education and employment history.The purpose of the interview was to exam-ine subjects' discourse style using a formatand scoring system similar to that ofthe AAIbut concerning a domain theoretically unre-lated to attachment.

Subjects were asked a series of ques-tions regarding their educational and em-ployment history, for specific memories ofemployment-related events, and about themeaning and importance of work in theirlives. They were also asked about periodsout of work, job loss, jobs for which theywere most and least suited, and their futuregoals. As much as possible, the interviewavoided eliciting discussion of interpersonalrelationships in the workplace so as to mini-mize overlap of topic with the AAI. Home-mEiking was considered a form of employ-ment. When discussing homemaking in theinterview, the interviewer maintained thefocus on activities rather than discussingspecific characteristics of the relationshipswith husband and children, again with thepurpose of keeping content distinct from theA/J.

The interviews were transcribed and, aswith the AAI, all identifying names andplaces were deleted from the typed copy.Eaich interview was rated on 13 five-pointscales (Crowell et al., 1992). Seven scalesreflect feelings and attitudes about educa-tional/work related experiences: balance be-tween work and home, investment in work,organized approach to employment-relateddecisions, satisfaction with the current worksituation, realistic evaluation of perfor-mance, sense of achievement or personalgrowth, and perceived employment-relatedpressures. Seven scales were derived fromthe AAI scoring system' and refiect discoursestj'le: coherency of transcript, insistence

upon inability to recall events, active angerexpressed, passivity of discourse/thought,idealization, and disorganized or disorientedresponses to work-related traiuma.

Based on the overall pattern ofthe inter-view and guided by scale ratings, subjectswere assigned to one of four groups: (1) Inte-grated about work, (2) Overinvolved withwork or overemphasis on employment-related experiences, (3) Uninvolved withwork, and (4) Aimlessness/disorganizationin the approach to work. Subjects classifiedas Integrated were characterized by theirability to put forward a coherent, organizeddiscussion about their work-related experi-ences. They openly discussed the balancebetween work and home life, had primarilypositive work experiences, but any dissatis-faction or work-related problem was clearlyand directly discussed. Subjects classified asOverinvolved were characterized by a pre-occupation with their job or some aspect ofit. The subjects tended to have difficultymaintaining a balanced discussion becausethe focus was continuously drawn to the job.They tended to be pressured or worriedabout their job, and their discourse was char-acterized as angry, pressured, or having anenthusiastic overly childish quality. Sub-jects classified as Uninvolved tended to dis-miss or derogate their work or some aspectof it or idealized their jobs as being fine ortypical without evidence for this character-ization. In either case, the subjects had littleapparent investment in their jobs, present-ing past and current jobs as irrelevant or un-important and demonstrating a minimalsense of achievement. The participants clas-sified as Aimless were distinguished by theirmarkedly incoherent or confused inter-views. Inconsistencies were pervasive, andthe subject's discourse and work-related his-tory were difficult to follow. Fifty subjectsreceived EEI classifications (two were un-available for the interview and one inter-view could not be transcribed because of apoor tape recording).

The EEI and AAI transcripts were as-signed different identification numbers inorder to insure blind scoring (two ofthe fourEEI coders were also coders for a subset ofthe AAI transcripts). All coders were un-aware of all other information about the sub-jects. Interrater agreement for four classifi-

die fastest subjects cannot attempt every item (Jensen, 1980, p. 135). The Henmon-Nelson meetsthese criteria (Jensen, 1980), and in this sample, 12 subjects completed 80 or more test items,and seven completed all 90 items.

2590 Child Development

cations on 36% of tbe cases (n = 19) was89%, k = .73, p < .01, for four coders (BO'C,BG, JC, and DT). Classification disagree-ments were resolved by a tbird coder. Thedistribution of the four major classificationswas: Integrated, n = 33 (66%), Over-involved, n = 6 (12%), Uninvolved, n = 10(20%), and Aimless, n = 1 (2%).

Social desirability: The Social Desir-ability Scale (SDS; Crowne h- Marlowe,1964).—The SDS assesses an individual'stendency to bias self-reports in a way thatpresents himself or herself in a socially de-sirable light. The test consists of 33 state-ments scored on a seven-point Likert scaleranging from strongly agree to strongly dis-agree. Cronbach's alpha was .58 in tbisstudy, showing moderate internal consis-tency as was found in the sample reportedby Bakermans-iCranenberg and van IJzen-doorn (1993).

