Child Development, August 1997, Volume 68, Number 4, Pages 571-591

Sensitivity and Attachment: A Meta-Analysis on Parental Antecedents of Infant Attachment

Marianne S. De Wolff and Marinus H. van IJzendoorn

This meta-analysis included 66 studies (N = 4,176) on parental antecedents of attachment security. The ques- tion addressed was whether maternal sensitivity is associated with infant attachment security, and what the strength of this relation is. It was hypothesized that studies more similar to Ainsworth's Baltimore study (Ainsworth, Blehar, Waters, & Wall, 1978) would show stronger associations than studies diverging from this pioneering study. To create conceptually homogeneous sets of studies, experts divided the studies into 9 groups with similar constructs and measures of parenting. For each domain, a meta-analysis was performed to describe the central tendency, variability, and relevant moderators. After correction for attenuation, the 21 studies (N = 1,099) in which the Strange Situation procedure in nonclinical samples was used, äs well äs precedmg or concurrent observational sensitivity measures, showed a combined effect size of r(l,097) = .24. According to Cohen's (1988) conventional cnteria, the association is moderately strong. It is concluded that in normal settings sensitivity is an important but not exclusive condition of attachment security. Several other dimensions of parenting are identified äs playing an equally important role. In attachment theory, a move to the contextual level is required to interpret the complex transactions between context and sensitivity in less stable and more stressful settings, and to pay more attention to nonshared environmental influences.

INTRODUCTION worth's pioneering study (Ainsworth & Wittig, 1969), and in the past few decades it has been used world- In the first volume of his trilogy, Attachment and Loss, wide (Thompson, in press; van IJzendoorn & Kroon- John Bowlby (1969) signaled an urgent need to deter- enberg, 1988). Ainsworth and her colleagues assessed mine the antecedent conditions that influence the de- a great variety of dimensions of maternal behavior velopment of attachment. Bowlby (1969) suggested at home. In particular, four rating scales (sensitivity, that one of the conditions contributing to the devel- acceptance, cooperation, and accessibility) were opment of a secure attachment relationship may be found to be strongly related to attachment security, the attachment figure's sensitivity in responding to and the authors concluded that "the most important the baby's Signals: When infants experience that their aspect of maternal behavior commonly associated social initiatives are successful in establishing a recip- with the security-anxiety dimension of infant attach- rocal interchange with the mother, it is likely that an ment is manifested in different specific ways in dif- active and happy interaction between the couple will ferent situations, but in each it emerges äs sensitive ensue and that a secure attachment relationship will responsiveness to infant Signals and communica- develop. tions" (Ainsworth et al, 1978, p. 152). After more than 25 years of research on the ante- Two decades after the Baltimore study, however, cedents of attachment security, we may now be in there is still great controversy over the parental ante- a position to answer Bowlby's question. Is parental cedents of the Strange Situation attachment classifi- sensitivity indeed an important condition for the de- cations. Gewirtz and Boyd (1977) and Lamb, Thomp- velopment of a secure attachment relationship be- son, Gardner, and Charnov (1985) sparked this tween infant and parent? controversy by arguirig that in her exploratory study Ainsworth and her colleagues were the first to ex- Ainsworth had overgeneralized from the findings of amine the relation between parental behavior in the her small sample. Lamb et al. contended that Ains- home and security of attachment (Ainsworth, Blehar, worth et al.'s (1978) "exciting hypotheses about the Waters, & Wall, 1978). They observed 26 middle-class specific antecedents of Strange Situation behavior re- mother-infant dyads from Baltimore throughout the main unproven except in their most general form" first year of life; more than 70 hr of observation were (p. 97). Although many studies had indicated that the spent in each home. At the time of the infant's first infant's prior experiences at home were indeed re- birthday, mother and infant came to the laboratory lated to their Strange Situation behavior, according to for assessment in the Strange Situation. This stan- dardized procedure for assessing the infant-parent © 1997 by the Society for Research in Child Development, Inc. attachment relationship was developed in Ains- All nghts reserved. 0009-3920/97/6804-0016S01.00 572 Child Development

Lamb et al. it is still unclear which specific maternal ory, a narrative review may not contribute to resolu- behaviors are of formative importance for attachment tion of the debate because of its "subjective" and less security. In the first meta-analysis on attachment and systematic nature (Bretherton, 1985; Lamb et al., sensitivity, Goldsmith and Alansky (1987, pp. 811, 1985). Meta-analysis offers a way of bringing some 813) also concluded that "many of the studies . . . degree of order to a large and inconsistent body of replicate Ainsworth et al.'s (1978) original findings findings (Rosenthal, 1991) at the same time that it of the predictive power of maternal sensitivity when allows for the testing of specific hypotheses statisti- replication is evaluated in terms of statistical signifi- cally. For example, one might test the idea that stud- cance." However, they cautioned that the actual size ies which show a greater similarity to the Baltimore of the predictive effect of maternal sensitivity is much study show stronger associations between attach- smaller than once was believed, suggesting only a ment and sensitivity. Almost 10 years after Gold- weak relation between attachment security and pa- smith and Alansky's (1987) meta-analysis of 13 stud- rental sensitivity. ies, it is time to take stock of the growing literarure Despite all of the skepticism, many authors con- on attachment and sensitivity and to focus on the tinue to embrace Bowlby's (1969) original proposi- core issue of variability among the pertinent stud- hon that maternal sensitivity is a crucial antecedent ies. More than 60 studies have investigated maternal of attachment security (Bretherton, 1985; Main, 1990; behavior in relation to attachment security. In the Sroufe, 1988). Recently, Isabella (1993) stated that at- current meta-analysis, we address three issues (Mül- tachment theory and research highlight maternal len, 1989): (1) Central tendency: What is the typical sensitivity äs an all-important characteristic of inter- strength of association between maternal behavior action that has been consistently linked to attachment and attachment security? (2) Variability: Is the set of security. The fact that some studies have yielded study results heterogeneous, in the sense that out- much weaker associations than Ainsworth's study comes are relatively variable across the studies? can, in this view, be attributed to methodological (3) Prediction: Can the Variation between studies be weaknesses of the replication studies. Some research- explained by study features that are relevant to the ers have restricted their observations to a single home controversy between the skeptical and orthodox po- visit, whereas others have used brief laboratory as- sitions? sessments of sensitivity instead of extensive home In particular, we test the "orthodox" hypothesis observations (e.g., Frodi, Grolnick, & Bridges, 1985). that studies which more closely resemble the original Some studies included Interviews to assess positive Baltimore study show the strengest associations be- parental attitude toward the infant (Benn, 1986), tween parenting and attachment, whereas studies de- whereas others focused on the frequency of physical viating from this intensive, naturalistic longitudinal contact (Kerns & Barth, 1995). Although these ap- study yield less impressive results. More specifically, proaches may tap into some dimension of a broad the following hypotheses are tested: (1) Stronger as- concept of parenting that is pertinent to the develop- sociations between maternal behavior and attach- ment of attachment, they may not capture the origi- ment security are found in studies defining maternal nal concept of sensitivity. Studies designed according behavior äs sensitivity äs compared with studies de- to the basic features and conceptualizations of Ains- fining maternal behavior differently. (2) Home-based worth's Baltimore study (e.g., Belsky, Rovine, & Tay- studies show stronger associations between attach- lor, 1984; Grossmann, Grossmann, Spangler, Suess, & ment and maternal behavior than do laboratory stud- Unzner, 1985; Isabella, 1993) have yielded results that ies. (3) Long-term home observation studies show are closer to the original findings (Pederson et al., stronger relations between maternal behavior and in- 1990). This latter view, which we will refer to äs the fant attachment than do short-term, home-based "orthodox" position, can be summarized äs follows: studies. (4) Assessments of maternal behavior during (1) The mother's interactive behavior, in particular, the first year of life show stronger associations with her sensitivity, is considered to be the primary deter- attachment security than do assessments after age 1. minant of attachment quality; (2) however, if this as- (5) The longer the time interval between the assess- sociation is to be detected, the observations of mater- ment of maternal behavior and the Strange Situation nal behavior must be sufficiently intensive and procedure, the weaker the association between sensi- reliable, and the observed dimensions of maternal be- tivity and attachment (see Goldsmith & Alansky, havior must be conceptually close to sensitivity. 1987). Moreover, we will test whether the association In this meta-analysis, we integrate the available between sensitivity and attachment is dependent on studies on parenting and attachment in a quantitative contextual factors such äs the socioeconomic or manner. In a controversial field like attachment the- (non-)clinical Status of the families involved. The De Wolff and van IJzendoorn 573 orthodox hypothesis would emphasize a relatively The first group, Sensitivity, included all constructs ,onU'\t-tree Interpretation of this association. It may that conformed to the original definition of Mary .iKo be argued, however, that contextual factors Ainsworth and her colleagues (Ainsworth, Bell, & p.irth m ernde the influence of sensitivity on the de- Stayton, 1974): the mother's ability to perceive the in- \i'lopment of attachment, and that in lower-class or fant's Signals accurately, and the ability to respond to »limail samples the association between sensitivity these Signals promptly and appropriately. Constructs and Attachment would be weaker. that had been assessed with Ainsworth's rating scale In the first study, we report on the construction (Ainsworth et al., 1974) or a rating scale that was ex- ot homogeneous sets of concepts related to maternal plicitly based on Ainsworth's original rating scale intiTcKtn e behavior, through a new method of expert were also assigned to this first group. The second M>rting and rating procedures. Meta-analysis has of- group of constructs was maternal Contiguity of Re- ·. n Iven cnticized for mixing apples and oranges. sponse, which we defined äs promptness or, more i. .in-tul, systematic conccptunl analysis is required to generally, frequency of response to the infant's sig- ι si.iblish sets of similar studies that can be included nals. The most important factor distinguishing conti- •,n separate meta-analyses (Cooper & Hedges, 1994). guity of response from sensitivity is the absence of l bis study paves the way for the second study in any qualitative assessment of the mother's behavior. \\hit. h the methods and results of a series of meta- Only promptness or frequency of the mother's re- ,m.\l\ses on the association between various aspects sponses contribute to contiguity of response; appro- ot nidtornal Interactive behavior and attachment se- priateness is irrelevant. In the literature, contiguity cire described. In the second study, the crucial of response is often referred to äs "responsiveness." ot liomogeneity will be addressed statistically. Because the term responsiveness is sometimes also used to indicate sensitivity (e.g., "sensitive respon- siveness"), the concept "contiguity" was preferred to STUDY l avoid confusion (see Bornstein, 1989). The third group consisted of all constructs that referred to qual- Method ity or quantity of Physical Contact. The fourth group, Bivause the grouping of predictor variables may Cooperation, included constructs that bore on the ha\ L· dn important effect on the outcome of a meta- presence or absence of intrusive or interfering mater- .m.iKsis, experts were asked to categorize aspects of nal behavior. This concept was originally defined and nuiU'rnal interactive behaviors. In sorting the con- operationalized by Ainsworth (Ainsworth et al., stnuts, the experts were blind to the effect sizes asso- 1974). The second author independently sorted 15 ii.iti-d uith the specific constructs. In the total set of studies in the four categories. The percentage agree- jvitment research papers (see Study 2), 55 different ment was 93%, which documented the self-evident vonstriicts were identified, all referring to various as- nature of this preliminary step in the sorting proce- (vcts c-,t maternal behavior, for example, maternal dure. ΚκΚ· contact, maternal involvement, support, stimu- Fifteen constructs fitted into one of the four l.ition, sensitivity, verbal responsiveness, delight in groups; 40 constructs referring to other aspects of ma- t tu· mteraction, mutuality between mother and child, ternal interactive behavior remained to be classified. •nid trequency of positive responses to child. For each Two methods were used to create conceptually ho- Minstruct, we also identified the formal definition mogeneous groups of concepts: A sorting task in th.it \\ ds presented in the introduction or method sec- which the experts sorted the concepts into a few sub- tn'ns ot the research report, äs well äs the method groups, and a rating task in which the experts rated "t assossing that particular construct. On the basis of each concept in terms of its similarity to Ainsworth's shi-si di-scriptions, the experts were asked to sort the sensitivity construct and in terms of its importance However, because the task of sorting 55 for the development of attachment relationships. cts would have been extremely time consum- "X UTI> preliminarily assigned to one of the four in attachment research were asked to participate in ^Mi^httorward categories. These four groups en- the study. Experts were defined äs persons who had ^mp,isv.ed the following aspects of maternal behav- been actively involved in attachment research for r "t' (D Sensitivity, (2) Contiguity of Response, (3) several years and who were at least participating in l'lusuMl Contact, and (4) Cooperation. a graduate program in the behavioral sciences. Com- 574 Child Development plete data were obtained for 19 respondents (12 analysis in the case of categorical data. A graphical women). Because the sorting task was rather time configuration is constructed in which variables (i.e., consuming, four persons were unable to comply with the experts who sorted the concepts) and categories our request. Four nonrespondents refused for other (i.e., the groups into which the concepts were sorted) personal reasons. Almost half of the sample had ob- are being represented. In calculating the coordinates tained a doctoral degree (n — 9); the remaining partic- of variables and categories, HOMALS uses a distance ipants were graduate students. The respondents had model: Those concepts that are sorted more fre- been achvely involved in the attachment research for quently into the same group are represented in 5 years on average (SD = 3.89). the configuration relatively close to each other. HOMALS has proved to be a fruitful method for ana- lyzing sorting data of large numbers of Stimuli (see Materials Van der Kloot & Van Herk, 1991; Verkes et al., 1989). The Similarity Sorting Task was constructed to in- Following Verkes et al. (1989), we also performed a vestigate how the experts evaluated the conceptual centroid cluster analysis (Everitt, 1974) on the dis- similarities between the concepts. The sorting task tances between the 40 concepts in the HOMALS con- consisted of 40 small cards (15 X 10.5 cm), on each figuration to facilitate the Interpretation of its results. of which a particular concept was presented äs well äs a conceptual definition of that concept and, if nec- essary, the way in which it was assessed. On the Results cards, any Information about the source of the con- The Similarity Sorting Task cept was absent to keep the sorters blind äs to the effect sizes related to the concepts. The experts were Concepts that were sorted by an expert into the instructed to sort the cards into maximally 10 groups same group received the same numerical code. Con- of relatively similar concepts. The maximum of 10 cepts that were coded into different groups were groups was chosen to facilitate the sorting task and given different numerical codes. Those concepts that to guarantee the statistical power of the subsequent were not grouped with any other concept were analyses (Verkes, Van der Kloot, & Van der Meij, treated äs missing data. In this way a 40 (concepts) 1989). The rating task consisted of the same set of 40 X 19 (experts) data matrix was created. cards with conceptual definitions. We now asked the HOMALS yielded a two-dimensional configura- experts to rate each concept on a 7 point scale in tion that is shown in Figure l. The configuration rep- terms of its similarity to Ainsworth's sensitivity con- resented the similarities and dissimilarities of the 40 struct, which was presented along with the cards. concepts for maternal interactive behavior. The two This scale ranged from "not any similarity" (1) to dimensions had satisfactory eigenvalues of .87 and "exact similarity" (7). Besides the similarity to sensi- .81. This means that the 40 concepts are clearly differ- tivity, we also asked the experts to rate each concept entiated by the two dimensions of the HOMALS so- in terms of its importance for the development of an lution. To support the Interpretation of the HOMALS attachment relationship. Experts could indicate the configuration, we performed a hierarchical cluster concept's importance on a 7 point rating scale that analysis on the distances between all concepts. We ranged from "not at all important" (1) to "very im- used the linkage coefficient to assess the appropriate portant" (7). number of clusters to describe the data optimally (Aldenderfer & Blashfield, 1984). Five clusters were constructed, which are also presented in Figure 1. Procedure Synchrony can be defined äs "the extent to which The two tasks and a written instruction were sent interaction appeared to be reciprocal and mutually to the experts. We asked the experts to complete the rewarding" (Isabella, Belsky, & Von Eye, 1989, p. 13). Sorting Task before starting to accomplish the Rating Asynchronous instances of matemal and infant be- Task. havior are "those considered to reflect one-sided, un- responsive, or intrusive behavioral exchanges" (Isa- bella & Belsky, 1991, p. 376). "Positive mutuality," a Data Analysis concept that was applied by Kiser, Bates, Maslin, and The sorting data were analyzed with Homo- Bayles (1986), exemplifies the cluster Mutuality. Posi- geneity analysis using alternating least squares tive mutuality is a construct that consists of the fol- (HOMALS; Gifi, 1990). We used the SPSS-PC pro- lowing maternal behaviors: Number of "positive ex- gram HOMALS (SPSS Inc., 1990). HOMALS can be changes where both mother and infant attend to the considered äs an equivalent to principal components same thing," and "the mother's skill at modulating De Wolff and van IJzendoorn 575

