Disorganized Infant Attachment Classifieationand Maternal Psychosocial Problemsas Predictors of Hostile-Aggressive Behaviorin the Preschool Classroom

Karlen Lyons-Ruth, Lisbeth Alpern,and Betty RepaeholiHarvard Medical School

LYONS-RUTH, KAHLEN; ALPEHN, LISBETH; and REPACHOLI, BETTY. Disorganized Infant AttachmentClassification and Maternal Psychosocial Problems as Predictors of Hostile-Aggressive Behaviorin the Preschool Classroom. CHILD DEVELOPMENT, 1993, 64, 572-585. Tbis study of 62 low-income families examined the relation between maternal and infant measures assessed at 18months infant age and child behavior problems at age 5 as rated by prescbool teachers. Theinfancy assessments included measures of motber-infant interaction, maternal psycbosocial prob-lems, infant cognitive development, and infant attachment security, including the disorganized/disoriented classification. Tbe strongest single predictor of deviant levels of hostile bebaviortoward peers in the classroom was earlier disorganized/disoriented attachment status, witb 71%of hostile preschoolers classified as disorganized in tbeir attachment relationships in infancy.Maternal psychosocial problems independently predicted hostile aggression in preschool andcombined additively with infant attacbment security in prediction. Results are discussed inrelation to tbe asymmetry of forward and backward prediction tbat characterized tbe findingsand in relation to the potential significance of disorganized attachment behavior as a precursorto later maladaptation.

Mapping the developmental pathways rity and behavior problems during the pre-of infants at risk for social maladaptation is school years, and these studies have variedan important priority in the field of develop- considerably in both methodology and re-mental psychopathology. In previous stud- suits (Bates, Maslin, & Frankel, 1985; Erick-ies of low-risk samples, security of attach- son, Sroufe, & Egeland, 1985; Goldberg,ment emerged as one important, broad-band Corter, Lojkasek, & Minde, 1990; Lewis,organizational construct in theories of early Feiring, McGuffog, & Jaskir, 1984). Only insocial development. A number of longitudi- the poverty sample of Erickson et al. (1985)nal studies dealing with not-at-risk popula- was a significant relationship reported be-tions have found that early attachment pat- tween attachment security and later behav-terns have significant potential for ior problems that was consistent for bothpredicting later social behaviors up to age 6 sexes,and beyond, both in relation to parents(Londerville & Main, 1981; Main, Kaplan, & Most of these studies examining longi-Cassidy, 1985; Matas, Arend, & Sroufe, tudinal continuity coded only the three orig-1987) and in relation to peers (Arend, Gove, inally described attachment patterns, which& Sroufe, 1979; Oppenheim, Sagi, & Lamb, were developed to describe variations in in-1988; Waters, Wippman, & Sroufe, 1979). fant behavior among adequately functioningHowever, in regard to the prediction of seri- families. During the past decade, a numberously maladaptive social behavior, the re- of investigators have published evidencesuits of previous studies have been less con- that the three original attachment classifica-sistent. Only a few studies have examined tions do not capture the distinct organiza-the relation between infant attachment secu- tional features of infant attachment behavior

This study was supported by NIMH Grant 35122, by grants from the A. L. Mailman FamilyFoundation and an anonymous foundation, and by a Bunting Institute Fellowship to the firstauthor. We would like to tbank Kristine Zacharsion for her help witb tbe data collection andRupa Roy for ber assistance in data analysis. Requests for reprints should be sent to the firstauthor at tbe Department of Psychiatry, Harvard Medical School, Gambridge Hospital, 1493Cambridge Street, Gambridge, MA 02139.

[Child Development, 1993, 64,572-585. © 1993 by the Society for Research in Child Developjnent, Inc.All rights reserved. 0009-3920/93/6402-0007$01.00]

Lyons-Ruth, Alpern, and Repacholi 573

among socially-at-risk subpopulations (Crit-tenden, 1985; Lyons-Ruth, Connell, Zoll, &Sitahl, 1987; Main & Weston, 1981; Radke-Yarrow, Cummings, Kuczynski, & Chapman,1985; Spieker & Booth, 1988). These reportsled Main and Solomon (1990) to developclassification criteria for a new disorganized/disoriented, or D, attachment category, in-cluding in the coding scales most of the atyp-ical behaviors observed in earlier studies.

In Main and Solomon's (1990) view, dis-organized/disoriented attachment behaviordoes not represent a fourth organized strat-egy for maintaining access to the attachmentfigure under stress, as do the three originalattachment patterns. Instead, in their view,confiicting behavioral tendencies are acti-vated in the infant and compete for expres-sion, resulting in incomplete or contradic-tory actions or the display of combinations ofbehaviors from two or more usually distinctorganized strategies. Disorganization can oc-cur in the context of an otherwise secure in-fant attachment strategy or in the context ofinsecure strategies characterized by the re-striction (avoidant pattern) or augmentation(resistant pattern) of attachment-related be-havior and affect. Therefore, Main and Solo-mon also recommended that a best-fitting al-ternate classification be assigned to allinfants, indicating the predominant underly-ing attachment strategy.

The disorganized form of infant attach-ment behavior appears infrequently amonginfants from middle-class samples, constitut-ing only 13% of 268 attachment classifica-tions in the Berkeley sample of Main andSolomon (1990). Disorganized attachmentpeitterns increase in frequency as the sever-ity of social risk factors increases, character-izing 28% of infants from multiproblem fam-ilies receiving supportive services (Spieker& Booth, 1988), 54% of infants of low-income mothers with serious depressivesymptoms and no services (Lyons-Ruth,Connell, Grunebaum, & Botein, 1990), and82% of infants from maltreating families(Carlson, Cicchetti, Barnett, & Braunwald,1989).

The patterning of the disorganized be-havior itself also appears to differ from low-to high-risk settings. A majority of disorga-nized infants in low-risk settings displayunderlying secure strategies, with particu-larly few disorganized-avoidant patterns. Inhigh-risk settings, a majority of disorganizedinfants are assigned insecure alternate clas-sifications, particularly insecure-avoidant al-

ternate classifications (Ainsworth & Eich-berg, 1991; Carlson et al., 1989; Lyons-Ruthet al., 1990; Main & Solomon, 1990; O'Con-nor, Sigman, & Brill, 1987; Spieker & Booth,1988; see Lyons-Ruth, Repacholi, McLeod,& Silva, 1991, for review).

Given that rates of disorganized infantattachment behavior are elevated in high-risk social environments that are similar tothose that produce elevated rates of child-hood psychopathology (Rutter et al., 1975),disorganization of attachment strategies,rather than insecurity per se, is likely to rep-resent an early precursor of behavior prob-lems. However, the relation between disor-ganized attachment behavior in infancy andlater behavior problems has not yet beenevaluated. Assessing this relation in socially-at-risk samples is particularly importantsince rates of disorganized behavior risesharply among infants in high-risk environ-ments. One goal of the current study was toevaluate the contribution of the disorga-nized attachment category to the predictionof teacher-rated behavior problems in pre-school and kindergarten among a sample ofinfants at social risk. While reliance on a sin-gle outcome measure limits the scope of thestudy, teacher ratings of child behavior prob-lems in the classroom provide an assessmentof adaptive functioning in an ecologicallyvalid context of great importance to thechild's subsequent development. Teacher-rated behavior problems at age 5 have alsobeen shown to predict receipt of guidanceservices in third grade, in a large-scale lon-gitudinal study of working-class children(Reinherz, Gordon, Morris, & Anastas,1983).