Social adjustment: The Social Adjust-ment Scale (SAS; Weissman <b- Paykel,1974).—Tbe SAS is a semi-structured inter-view wbicb was designed for use in an epi-demiological study of depression in women(Weissman & Paykel, 1974). The SAS as-sesses adaptive functioning from excellentto severe impairment regardless of occupa-tion and discriminates subjects in psychiat-ric samples (with both mild and severe dis-orders) from those in general populations,and subjects with mild psychiatric disordersfrom those with severe disorders (Bothwell& Weissman, 1977; Mostow & Newberry,1975; Weissman, Kasl, & Klerman, 1976;Weissman & Paykel, 1974). The global ad-justment score derived from tbe scale hassufficient sensitivity to be used as an assess-ment of therapeutic outcome (Follette, Alex-ander, & Follette, 1991; Resnick &Schnicke, 1992; Watkins et al., 1993).

The SAS assesses functioning in the fol-lowing domains: (1) work/management offinances, (2) social and leisure activities, re-lationships with (3) spouses, (4) children,and (5) extended families, and (6) overall ad-justment. Adjustment in each domain is as-sessed by asking the subject an average offive questions about ber functioning and be-havior in the preceding 2 months within thedomain, for example, specific activities withfriends, arguments with children, instru-mental assistance from extended family orspouse, ability to conduct household chores,and pursuit of^hobbies. As sucb, it is consid-ered a measure of broad social adjustmentand functioning. Each response within a do-

main is scored on a five-point scale by theinterviewer during the interview. Using in-dividual response scores, the subject isgiven an adjustment score for the domain us-ing a seven-point scale. Upon completingthe interview, the interviewer assigns anoverall adjustment score on a seven-pointscale using all the domain scores and theimpression the interviewer has gained of thesubject's global functioning (1 = excellentfunctioning in all areas, 7 = marked or se-vere maladjustment in all areas). The globalscore allows tbe interviewer to take into ac-count information given by the subjectwhich does not readily fit the domains ad-dressed. For example, obsessive/compul-sive behavior is not directly assessed by theinterview, but the interviewer can take a re-port of this bebavior into account in theglobal assessment

The interviews were audiotaped, and 15(28%) were scored independently for in-terrater agreement from the tapes. Theagreement for overall adjustment was r =.76, p < .001, and domain agreement scoreswere as follows: work, r = .55, p = .04, so-cial leisure, r = .65, p = .01, extended fam-ily, r = .78, p = .001, marital, r = .90, p <.001, and parenting, r = .55, p = .04. T'hisrepresents one coder's agreement (CC-D)with the six interviewers (MC, MF, OF, YG,GP, HP) who did the original interviews.

ProcedureParticipants were recruited through lo-

cal preschools and from the community byadvertisement. The subjects were seen ontwo occasions. Most interviews were con-ducted in our laboratory, but a few subjectspreferred to be tested at home. The AAI, theassessments of social desirability, and socialadjustment interview were completed in thefirst session. The Employment ExperiencesInterview and the assessment of intelligencewere administered during a second session.The tasks for each session were selected tobalance tbe amount of time the participantsspent in each session and to minimize anycarryover of discourse style that might resultif the AAI and the EEI were administeredtogether. Because the measures involved arequite complex, the amount of work consider-able in eacb session, and tbe response for-mats of the various measures very diverse,neither the order of the psychometric mea-sures nor the order of AAI and EEI wereconsidered significant factors in the design.The AAI and the EEI were done by differentinterviewers. All interviews were conductedby researcbers fully trained in AAI inter-

Crowell et al. 2591

viewing techniques, and two were formallytrained in the scoring ofthe AAI. All assess-ments were scored blind to the results oftheother measures.

ResultsOne-way analyses of variance (ANO-

VAs) Were used to examine relations of the'four AAI classifications to major demo-graphic and discriminant validity variables.^We also conducted three theoretically sig-nificant planned comparisons: Secure versusAll Insecure (dismissing, preoccupied, unre-solved). Dismissing versus Preoccupied,and Unresolved versus Other Insecure (dis-missing, preoccupied). The discriminantvalidity of the Secure versus Insecure clas-sifications is central to the attachment-rekitedness of AAI classifications. The Dis-missing versus Preoccupied comparisonaddresses the discriminant validity of dis-tinctions among insecure subjects. The Un-resolved versus Other Insecure analyses ad-dresses the attachment-relevance of theUnresolved classification. Because one re-quirement for the Unresolved classificationis loss or trauma, its discriminant validitywith respect to general social adjustment isof particular importance.

Although most research with the AAIhas focused on the discrete attachment clas-sifications (secure, dismissing, preoccupied,and unresolved) and subclassifications,Ffy:fe and Waters (1996) have demonstratedthat Main and Goldwyn's (1994) discoursecoherence scale is highly related to the se-cure versus insecure dichotomy in the dis-crete classifications. The secure versus inse-cure distinction is central to many uses ofthe AAI. In addition, continuous variablesoften provide greater resolution than broadclassifications. They also lend themselves toa wide range of data analytic strategies thatoften have greater statistical power than thenonparametric methods most often usedwith classification data. In light of these con-siderations, we used Pearson correlations toexamine relations of the discourse coher-

ence scale score to the discriminant validityvariables.