poaltlve-affact accaption alf«ctlvHy attltuda(2) dallghl warm(h(1) warmth(2) aflaclionate-C'

amotion-expreaaion

Integration

aflactiva-qualtty

POSITIVE ATTITÜDE

supportlve-pressnce EMOTIONAL SUPPORT quallty-supporl

Figure 1 Two-dimensional representation of 40 concepts in the HOMALS solution the baby's arousal, her entertainment value, and her helping the child feel comfortable, and (b) being in- responsiveness to the infant's cues" (p. 71); it also in- volved äs manifested by the attentiveness to the child cludes some infant behaviors: "Expression of posi- and to the task." Affective Quality (Zaslow, Ra- tive affect, nonavoidance, active maintenance of the binovich, Suwalsky, & Klein, 1988) is an important interaction, and amount of gazing at the mother" construct in the cluster of Positive Attitüde. Zaslow et (p. 71). A central concept in the cluster Support is Sup- al. (1988, p. 290) defined this concept äs "the mother's portive Presence, which was introduced by Erickson, expression of positive affect to the baby, the mother's Sroufe, and Egeland (1985). Matas, Arend, and Sroufe expression of negative affect to the baby, and the de- (1978, p. 350) defined this concept äs follows: "The gree to which mother and infant engaged in recipro- extent to which the mothers appeared attentive and cal interactions." Finally, Stimulation can be de- available to the children and supportive to their ef- scribed äs "any action on the part of the mother forts. A high score on supportive presence involved directed toward her baby" (Miyake, Chen, & meeting two criteria: (a) Providing a secure base by Campos, 1985, p. 292). 576 Child Development