A second goal of the study was to evalu-ate the longitudinal prediction availablefrom three maternal risk indicators in in-fancy: maternal lack of involvement with theinfant, maternal hostile intrusiveness, andthe presence of serious maternal psycho-social problems. Psychosocial problems in-cluded depressive symptoms, child mal-treatment, or a history of psychiatrichospitalization. All maternal measures werepredicted to relate to later behavior prob-lems, given their concurrent associationwith childhood psychopathology in cross-sectional studies (Aber, Allen, Carlson, &Cicchetti, 1989; Hammen, Burge, & Stans-bury, 1990; Richman, Stevenson, & Graham,1982). We also wished to examine whetherthe maternal variables would account equallywell for variance related to attachment statusand whether attachment status might inter-

574 Child Developnient

act with maternal risk factors to protectagainst or potentiate later problems (seeMasten, 1989; Morisset, Barnard, Green-berg, Booth, & Spieker, 1990; Rutter, 1985).The contribution of infant mental develop-ment scores to the later prediction of behav-ior problems was also investigated, given re-cent reports from high-risk samples relatinginsecure or disorganized attachment rela-tionships to slowed cognitive development(Lyons-Ruth et al., 1991; Morisset et al.,1990) and relating cognitive development at24 and 42 months to behavior problems atage 5 (Erickson et al., 1985).

Given the diversity of previous findings,no prediction was made concerning the typeof later problem behavior likely to be associ-ated with disorganized/disoriented attach-ment behavior, lowered mental develop-ment scores, or the maternal risk factors.Teacher ratings of child behavior in theclassroom were analyzed both for the overallbehavior problem score and for aggressive,anxious, and hyperactive behavior prob-lems.

MethodSubjects

The subjects for the study were 62 pre-school children from low-income families,ranging in age from 49 to 71 months, meanage 59 months. Twenty-two males and 12females were between 48 and 59 months ofage, and 15 males and 13 females were be-tween 60 and 71 months of age.

The subjects were recruited from among78 low-income families who had partici-pated in a previous study of the impact offamily risk factors on infant development(Lyons-Ruth et al., 1990). Two percent of theoriginal sample could not be reinterviewedbecause they had moved too far away, 11%could not be relocated, and 5% refused toparticipate. Two additional families hadagreed to participate, but their childrenwere not enrolled in preschool.

Families were characterized by the fol-lowing social risk factors: 18% of the moth-ers had experienced psychiatric hospitaliza-tion; 16% of families had a documentedhistory of child maltreatment; 83% of fami-lies were female-headed households; 45% offamilies were supported by AFDC; 23% offamilies included a minority parent; and46% of mothers were not high school grad-uates.

During the infant study, home-visitingservices were provided to 23 families until

infants were 18 months old. All infant assess-ments were gathered at 18 months after thehome visiting services had ended and repre-sent maternal and infant status postinter-vention.

An additional group of 182 children(mean age 60 months) attending the sameclassrooms as the study children were alsoassessed by their teachers. These childrenwere selected by matching all study chil-dren with the three same-sex classmatesnearest in birthdate. Four controls wereomitted through teacher error. Classmatescores were included to control for potentialdifferences among teachers in their use ofthe rating scales based either on variationsin child age or on idiosyncratic baseline dif-ferences among teachers.

Assessment Procedures—InfancyMaternal and infant functioning was as-

sessed when the child was 18 months old.The observational measures had also beencollected at 12 months.

Demographic data.—A measure of cu-mulative demographic risk was created bysumming the presence of the following sixfactors coded from maternal interviews:mother's minority race, mother not a highschool graduate, mother under 20 at thebirth of her first child, mother a singleparent, family supported by governmentassistance, and three or more children underage 6.

Maternal psychosocial problems.—Ma-ternal psychosocial problems were coded aspresent if the mother had a documented his-tory of child maltreatment, a history of inpa-tient psychiatric hospitalization, or reporteddepressive symptoms on the CES-Depres-sion Scale over the validated cutoff point forpossible clinical disorder. All other mothersreceived a score of 0. The CES-DepressionScde was administered verbally to motherswhen infants were 18 months old. TheCES-D is a 20-item questionnaire concern-ing depressive symptoms during the pastweek (Radloff, 1977). It has been well vali-dated in large-scale epidemiologic studies(Myers & Weissman, 1980; Weissman, Sho-lomskas, Pottenger, Prusoff, & Locke, 1977).Further differentiation among the threetypes of problems was not attempted due tothe degree of overlap among problem types.Among mothers with psychiatric histories,55% were also depressed and 46% also hadmaltreatment histories. Among maltreatingmothers, 80% were also either depressed orhad psychiatric histories.

Lyons-Ruth, Alpern, and Repacholi 575

Maternal hehavior at home.—Naturalis-tic mother-infant interaction was videotapedat home for 40 min when the infants wereawake and alert. Maternal behavior wascoded in 10 4-min intervals on 12 five-pointrating scales and one timed variable, includ-ing sensitivity, warmth, verbal communi-cation, quality and quantity of comfortingtouching (physical contact in tbe service ofcommunicating affection, "touching base,"or reducing distress), quality and quantity ofcaretaking touching, interfering manipula-tion, covert hostility, anger, disengagement,flatness of affect, and time out of room,rounded to the nearest half minute. Coderswere blind to all other data on the families.Interobserver reliabilities, computed on arandomly selected 20% of the 40-min video-tapes, yielded percentages of agreement,calculated within one point, above 90% forall scales, with a mean of 91%. Cohen'skappa coefficients ranged from .45 to .81, allp < .001. Additional information on codingprocedures is available in Lyons-Ruth et al.(1987). Principal components analyses of thescales, conducted during the previous infantportion of the study, had yielded two mainfactors. Factor 1, labeled maternal involve-ment, accounted for 38% of the variance andincluded negative loadings (<.5O) for mater-nal disengagement and positive loadings formaternal sensitivity, warmth, verbal commu-nication, and quantity of comforting touch.Factor 2, labeled hostile-intrusiveness, ac-counted for 26% of the variance and in-cluded negative loadings for quality of com-forting touch and quality of caretaking touchand positive loadings for covert hostility, in-terfering manipulation, and anger. For fivemothers who were missing 18-months homedata, identical 12-months home data weresubstituted.

Infant development.—The BayleyScales of Infant Development, Mental andMotor Scales, were administered to each in-fant in a laboratory visit before assessmentin the Ainsworth Strange Situation. For twoinfants who were missing the 18-monthsBayley assessment, the 12-months scoreswere substituted.