Demographic VariablesNo differences were found among AAI

classifications and the demographic vari-ables of age, number of children, work sta-tus, or socioeconomic status.

AAI Classifications, Intelligence, andEducation

Analyses addressed the relation be-tween AAI classifications and educationallevel and intelligence scores. There was nosignificant difference among AAI classifica-tions for years of education or educationallevel attained.

Analyses of variance addressed the rela-tion between attachment classifications andintelligence test scores, first examining theassociation with the number of items com-pleted and then number of items correctlyanswered. No relation was found betweenAAI classification and number of items com-pleted, suggesting that personal speed intest taking did not differ among tbe fourgroups. Nor was any significant differencefound among the four AAI classifications innumber of items correctly completed, that is,raw IQ scores (see Table 2).

Significant results were found for two oftbe planned comparisons. Women classifiedas Secure scored higher than women in theInsecure group. Women classified as Dis-missing scored higher on the intelligencetest than women classified as Preoccupied.Because all of the variance appeared to becarried by the Preoccupied group, analyseswere repeated without the Preoccupied sub-jects. Results found no differences betweenthe Secure and Insecure subjects in rawscores when the Preoccupied subjects wereomitted from analyses. Thus it appears thatthe significant results were because womenin tbe Preoccupied group bad significantlylower IQ scores than women in the otherthree groups. Because the number of Preoc-cupied subjects in this sample is small, thisissue deserves further study. Nonetheless,

^ ANOVAs of the discriminant variables were also conducted using the three major classifi-cations. No differences were found between AAI classifications and the demographic variablesof age, number of children, work status, or socioeconomic status. Likewise, no significant differ-ences were found among the three groups for scores on global social adjustment or social desir-ability. An ANOVA of the three classifications with intelligence scores was significant, F(2, 49)= 3.52, p s .05. The mean score of Secure subjects (n = 30) was 56.2 (SD 14.1), of Dismissingsubjects (n = 16) was 51.9 (SD 16.0), and of Preoccupied subjects (n = 6) was 38.7 (SD 15.1).Post hoc comparisons revealed that the Preoccupied subjects scored significantly lower thansubjects classified as Secure, and there was a trend for them to score lower than the Dismissingsubjects. Ghi-square analysis found no relation between AAI using a three classification systemand the EEI classifications.

eiOin

ao

IQ

I

oo

<S

ao

ao

O

PH

Ua01

I

_e ft

B 01

00

co"CO 1©oq oqi-H

*

3.5

*cq

5.5

en

z'*mCO

en

z'enZ

0) ,-^> 00

a >

'X3o>

s

en

Zen2

CO

zen

ZenZ

SCD

en en

Z Z

enZ

I-H

lO

enZ 00

CO

oias_^_co05 d d '^

p 05 --H oqCO «5 l r j - *i n •—I CO 1—1

m -H -^

_•&• ^

.S01

" 3 ^•go

(D

I .•s s ! VI VI ! a

« + * * *

Crowell et al. 2593

b€!cause the present result is statistically sig-nificant and contradicts results from othersamples (e.g., Bakermans-Kranenburg & vanIJzendoorn, 1993; Rosenstein & Horowitz,1993; Sagi et al., 1994; Ward et al., 1991), werecommend that IQ be included as a covari-ate in research with the AAI until the issueis resolved.

AAI Classifications and Discourse Style inthe Employment Experience Interview

Before examining the relations betweenthe AAI and Employment Experience Inter-view (EEI) classifications, the employmenthistory of the women was examined, as wasthe relation between the EEI and demo-graphic variables and intelligence scores.These analyses were conducted to assesswhether the EEI was tapping separate as-pects of the subjects' lives from the AAI, aswas intended (a particular concern withthose who were currently homemakers). Allofthe subjects had been employed full-timeoutside the home prior to having children,and currently 26 were working part- or full-time outside the home. On average, thewomen had held five jobs outside the home.There was no significant difference in thenumber of jobs beld between subjects whowere currently employed outside the homeand those who were currently homemakers.

The EEI classifications were unrelatedto subjects' age or IQ scores, but there wasa trend for a relation between EEI classifi-cation and average number of years of edu-cation, F(3, 45) = 2.68, p = .058, Integrated= 15.6 years, Overinvolved = 17.3 years,Uninvolved = 13.3 years. Aimless = 13.0years. Post hoc analyses (Tukey honest sig-nificant difference test) revealed that therewas a tendency {p = .09) for subjects classi-fied as Overinvolved to have more years ofeducation than those classified as Unin-

volved. Coherence on the EEI was relatedto the years of education reported by thesubjects, r = .31, p < .05. Tbere was no rela-tion between work status or SES and EEIclassification.