The Rating Task terized searches, manual search procedures, and con- sultation of other scientists working in this field. In a preliminary principal components analysis for The following computerized abstracting Services categoncal data (PRINCALS; Gifi, 1990), it was estab- were used to locate studies: Psychological Abstracts hshed that the expert raters vvere unidimensional in (from 1974 on), Educational Resources Information their ratings on the two rating scales (Verkes et al., Center (from 1983 on), and Social Sciences Citation 1989). The sorting data revealed five distinct clusters Index (from 1983 on). Furthermore, we used "World of parenting concepts. For each cluster, we computed Catalog," a database of the On-Line Contents Library the mean ratings. The five cluster means vvere found Center (OCLC). The following keywords were used to differ significantly from each other: On the first in different combinations: attachment, childrearing rating scale, similarity to Ainsworth's sensitivity con- practices, infant, mother, mothering, mother-child in- cept, the overall F(4, 35) = 5.52, p = .0015; on the teractions, mother-child relations, parent-child rela- second rating scale, importance for attachment devel- tions, parenting, responsiveness, and sensitivity. opment, the overall F(4, 35) = 5.03, p = .003. Multiple To Supplement these Computer searches, we used comparison tests, using the Tukey-HSD procedure, several manual search procedures. Reference lists mdicated that this overall difference between the five from existing reviews were inspected (Goldsmith & clusters could be ascribed to two clusters that were Alansky, 1987; Lamb et al., 1985; Lambermon, 1991), located at the outer part of the HOMALS configura- äs were reference lists of the reviewed articles. In ad- tion: Positive Attitüde and Stimulation. Concepts be- dition, we located unpublished dissertations through longing to one of these two broader concepts re- a manual search of Dissertation Abstracts International ceived significantly lower ratings on both rating (from 1980 on). To locate unpublished research, we scales than the concepts belonging to the other clus- also worked through several volumes of Conference ters. abstracts (ICIS, SRCD). Finally, we asked two promi- On the basis of this outcome, we distinguished in nent researchers in the attachment field, Drs. M. Main the total set of 40 concepts between a More Optimal and J. Belsky, to suggest additional studies. They Group, consisting of the clusters Synchrony, Mutual- were able to mention three additional studies. One ity, and Emotional Support, and a Less Optimal of the reviewers mentioned three additional studies Group of concepts, consisting of the clusters Positive (Fagot & Kavanagh, 1990; Malatesta, Culver, Tes- Attitüde and Stimulation. On the rating scales, the man, & Shepard, 1989; Seifer, Schiller, Sameroff, two global clusters vvere significantly different. On Resnick, & Riordan, 1996). After Submission of the the similarity scale, the group means vvere M = 4.10 manuscript, we came across two very recent studies and M = 3.09 for the More Optimal and Less Optimal (Gunnar, Brodersen, Nachmias, Buss, & Rigatuso, clusters, F(l, 35) = 16 34, p = .001. On the importance 1996; NICHD Early Child Care Network, 1996). scale, the mean rating for the More Optimal Cluster Each study had to meet three criteria for inclusion was M = 4.79 versus M = 4.30 in the Less Optimal in the meta-analysis. The first criterion was that the Group, F(l, 35) = 4.89, p = .02. study contained a measure of the mother's behavior In sum, we started with the more general construct toward her infant, and a measure of the infant's at- of maternal interactive behavior, and then distin- tachment security. The vast majority of maternal be- guished, partly with the help of experts, nine differ- havior measures have been derived from observa- ent groups, each referring to a narrower construct. tions of mother-infant interaction. In a few studies, For each group we will conduct a separate meta- maternal behavior was measured using question- analysis. The empirical approach of conceptual anal- naires or Interviews with the mother. For example, ysis through expert ratings enabled us to system- Benn (1986) and Bretherton, Biringen, Ridgeway, atically differentiate between several dimensions Maslin, and Sherman (1989) assessed maternal sensi- of parenting behavior independent of the meta- tivity in an interview with mother, whereas Izard, analytic results. In Study 2, we describe the methods Heynes, Chisholm, and Baak (1991) used a questiOn- and results of the subsequent meta-analyses. naire to assess the mother's style of emotional expres- sion. In the assessment of attachment security, the STUDY 2 Strange Situation procedure was, of course, used in Method most studies. In some studies, the original procedure was changed because the researchers adapted the Selection of the Studies procedure to a home Situation, or because they To identify studies for inclusion in the meta- wanted to limit the infant's stress (see Capps, Sig- analysis, we applied three search strategies: Compu- man, & Mundy, 1994). In other studies, the Waters De Wolff and van IJzendoorn 577 and Deane (1985) Attachment Q-Sort for observers of quality of research in three ways. First, the overall was used instead of the Strange Situation (Vaughn & quality of the study was estimated in terms of the Waters, 1990). We did not include studies using the quality of the publication medium. For Journals, the Mother Attachment Q-Sort, because a recent meta- "impact factor," defined äs the average number of analysis clearly showed its lack of convergent and citations to the papers in a Journal, was considered discriminant validity (van IJzendoorn, Vereijken, & äs a proxy of quality (Garfield, 1979). The impact of Riksen-Walraven, in press). Conference papers and dissertations was set at zero. The second criterion for inclusion was that each The overall quality of the studies was then included study should report an association between maternal in the meta-analyses äs a moderator variable. Second, behavior and infant attachment, so that effect sizes we tested the influence of publication Status (pub- could be calculated. Even if only nonnumerical infor- lished versus unpublished) on the size of the com- mation was provided (i.e., "no relation" or "signifi- bined effects of the associations between attachment cant relation"), we included the study in the meta- and aspects of parenting. Third, we provided com- analysis and estimated the effect sizes (Müllen, 1989). bined effect sizes for associations between attach- In some studies in which both maternal behavior and ment and sensitivity separately for the total set of quality of attachment were assessed, however, the re- pertinent studies, äs well äs for the subset of studies searchers did not compute any association between using the Standard assessment of attachment, that is, the two assessments, nqr did they report on the asso- the Strange Situation procedure, antecedent or con- ciation (see Easterbrooks & Goldberg, 1985). These current obse'rvational assessment of sensitivity, and studies had to be excluded. nonclinical participants. These studies more closely Our third criterion concerned Intervention studies, resembled the original Baltimore study in their as- The central hypothesis in our meta-analysis refers to sessments of predictor and outcome variables. the association between maternal behavior—äs it A total of 66 studies were identified that together occurs naturally—and infant attachment. Conse- involved 4,176 mother-infant pairs. The meta- quently, if maternal behavior was experimentally in- analytic database included 54 published articles, äs fluenced through Intervention or therapy, we only well äs 12 unpublished Conference presentations or included data for the nontreated control group. Thus, dissertations. A common criticism of meta-analysis is only those Intervention studies that reported separate that studies with significant findings are overrepre- data for the control group(s) could be included in the sented in the meta-analytic database because studies meta-analysis (for a meta-analysis of Intervention with significant findings are more likely to be pub- studies, see van IJzendoorn, Juffer, & Duyvesteyn, lished than studies with nonsignificant findings. It is 1995). assumed that many studies with effect sizes of zero Overall, we adopted a fairly liberal inclusion strat- remain unpublished in file drawers. If a rneta-analyst egy. No studies were excluded from the meta- relied exclusively on published studies, this could re- analysis on the basis of flawed design. Both Hedges sult in an overestimation of population effect sizes. (1986) and Müllen (1989) advise against excluding Rosenthal (1979) identified this äs the "file drawer" studies of low quality, because it is very difficult to problem. The best solution to the file drawer problem assess a diffuse dimension like "study quality" di- is simply to open the drawers and include äs many rectly and appropriately. Instead of using study qual- unpublished studies äs possible (Rosenthal, 1991). ity äs a criterion for inclusion in the meta-analysis, Unpublished dissertations were located through a the meta-analyst is advised to include studies of manual search of Dissertation Abstracts International varying quality. More detailed Information on study (from 1980). We also searched through several Con- quality should be included in the coding System, so ference proceedings (ICIS, SRCD) in which abstracts that the influence of study characteristics on effect of presented papers were published. size can be examined in the meta-analysis. In the area Another way of dealing with the "file drawer" of attachment research, the quality of the studies re- problem is to estimate the magnitude of the problem mains a hotly disputed issue. The main body of stud- by calculating the minimum number of unpublished ies on attachment has been published in leading Jour- studies with null results that would be required to nals such äs Child Development (k = 21), Developmental turn a significant meta-analytic finding into a nonsig- Psychology (k = 7), Infant Behavior and Development (k nificant one. This number of imaginary unpublished ~ 6), the Monographs oftlie SRCD (k = 3), and in other studies is called the "fail safe number" and was intro- refereed Journals. A number of studies, however, duced by Rosenthal (1979). If this fail safe number have been published in less prestigious Journals, or is relatively small, a file drawer problem may exist. remained unpublished. We have addressed the issue Calculation of the fail safe number is no definitive 578 Child Development solution to the file drawer problem. It only "estab- mogeneity test is significant, Johnson, Müllen, and lishes reasonable boundaries on the file drawer prob- Salas (1995) encouraged the meta-analyst to check for lem and estimates the degree of damage to the re- significant moderator variables that may partly ac- search conclusion that could be done by the file count for the Variation across studies (see also Rosen- drawer problem" (Rosenthal, 1991, p. 104). thal, 1995). To determine whether a study feature sig- nificantly explains a part of the Variation in effect sizes, we used Rosenthal's method of focused com- Calculation of Effect Sizes parison of combined effect sizes (Müllen, 1989). In the meta-analysis, Pearson's product-moment Last, blocking was used to reveal moderators in correlation coefficient (r) was used äs the effect size the dataset. Blocking involves grouping study out- estimate. An effect size indicates the magnitude of comes on the basis of a potential moderator variable. the association between two variables, disregarding Within each level of the moderator variable, a com- sample size. If a study reported means and Standard bined effect size can be computed, and the signifi- deviations, one-directional t values were computed cance of the differences can be tested (Müllen, 1989). and transformed into r using Schwarzer's (1989) algo- rithms. If no means and Standard deviations were Multiple Outcomes within the Primary Study available, the reported test statistics (t, F, or chi- square) or the one-directional p value were trans- In meta-analysis, the study itself is the unit of anal- formed into r \vith Mullen's (1989) Computer pro- ysis. Because the meta-analytic procedures assume gram. independence of units of analysis, we had to deal Because only F tests with l degree of freedom in with the problem of multiple outcomes from single the numerator are appropriate for inclusion in meta- studies. In this meta-analysis, there were three types analysis, a contrast F with l degree of freedom was of multiple outcomes. computed on the basis of the global F value using the First, in several studies the same maternal behav- Contrast Analysis procedures (Rosenthal & Rosnow, ior was assessed at multiple points in time. In these 1985). Global F tests can be converted to contrast Fs cases we computed a combined effect size over the only if group means are available. We applied con- various measurements; the combined effect size was servative estimation procedures if a study only re- treated äs the outcome of such a study. For example, ported "no significant effect" (p = .50), or "a signifi- Isabella (1993) assessed maternal sensitivity at l, 4, cant effect" (p = .05) in the case of univariate and 9 months of age. The association between mater- relations (Müllen, 1989). nal sensitivity and the Strange Situation outcome To compute combined effect sizes, each correlation was reported for each age separately. In the meta- coefficient was transformed to a Fisher's Z to make analysis, we took the mean outcome across the three the sampling distribution of r more approximate to times of assessment (see Goldsmith & Alansky, 1987). a Gauss curve. The distribution of r be~omes nonlin- Second, several studies examined the relation be- ear at the extreme ends of the scale (Müllen, 1989). tween a variety of maternal behaviors and attach- Furthermore, in Computing the combined effect sizes, ment security. In the first study, we identified nine individual effect sizes were weighted by sample size, categories of predictor variables, each referring to a because correlations become more stable äs sample specific aspect of maternal behavior (see Study 1). size increases and because effect sizes based on large After determining the kind(s) of behavior that had samples deviate less from the population effect size been examined in each study, we assigned each be- than those based on smaller samples (Müllen, 1989; havior to one of our nine categories of maternal be- Rosenthal, 1991). As an indicator of accuracy, we havior. If two maternal behaviors were classified in computed the 95% confidence interval around the the same category, we computed the mean effect size. mean effect size, using Schwarzer's (1989) program. For example, Goldberg, Perotta, Minde, and Corter's Combining estimates of effect sizes across studies (1986) measures of acceptance, delight, perception of is reasonable if the studies have a common popula- the baby, and attitude were all assigned to the cate- tion effect size. If this is the case, estimates of effect gory Positive Attitude toward the infant. We there- sizes will differ only because of unsystematic sam- fore computed a mean effect size over these four be- pling error. The crucial question is thus whether the haviors. underlying dataset is sufficiently homogeneous. Third, because many studies assessed different as- Hedges and Olkin (1985) äs well äs Rosenthal (1991) pects of maternal behavior in their primary analyses, advised the use of a test for homogeneity to deter- data from a single study were often included in sev- mine to what extent effect sizes are relatively con- eral domains of maternal behavior. For example, the stant across studies. Regardless of whether this ho- study of Goldberg et al. (1986) provided data for al- De Wolff and van IJzendoorn 579 most every domain of maternal behavior. In case of tion Status (published, unpublished) and year of pub- combining domains into the more and less optimal lication. Second, we coded several characteristics of clusters, we computed means over the study out- the sample: sample size, whether the sample was spe- comes for the domains so that each study had only cial (i.e., whether the mother or the child suffered one outcome in the meta-analysis. from mental or physical handicaps, and whether the Fourth, some studies were conducted by the same family conditions were deviant), socioeconomic sta- research team and included the same samples (Isa- tus of the sample (middle class, lower class, or hetero- bella & Belsky, 1991, and Isabella et al, 1989; Goos- geneous), and whether the infants were all firstborn. sens, 1987, and van IJzendoorn, Kranenburg, Zwart- The most important variables concerned study de- Woudstra, Van Busschbach, & Lambermon, 1991; sign. A total of eight design variables were coded: Main, Tomasini, & Tolan, 1979, and Londerville & (1) the techniques for measuring maternal behavior Main, 1981; Miyake et al., 1985, and Nakagawa, (global rating scales, specific behavioral codings Lamb, & Miyake, 1992; Bates, Maslin, & Frankel, based on time or event sampling, or nonobserva- 1985, and Kiser et al., 1986, and Frankel & Bates, tional techniques like Interviews and questionnaires); 1990). In these cases we also computed a mean effect (2) the exact duration of the observation, in minutes; size. In other words, studies involving the same sam- (3) location where the mother-infant interaction was ple were represented by only one effect size in the observed (home or laboratory); (4) the age of the child set of meta-analyses. at the time of the observation; (5) whether the Strange Situation or an alternative measure was used to as- sess attachment security; (6) the age of the child at the Coding the Variables time of assessment of attachment security; (7) time The outcome variable, attachment security, did interval in months between the two assessments; and not need to be coded because it is a sharply defined (8) whether the assessments of maternal behavior construct. In the majority of the studies, attachment and infant attachment had been conducted com- security was assessed by means of Ainsworth and pletely independently. Wittig's (1969) Strange Situation procedure. In four If maternal behavior was assessed at different in- studies, shortened versions of the Strange Situation fant ages, we computed the mean age. For example, procedure were applied (Bohlin, Hagekuli, Germer, in Belsky et al.'s (1984) study, maternal behavior was Andersson, & Lindberg, 1989; Capps et al., 1994; observed at l, 3, and 9 months of age. The Strange Lewis & Feiring, 1989; Persson-Blennow, Binett, & Situation was assessed at 12.5 months. In this case, McNeil, 1988). A total of five studies utilized alterna- we coded the age of the child at the maternal behav- tive assessments of attachment, like the Attachment ior observation (moderator 4) äs 4.3 months; the time Story Completion Task (Altman, Monk, Jones, & interval between predictor and outcome assessments Sosa, 1993; Goodman, Andrews, Jones, Weissman, & (moderator 7) was coded äs 8.2 months. Weisman, 1993) and the Attachment Q-Sort (Kerns Two persons coded the moderator variables. After & Barth, 1995; Pederson et al., 1990; Pederson & a training phase, we performed a reliability check in Moran, 1996). Recently, van IJzendoorn et al. (in which each coder independently coded 15 studies. press) showed that the observer Attachment Q-Sort The mean percentage of agreement was 92%. The is a reliable and v^lid measure of attachment security, coding of variables such äs publication year, age of and they argued that in several respects the Attach- the mother, parity, age of the child at predictor, and ment Q-Sort is even preferable to the Strange Situa- outcome assessment appeared to be straightforward tion procedure. In the same meta-analysis, the (for these variables, intercoder agreement was 100%). mother Attachment Q-Sort was shown to be an in- Most problematic was the coding of the measure- valid self-report attachment measure. Studies with ment technique used to assess maternal behavior the mother version of the Attachment Q-Sort were (global versus specific). In this case, intercoder agree- excluded, therefore, from the current meta-analyses. ment was 67%. Furthcrmore, ve focused in the meta-analyses on at- tachment security (using either the secure-insecure dichotomous variable or security ratings), and we de- Results cided not to include the traditional split between in- Central Tendency and Variability secure attachment classifications because only part of the studies report on the difference between the inse- Table l presents the mean weighted effect sizes for cure classifications. each domain of maternal behavior, äs well äs confi- We coded a number of study characteristics. First, dence intervals, homogeneity tests, and fail safe num- some background variables were coded, like publica- bers. The data pertain to all the studies except the 580 Child Development