Infant attachment security.—Within 2weeks of the home videotaping, mothers andinfants were videotaped in the AinsworthStrange Situation (Ainsworth, Blehar, Wa-ters, & Wall, 1978). In this procedure theinfant is observed in a playroom during aseries of eight 3-min episodes in which themother leaves and rejoins the infant twice.Videotapes were coded for attachment-related behaviors and for the three attach-

ment classifications as described by Ains-worth et al. (1978) and for disorganized/disoriented behaviors as described by Mainand Solomon (1990). The three original at-tachment classifications (secure, avoidant,ambivalent) were assigned by both a compu-terized multivariate classification proceduredeveloped on the original Ainsworth data(Connell, 1976; see also reference in Rich-ters, Waters, & Vaughn, 1988) and a codertrained by M. Main. Agreement between thetwo sets of classifications on the full 18-months N of 72 was 86%. Seventy-five per-cent of the disagreed-upon tapes were laterfound to meet criteria for the disorganized/disoriented category. Agreement on thedisorganized/disoriented classification be-tween M. Main and the third author for 32randomly selected tapes was 83%. Classifi-cation data can be seen in Table 3.

For regression analyses, security of at-tachment was ordered from 1 to 3, with se-cure = 3, avoidant = 2, and disorganized/disoriented = 1. In keeping with other in-vestigations (Morisset et al., 1990; Vondra &Shaw, 1991), disorganized attachment strate-gies were given the least secure rankingbased on their differentially high frequen-cies among the most impaired parent-infantrelationships, particularly those in whichmaltreatment has been independently docu-mented. For five infants missing the 18-months attachment data, 12-months attach-ment data were substituted.

Assessment Procedures—Age 5Teacher ratings of child behavior prob-

lems.—The Preschool Behavior Question-naire (PBQ) (Behar & Stringfield, 1974a) wascompleted by teachers for the 62 children inthe study and the three same-sex classmatesnearest in age to the study child. A single,mean classmate control score was computedfor each study child. Fifty-one teachers ratedthe children. Half the children were in class-rooms located in public schools, 18 attendedHead Start, 10 went to private or parochialpreschools, and 3 were in public daycare.

The PBQ represents a modified versionof Rutter's (1967) Children's Behavior Ques-tionnaire, developed for children in the 3—6-year-old age range. Total score ranges from0 to 60. Test-retest values, interrater reliabil-ities, and standardization sample character-istics are all acceptable (see Behar &Stringfield, 1974b). Factor analytic studiesextracted three factors: hostile, anxious, andhyperactive (Behar, 1977; Bebar & String-field, 1974b). Cutoff scores w^hich maximallydiscriminated normal and disturbed groups

576 Child Development

were 17 for the total score, 8 for hostile be-havior, 5 for anxious behavior, and 5 for hy-peractive behavior (Behar & Stringfield,1974b). Cutoff values indicated the upper10% of scores among the normal validationgroup. Scores over the cutoff point cannot beconsidered comparable to a clinical disorderbut merely denote deviance from normativebehavior.

Results

Control AnalysesAge.—There were no age effects (4

years vs. 5 years) on the four problem scalesfor either subjects, F(l, 60) = .27-2.58, allp = N.S., or classmates, F(l, 60) = .02-1.89,all p = N.S. Age effects over this range werenot observed in the original PBQ validationstudy (Behar & Stringfield, 1974b).

Teacher baseline effects.—Significantcorrelations between teachers' ratings ofclassmates and ratings of subjects occurredfor total problems, r = .28, p < .01, anxiousbehavior, r = .36, p < .002, and hyperactivebehavior, r = .25, p < .05, suggesting poten-tial teacher baseline differences in ratingthese problem behaviors. Only the scores forhostile behavior showed no significant asso-ciation, r = .11, p < .20. To control forteacher baseline effects and any residualchild age variance, mean classmate controlscores were entered first as covariates in allanalyses. "̂

Demographic variables.—Within therestricted economic range represented inthis sample, variation in cumulative demo-graphic risk was not significantly related tothe outcome variables and was not includedin further analyses, partial correlations withclassmate scores controlled: total problems,r = .12, hostile, r = .16, hyperactive, r =.08, anxious, r = .04, all N.S. There was asignificant sex difference only on the hyper-active scale, F(l, 59) = 6.08, p < .02, withmales scoring higher than females. There-fore, sex was included as a covariate in anal-yses of hyperactive scores.^

Because some investigators have re-ported interaction effects between sex andattachment (LaFreniere & Sroufe, 1985;Lewis et al., 1984), sex x attachment AN-COVAs were also computed with classmatescores covaried. None were significant, totalproblems, F(2, 55) = .46, N.S., hostile, Fi2,55) = .10, N.S., anxious, F(2, 55) = 2.40,N.S., hyperactive, F(2, 55) = .33, N.S.

Prediction from Assessments in InfancyFive summary measures of maternal and

infant functioning at 18 months were ana-lyzed in relation to later behavior problemscores: infant security of attachment (scored1-3, with 3 as secure), maternal psychoso-cial problems (scored present or absent), ma-ternal hostile-intrusive behavior toward theinfant at home (continuous factor score), ma-ternal involvement with the infant at home(continuous factor score), and infant mentaldevelopment (continuous score).

These variables were first assessed foroverall significance of prediction by enter-ing the five infancy measures as a group intomultiple regression analyses of each of thefour preschool behavior problem scores (to-tal problems, hostile behavior, anxious be-havior, hyperactive behavior). Classmatecontrol scores were entered first and, for hy-peractive scores only, child sex was enteredsecond. The five infant risk variables as agroup accounted for a significant proportionof the variance in hostile behavior, F chg(5,55) = 2.50, p < .04, variance explained =18%. The variance accounted for in totalproblems, 16%, failed to reach significance,F chg(5, 55) = 2.25, p < .06. The infant vari-ables also failed to predict anxious behaviorand hyperactive behavior. Table 1 displaysthe partial correlations between the infancypredictors and the four behavior problemscores, with classmate scores covaried and,for hyperactive scores only, sex covaried. Inthe case of the dichotomous psychosocialproblems scores, ANCOVAs with classmatescores covaried were computed on the be-havior problem scores. Strength of the asso-

^ Teacher baseline effects proved to be largely independent of the effects related to theinfant assessments. Risk-related subgroups in which subjects received the highest mean ratingswere not the same subgroups in which classmates received high ratings, working against thealternate possibility that correlation in ratings primarily reflected true classroom differences. Allsigniflcant results remained the same when the covariate control was omitted. However, withoutthe inclusion of the covariate control, scattered and difficult-to-interpret significant results ap-peared on the anxious scale.

^ Sex differences in hostile behavior are not consistently found in epidemiological samplesof preschoolers (see Richman et al., 1982) but tend to emerge in these same samples by schoolage.