Cohen's kappa was used to examine therelation between the AAI and EEI, usingfour classifications for each interview. Nosignificant relation was found between AAIand EEI classifications for the sample as awhole (see Table 3). In addition, there wasno correspondence between the EEI andAAI in the sample of women who were full-time homemakers. The correlation betweencoherence scores on the two interviews wasnonsignificant (see Table 4). The results sug-gest that the AAI is not measuring a trait-likediscourse style.

AAI Classifications and Social DesirabilityAs predicted, results from ANOVAs and

planned comparisons revealed no significantdifferences in social desirability scoresamong AAI classifications (see Table 2).

AAI Classifications and Social AdjustmentNinety-six percent of the subjects re-

ceived scores of 1, 2, or 3 on the SAS globaladjustment scale, indicating excellent togood social adjustment for the sample as awhole. The ANOVA revealed a significantdifference among the groups in global ad-justment score, and planned comparisonsshowed that women classified as Securewere given significantly better overall ad-justment scores than women classified as In-secure (see Table 2). There were no differ-ences among the insecure groups on overalladjustment scores. Before examining thespecific domains of social adjustment, a MA-NOVA of the domain scores was conductedand found significant, Rao's R{df 15, 119) =2.26, p — .008. Significant differences be-

EEI classifications;

Uninvolved

Aimless

Total

NOTE.—kappa =

Secure

19520

26

.06, N.S.

TABLE 3

AAI AND EEI CORRESPONDENCE

Dismissing

9221

14

AAI CLASSIFICATIONS

Preoccupied

2020

4

Unresolved

3300

6

Total

331061

50

2594 Child Development

tween women classified as secure and thoseclassified as insecure were subsequentlyfound in the domains of work and the ex-tended family (see Table 2). Furthermore,women classified as dismissing received sig-nificantly better scores in functioning atwork and in the extended family than thoseclassified as preoccupied. No differenceswere found between the subjects classifiedas unresolved and those in the other two in-secure groups.

Because intelligence is a possible con-tributing factor in social adjustment, we per-formed an ANCOVA of AAI classificationand the overall social adjustment score withIQ as a covariate. Results indicated the rela-tions between AAI classification and socialadjustment overall and the domain scoreswere not significant when IQ was used asa covariate, F(l, 49) = 2.19, N.S. Plannedcomparisons with IQ score used as a covari-ate found that subjects classified as Securescored significantly bigher than Insecuresubjects in global social adjustment, F(l, 47)= 6.30, p = .02, witb no differences amongthe Insecure subjects. Witb respect to spe-cific domain scores with IQ factored out,subjects classified as Insecure scored sig-nificantly lower tban the Secure group, F(l,47) = 9.73, p = .003, and the Preoccupiedsubjects scored lower than Dismissing sub-jects, F(l, 47) = 5.86, p = .019, in the do-main of work/management of finances.

To examine the size of the effect of at-tachment status on social adjustment, omega

square analyses were conducted. Omegasquare provides a measure of relative treat-ment magnitude which is responsive to thestrength of association between the vari-ables and is independent of sample size(Keppel, 1982). Thus it provides useful in-formation even when an F score is not sig-nificant. Omega square ranges from .00 to1.00, with a "medium" effect being repre-sented by o) = .06, and a "large" effectby (1) = .15 or greater (Keppel, 1982, p.92). The relation between AAI classificationand overall social adjustment score was to= .14, and with intelligence score factoredout, was (0 = .10, indicating a moderaterelation between social adjustment and AAIclassification. These results are consistentwith the predicted relation between attach-ment security and social adjustment. Theyare not strong enough to suggest a generalsocial adjustment interpretation of the AAIwhen it is used in nonclinic samples. Therelation between general social adjustmentand AAI classifications may be very differentin clinical samples.

Discriminant Validity of the AAIDiscourse Coherence Scale

Because the discourse coherence scaleis potentially valuable as a way of scoringadult attachment security as a continuousvariable, we examined relations betweenthis summary scale and the discriminantvariables discussed above (see Table 4).

There was a relation between coher-ence scored from the AAI and subject's age

TARLE 4

CORRELATIONS BETWEEN AAI COHERENCE OF TRANSCRIPT ANDDEPENDENT VARIABLES

Dependent Variables (n = 53)

Demographics:AgeYears of education

Intelligence scoreSocial desirabilityCoherence on EEISocial adjustment:

ClobalWork/iinancial managementSocial/leisureExtended familyMaritalParental

* p £ .05.**p£ .01.*** p s .001.

AAI Coherencer

.32*

.35*

.42**

.10

.10

.46***

.49***

.30*

.29*

.49***

.31*

Crowell et al. 2595

and years of education. There was also amoderate relation between AAI coherenceand raw IQ scores. Interestingly, however,the mean scores for coherency of transcriptdid not differ for subjects classified as Dis-missing (coberence = 3.9) versus Preoccu-pied (coherence = 3.7), suggesting that theability of subjects in the two groups to givean integrated account of childhood wasequally poor. The relation between AAI co-herence and social adjustment scale scoreswas significant for overall social adjustmentand for the domains of social adjustment.There was no relation between AAI coher-ence score and scores of social desirability.