Table l Parenting and Attachment: Meta-Analytic Findings for Each Domain of Parental Behavior

Effect 95% Studies Participants Size Confidence Homogeneitv Fail Safe Parental Behavior (N) (N) (r) P Interval ~/L! ' (N)

Sensitivitv 30 1,666 .22 9.12 E-15 .18-.27 43.5" 861.8 Ainsworth-scjle 16 837 .24 1.55 E-09 .17-.30 18.8 238.9 ContiguiU öl" response 14 825 .10 .01 .03-. 17 20.4 38.7" Phvsical contact 9 637 .09 .04 .01-. 17 7.8 14.1b Cooperation 9 493 .13 .007 .03-.21 11.1 17.21' More optimal group 28 1,928 .19 7.77 E-08 .14- .23 63.7·' 735.6 Svnchronv 6 258 .26 .0001 .14-.37 11.9' 35.5b Mutualitv .·) 168 .32 .00003 .18-.46 9.3·' 17.6" Support ->-> 1,664 .16 7.41 E-06 .11-.21 37.1 328.2 Less optimal group 24 1,233 .19 7.88 E-10 .14-.26 30.5 377.8 Attitudj 21 1,092 .18 3.02 E-08 .14-.25 29.4 273.2 Stimulation 9 422 .18 .0001 .10-.29 7.5 32.0b

Total set 123' 7,225 .17 8.24 E-24 .15-. 19 198·' 9,923.5 Random set 66 4,176 .19 1.83 E- 17 .16-.22 139·' 3,957.3

Data set is heterogeneous. File drawer problem is indicated. Total \vithout overlapping studies. pioneering Baltimore study (Ainsworth et al., 1978) tion, a similar combined effect size was found: r(420) that inspired so many replication studies. Although = .18 (k = 9,N = 422). For Positive Attitüde, the effect inclusion of the Baltimore study would not change size was r(l,090) = .18 (k = 2l, N = 1,092), and for any of our results drastically, we decided to exclude Emotional Support this figure was r(l,662) = .16 (k this study for two reasons. First, it was in the Balti- = 22, (V = 1,664). In the set of studies on attachment more study that central measures for sensitivity and and Synchrony, the combined effect size appeared to attachment were developed and the hypothesis be similar to the outcome in the domain of Sensitiv- about the association between attachment and sensi- ity: r(256) = .26 (k = 6, N = 258). All effect sizes were tivity was generated and specified. Later studies can significant at the alpha = .05 level. be considered äs replications and extensions. Second, Besides meta-analyses on each domain, we also the Baltimore study showed very strong associations performed two meta-analyses on a broader concep- between attachment and several aspects of parenting, tual level. Overall effect sizes, however, did not differ whereas later studies generally showed less strong between the more optimal and the less optimal düs- relations; in many respects the Baltimore study oc- ter: »-(1,926) = .19 (k = 28, N = 1,928) in the more cupies a somewhat outlying position (see Tables 3 optimal group, and r(l,231) = .19 (k = 24, N = 1,233) and 4). in the less optimal group. On the most global level, As can be derived from Table l, the combined ef- we also computed an Overall effect size of the studies, fect size for the association between maternal sensi- irrespective of their cluster membership: mean tivity and attachment was r(l,664) = .22 (k = 30, N r(7,223) = .17 (k = 123, N = 7,225). This global ap- = 1,666). However, when we adopted a more strict proach increased, of course, the heterogeneity of the definition of the predictor variable and included only set of studies, and it also led to the inclusion of de- studies that measure sensitivity using Ainsworth's pendent outcomes of the same studies. Therefore, we original rating scale, the effect size increased to r(835) randomly selected one effect size per study, in case = .24 (k = 16,N = 837). The correlation between ma- of multiple outcomes. The resulting overall effect size ternal behavior and infant attachment was highest was r(4,174) = .19 (k = 66, N = 4,176). In this random for the set of studies on Mutuality, r(166) = .32. How- set of 66 effect sizes, we tested whether the combined ever, this meta-analysis involved only three studies effect size of the sensitivity cluster was significantly (,V = 168). Weaker effect sizes were found in the set different from the other clusters. Because eight com- of studies on Contiguity of response, r(823) = .10 parisons were made, the Bonferroni corrected alpha (k = 14, N = 825), and Physical Contact, r(635) = .09 level was used. The Sensitivity cluster appeared to (k = 9, N = 637). The combined effect size for Cooper- show a significantly larger effect size than the Conti- ation was ;-(491) = .13 (k = 9, N = 493). For Stimula- guity of response cluster (p = .004). De Wolff and van IJzendoorn 581

The fail safe numbers indicated possible file Support, the disjoint cluster analysis set the Matas et drawer problems in several domains of maternal be- al. (1978) study apart from the other studies which havior: Contiguity of response, Physical Contact, Co- constituted a homogeneous set, χ:(20, Ν = 1,616) = operation, Synchrony, Mutuality, and Stimulation. 18.4, p = .56. The 21 studies in this set showed a com- The combined effect sizes for these domains of par- bined effect size of r(l,614) = .14. enting need to be considered carefully, äs they re- The more optimal group of 28 studies appeared quire further corroboration. In the other domains, fail to become homogeneous after exclusion of the Benn safe numbers exceeded Rosenthal's (1991) critical (1986), Isabella et al. (1989), and Matas et al. (1978) value (5 X k + 10). These effect sizes can be consid- studies that were identified through the disjoint clus- ered robust. For example, 862 studies with null re- ter analysis. The remaining set showed a combined sults in the file drawers of disappointed researchers effect size of r(l,818) = .15 (k = 25); χ:(24, Ν = 1,820) would be required to make the effect size for the asso- = 22.0, p = .57. In five of the domains of maternal ciation between sensitivity and attachment nonsig- behavior (Positive Attitüde, Stimulation, Contiguity nificant. of response, Cooperation, and Physical Contact), ef- Table l also presents the test statistic chi-square, fect sizes appeared relatively constant across studies, which indicates whether the effect sizes within the even though some of the meta-analyses included a domain were homogeneous. The combined effect large number of studies. Effect sizes in the "less opti- size for heterogeneous clusters constitutes problem- mal group" were also found to be homogeneous. We atic estimates of the popuiation effect size. Therefore, nevertheless performed moderator analyses in these we tried to find homogeneous subsets of studies us- domains because a nonsignificant homogeneity test ing disjoint cluster analysis (Müllen, 1989; Schwarzer, "does not mean that there is no interesting variability 1989). For the Sensitivity domain, we found two sig- in the research domain" (Müllen, 1989, p. 102). nificantly disjoint clusters (alpha was set at .05). One small and heterogeneous cluster consisted of the Prediction study by Capps et al. (1994) on autistic children, and the study by Benn (1986), who used an interview to In Table 2, the moderators and their : \ alues are assess sensitivity. The remaining studies were homo- reported for each subset. Agam, the Baltimore study geneous, χ2(27, N = 1,621) = 30.6, p = .29, and the (Ainsworth et al., 1978) was not included because of combined effect size was r(l,619) = .21 (k = 28). In- its exploratory nature. cluding studies similar to the Ainsworth Baltimore Publication Status and impact factor were not sig- study in terms of (1) the use of the Strange Situation nificant moderators in any of the nine domains of procedure, (2) the application of observational mea- analysis. In this respect, then, the data do not point sures of sensitivity, (3) nonclinical participants, and to a file drawer or quality problem. Year of publica- (4) sensitivity assessments preceding or concurrent tion also was not significant in any domain. The du- with the attachment assessment, we found a com- ration of the maternal behavior assessment, the use bined effect size of r(l,097) = .20 (k = 21), and this of the Strange Situation procedure or alternative at- set of studies was homogeneous, χ2(20, Ν = 1,099) = tachment measures, the independence of parenting 25.2, p = .19. and attachment assessments, and whether the sample For the Synchrony domain, disjoint cluster analy- consisted exclusively of firstborn infants did not sis showed two different clusters: The study of Isa- emerge äs significant moderators äs well. In meta- bella et al. (1989) was set apart from the other studies. analyses, sample size offen is a significant moderator Isabella et al. 's (1989) study included an equal num- because smaller samples usually tend to show ber of avoidant (n = 10), secure (n = 10), and ambiva- strenger effect sizes than larger samples (see lent (n = 10) infants, selected from a larger sample Amato & Keim, 1991). In our case, sample size was of 51 infants. In particular, the secure group con- a significant moderator in four domains: Sensitivity tained only the "most secure" B2 and B3 patterns of (z = 2.02), Synchrony (z = 3.39), Mutuality (z = 2.36), attachment (Ainsworth et al., 1978). The remaining and Positive Attitüde (z = 2.06). Smaller samples in- Synchrony studies constituted a homogeneous set, deed showed stronger effect sizes. The remaining χ2(4, N = 228) = 3.4, p = .39, and their combined moderators will be discussed separately for each do- effect size was r(226) = .19 (k = 5). For the domain main. of Mutuality studies, the disjoint cluster analysis dis- Sensitivity. In Table 3, a stem and leaf display of criminated the Smith and Pederson (1988) study from the effect sizes in the most important domain, Sensi- the two other studies. This latter set, however, ap- tivity, is presented. It shows the somewhat outlying peared to remain heterogeneous. In the domain of position of the Baltimore study, äs well äs the Seifer 582 Child Development