Lyons-Ruth, Alpern, and Repaeholi 577

TABLE 1

STRENGTH OF ASSOCIATIONS BETWEEN ASSESSMENTS IN INFANCY AND PRESCHOOLBEHAVIOR PRORLEM SCORES

PRESCHOOL BEHAVIOR QUESTIONNAIRE SCORES (n = 62)

TYPE OF ASSESSMENT Total Hostile Hyperactive AnxiousIN INFANCY Problems Behavior Behavior Behavior

Continuous PBQ scores:"Sec attach -.26* -.29** - .21 - .04Psych probs .22 .26* .07 .14Matbeh-hi 21 .23* .05 .11Matbeh-inv - .10 .04 - .09 - .18Mental dev -.34** -.29** -.24* -.26*

PBQ deviant classifications:''Sec attach -.30* -.36** - .19 - .06Psych probs 29* .31* .18 .13Matbeh-hi 34** .24* .18 .02Matbeh-inv .06 .06 .08 - .18Mental dev - .21 - .15 -.22* - .14

NOTE.—Sec attach = security of attachment (3 = secure, 2 = avoidant, 1 = disorganized); psych probs =maternal psychosocial problems (0 = absent, 1 = present); mat beh-hi = maternal home behavior—hostile intru-siveness; mat beh-inv = maternal home behavior—involvement.

" For continuous PBQ scores, strength of association is indexed by partial r with classmate score (and sex forhyperactive scores) covaried.

' For deviant classifications and continuous infancy variables, strength of association is indexed by beta withsignificance assessed by F test and mean classmate score covaried; for deviant classifications and dichotomous infancyvariables, strength of association is indexed by phi, with significance assessed by chi square. Beta and phi are measuresof tbe strength of association for effects tested by F or cbi square statistics.

* p < .05.**p< .01.

ciation assessed by the F test is indexed inthe table by beta coefficient.

Since the continuous behavior problemscores used in the regression analyses didnot reveal whether the problems being re-ported were severe enough to fall within therange of risk for clinical disorder, childrenwere also classified as deviant or nondeviantaccording to the PBQ cutoff scores. Hostilebehavior scores in the deviant range werealso significantly predicted by the five in-fancy variables, as assessed by multiple dis-criminant function analysis, with classmatescores controlled, F(5, 55) = 2.33, p < .04,Wilks's lambda = .80, variance explained =20%. Table 1 also displays measures of thestrength of association (beta or phi, see notesto Table 1) between the infancy variablesand later deviant behavior classifications.

Given the significant prediction of hos-tile behavior, the relations between individ-ual infancy variables and later hostile behav-ior were examined further. As seen in Table1, predictors of the dichotomous deviancescores were similar to the predictors of theenilre range of problem scores, with the ex-ception of infant mental development. Men-tal development scores were related primar-

ily to lower-level problem behaviors notsevere enough to classify in the deviantrange. Contrary to expectation, maternalinvolvement in infancy was unrelated toproblem behavior.

Maternal involvement in infancy wassignificantly related to infant mental devel-opment scores at 18 months, r = .35, p <.002, so there may have been an indirect in-fiuence of maternal involvement on laterproblem behavior. This set of relations in-volved milder forms of problem behavior notfalling in the deviant range, however.

As Table 1 also shows, security of at-tachment in infancy exhibited the strongestrelation with hostile behavior in preschool.However, the scaled scores for attachmentsecurity used in the regression analysis donot reveal whether, compared to secure in-fants, avoidant infants and disorganized in-fants were at equally elevated risk for laterhostile behavior. Therefore, avoidant anddisorganized infants were compared sepa-rately to secure infants on both the continu-ous hostile behavior scores, using ANCOVAwith planned orthogonal comparisons, andon the dichotomous hostile deviance scores,using chi square analyses. These analyses

578 Child Development

revealed that hostile behavior scores of dis-organized infants differed significantly fromthose of securely attached infants, deviancescore, x^(l, iV = 50) = 6.20, p < .01, <{> =.40, continuous score orthogonal contrast,F(l, 47) = 4.80, p < .03, (3 = .30. Hostilebehavior scores of avoidant infants did notdiffer significantly from those of secure in-fants, deviance score, x (̂l> N = 35) = 1.71,N.S., continuous score contrast, F(l, 32) =.10, N.S.'* The proportions of infants in eachattachment classification who later exhibiteddeviant levels of hostile behavior are shownin the upper half of Table 2.

The marginal totals in Table 2 revealthat children were significantly more likelyto be classified as disorganized in infancy(n = 27) than to be classified as hostile inpreschool (n = 17), McNemar Q = 4.05, p< .05. Furthermore, at the level of predic-tion for a given individual, the predictive as-sociation between attachment classificationand later hostile behavior problems wasasymmetrical as assessed by the lambda sta-tistic. Lambda indicates percent reductionin error of prediction of the subject's statuson one categorical variable that is achievedby knowing the subject's status on a secondcategorical variable (Hays, 1963) and pro-vides additional information about the formof the significant association between attach-ment status and hostile behavior.^ Lookingforward from infancy to preschool, knowl-edge of infant attachment classification doesnot reduce the rate of error in prediction ofpreschool deviance status, lambda = 0%.Because a majority of the children in all at-tachment groups are nondeviant, the pre-dicted outcome for a child in any attachmentgroup is nondeviant behavior. However,looking backward from preschool status toinfant attachment classification, knowledgeof preschool deviance status results in 17%fewer errors in predicting prior attachmentclassification, lambda = 17%. As can be seenin the lower half of Table 2, for a child withdeviant levels of hostility in preschool, themost likely earlier attachment classificationwas disorganized, while for a nondeviantpreschooler the most likely earlier classifi-cation was secure. Seventy-one percent ofchildren with deviant levels of hostile be-

havior at age 5 had been disorganized intheir attachment behavior in infancy, com-pared to only 33.3% of nondeviant children.Preschoolers with highly hostile behaviorwere six times more likely to have been clas-sified as disorganized than to have been clas-sified as secure.

To further evaluate the utility of the dis-organized attachment category, the rates ofhostile behavior were also examined usingonly the three original attachment catego-ries, including the alternate forced classifi-cations for all D infants. As shown in Table3, a majority of the children categorized asdisorganized, 59.2% or 16 of 27, were avoid-ant in their alternate forced classifications,which is consistent with other studies ofhigh-risk populations (Carlson et al., 1989;Spieker & Booth, 1988). Later hostile behav-ior was less specifically associated with thethree original attachment strategies thanwith the disorganization of those strategies,as also shown in Table 3. Using the originalthree-category system (ambivalent group ex-cluded because of small cell size), ANCO-VAs comparing secure and avoidant groupson the four continuous problem scores (total,hostile, anxious, and hyperactive) yielded Fvalues all <1.0, N.S., with classmate scorescovaried; for the four deviant classificationscores, secure versus avoidant chi-squareswere all under 1.96, N.S.

The maternal psychosocial problemsvariable was also examined further. To eval-uate whether maltreatment alone was re-sponsible for the relation between maternalpsychosocial problems and hostile child be-havior, the relation between these variableswas reanalyzed with the 10 maltreatingmothers removed. The association remainedsignificant, hostile continuous score, Fchg(l, 49) = 5.68, p < .02, 3 = .33, deviancestatus, x^(l, N = 52) = 5.15, p < .02, <i> =.31. Among the 52 nonmaltreating mothers,serious depressive symptoms at 18 monthsinfant age constituted the most prevalentmaternal psychosocial problem. Nineteen ofthe 22 mothers with psychosocial problemsother than maltreatment reported high lev-els of depressive symptoms when their in-fants were 18 months old, with three of these

' Since the security of attachment rankings might have obscured a significant differencebetween secure and avoidant children in the regression analyses, secure vs. avoidant ANOVAswere also computed for total, anxious, and hyperactive scores. None approached significance, allF(l, 34) < .20, N.S.