Last, to examine the relation betweendiscourse style on another topic and AAI co-herence, a one-way ANOVA was done usingthe four classifications from the EEI and theAAI coherence scores. No significant rela-tion was found between EEI classificationand AAI coberence. Planned comparisons ofSecure versus Otber, Uninvolved versusOverinvolved, and Aimless versus Unin-volved plus Overinvolved were also not sig-nificant. The correlation between AAI co-herence score and EEI coherence score wasnot significant.

DiscussionIn summary, the relations between the

AAI classifications and assessments of dis-course style and social desirability were notsignificant. Social adjustment did sbow a re-lation to AAI classifications, and, contrary toexpectations, the Preoccupied group dif-fered from the other classifications in theirrav/ scores on the intelligence test. Signifi-cant relations were also found between AAIcoherence of transcript scores and a numberof variables, including subjects' age, yearsof education, intelligence scores, and socialadjustment.

Attachment theory and empirical evi-dence suggest a relation between secure at-tachment and confident exploration and so-cial competence. We were interested in therelation between adult attachment statusand the broad construct of social adjustmentin this sample of women who overallshowed good to excellent functioning, andthus adaptive behavior across a variety of do-mains was assessed. The examination of ef-fect size showed a moderate relation be-tween the AAI classifications and socialadjustment even when the effect of intelli-gence was taken into account. This finding isconsistent with the idea that the attachment

system is related to social competence andadaptation even when a broad definition ofadjustment is used and psychopathology isnot a factor. However, with regard to thequestion of the discriminant validity of theAAI, it appears that the AAI is not equivalentto a measure of social adjustment.

The finding that intelligence scoreswere related to AAI classification was unex-pected and contrasts with findings fromother studies. One possible explanation liesin the assessment of intelligence. The intel-ligence assessment used in the study is awell-validated measure, but it has a heavyemphasis on verbal abilities and can beviewed primarily as a measure of scholasticpotential. Unfortunately, the nature of thetest does not allow for diagnostic break-down. We have used the test in a large sam-ple of engaged couples, and no relation wasfound between intelligence and attacbmentclassification (Treboux, Crowell, & Waters,1994), suggesting a sample specific result.Subjects classified as Preoccupied are rela-tively rare (van IJzendoorn & Bakermans-Kranenburg, 1996), and in some circum-stances, they may be less capable thansubjects in other groups, much as Insecure/resistant children have been found to be lesssocially competent than Secure or Insecure/avoidant children and show the lowest lev-els of mastery motivation (Maslin-Cole &Spieker, 1990; Sroufe, 1983).

A second finding of interest with respectto intelligence is the correlation between co-herence of transcript scores and intelligencescores and years of education. The correla-tion is not great enough to suggest that theAAI or its coherency scale are alternativeways to measure intelligence, but it is notentirely surprising that the ability to speakin an organized way migbt be related to ver-bal intelligence and education (Liu et al.,1995).

As noted above, our findings witb re-spect to intelligence testing contrast withthe published findings of Bakermans-Kranenburg and van IJzendoorn (1993) andSagi and colleagues (1994), who report onDutch and Israeli samples, respectively. Adirect comparison of our study to these twostudies is difficult because of differencesacross samples in assessments of intelli-gence, in reporting of essential informationwhich would facilitate comparisons acrossstudies such as the variability of scores,means, and standard deviations, and finally,in the analyses selected. For example, the

2596 Child Development

other two samples are non-native-Englishspeaking. Intelligence tests in one case werestandardized (Bakermans-Kranenburg & vanIJzendoorn, 1993), but in the other, stan-dardized intelligence tests were not avail-able for the population so scores on tests foradmission to university were used in analy-ses (Sagi et al., 1994). The Israeli study uti-lized a sample purposely selected for highintelligence, and mean scores for the fourgroups were not reported (Sagi et al., 1994),and the range of scores was not reported foreither sample. Neither study reported the re-lation of coherence of transcript to intelli-gence (in part because the classificationswere viewed as the "center of theoreticaland empirical studies on adult attachment"(Bakermans-Kranenburg & van IJzendoorn,1993, p. 873). Despite differences among thestudies, taken together they indicate that theAAI is certainly not primarily a measure ofmental ability. Process level analyses of therole mental abilities play in adults' abilitiesto present coherent discourses about theirattachment histories would be useful addi-tions to our knowledge of the AAI's discrimi-nant validity and to understanding of howattachment representations are related to so-cial perception and behavior. In the mean-time, we recommend including a measure ofmental ability as a covariate in research withthe AAI.