Table 2 The Effect of Moderators in Different Parenting Domains (z Values)'

St'nsitn it\ Less Optimal Cluster 1VKMC wpimiai V_ILIS>ltl \inbworth Contiguitv ot Phvsical Positive Total Scale Response Contact Cooperation Svnchrony Mutuahtv Support Attitüde Stimulation Moderator (fc - 30) (fc - 16) (fc = 14) (fc -9) (fc = 9) (fc = 6) ' (fc = 22) (fc = 21) (fc = 9)

Publikation 1 89 91 «7 97 " b 1 49 21 61 b [mpatt tactor 14 28 82 91 09 134 46 26 27 61 U'ar 97 92 14 31 52 172 17 167 65 81 Si/e 202* 28 148 1 76 81 3 39*** 236* 176 206* 26 SIS 241* 1 81 41 1 10 67 186 e 29 71 26 Climcal 84 17 66 13 210* 1 86 176 110 278" 51 hrstborni 28 57 136 79 76 187 h 19 112 11 Global 87 — 234* 14 259" b b 112 209* 24 Duration 27 99 30 19 78 1 23 149 142 46 1 17 Home lab 20 65 1 26 1 65 163 133 149 93 170 233* \ge 256** 64 258" 1 19 62 25 219* 131 76 189 SSP 1 95 1 32 144 1 86 60 b h 05 98 b \ge SSP 231* 1 35 95 102 68 67 176 12 21 88 Tiim mterxal 269" 79 297** 128 33 40 176 80 182 220* Independent 35 49 — — r 1 14 b b 46 b b

1 All computations without the Amsworth et al (1978) sample ' \o \anation in the moderator variable *p < 05, **p < 01, *** p < 001 et al. (1996) study with a negative correlation be- neous socioeconomic Status (e.g., variability from tween sensitivity and attachment. lower to higher class) were excluded from these anal- Socioeconomic Status appeared to be a significant yses. moderator tor the 30 studies assessmg sensitivity The age of the infants at the time of the sensitivity äs defaned by Ainsworth et al (1978) (z = 2.41; see assessment was also a significant moderator (z = Table 2). In the 18 middle-class samples, the effect 2.56). Contrary to our expectation, samples with size was r(886) = 27 (N = 888), whereas in the eight older infants (older than l year) showed stronger ef- lower-class samples, this figure was r(650) = .15 (N fect sizes, r(l, 062) = .27 (k = 11) than samples with = 652) Samples without clearly specified or homoge- younger infants, r(602) = .20 (k = 19). Also, the age of the infants at the time of the attachment assess- ment was a significant moderator (z = 2.31). Samples Table 3 Stern and Leaf Display of Effect Sizes for the Associa- with older infants showed stronger effect sizes, r(847) tion between Sensitivity, Contiguity, and Attachment (r) = .25 (k = 16) than samples with younger infants,

Leaf r(817) = .19 (fc = 14). A shorter time interval between the sensitivity and attachment assessments led to Stern Sensitiv itvh Contiguity stronger effect sizes (z = 2.69) in the total set of sensi- tivity studies. 9 Contiguity of response. In Table 3, a stem and leaf 8 display of the effect sizes for the Contiguity studies 7 8 6 Λ 8 is presented. Three moderators were important: ΐ 1 0 Whether contiguity of response was assessed glob- 4 1, \ 4 8 ally (z = 2.34), the age of the infant at the contiguity 3 0, 2, 3, 4 assessment (z = 2.58), and the time interval between 2 0, 1, 3, Λ 6, 6, 7 3, 9 the contiguity and attachment assessments (z = 2.97). 1 2, 4, 4, 5, 6, 7, 8, 9 U, 5 0 0, 0, 4, 4 0, 0, 1, 5, 6, 7, 8 More global assessments, r(363) = .08 (fc = 8), showed - 1 τ 7 weaker associations than more specific assessments, r(458) = .15 (fc = 8). Contrary to the moderator results ' The stem contams the hrst digit of the correlation coefficient, in the domain of sensitivity, a longer time interval, u hereas the leat contams the second digit r(352) = .16 (fc = 6), led to a stronger effect size than Lnderscored tigureb mdicate studies vwth the Ainsworth et al (1978) rotmg scale for sensitiv itv a shorter interval, r(469) = .06 (fc = 9), and samples Amsuorth et al (1978) with younger infants, r(501) = .13 (fc = 7), yielded De Wolff and van IJzendoorn 583