•* The lamhda statistic has no associated significance level heyond the significance of theoverall association represented in the 2 X 3 table, x^(2, N = 62) = 8.02, p < .02.

Lyons-Ruth, Alpern, and Repaeholi 579

TABLE 2

INFANT ATTACHMENT CLASSIFICATION AND PRESCHOOL HOSTILEBEHAVIOR: FORWARD AND BACKWARD PREDICTION

HOSTILE BEHAVIOR

Not deviant Deviant(%) (n = 45) (%) (n = 17)

Looking forward:"Secure 91 (21) 9 (2)Avoidant 75 (9) 25 (3)Disorganized 56 (15) 44 (12)

Looking backward:*"Secure 47 (21) 12 (2)Avoidant 20 (9) 18 (3)Disorganized 33 (15) 71 (12)

NOTE.—Cell n's in parenthesis." Row percentages shown; lambda forward = 0%.'' Column percentages shown; lambda backward = 17%.

also having past inpatient psychiatric hospi- between maternal psychosocial problemstalization. Only three mothers coded as hav- and later hostile behavior was asymmetrical,ing psychosocial problems had inpatient JCnowing the mother's problem status in in-psychiatric hospitalizations alone. Thus, ex- fancy did not reduce error in predicting acept for these last three mothers, a mother child's later deviance status, lambda = 0%.classified as having psychosocial problems However, looking backward, knowledge ofhad problems that were active during her whether the child's behavior was deviantchild's infancy, as manifested in continuous did reduce error by 23% in predicting mater-protective service involvement or in a high nal psychosocial problems; 76.5%, or 13 oflevel of depressive symptoms. As previously 17, mothers of highly hostile children hadreported, maternal depressive symptoms in psychosocial problems while only 44.4%, orthis sample showed a high level of stability 19 of 45, mothers of nondeviant children hadfrom 12 to 18 months infant age (Lyons- similar problems.Ruth, Zoll, Connell, & Grunebaum, 1986). . r J- • i

Asymmetry of prediction was not evalu-As was the case for disorganized attach- ated in relation to the continuous maternal

ment behavior, the predictive association hostile-intrusiveness scores due to the lack

TABLE 3

PERCENTAGES OF INFANTS IN EACH ATTACHMENT SUBGROUP WHOWERE DEVIANT OR NONDEVIANT IN PKESCHOOL: THREE- ANDFOUR-CATEGORY ATTACHMENT CLASSIFICATIONS COMPARED

Secure:SecureSecure-D

Avoidant:AvoidantAvoidant-D

Ambivalent:AmbivalentAmbivalent-D

NOTE.—Cell n's in parenthesis.

HOSTILE

Not Deviant(%) {n = 45)

81 (25)91 (21)50(4)61 (17)75 (9)50 (8)

100 (3). . . (0)100 (3)

BEHAVIOR

Deviant(%) (n = 17)

19(6)9(2)

50(4)39 (11)25(3)50(8)00(3). . (0)00(0)

580 Child Development

of a validated basis for establishing a classi-fication boundary for deviant maternal be-havior.

Relations among maternal and infantvariables.—A multiple discriminant func-tion analysis on the deviance scores for hos-tile behavior was conducted to evaluatewhether the contributions of the three infantpredictors were redundant, additive, or in-teractive. Redundancy between the two ma-ternal variables, in particular, appearedlikely because maternal hostile-intrusive be-havior was strongly correlated with maternalpsychosocial problems during the child'sinfancy, r = .48, p < .001 (see Lyons-Ruthet al., 1987, 1990). Classmate control scoreswere entered into the analysis first, followedstepwise in order of significance by attach-ment security, psychosocial problems, andhostile-intrusive behavior, followed by thethree interaction terms (attachment x prob-lems, attachment x behavior, problems xbehavior).

None of the three interaction effects ap-proached significance, with the F changevalues ranging from 1.53 to .023, all N.S.Both security of attachment and maternalpsychosocial problems made additive contri-butions to the prediction of hostile behavior,secattach F(l, 59) = 7.60, p < .01; psych-prob, F chg(l, 58) = 4.99, p < .03 (after secu-rity of attachment was entered). In a follow-up analysis, security of attachment was alsosignificant when entered after maternal psy-chosocial problems, secattach, F chg(l, 58)= 6.35, p < .02. Maternal hostile-intrusive

behavior was redundant with maternal psy-chosocial problems, accounting for no in-dependent variance once the psychosocialproblems variable was entered, F chg(l, 58)= .41, N.S. (after psychprob only). Attach-ment security and maternal problems to-gether accounted for 18% of between-groupsvariance, R = .43, Wilks's lambda = .82,X (̂2, N = 62) = 11.82, p < .003, and cor-rectly classified 76% of cases as to deviantlevels of hostile behavior, x^(l, ^ = 124) =8.85, p < .005. Regression analysis of contin-uous behavior problem scores yielded virtu-ally identical results: secattach, F chg(l, 59)= 5.71, p < .02, psychprob, F chg(l, 58) =4.03, p < .05, hostile, F chg(l, 57) = .22,N.S.; no interaction effects approached sig-nificance.

To assess whether maternal psychoso-cial problems were related to child behaviorprimarily through the covariance betweenmaternal problems and maternal hostile-intrusive behavior, a follow-up discriminantanalysis was performed entering maternalhostile-intrusive behavior before the othertwo variables. Maternal psychosocial prob-lems failed to account for significant addi-tional variance after hostile-intrusive behav-ior was entered, F chg(l, 58) = 2.98, p <.17, while attachment security continued tomake a significant contribution after bothmaternal variables were entered, Fchg(l, 57)= 6.45, p < .01. Regression analysis of con-tinuous scores yielded the same findings.Thus, maternal psychosocial problems ac-counted for variance in child behavior partly

TABLE 4PERCENTAGE OF INFANTS WHO ARE HOSTILE AND NOT-HOSTILE

IN PRESCHOOL CROSS-CLASSIFIED BY INFANT ATTACHMENT STATUSAND PRESENCE OF MATERNAL PSYCHOSOCIAL PROBLEMS

HOSTILE BEHAVIOR

Nondeviant Deviant(%) (n = 45) (%) (n = 17)

Maternal psycbosocial problemsabsent:

Secure 100 (14) 00 (0)Avoidant 80 (4) 20 (1)Disorganized 73 (8) 27 (3)

Maternal psychosocial problemspresent:

Secure 78 (7) 22 (2)Avoidant 71 (5) 29 (2)Disorganized 44 (7) 56 (9)

NOTE.—Cell n's in parentheses.