No relation was found between AAIclassifications and social desirability ratings.The findings were consistent with those ofBakermans-Kranenburg and van IJzendoorn(1993). It appears that subjects are not ableto predict what constitutes a "good" re-sponse to an AAI question and therefore arenot able to select one that is socially desir-able, even if so inclined. Thus it seems thatsubjects are responding to AAI questionswith the biases inherent in their workingmodels of attachment.

Our results indicate that the AAI is notmerely a measure of trait-like discoursestyle. However, a number of the discrimi-nant and demographic variables related toAAI coherence of transcript scores. The dis-course coherence concept is derived fromanalyses of conversational pragmatics(Grice, 1975) and refiects the individual's ca-pacity to be believable (no factual or logicalinconsistencies), give sufficient but not ex-cessive responses to the questions, and tobe clear; orderly, and relevant. Our findingssuggest tbat coherence of transcript, whilehighly related to attachment security, is alsorelated to other constructs or factors which

exert an impact on language, thought, andadaptive functioning, that is, age, education,and intelligence. The findings support theidea that despite its high correlation withoverall security, the coherence of transcriptscore should not be considered equivalentto the classification system (Main & Gold-wyn, 1991). Main and Goldwyn are in theearly stages of developing a "coherence ofmind" scale which takes into account "un-usual beliefs [which could refiect incoher-ence of thought, but which] may be statedcoherently, cooperatively, and from theviewpoint of the subject, trutbfully" (1994,p. 90). As such this scale attempts to separateout abilities which may be related to intelli-gence or experience but which are not di-rectly related to attachment organization.Future studies, especially in collaborationwith cognitive developmental psychologists,should investigate the capacity to give a co-herent narrative, its relation to attachmentorganization and cognitive abilities, and theinfiuence of experience and emotion on thedevelopment of those abilities (Oppenheim& Waters, 1995).

In conclusion, the Adult Attachment In-terview classifications were differentiated inthis study from factors which relate to adap-tation, language usage, and thought butwhich are not specific to the attachment sys-tem, in particular, social desirability and dis-course style. Although sensitive to social ad-aptation, the AAI does not appear to besimply a complex method of assessing socialadjustment. The relation between the AAIand intelligence in this study snggests thatin some samples there may be an interestingcorrespondence, but overall the results ofwork to date suggest that the AAI is notequivalent to an assessment of intelligence.Although new issues regarding the discrimi-nant validity of the AAI will arise as the mea-sure is applied in new populations and newresearch designs, the present study estab-lishes a significant degree of discriminantvalidity for the AAI in an important popu-lation.

References

Ainsworth, M. D. S., & Eichberg, C. (1991). Ef-fects on infant-mother attachment of mother'sunresolved loss of an attachment figure, orother traumatic experience. In P. Marris, J.Stevenson-Hinde, & C. Parkes (Eds.), Attach-ment across the life cycle (pp. 161-183). NewYork: Routledge.

Allen, J., & Crowell, J. A. (1995). Reports of par-ents, peers, pariners and self: Attachment

Crowell et al. 2597

and social competence. Paper symposiumpresented at the biennial meeting ofthe Soci-ety for Research in Child Development, Indi-anapolis.

Anastasi, A. (1982). Psychological testing. NewYork: Macmillan.

Bakermans-Kranenburg, M., & van IJzendoom,M. (1993). A psychometric study ofthe AdultAttachment Interview: Reliability and dis-criminant validity. Developmental Psychol-ogy, 29, 870-879.

Benoit, D., & Parker, K. (1994). Stability and trans-mission of attachment across three genera-tions. Child Development, 65,1444-1456.

Borman, E., Allen, J., Cole, H., & Hauser, S.(1996). Assessing adult attachment. Manu-script submitted for publication.

Bothwell, S., & Weissman, M. (1977). Social im-pairments four years after an acute depressiveepisode. American Journal of Orthopsychia-try, 47, 231-237.

Bowlby, J. (1973). Attachment and loss: Separa-tion. New York: Basic Books.

Bowlby, J. (1980). Attachment and loss: Loss.New York: Basic Books.

Bowlby, J. (1982). Attachment and loss: Attach-ment. New York: Basic Books.

Bowlby, J. (1988). A secure base: Parent-child at-tachment and healthy hum.an development.New York: Basic Books.

Bretherton, I. (1985). Attachment theory: Retro-spect and prospect. In I. Bretherton & E. Wa-ters (Eds.), Growing points of attachment the-ory and research (pp. 3-35). Monographs ofthe Society for Research in Child Develop-ment, 50(1-2, Serial No. 209).

Buros, O. K. (Ed.). (1965). The sixth mental mea-surements yearbook. Highland Park, NJ:Cryphon.

Crowell, J. A., & Feldman, S. S. (1988). Mothers'internal models of relationships and chil-dren's behavioral and developmental status:A study of mother-child interaction. Child De-velopment, 59, 1273-1285.

Crowell, J. A., & Feldman, S. S. (1991). Mothers'working models of attachment relationshipsand mother and child behavior during separa-tion and reunion. Developmental Psychology,27, 597-605.