somewhat strenger associations than samples with Table 4 Stem and Leaf Display of Effect Sizes for the Associa- older infants, r(320) = .06 (k = 7). tion between the Less Optimal Cluster (Positive Attitude, Stimu- Cooperation. Two moderators were significant (z = lation), the More Optimal Cluster (Synchrony, Mutuality, Sup- 2.10 and z = 2.59, respectively): Studies with special port) and Attachment (r) samples revealed weaker effect sizes, r(223) = .03 (k Leaf = 4), than studies with normal samples, r(266) = .20 (k = 5). Global assessments of Cooperation yielded Stem' Less Optimal Cluster More Optimal Cluster weaker effect sizes, r(360) = .05 (k = 6), than specific 9 assessments, r(357) = .32 (k = 2). These results need 8 to be interpreted with caution, however, because 7 5' 7b only nine studies were included in this dataset. For 6 6 2, 6, 6 this small dataset, a stem and leaf display is not pre- 5 sented. 4 5 7 3 1, 4, 9 1, 1, 5, 8 Positive Attitüde. Special samples showed signifi- 2 2 3 3, 4, 6, 6, 7 0 1 7 4 (S 7 cantly (z = 2.78) weaker associations between Posi- 1 1 ·> 346889 0, 2, 4, 5, 6, 7, 9 tive Attitüde and attachment, r(337) = .08 (k = 6), 0 0, 0, 1, 9 0, 0, 0, 0, 5, 9, 9 than normal samples, r(758) = .23 (k = 15). Stud- 1 ies using global attitude assessments, r(546) = .21 The stem contams the first digit of the correlanon coefficient, whereas the leaf wntams the second digit (k = 11), showed significantly stronger effect sizes b Ainsworth et al (1978) (z = 2.09) than studies with specific assessments, r(357) = .06 (k = 6). Stimulation. Contrary to our expectation, studies titude and Stimulation (k = 24). In Table 4, the stem on attachment and Stimulation at home revealed and leaf display of the effect sizes in this domain is weaker associations, r(229) = .07 (k = 5), than studies presented. The Ainsworth et al. (1978) study is some- carried out in the laboratory, r(169) = .31 (k = 3). what of an outlier. Significant moderators were: clini- Longer intervals again led to smaller effect sizes than cal Status (z = 2.50), time interval (z = 2.38), and shorter intervals (z = 2.20). This set of studies was global versus specific assessment of maternal behav- rather small. ior assessment (z = 2.03). In the 18 normal samples, The moderator analyses did not uncover signifi- the combined effect size was r(896) = .22, whereas in cant moderators in the domains of Physical Contact the six clinical samples it was r(333) = .10. Longer and Emotional Support. In the domain of Synchrony, time intervals led to somewhat smaller effect sizes only sample size appeared to be a relevant modera- than shorter time intervals (z = 2.38). Thirteen stud- tor. The domain of Mutuality consisted of only three ies in which maternal behavior was assessed globally studies; the Smith and Pederson (1988) study showed showed a combined r(681) = .22, and the six studies the strongest effect size, and the characteristics of this in which maternal behavior was assessed in a specific study determined the outcome of the moderator way yielded a combined r(342) = .06. analysis. Special samples. In two domains, Cooperation and Synchrony, Mutuality, and Emotional Support. The Positive Attitude, the clinical Status of the samples three domains Synchrony, Mutuality, and Emotional appeared to be a significant moderator: Normal sam- Support were considered to be the "more optimal" ples revealed stronger associations between attach- cluster. In Table 4, the stem and leaf display of the ment and parenting than did special samples. We hy- effect sizes in this domain is presented. There were pothesized that relatively "mild" factors such äs no negative effect sizes, and again, the Ainsworth et prematurity and adoption have no moderating effect al. (1978) study occupied a somewhat outlying posi- on study outcome, whereas more severe factors like tion. deafness, autism, and maltreatment lead to a weaker We performed a moderator analysis on this cluster association between maternal behavior and attach- because its outcome would be more robust (k = 28). ment. All of the studies using special or clinical Publication Status of the study (z = 2.23) was a sig- groups were combined (k = 10). These studies were nificant moderator. Seven unpublished studies divided into groups on the basis of whether the sam- yielded a combined effect size of r(861) = .11, ple involved was "mildly" clinical or more severely whereas the 21 published studies yielded a combined clinical. However, this distinction did not prove to be effect size of r(l,063) = .25. a significant moderator variable (z = .84, p = .20), Positive Attitude and Stimulation. The less optimal although the trend was clearly in favor of the hypoth- group was a combination of the domains Positive At- esis. The six mildly clinical samples (four premature 584 Child Development samples and two adoption samples) yielded a com- play (BESD) is an alternative Interpretation of effect bined effect size of r(265) = .26, and the four severely sizes. The BESD depicts an effect size (r) in terms of clinical samples (maltreatment, cleft palate, deafness, the improvement rate that is attributable to the pre- and autism) shovved a combined effect size of r(218) dictor variable. Applying this approach to the find- = .16. ings of the current meta-analyses, the correlation of r - .24 for those studies closely resembling the original Baltimore study represents an improvement in secu- Discussion rity from 38% to 62%; that is, infants whose mothers Maternal sensitivity, defined äs the ability to re- respond sensitively to their Signals improve their spond appropriately and promptly to the Signals of chance of developing a secure relationship from 38% the infant, indeed appears to be an important condi- to 62%, whereas infants whose mothers are less sensi- tion for the de\relopment of attachment security. For tive decrease their chance of developing a secure rela- the 30 pertinent studies, the combined effect size was tionship from 62% to 38%. This improvement rate can K l,664) = .22 (N = 1,666). Including only the 21 stud- hardly be considered trivial in a theoretical or practi- ies using the Strange Situation procedure in nonclini- cal sense, in particular when we compare this effect cal samples, äs well äs applying observational sensi- size with famous examples from medical research, tivity measures preceding or concurrent with the such äs the widely used heart failure reducing drugs attachment assessment, we found a combined effect Propranolol (r = .04) and aspirin (r = .03) (Gage, size of r(l,097) = .20 (/V = 1,099). Applying the 1996; Rosenthal, 1991). Hunter and Schmidt (1990) procedures of correcting We therefore cannot agree with Goldsmith and results for attenuation based on the reliabilities of the Alansky's (1987) conclusion that there is only a weak measures (mean reliability of the sensitivity mea- association between sensitivity and attachment. In 12 sures was .83, and mean reliability of the Strange Sit- studies, they found an overall effect size of r = .16 uation procedure was .81), we found a "true" popu- (without the Ainsworth et al. [1978] study), which is lation effect size for the association between remarkably similar to our overall combined effect attachment and sensitivity of r(l,097) = .24. The 16 size across all study outcomes, r(7,223) = .17. First, studies using the original Ainsworth et al. (1974) sen- on the basis of the BESD approach, we are inclined siti\ ity scale also showed a combined effect size of to be more impressed with this overall effect size than r(835)' = .24 (.V = 837). Cohen (1988, p. 82) has pro- Goldsmith and Alansky (1987) were. Second, our posed criteria for small, medium, and large effect combined effect size represents somewhat stricter sizes (d = .20, d = .50, and d = .80, respectively) corre- replication studies of the original Baltimore study. sponding to correlations of .10, .24, and .37, respec- Goldsmith and Alansky (1987) derived their com- tively. According to these criteria, the size of the asso- bined effect size from 12 studies in which a variety ciation between sensitivity and attachment in the of parenting measures had been examined in relation replication studies is medium. After riore than 25 to attachment security. These disparate measures years of research, Bowlby's (1969) important question were combined in a single meta-analysis. For exam- about the role of sensitivity in the development of ple, behavioral categories such äs the frequency of infant attachment can therefore be answered in the maternal looking or vocalizing were combined with affirmative. Ainsworth's rating scale for sensitivity. We showed Cohen's (1988) criteria are rather arbitrary, how- that Contiguity of response is significantly less ever, even according to the originator (Cohen, 1962). strongly associated with attachment security than Several authors argue that criteria for the statistical Sensitivity, and that the moderator analysis in the magnitude of effect sizes are not equivalent to criteria former domain leads to opposite outcomes, in partic- for their theoretical or practical importance (Abelson, ular with respect to infants' age and time interval be- 1995; Prentice & Miller, 1992; Sechrest & Yeaton, tween assessments. Hedges (1986) argued that the re- 1982). Rosenthal (1990) criticized simplistic Interpre- sults of meta-analyses using broad constructs may tation of the squared correlation coefficient in terms obscure important differences among narrower con- of percentage of explained Variation: "From under- structs subsumed under the broad construct. He rec- graduate days on we have been taught that there is ommended including broad constructs in the meta- only one proper thing to do when we see a correla- analysis, but distinguishing narrower constructs in tion coefficient: We must square it" (Rosenthal, 1990, the data and presentation of results. p. 775). This approach ignores, among other things, Third, moderately strong—and even weak—cor- the ceiling effect imposed by measurement error. Ro- relations may nevertheless indicate powerful causal senthal and Rubin's (1982) 'ßinomial Effect Size Dis- mechanisms. The current meta-analyses are based on De Wolff and van IJzendoorn 585 correlational studies—like the Baltimore study—and doorn, 1991). Paradoxically, the strength of the Balti- it is therefore impossible to derive causal conclusions more results may have inspired many researchers to from its outcome (Cook et al., 1992; Miller & Pollock, document the association between sensitivity and at- 1994; Stroebe & Diehl, 1991). Although most studies tachment, and at the same time its exploratory design on attachment and sensitivity are predictive in the and the size of its results prevented them from sense that earlier assessments of sensitivity were cor- strictly replicating the original effect size. Without related with subsequent attachment assessments, a the Baltimore study, the solid scientific fact of a mod- requirement of causality is the absence of a third fac- erately strong causal association between sensitivity tor explaining the association between sensitivity and and attachment would not have been established. attachment. In their meta-analysis of attachment In- The current meta-analyses qualify the original Bal- tervention studies, van IJzendoorn et al. (1995) timore results in yet another way. Sensitivity cannot showed that interventions are effective in enhancing be considered to be the exclusive and most important maternal sensitivity, and that, in particular, short- factor in the development of attachment. Several do- term interventions focusing on maternal sensitivity mains of maternal interactive behavior showed effect can enhance infants' attachment security. The com- sizes that were similar to those for the domain of Sen- bined effect size for the effectiveness of the attach- sitivity. For example, Mutuality and Synchrony were ment interventions was d = .48, which is a medium quite strongly associated with attachment security, effect (Cohen, 1988). In fact, the short-term interven- rs(166, 256) = .32 and .26, respectively, äs were Stim- tions may be considered to be a minimal manipula- ulation, Positive Attitüde, and Emotional Support. tion of the predictor that still accounts for significant Contrary to our expectation, the combined effect size variance in the criterion. Prentice and Miller (1992) of the more optimal cluster of studies on aspects of argue that this is a plausible reason for interpreting parenting more closely resembling sensitivity, effect sizes äs important even when they are small. r(l,926) = .19, did not differ from the outcome of the Combined with the outcome of the current meta- less optimal cluster, r(l,231) = .19. That is, aspects analyses, this attests to the important, but not exclu- of parenting only indirectly related to the sensitivity sive, causal role of sensitivity in the development of concept appear to play a similar role in the develop- infant attachment (Ainsworth et al., 1978; Bowlby, ment of attachment. The modest correlations be- 1969). tween the various aspects of parenting and sensitiv- ity leave room for unique and additional influences on attachment. In six studies, sensitivity was corre- GENERAL DISCUSSION lated with six other aspects of parenting: mean r(358) Although our results appear to support the "ortho- = .34 (N = 360). The original concept of sensitivity dox" position that maternal sensitivity is an impor- may not capture the only mechanism through which tant condition of attachment security, the outcome of the development of attachment is shaped (van IJzen- the Baltimore study itself cannot be considered to be doorn, 1995), and studies combining the promising replicated (Goldsmith & Alansky, 1987; Lamb et al., measures may provide more insight into the addi- 1985). In the Baltimore study, an effect size of r(21) tional explanatory value of these alternative ap- = .78 was found for the association between sensitiv- proaches over and above sensitivity. Sensitivity has ity and attachment, which is rather different from the lost its privileged position äs the only important combined effect size of the replication studies, causal factor. A multidimensional approach of par- r(l,097) = .24. Correcting the Baltimore correlation enting antecedents should replace the search for the for predictor and criterion unreliability (.89 and .95, unique contribution of sensitivity. It should be noted respectively) yields a corrected r(21) = .85. Logically, that in the current set of attachment studies the con- a correlation of this impressive magnitude seems to cept of parenting is virtually limited to the general indicate close similarity of the constructs and/or the domain of parental warmth and acceptance, rather assessments of attachment and sensitivity rather than than parental management and control. It is therefore an association between independent variables. Fur- unclear whether these latter aspects of parenting thermore, in a small sample, confidence boundaries would also contribute to the development of attach- around estimated correlations are broad, and outly- ment, in particular, after the first year of life. ing observations may be rather influential. Even rep- Contrary to our expectations, the duration of home lication studies with similar longitudinal and inten- observations was not related to the magnitude of the sive designs have failed to replicate the exceptionally association between sensitivity and attachment. It strong and striking results of the Baltimore study did not appear to matter whether studies were con- (e.g., Grossmann et al., 1985; Hubbard & van IJzen- ducted in the laboratory or in the home, except in the 586 Child Development case of maternal Stimulation. The use of the Standard velopment of emotional (in-)security from a family- Strange Situation procedure did not lead to different wide perspective (Cummings & Davies, 1996). The effect sizes compared to the use of alternative attach- transactions between social context or clinical condi- ment measures. The intensive naturalistic design of tions, on the one hand, and attachment on the other the Baltimore study does not seem to be essential for need more careful study to determine the role of sen- the strength of the association between parenting and sitivity, and other aspects of parenting and family life attachment found in the replication studies. It should in the development of attachment security more pre- be kept in mind, however, that only few replications cisely (Belsky & Cassidy, 1994; Egeland & Erickson, indeed used an exactly similar design. Furthermore, 1993; Sameroff & Chandler, 1975; Thompson, in in samples with younger infants, somewhat weaker press). In attachment theory, a move to the level of associations between sensitivity and attachment context may be necessary, so that the interaction be- tended to be found. Contrary to our hypothesis based tween maternal sensitivity and the accumulation of on the Baltimore study, we suggest post hoc that stresses and risk factors in lower-class or clinical these results are theoretically congruent with Bowl- groups can be taken into account. by's (1973) view of the development of attachment Even in "normal," nonclinical groups, sensitivity äs contextually labile and flexible in the early years. plays an important but not exclusive role in the emer- Because the security of attachment is a characteristic gence of attachment security. In cognitive develop- of the dyad more than of the infant in the early years, ment, genetics may constitute a ceiling effect for the the development of attachment can easily change influence of environmental factors, for which less direction when family life circumstances, childrear- than 50% of the Variation in individual differences ing arrangements or maternal sensitivity change appears to be left (Plomin, 1994). Attachment secu- (Thompson, in press; Thompson & Lamb, 1983; rity, however, does not seem to be genetically deter- Thompson, Lamb, & Estes, 1982). Sensitivity may be mined in any comparable way. The first, small-scale an important condition of attachment security only twin studies on attachment reported only 30%-50% when it remains stable across time, which may occur congruence of attachment classifications (Minde, only in a stable social context (Lamb et al., 1985; Corter, Goldberg, & Jeffers, 1990; Szajnberg, Skrin- Sroufe, 1988). jaric, & Moore, 1989; Vandell, Owen, Wilson, & Hen- In this respect, it is important to note that the asso- derson, 1988). A secondary analysis of the available ciation between maternal behavior and infant attach- twin data (N = 56) did not support the idea of a ge- ment is significantly weaker in studies of lower-class netic basis for individual differences in attachment or clinical samples. The measures of maternal behav- security (Ricciuti, 1993). Only the variable represent- ior and attachment security have been developed and ing the split between the Strange Situation subclassi- validated in nonclinical, middle-class samples, and fications A1-B2 and B3-C2 showed significant genetic they may be less valid in lower-class and clinical sam- influence. This may be due to temperament (e.g., ples. We also want to suggest that the formation of emotional reactivity), which has been shown to be attachment relationships under complex lower-class associated with this split (Belsky & Rovine, 1987). Al- or clinical conditions may not be adequately ex- though more research in this area is needed, we sug- plained in a monocausal and linear way. In their gest that attachment security is especially liable to quasi-experimental study of family-based and com- nongenetic, environmental influences. Nevertheless, munal kibbutzim, Sagi and his colleagues (Sagi, van behavior genetics may inspire a move toward the IJzendoorn, Aviezer, Donnell, & Mayseless, 1994) contextual level in a specific sense. One of the most showed that maternal sensitivity may be overridden intriguing findings of behavior genetics is the crucial by an unfavorable childrearing arrangement in role of nonshared environment for child develop- which infants have to sleep away from home, and ment. Unrelated adoptive siblings appear to develop therefore often develop insecure attachments. In a quite differently, even though they are raised in the similar vein, it may be expected that the strains and same, shared family environment. For example, the Stresses of lower-class life or the problems presented correlation between IQ of adoptive siblings is ap- by clinical conditions may overburden potentially proaching zero when they grow older (Plomin, 1994). sensitive mothers. Davies and Cummings (1994), for In the case of parenting and attachment, the concept example, suggested that unresolved marital conflicts of nonshared environmental influences may, for ex- may have a profound negative effect on the chil- ample, lead to more emphasis on the family System dren's emotional security even if the bond with the and on life events. Although parents may interact mother is balanced and her interaction style toward equally sensitively with both siblings at the same age, the child is sensitive. They proposed to study the de- the older sibling also experiences the parents inter- De Wolff and van IJzendoorn 587 acting sensitively with the younger sibling—which (1978). Patterns of attachment: A psychological study of the is a unique and potentially powerful experience that Strange Situation. Hillsdale, NJ: Erlbaum. never has been studied thoroughly (Dünn & Plomin, Ainsworth, M. D. S., & Wittig, B. A. (1969). Attachment and 1990). Life events such äs parental loss of attachment exploratory behavior of one-year-olds in a Strange Situ- ation. In B. M. Foss (Ed.), Oetenninants ofinfant behaviour figures may affect siblings in different ways, de- (Vol. 4, pp. 11-136). London: Methuen. pending on their age. If parents suffered bereave- Aldenderfer, M. S., & Blashfield, R. K. (1984). Cluster analy- ment within 2 years after the birth of a sibling, this sis. Beverly Hills, CA: Sage. sibling—but not the other siblings—may develop Altman, S., Monk, C., Jones, P., & Sosa, L. (1993, March). dissociative tendencies that may be related to inse- Chüdren's working models of attachment at 42 months and cure-disorganized attachment (Hesse & van IJzen- maternal attitudes and behavior. Poster presented at the doorn, 1996; Liotti, 1992). After more than 25 years meetings of the Society for Research in Child Develop- of research on the important dimension of sensitivity, ment, New Orleans, LA. [7].* a move to the level of (nonshared) context should in- Amato, P. R., & Keith, B. (1991). Parental divorce and the spire the next wave of studies on the antecedents of well-being of children: A meta-analysis. Psychological attachment security. Bulletin, 110, 26-46. Anisfeld, E., Casper, V., Nozyce, M., & Cunningham, N. (1990). Does infant carrying promote attachment? An ACKNOWLEDGMENTS experimental study of the effects of the increased physi- cal contact on the development of attachment. Child De- Support for this study was provided by a PIONEER velopment, 61, 1617-1627. [l, 2].* award from the Netherlands' Organization for Scien- Antonucci, T. C., & Levitt, M. J. (1984). Early prediction of tific Research (NWO: PCS 59-256) and by a fellow- attachment security: A multivariate approach. Infant Be- ship from the Netherlands Institute for Advanced havior and Development, 7, 1-18. [3, 5].* Study in the Humanities and Social Sciences (NIAS) Bates, J. E., Maslin, C. A., & Frankel, K. A. (1985). Attach- to Marinus van IJzendoorn. The authors gratefully ment security, mother-child interaction, and tempera- ment äs predictors of behavior-problem ratings at the acknowledge the thoughtful comments of members age of three years. In I. Bretherton & E. Waters (Eds.), of the editorial board and reviewers on an earlier Growing points in attachment theory and research (pp. draft of this paper. 167-193). Monographs of the Society for Research in Child Development, 50(1-2, Serial No. 209). [2, 7, 8].* Belsky, J., & Cassidy, J. (1994), Attachment: Theory and evi- ADDRESSES AND AFFILIATIONS dence. In M. Rutter, D. Hay, & S. Baron-Cohen (Eds.), Corresponding authors: Mariarme S. De Wolff and Developmental principles and dinical issues in psychology Marinus H. van IJzendoorn, Center for Child and psychiatry (pp. 373-402). Oxford: Blackwell. Belsky, ]., & Rovine, M. (1987). Temperament and attach- and Family Studies, Leiden University, P.O. Box ment security in the Strange Situation: An empirical rap- 9555, NL-2300RB Leiden, The Netherlands; e-mail: prochement. Child Development, 58, 787-795. [email protected]. Belsky, J., Rovme, M., & Taylor, D. G. (1984). The Pennsyl- vania Infant and Family Development Project, III: The