Lyons-Ruth, Alpern, and Repaeholi 581

TABLE 5

MEAN BEHAVIOR PROBLEM SCORES FOR CHILDREN WITH EARLY RISK FACTORS COMPAREDTO CLASSMATE SCORES AND PBQ VALIDATION SAMPLE SCORES

PRESCHOOL BEHAVIOR QUESTIONNAIRE SCORES

Total Hostile Anxious HyperactiveN Problems Behavior Behavior Behavior

PBQ validation sample:Disturbed sample 102 21.3 6.2 6.7 4.8Normal sample 476 8.1 2.7 2.0 2.2

Current study sample:Classmate controls 182 8.8 3.0 2.6 2.1Subjects:

Neither risk factor 19 7.8 1.5 2.8 2.4Psychprob only 16 12.6 3.8 4.3 2.6D attach only 11 12.2 3.7 3.8 2.9Both risk factors 16 16.4 6.8 3.7 4.0

NOTE.—Psychprob = maternal psychosocial problems; D attach = disorganized attachment classification.

because mothers with psychosocial prob-lems were more hostile and intrusive inearly interactions with their infants. Thepsychosocial problems variable was asixonger predictor of later hostile behavior,however, as was shown in Table 1, and incombination with attachment status resultedin more accurate classification of deviant be-havior than did the combination of attach-ment security and maternal behavior, 76%compared to 69%.

The additive effect of maternal psycho-social problems and disorganized infant at-ta.chment behavior on the rate of deviant be-havior in preschool is shown in Table 4. Ascan be seen, if the infant displayed disorga-nized attachment behavior and the motherhad psychosocial problems, a majority of thechildren, 56%, exhibited deviant levels ofhostile behavior in preschool.

Means for the continuous problemscores by the two infant risk factors areshown in Table 5, in comparison to class-mate scores and PBQ norms. Despite thelower socioeconomic status of the subjects'school districts, classmates' mean scores onthe four PBQ problem scales were similar tomeans for the normal PBQ validation sam-ple, as shown. As can also be seen, scores forstudy subjects with neither risk factor wereslightly lower than scores for low-incomeclassmates and lower than scores for the nor-mal PBQ validation group. In contrast,scores for study children with both early riskfactors exceeded scores for their own low-income classmates and approached, or in thecase of hostile behavior exceeded, the scores

for preschoolers in the disturbed PBQ vali-dation group.

Diseussion

Results from the study extend the find-ings of previous longitudinal work relatinginfant attachment security to later maladap-tive social behavior with peers. The inclu-sion of the disorganized/disoriented attach-ment category in the classification systemrevealed that children with disorganized at-tachment histories accounted for 71% of thecases of serious hostile behavior in pre-school. At the same time, this subgroup ofhighly aggressive children remained a mi-nority of all children classified in infancy asdisorganized in their attachment relation-ships.

The obtained association between dis-organized attachment behavior and later be-havior problems may contribute to the previ-ous mixed pattern of findings in longitudinalstudies that did not include a disorganizedattachment category. In low-social-risk sam-ples such as those of Bates et al. (1985) andGoldberg et al. (1990), where no relation be-tween infant attachment security and laterproblems were found, and that of Lewis etal. (1984), where predictive effects occurredonly in boys, a low rate of disorganized in-fant behavior would be expected. In con-trast, in the high-risk sample of Erickson etal. (1985), where a significant relationshipbetween infant attachment security and latermaladaptation was reported regardless ofgender, a high rate of disorganized behaviorwould be predicted, particularly disor-

582 Child Development

ganized-avoidant behavior (Garlson et al.,1989; Lyons-Ruth et al., 1990).

One subgroup of disorganized/disori-ented infants who did develop behaviorproblems could be accounted for by the si-multaneous presence of psychosocial prob-lems affecting their mothers. The effects ofattachment status and maternal psychosocialproblems were independent and were addi-tive rather than interactive, with no protec-tive or potentiating relations between thetwo sources of risk. However, previous anal-yses of the infant data (Lyons-Ruth et al.,1990) had indicated that infants of motherswith psychosocial problems were less likelyto develop secure attachments in the firstplace, suggesting both direct and indirectpathways for the infiuence of maternal prob-lems on later child adaptation.

The finding that disorganized infant at-tachment behaviors and serious maternalpsychosocial problems were part of the earlyfamily context of children who later dis-played hostile behavior converges with andextends the larger literature on conductproblems. Maternal depressive symptomsand their correlate, maternal hostility, havebeen identified as frequent concurrent cor-relates of child conduct problems during theschool years (Hammen et al., 1990; Richmanet al., 1982; Stewart, deBlois, & Cummings,1980; Webster-Stratton & Hammond, 1988;see Downey & Goyne, 1990, for review).The present data further suggest that mater-nal depressive symptoms and maternal hos-tility may have been present from early inthe lives of children with conduct problemsand that the conduct-disordered child's reg-ulation of attachment-related affects in in-fancy may have been disorganized.

The observed asymmetry between for-ward and backward prediction sheds somelight on the differing theoretical modelsheld by clinicians and developmental re-searchers concerning the extent of continu-ity in early development. In the clinicalarea, a long-standing psychodynamic theo-retical model has postulated significant con-tinuity in maladaptive relationship patternsfrom infancy to early childhood, whileamong infant researchers prediction from in-fancy to later behavioral outcomes has ap-peared much more modest. The present datasuggest that this difference in perspectivemay stem from a legitimate asymmetry in thebackward and forward predictive associationbetween behavioral difficulties in infancyand clinically significant behavioral prob-

lems in childhood, a common type of back-ward and forward asymmetjry that exists inthe prediction of other disorders such as theassociation between lung cancer and smok-ing. Clinicians may be able to predict atgreater-than-chance levels that children pre-senting at age 5 with clinically s^nificanthostile behavior are likely to havfe experi-enced both maternal psychosocial problemsand disorganized attachment relationshipsin infancy. However, based on the samedata, developmental observers in infancycannot predict forward with the same degreeof confidence. It remains to be seen whethermore differentiated assessments of familyenvironment or of subtypes of disorganizedinfant behavior might provide more preciseforward prediction from the infancy period.

Bowlby (1988) has theorized that oncea representational model of the attachmentrelationship is established in infancy, it willtend to maintain organization over time andexert an infiuence on later behavior inde-pendent of other environmental sources.One major contribution of previous attach-ment research has been the description ofpatterns of self-regulatory behaviors in in-fancy that were related to, but distinct from,patterns of caregiving. The finding that dis-organized attachment status accounted forvariation in later behavior independent ofmaternal variables is consistent with thisview that self-regulatory processes need tobe conceptualized as partially independentof the larger relational context. This tenet ofBowlby's theory also has important implica-tions for the design of intervention effortsand deserves further evaluation in longi-tudinal studies (see also Renken, Egeland,Marvinney, Mangelsdorf, & Sroufe, 1989;Sroufe, Egeland, & Kreutzer, 1990, for re-lated analyses).

As Sroufe (1989) has pointed out, one ofthe key premises of a developmental ap-proach to psychopathology is that precursorsto later maladaptation may be identified wellbefore a true disorder develops. Disorga-nized attachment behavior deserves furtherstudy as one potential infant precursor oflater maladaptation. A more process-orient-ed account is now needed of the particularaspects of high-risk environments that inter-fere with the infant's implementation of or-ganized strategies for regulating attachment-related affects. While less optimal maternalbehavior has been associated with disorga-nized infant attachment patterns, there arestill few data available to evaluate why thesenegative interactive patterns result in disor-

Lyons-Ruth, Alpern, and Repaeholi 583

ganized rather than organized insecure strat-egies (Hann, Gastino, Jarosinski, & Britton,1991; Lyons-Ruth et al., 1991; Main et al.,1985; Spieker & Booth, 1988; but see alsoMain & Hesse, 1990).