Crowell, J. A., Treboux, D., & Waters, E. (1992).The Employment Experiences Interview andScoring System. Unpublished manuscript.State University of New York at Stony Brook.

Crowell, J. A., Waters, E., Treboux, D., O'Connor,E., & Colon-Downs, C. (1995, March). Stabil-ity of attachment representations and experi-ences of life events in newly married couples.Paper presented at the biennial meeting ofthe Society for Research in Child Develop-ment, Indianapolis.

Crowne, D. P., & Marlowe, D. (1964). A new scaleof social desirability independent of psycho-pathology. Journal of Consulting Psychology,24, 349-354.

Elicker, J., Englund, M., & Sroufe, L. A. (1992).Predicting peer -competence and peer rela-tionships in childhood from early parent-child relationships. In R. Parke & G. Ladd(Eds.), Family-peer relationships: Modes oflinkage (pp. 77-106). Hillsdale, NJ: Erlbaum.

Ffyfe, C , 6s: Waters, E. (1996, August). Empiricalclassification of adult attachment status us-ing the Adult Attachment Interview. Paperpresented at the fourteenth biennial meetingof the International Society for the Study ofBehavioral Development, Quebec City.

FoUette, V., Alexander, P., & Follette, W. (1991).Individual predictors of outcome in grouptreatment for incest survivors./oumai of Con-sulting and Clinical Psychology, 59,150—155.

Fonagy, P., Steele, H., & Steele, M. (1992, July).A prospective longitudinal study of adult at-tachment patterns on infant attachment andchild development. Presented at the twenty-fifth International Congress of Psychology,Brussels.

George, C , Kaplan, N., & Main, M. (1985) Theadult attachment interview. Unpublishedmanuscript. University of California atBerkeley.

Grice, H. P. (1975). Logic and conversation. In P.Cole & J. L. Moran (Eds.), Syntax and seman-tics: III. Speech acts (pp. 41-58). New York:Academic Press.

Grossmann, K., Fremmer-Bombik, E., Rudolph,J., & Grossmann, K. (1988). Maternal attach-ment representations as related to patterns ofinfant-mother attachment and maternal careduring the first year. In R. A. Hinde & J. Ste-venson-Hinde (Eds.), Relationships withinfamilies: Mutual influences (pp. 241-260).Oxford: Clarendon Press.

Hauser, S. T. (1992, March). Adolescent attach-ment and parental ego development anteced-ents of young adult attachment representa-tions: Longitudinal studies. Paper presentedat the Society for Research in Adolescence,Washington, DC.

Hauser, S. T. (1976). Loevinger's model and mea-sure of ego development: A critical review.Psychological Bulletin, 83, 928-955.

Hollingshead, A. B. (1975). Four-Factor Index ofSocial Status. Unpublished manuscript, YaleUniversity.

Jensen, A. J. (1980). Bias in mental testing. NewYork: Free Press.

Keppel, G. (1982). Design and analysis: A re-searcher's handbook (2d ed.). EnglewoodClifTs, NJ: Prentice Hall.

Kerns, K. A. (1995). Individual differences in

2598 Child Development

friendship quality and their links to child-mother attachment. In W. Bukowski, A. New-comb, & W. Hartup (Eds.), The company theykeep: Friendship in childhood and adoles-cence. New York; Cambridge UniversityPress.

Kling, J., Davis, W., & Knost, E. (1978). Henmon-Nelson IQ scores as predictors of WAIS FullScale IQ in alcoholics. Journal of ClinicalPsychology, 34, 1001-1002.

Lamke, T. A., & Nelson, M. J. (1973). The Hen-mon-Nelson Tests of Mental Ability. Boston;Houghton Mifflin.

Liu, X.-D., Colon-Downs, C , Lord, C , Wang, C ,& Crowell, J. (1995). Experience and the de-velopment of AAI Coherence. Poster pre-sented at the biennial meeting of the Societyfor Research in Child Development, India-napolis.

Main, M., & Goldwyn, R. (1991). Adult attach-ment rating and classification system. Ver-sion 5.0. Unpublished manuscript. Universityof California at Berkeley.

Main, M., & Goldwyn, R. (1994). Adult attach-ment rating and classification system,. Ver-sion 6.0. Unpublished manuscript. Universityof California at Berkeley.

Main, M., Kaplan, N., & Cassidy, J. (1985). Secu-rity of infancy, childhood, and adulthood; Amove to the level of representation. In I.Bretherton & E. Waters (Eds.), Growingpoints of attachment theory and research (pp.66-106). Monographs of the Society for Re-search in Child Development, 50(1—2, SerialNo. 209).