1 origins of individual differences in infant-mother at- REFERENCES tachment: Maternal and infant contributions. Child De- Abelson, R. P. (1995). Statistics äs principled argument. Hills- velopment, 55, 718-728. [2, 5, 7, 9].* dale, NJ: Erlbaum. Benn, R. K. (1986). Factors promoting secure attachment re- Ainsworth, M. D. S., Bell, S. M., & Stayton, D. J. (1974). In- lationships between employed mothers and their sons. fant-mother attachment and social development: "So- Child Development, 57, 1224-1231. [l, 8].* cialization" äs a product of reciprocal responsiveness to Bohlin, G., Hagekull, B., Germer, M., Andersson, K., & Signals. In M. P. M. Richards (Ed.), The Integration of a Lindberg, L. (1989). Avoidant and resistant behaviors äs child into a social world (pp. 99-135). Cambridge: Cam- predicted by maternal Interactive behavior and infant bridge University Press. temperament. Infant Behavior and Development, 12, 105- Ainsworth, M. D. S., Blehar, M. C., Waters, E., & Wall, S. 117. [l, 2, 3, 4].* Booth, C. L., Rose-Krasnor, L., & Rubin, K. H. (1991). Re- lating preschoolers' social competence and their moth- 1. References marked with an asterisk indicate studies in- ers' parenting behaviors to early attachment security cluded in the meta-analyses. We also indicated after these refer- ences the numbers of the meta-analytic domains for which the and high-risk Status. Journal of Social and Personal Rela- particular study has provided data. The numbers refer to the tionships, 8, 363-382. [7].* following meta-analytic domains: (1) Sensitivity, (2) Respon- Bornstein, M. H. (Ed.). (1989). Maternal responsiveness: siveness, (3) Physical Contact, (4) Cooperation, (5) Synchrony, Characteristics and consequences. New Directions for (6) Mutuality, (7) Support, (8) Attitüde, and (9) Stimulation. Child Development. San Francisco: Jossey-Bass. 588 Child Development

Bowlby, J (1969) Attachment and löst, Vol l Attachment and chüd adjustment An emotional secunty hypothe- (rev ed ) Harmondsworth Penguin sit, Psyüiological Bulletin Π 6, 387-411 Bon Ibv J (1973) Attachment and /oss Vol 2 Separation Del Carmen, R, Pedersen, F , Huffman, L C , & Bryan, Υ Ainieti/and an^er Harmondsworth Penguin E (1993) Dyadic distress management predicts subse- Bretherton, I (1985) Attachment theory Retrospect and quent secunty of attachment Infant Behavwr and Devel- prospect In I Bretherton & E Waters (Eds ), Growing opment 16 131-147 [2,8,9]* points in attachment theorv and research (pp 3-38) Dünn, J , & Plomin, R (1990) Separate Iwes Why siblmgs are Vlono^iapli^ of the Soc/t/t/ for Re^earüi in Cluld Devclop so different New York Basic intnt 50(1-2 Senal No 209) Easterbrooks, M A , & Goldberg, W A (1985) Effects of Bretherton, I Bmngen, Z , Ridgewav, D , Maslm, C , & early maternal employment on toddlers, mothers, and Sherman, M (1989) Attachment The parental perspec fathers Developmental Psychology 21, 774-783 tive Infant Mental Health Journal IQ 203-221 [1]* Egeland, B, & Enckson, M F (1993) Attachment theory Burchmal, M R , Brvant, D M, Lee, M W, & Ramev, C and fmdings Implications for prevention and Interven- T (1992) Earlv dav care, mrant-mother attachment, and tion In S Kramer & H Parens (Eds ), Prevention m men- maternal responsiveness in the mfant's first year Earli/ tal health Now tomoirow ever7 (pp 21-50) Northvale, Childhood Re^eanh Quarterh/ 7 383-396 [7, 8] * N] Jason Aronson Butcher, P R , Kah erboer, A Γ , Minderaa, R B, Van Door- Egeland, B , & Farber, E A (1984) Infant-mother attach- maal, E F , & Ten Wolde, Υ (1993) Rigidity, sensitiv ity, ment Factors related to its development and changes and qualitv of attachment The role of maternal ngiditv over time Child Development, 55, 753-771 [1] * in the early socio-emotional development of premature Enckson, M F, Sroufe, L A, & Egeland, B (1985) The infants Acta P^i/chtatnca Scandinainca, §S(Suppl 374), 4- relationship between quality of attachment and be- 38 [l, 8]* havior problems in preschool in a mgh-nsk sample In Caldvvell, B M & Bradlev R H (1984) Administration I Bretherton & E Waters (Eds), Growing points m at- maniial Home Obsei cation for Measiirement of the Environ tachment theory and research (pp 147-166) Mono nullt Little Rock Department of Psychology, Umversity graphb of the Society for Research m Child Development, ot Arkansas 50(1-2, Serial No 209) Camfield, E , Brovvnell C , Tavlor, P , Dav, N , Brown E Everitt, B (1974) Cluster analysis London Heineman Edu- & Kratzer L (1991 April) Mediator of rdationt, be- cational Books ticven matemal belnnnor at 4 montli^ and infant attachment Fagot,B I, & Kav anagh, K (1990) The prediction of antiso- at 12 month^ Poster presented at the meetmgs ot cial behavior from avoidant attachment classificahons the Society tor Research in Child Development, Seattle Child Development 61, 864-973 [7, 8]* [2 71* Fracasso, M P, Busch-Rossnagel, N A, & Fisher, C B Capps, L , Sigman, M , & Mundv P (1994) Attachment se- (1994) The relationship of maternal behavior and accul- cuntv in children with autism Dicelopment and Psi/t/fo turation to the quality of attachment in Hispamc infants patholow 6 249-261 [1]* hvmg in New York city Hispamc Journal of Behavioral Sci- Cohen, J (1962) The statistical povver of abnormal social ences 16 143-154 [1]* psvchologv research A revievv Jouinal of Abnormal and Frankel A , & Bates, J E (1990) Mother-toddler problem Social P^i/üiologif 65 145-153 solving Antecedents in attachment, home behavior, and Cohen, J (1988) Statistical powei analysis for the behavioral temperament Child Development, 61, 810-819 [7] * suences (rev ed ) New York Academic Press Frodi, A , Grolnick, W , & Bndges, L (1985) Maternal cor- Cook, T D, Cooper, H , Cordrav, D F, Hartman, H, relates of stabihty and change m infant-mother attach- Hedges, L V , Louis, T A & Mosteller, F (1992) Meta ment Infant Mental Health Journal, 6, 60-67 [1] * «Hfl/i/s/s for evplanation A casebook New York Russell Gage, N L (1996) Confrontmg counsels of despair for the Sage behavioral sciences Editcational Researcher, 25, 5-15 Cooper, H & Hedges, L V (Eds ) (1994) The handbook for Garfield, E (1979) Citationmdexmg Its theory and apphcatwn reseanh ^i/ntliesis New York Russell Sage m bcience technohgy and humanities New York Wiley Cox, M J , Owen, M T Henderson, V K , & Margand, N Gewirtz, J L, & Boyd, E F (1977) Does maternal re- A (1992) Prediction of infant-father and infant-mother sponding imply reduced infant crying7 A crinque of the attachment Dcvelopmental Psi/t//o/o?i/ 28 474-483 [3, 7, 1972 Bell and Ainsworth report Child Development, 48, 8]* 1200-1207 Crockenberg, S B (1981) Infant irritabihtv, mother respon- Gifi, A (1990) Νοη-lmear multivanate analysis New York siveness, and social support influences on the secuntv Wiley of intant-mother attachment Child Development 52 857- Goldberg, S, Perotta, M , Mmde, K, & Corter, C (1986) 865 [2]* Maternal behavior and attachment in low-birth-weight Cummings, E M , & Dav les P T (1996) Emotional secu- twins and smgletons Child Development, 57, 34-46 [l, 2, ntv äs a regulatory process m normal development and 3, 4, 7, 8, 9] * the development of psvchopathology Dicelopment and Goldsmith, H H, & Alansky, J A (1987) Maternal and Psi/diopathologi/ 7 123-139 infant predictors of attachment A meta-analytic review Davies, P T & Cummmgs, E M (1994) Mantal conflict Journal of Consulting and Clinical Psychology, 55, 805-816 De Wolff and van IJzendoorn 589