Equally important, evidence to datesuggests that the behavioral patterns of chil-dren with disorganized attachment relation-ships may be particularly discontinuous overearly development. Disorganized infant at-tachment strategies increase in frequencyfrom 12 to 18 months of age among high-riskinfants (Beckwith & Rodning, 1991; Lyons-Ruth et al., 1991), so that attachment assess-ments at 18 months rather than 12 monthsmay be particularly important for predictionof later problems. This increase in disorga-nized strategies is apparently followed by aprocess of reorganization during the pre-school years into one of two alternate formsof controlling behavior, controlling-care-giving or controlling-punitive (Main et al.,1985; Speltz, deKlyen, & Endriga, 1991).Ghildren who display controlling-punitivebehaviors with their parents may be thosemost likely to display hostile behavior to-ward peers in the classroom, though fewdata are yet available to evaluate this link.These gaps in current knowledge need tobe addressed if the predictive associationsfound in the current data are to be more fullyunderstood.

ReferencesAber, J. L., Allen, J. P., Carlson, V., & Ciccbetti,

D. (1989). The effects of maltreatment on de-velopment during early childhood: Recentstudies and their theoretical, clinical, and pol-icy implications. In D. Cicchetti & V. Carlson(Eds.), Child maltreatment: Theory and re-search on the causes and consequences ofchild abuse and neglect (pp. 579-619). Cam-bridge: Cambridge University Press.

Ainsworth, M. D. S., Blehar, M., Waters, E., &Wall, S. (1978). Patterns of attachment. Hills-dale, NJ: Erlbaum.

Ainsworth, M. D. S., & Eichberg, C. G. (1991).Effects on infant-mother attachment ofmother's unresolved loss of an attachmentfigure or other traumatic experience. In P.Marris, J. Stevenson-Hinde, & C. Parkes(Eds.), Attachment across the life cycle (pp.160-186). New York: Routledge.

Arend, R., Gove, F., & Sroufe, L. A. (1979). Conti-nuity of individual adaptation from infancy tokindergarten: A predictive study of ego-resiliency and curiosity in preschoolers.Child Development, 50, 950-959.

Bates, J. E., Maslin, C. A., & Frankel, K. A. (1985).

Attachment security, mother-child interactionand temperament as predictors of bebaviorproblem ratings at age three years. In I. Breth-erton & E. Waters (Eds.), Growing points ofattachment theory and research. Monographsof the Society for Research in Child Develop-ment, 50(1-2, Serial No. 209), 167-193.

Beckwith, L., & Rodning, C. (1991, April). Stabil-ity and correlates of attachment classificationsfrom 13 to 36 months in a sample of preterminfants. In R. S. Marvin & J. Cassidy (Chairs),Attachment during the preschool years: Ex-amination of a new measure. Symposiumconducted at the biennial meeting of the Soci-ety for Research in Child Development, Se-attle.

Behar, L. B. (1977). The preschool bebavior ques-tionnaire. Journal of Abnorm,al .Child Psy-chology, 5, 265-275.

Behar, L. B., & Stringfleld, S. (1974a). Manual forthe Preschool Behavior Questionnaire.(Available from Dr. Lenore Behar, 1821Woodburn Road, Durbam, NC 27705.)

Behar, L. B., & StringBeld, S. (1974b). A behaviorrating scale for the preschool child. Develop-mental Psychology, 10, 601-610.

Bowlby, J. (1988). A secure base. New York: Basic.Carlson, V., Cicchetti, D., Barnett, D., & Braun-

wald, K. (1989). Disorganized/disoriented at-tachment relationships in maltreated infants.Developmental Psychology, 25, 525—531.

Connell, D. B. (1976). Individual differences inattachment: An investigation into stability,implications, and relationships to structureof early language development. Unpublisheddoctoral dissertation, Syracuse University,Syracuse, NY.

Crittenden, P. M. (1985). Maltreated infants: Vul-nerability and resilience. Journal of ChildPsychology and Psychiatry, 26, 85-96.

Downey, G., & Coyne, J. C. (1990). Children ofdepressed parents: An integrative review.Psychological Bulletin, 108, 50-76.

Erickson, M., Sroufe, L. A., & Egeland, B. (1985).The relationship between quality of attach-ment and behavior problems in preschool ina high-risk sample. In I. Bretherton & E. Wa-ters (Eds.), Growing points of attachment the-ory and research. Monographs of the Societyfor Research in Child Development, 50(1-2,Serial No. 209), 147-166.

Goldberg, S., Corter, C , Lojkasek, M., & Minde,K. (1990). Prediction of behavior problems in4-year-olds bom prematurely. Developmentand Psychopathology, 2, 15-30.

Hammen, C , Burge, D., & Stansbury, K. (1990).Relationship of mother and child variables tochild outcomes in a high-risk sample: Acausal modeling analysis. DevelopmentalPsychology, 26, 24-30.

584 Child Development

Hann, D. M., Castino, R. J., Jarosinski, J., & Brit-ton, H. (1991, April). Relating mother-toddlernegotiation patterns to infant attachment andmaternal depression with an adolescentmother sample. In J. D. Osofsky & L. Hubbs-Tait (Chairs), Consequences of adolescentparenting: Predicting behavior problems intoddlers and preschoolers. Symposium con-ducted at the biennial meeting of the Societyfor Research in Child Development, Seattle.

Hays, W. L. (1963). Statistics'for psychologists.New York: Holt, Rinehart & Winston.

LaFreniere, P., & Sroufe L. A. (1985). Profiles ofpeer competence in the preschool: Interrela-tions between measures, influence of socialecology, and relation to attacbment history.Developmental Psychology, 21, 56-69.

Lewis, M., Feiring, C , McGuffog, C , & Jaskir, J.(1984). Predicting psycbopathology in six-year-olds from early social relations. ChildDevelopment, 55, 123-136.

Londerville, S., & Main, M. (1981). Security ofattachment, compliance, and maternal train-ing methods in the second year of life. Devel-opmental Psychology, 17, 289-299.

Lyons-Ruth, K., Connell, D. B., Grunebaum, H.,& Botein, S. (1990). Infants at social risk: Ma-ternal depression and family support servicesas mediators of infant development and secu-rity of attachment. Child Development, 61,85-98.

Lyons-Ruth, K., Connell, D. B., Zoll, D., & Stahl,J. (1987). Infants at social risk: Relationsamong infant maltreatment, maternal behav-ior, and infant attachment behavior. Develop-mental Psychology, 23, 223-232.

Lyons-Ruth, K., Repaeholi, B., McLeod, S., &Silva, E. (1991). Disorganized attachment be-bavior in infancy: Short-term stability, mater-nal and infant correlates and risk-related sub-types. Development and Psychopathology, 3,377-396.