Maslin-Cole, C , & Spieker, S. (1990). Attachmentas a basis for independent motivation; A viewfrom risk and non-risk samples. In M.Greenberg, D. Cicchetti, & E. M. Cummings(Eds.), Attachment in the preschool years:Theory, research and intervention (pp. 245-272). Chicago; University of Chicago Press.

Masten, A., Best, K., & Garmezy, N. (1990). Resil-ience and development; Contributions fromthe study of children who overcome adver-sity. Development and Psychopathology, 2,425-444.

Mostow, E., & Newberry, P. (1975). Work role anddepression in women; A comparison of work-ers and housewives in treatment. AmericanJournal of Orihopsychiatry, 45, 538-548.

Oppenheim, D., & Waters, H. (1995). Narrativeprocesses and attachment representations; Is-sues of development and assessment. In E.Waters, B. Vaughn, G. Posada, & K. Kondo-Ikemura (Eds.), Garegiving, cultural, and cog-nitive perspectives on secure-base behaviorand working models; New growing points ofattachment theory and research (pp. 197—

215). Monographs of the Society for Researchin Child Development, 60(2-3, Serial No.244).

Pearson, J., Gohn, D., Cowan, P., & Cowan, C. P.(in press). Earned- and continuous-security inadult attachment; Relation to depressivesymptomatology and parenting style. Devel-opmental Psychology.

Posada, G., Waters, E., & Crowell, J. (1995). Is iteasier to use a secure mother as a secure base?Attachment q-sort correlates of the Adult At-tachment Interview. In E. Waters, B. Vaughn,G. Posada, & K. Kondo-Ikemura, (Eds.), Care-giving, cultural, and cognitive perspectiveson secure-base behavior and working models;New growing points of attachment theory andresearch (pp. 133-145). Monographs of theSociety for Research in Child Development,60(2-3, Serial No. 244).

Resnick, P., & Schnicke, M. (1992). Cognitive pro-cessing therapy for sexual assault victims.Journal of Consulting and Clinical Psychol-ogy, 60, 748-756.

Rosenstein, D. S., & Horowitz, H. A. (1993). Work-ing models of attachment in psychiatricallyhospitalized adolescents: Relation to psycho-pathology and personality. Paper presentedat the sixtieth meeting of the Society for Re-search in Child Development, New Orleans.

Sagi, A., van IJzendoorn, M., Scharf, M., Koren-Karie, N., Joels, T., & Mayseless, O. (1994).Stability and discriminant validity of theAdult Attachment Interview; A psychometricstudy in young Israeli adults. DevelopmentalPsychology, 30, 771-777.

Sroufe, L. A. (1983). Infant-caregiver attachmentand patterns of adaptation in preschool: Theroots of maladaptation and competence. In M.Perlmutter (Eds.), Minnesota symposium inchild psychology (pp. 41-81). Hillsdale, NJ;Erlbaum.

Sternberg, R., Conway, B., Ketron, J., & Bernstein,M. (1981). Peoples' conceptions of intelli-gence. JournaZ of Personality and Social Psy-chology, 41, 37-55.

Thorndike, R. M., Cunningham, G. K., Thorndike,R. L., & Hagen, E. P. (1991). Measurementand evaluation in psychology and education.New York; Macmillan.

Treboux, D., Crowell, J., & Waters, E. (1994). IQand attachment security in 151 young adults.Unpublished raw data, SUNY-Stony Brook.

van IJzendoorn, M. (1992). Intergenerationaltransmission of parenting; A review of studiesin non-clinical populations. DevelopmentalReview, 12, 76-99.

van IJzendoorn, M., & Bakermans-Kranenburg,M. (1996). Attachment representations inmothers, fathers, adolescents and clinical

Crowell et al. 2599

groups: A meta-analytic search for normativedata. Journal of Clinical and Consulting Psy-chology, 64, 8-21.

Ward, M. J., Botyanski, N. C , Plunket, S. W., &Carson, E. A. (1991, April). Concurrent valid-ity of the AAI for adolescent mothers. Paperpresented at the biennial meeting ofthe Soci-ety for Research in Child Development, Se-attle.

Waiters, E., & Sroufe, L. A. (1983). Social compe-tence as a developmental construct. Develop-mental Review, 3, 79-97.

Waiters, E., Wippman, J., & Sroufe, L. A. (1979).Attachment, positive affect, and competence

in the peer group: Two studies in constructvalidation. Child Development, 49, 821-829.

Watkins, J., Leber, W., Imber, S., Collins, J.,Elkin, I., Pilkonis, P., Sotsky, S., Shea, M., &Glass, D. (1993). Temporal course of changein depression. Journal of Consulting andClinical Psychology, 61, 858-864.

Weissman, M., Kasl, S., & Klerman, G. (1976). Fol-low-up of depressed women after mainte-nance treatment. American Journal of Psychi-atry, 133, 757-760.

Weissman, M., & Paykel, E. (1974). The depressedwoman: A study of social relationships. Chi-cago: University of Chicago Press.