Lyons-Ruth, K., Zoll, D., Connell, D. B., & Gru-nebaum, H. (1986). The depressed motherand her one-year-old infant: Environmentalcontext, mother-infant interaction and attach-ment, and infant development. In E. Tronick& T. Field (Eds.), Maternal depression andinfant disturbance (pp. 61—82). San Fran-cisco; Jossey Bass.

Main, M., & Hesse, E. (1990). Parents' unresolvedtraumatic experiences are related to infantdisorganized attachment status: Is frightenedand/or frightening parental bebavior the link-ing mechanism? In M. Greenberg, D. Cic-chetti, & E. M. Cummings (Eds.), Attachmentin the preschool years: Theory, research andintervention (pp. 161-184). Chicago: Univer-sity of Chicago Press.

Main, M., Kaplan, N., & Cassidy, J. (1985). Secu-

rity in infancy, childhood and adulthood: Amove to the level of representation. In I.Bretherton & E. Waters (Eds.), Growingpoints of attachment theory and research.Monographs of the Society for Research inChild Development, 50(1-2, Serial No. 209),66-104.

Main, M., & Solomon, J. (1990). Procedures foridentifying infants as disorganized/disori-ented during the Ainsworth Strange Situa-tion. In M. Greenberg, D. Cicchetti, & E. M.Cummings (Eds.), Attachment in the pre-school years: Theory, research and interven-tion (pp. 121-160). Chicago: University ofChicago Press.

Main, M., & Weston, D. (1981). The quality of thetoddler's relationship to mother and to father:Related to conflict behavior and the readinessto establish new relationships. Child Devel-opment, 52, 932-940.

Masten, A. S. (1989). Resilience in development:Implications of the study of successful adapta-tion for developmental psychopathology. InD. Cicchetti (Ed.), The emergence of a disci-pline: Rochester symposium, on develop-mental psychopathology (pp. 261-294).Hillsdale, NJ: Erlbaum.

Matas, L., Arend, R. A., & Sroufe, L. A. (1987).Continuity of adaptation in the second year:The relationship between quality of attach-ment and later competence. Child Develop-ment, 49, 547-556.

Morisset, C. E., Barnard, K. E., Greenberg, M. T.,Booth, C. L., & Spieker, S. J. (1990). Environ-mental influences on early language develop-ment: The context of social risk. Developmentand Psychopathology, 2, 127-149.

Myers, J. K., & Weissman, M. M. (1980). Use of aself-report symptom scale to detect depres-sion in a community sample. American Jour-nal of Psychiatry, 137, 1081-1084.

O'Connor, M., Sigman, M., & Brill, N. (1987). Dis-organization of attachment in relation to ma-ternal alcohol consumption. Journal of Clini-cal and Consulting Psychology, 55, 831-836.

Oppenheim, D., Sagi, A., & Lamb, M. (1988). In-fant-adult attachments on the kibbutz andtheir relation to socioemotional developmentfour years later. Developmental Psychology,24, 427-433.

Radke-Yarrow, M., Cummings, E. M., Kuczynski,L., & Chapman, M. (1985). Patterns of attach-ment in two- and three-year-olds in normalfamilies and families with parental depres-sion. Child Development, 56, 884-893.

RadlofF, L. (1977). The CES-D scale: A self-reportdepression scale for research in the generalpopulation. Applied Psychological Measure-ment, 1, 385-401.

Reinherz, H., Gordon, A., Morris, K., & Anastas,

Lyons-Ruth, Alpern, and Repaeholi 585

T. (1983). Who shall be served? Issues inscreening for emotional and behavioral prob-lems in school. Journal of Primary Preven-tion, 4, 73-95.

Renken, B., Egeland, B., Marvinney, D.,Mangelsdorf, S., & Sroufe, A. (1989). Earlychildhood antecedents of aggression and pas-sive-withdrawal in early elementary school.Journal of Personality, 5, 257-281.

Richman, N., Stevenson, J., & Graham, P. (1982).Preschool to school: A behavioral study. Lon-don: Academic Press.

Richters, J. E., Waters, E., & Vaughn, B. E. (1988).Empirical classification of infant-mother rela-tionships from interactive behavior and cry-ing during reunion. Child Development, 59,512-522.

Rutter, M. (1985). Resilience in the face of adver-sity: Protective factors and resistance to psy-chiatric disorder. British Journal of Psychia-try, 147, 598-611.

Rutter, M. (1967). A children's behavior question-naire for completion by teachers: Preliminaryfindings. Journal of Child Psychology andPsychiatry, 8, 1-11.

Rutter, M., Yule, B., Quinton, D., Rowlands, O.,Yule, W., & Berger, M. (1974). Attainment andadjustment in two geographical areas: III.Some factors accounting for area differences.British Journal of Psychiatry, 126, 520-533.

Speltz, M., deKlyen, M., & Endriga, M. C. (1991,April). Controlling patterns of reunion behav-ior in preschool children with disruptive be-havior. In K. Lyons-Ruth (Chair), Disorga-nized attachment behavior from infancy toage six: Maternal context, representationalprocesses, and child maladaptation. Sympo-sium conducted at the biennial meeting of theSociety for Research in Ghild Development,Seattle.

Spieker, S. J., & Booth, C. (1988). Maternal ante-

cedents of attachment quality. In J. Belsky &T. Nezworski (Eds.), Clinical implications ofattachment (pp. 300-323). Hillsdale, NJ:Erlbaum.

Sroufe, L. A. (1989). Pathways to adaptation andmaladaptation: Psychopathology as develop-mental deviation. In D. Cicchetti (Ed.), T/ieemergence of a discipline: Rochester sympo-sium on developmental psychopathology(Vol. 1, pp. 13-40). Hillsdale, NJ: Erlbaum.

Sroufe, L. A., Egeland, B., & Kreutzer, T. (1990).The fate of early experience following devel-opmental change: Longitudinal approaches toindividual adaptation in childhood. Child De-velopment, 61, 1363-1373.

Stewart, M. A., deBlois, C. S., & Cummings, G.(1980). Psychiatric disorder in the parents ofhyperactive boys and those with conduct dis-order. Journal of Child Psychology and Psy-chiatry, 21, 283-292.

Vondra, J. L, & Shaw, D. S. (1991, April). Pre-dicting infant attachment: Maternal and eco-logical factors. Paper presented at the bien-nial meeting of tbe Society for ReseEirch inChild Development, Seattle.

Waters, E., Wippman, J., & Sroufe, L. A. (1979).Attachment, positive affect, and competencein the peer group: Two studies in constructvalidation. Child Development, 50, 821-829.

Webster-Stratton, C , & Hammond, M. (1988). Ma-ternal depression and its relationship to lifestress, perceptions of child behavior prob-lems, parenting behaviors, and child conductproblems. Journal of Abnormal Child Psy-chology, 16, 299-315.

Weissman, M., Sholomskas, D., Pottenger, M.,Prusoff, B., & Locke, B. (1977). Assessing de-pressive symptoms in five psychiatric popula-tions: A validation study. American Journal ofEpidemiology, 106, 203-214.