Goodman, G., Andrews, T., Jones, S., Weissman, }., & baby], Amersfoort, The Netherlands: Academische L'it- Weissman, A. (1993). Maternal play behaviorand children'* geverij. [l, 4].* rcprescntations äs Mediators of developmcntal ontconies Kerns, K. A., & Barth, J. M. (1995). Attachment and play: among Ingh-risk dyads. Manuscript in preparation. [1].* Convergence across components of parent-child rela- Goossens, F. A. (1987). Kwaliteit van de hechtingsrelatie tionships and their relations to peer competence. Journal van jonge kinderen aan hun moeder, vader en creche- ofSocml and Personal Relationships, 12, 243-260 [3].* leidster. [Quality of infants' attachment to father, Kiser, L. ]., Bates, J. E., Maslin, C. A., & Bayles, K. (1986). mother, and caregiver in daycare]. Nederlands Tijdsclmft Mother-infant play at six months äs a predictor of at- voor de Psychologie, 42, 308-320. [1].* tachment security of thirteen months. Journal of the Grossmann, K., Grossmann, K. E., Spangler, G., Suess, G., & American Academy of Child Psychiatry, 25, 68-75. [6].* Unzner, L. (1985). Maternal sensitivity and newborns' Krentz, M. S. (1982). Qualitative differcnces between mother- orientation responses äs related to quality of attachment child and caregivcr-cluld attacliment of infant in fainily day in northern Germany. In I. Bretherton & E. Waters care. Unpublished doctoral dissertation, California (Eds.), Growing points in attachment theory and re- School of Professional Psychology. [2, 7].* search (pp. 233-268). Monograplis of the Society for Re- Kveton, E. M. (1989). The quality of infant-parent attach- search in Child Development, 50(1-2, Serial No. 209). [1].* ment at 17 months äs related to the quality of parent- Gunnar, M. R., Brodersen, L., Nachmias, M., Buss, K., & infant interaction and prenatal parenting attitudes (Doc- Rigatuso, J. (1996). Stress reactivity and attachment se- toral dissertation, Virginia Consortium for Professional curity. Developmental Psychobiology, 29, 191-204. [1].* Psychology, 1989). Dissertation Abstracts International, 50, Hedges, L. V. (1986). Issues ip meta-analysis. Revieiv of Re- 5884B. [1]; search in Education, 13, 353-399. Lamb, M. E., Hopps, K., & Elster, A. B. (1987). Strange Situ- Hedges, L. V., & Olkin, I. (1985). Statistical methods for meta- ation behavior of infants with adolescent mothers. Infant analysis, Orlando, FL: Academic Press. Behavior and Development. W, 39-48. [2, 5, 8, 9].* Hesse, E., & van IJzendoorn, M. H. (1996). Parental loss of Lamb, M. E., Thompson, R. A., Gardner, W., & Charnov, dosefamüy membcrs within two years preceding or folhm'ing E. L. (1985). Infant-mother attacliment: The origins and de- participants' birth is related to absorption. Unpublished vclopmental significance of individiial differences in Strange manuscript, Center for Child and Family Studies, Situation behavior. Hillsdale, NJ: Erlbaum. Leiden University. Lambermon, M. W. E. (1991). Video offoldei ? Körte- en lange- Hoeksma, J. B., & Koomen, M. M. Y. (1991). Development of termijn effecten ran voorlichting over vroegknuleiliike op- early mother-child internctions and attachment. Unpub- vocding. [Video or booklet? Short-term and long-term lished doctoral dissertation, Free University of Amster- effects of Information about early childhood educa- dam, The Netherlands, [l, 6].* tion]. Unpublished doctoral dissertation, University of Hubbard, F. Ο. Α., & van IJzendoorn, M. H. (1991). Mater- Leiden, The Netherlands, [l, 7].* nal unresponsiveness and infant crying across the first Lederberg, A. R., & Mobley, C. E. (1990). The effect of hear- nine months: A naturalistic longitudinal study. Infant ing impairment on the quality of attachment and Behavior and Development, 14, 299-312. mother-toddler interaction. Child Development, 61,1596- Hunter, J. E., & Schmidt, F. L. (1990). Methods of meta- 1604. [4, 7, 8, 9].* analysis: Correctmg error and bias in researdi findtngs. Bev- Lewis, M., & Feiring, C. (1989). Infant, mother, and mother- erly Hills, CA: Sage. infant interaction behavior and subsequent attachment. Isabella, R. A. (1993). Origins of attachment: Maternal inter- Child Development, 60, 831-837. [2, 3].* active behavior across the first year. Child Development, Lieberman, A. F., Weston, D. R., & Pawl, J. H. (1991). Pre- 64, 605-621. [l, 8, 9].* ventive Intervention and outcome with anxiously Isabella, R. A., & Belsky, J. (1991). Interactional synchrony attached dyads. Child Development, 62, 199-209. [7, 9].* and the origins of infant-mother attachment: A replica- Liotti, G. (1992). Disorganized/disoriented attachment in tion study. Child Development, 62, 373-384. [5].* the etiology of the dissociative disorders. Dissociation, 5, Isabella, R. A., Belsky, J., & Von Eye, A. (1989). Origins of 196-204. infant-mother attachment: an examination of interac- Londerville, S., & Main, M. (1981). Securit) of attachment, tional synchrony during the infant's first year. Develop- compliance, and maternal training methods in the sec- mental Psychology, 25, 12-21. [5]* ond year of life. Developmental Psychology, 17, 289-299. Izard, C. E., Heynes, O. M., Chisholm, G., & Baak, K. (1991). [4].*' Emotional determinants of infant-mother attachment. Lyons-Ruth, K., Connell, D. B., Zoll, D., & Stahl, J. (1987). Child Development, 62, 906-917. [8].* Infants at social risk: Relations among infant mal- Johnson, B. T., Müllen, B., & Salas, E. (1995). Comparison treatment, maternal behavior, and infant attachment be- of three meta-analytic approaches. Journal of Applied Psy- havior. Developmental Psychology, 23, 223-232. [l, 3, 4, chology, 80, 94-106. 8].* Juffer, F. (1993). Verbanden door adoptic: Een experimentee! Main, M. (1990). Cross-cultural studies of attachment orga- onderzoek bij gezinnen inet een adoptiebaby. [Attached nization: Recent studies, changing methodologies, and through adoption: An experimental study on attach- the concept of conditional strategies. Human Develop- ment and competence in families with an adopted ment, 33, 48-61. 590 Child Development

Viain, M , Tomasim, L, & Tolan, W (1979) Differences 1993 Genetic research and Identification of environmen- among mothers of infants judged to differ in secunty tal mfluences Journal of Child Psychology and Psychiatry, Dccdofimmlal P^i/diologi/ 15 472-473 [l, 3, 8] * 35 817-834 Malatesta, C Z, Culver, C, Tesman, C & Shepard, B Prentice, D A , & Miller, D A (1992) When small effects (1989) The development of emotion expression dunng are impressive Psydiological Bulletin, 112, 160-164 the first two years of hfe Monographs for tlic Soueti/ for Ricciuti, A E (1993) Child-mother attachment A twm Risentt.il in Child Development 54(1-2, Senal No 219) studv Dibbeitation Abbtracts International, 54, 3364 [8]* Roggman, L A , Langlois, J H , & Hubbs-Tait, L (1987) Vlangelsdort, S , Gunnar, M , Kestenbaum, R , Lang, S , & Mothers, infants, and toys Social play correlates of at- Andreas D (1990) Infant proneness to distress temper- tachment Infant Behamor and Development, W, 233-237 ament, maternal personahtv, and mother-mfant attach- [5, 9] * ment Associations and goodness of fit Child Develop- Rosenboom, L G (1994) Gemengde gezmnen, gemengde ment 61 820-831 [7]* gevoclens? Hechtmg en competentie van adoptiebaby'b in Vlatas L , Arend, R A , & Sroufe, L A (1978) The relation- gezmnen met biologisch eigen kinderen [Mixed famihes, ship between quahty of attachment and later compe- mixed feelmgs7 Attachment and competence of adopted tence Child Development 49 547-556 [7] * babies m families with biological children] Unpub- Vleij, ] (1991) Sonde ondei Henning geheditheid^kwaliteit en lished doctoral dissertation, Umversity of Utrecht, The L'ioegkindeilijke conipetentie ontwikkehng [Social support, Netherlands [l, 7] * quahtv ot attachment, and earlv dev elopment of compe- Rosenthal, R (1979) The "file drawer problem" and toler- tence in infants] Unpublished doctoral dissertation, ance for null results Psydiological Bulletin, 86, 638-641 Umversitv of Nijmegen, The Netherlands [1] * Rosenthal, R (1990) How are we doing in soft psychology7 Miller, N , & Pollock, V E (1994) Meta-analytic svnthesis American Ρί,ι/chologtst, 45, 775-777 for theorv de\ elopment In H Cooper & L V Hedges Rosenthal, R (1991) Meta-analytic procedures for social re- (Eds ), The handbook of research si/j/f/iesis (pp 457-483) bearch Newbury Park, CA Sage New York Russell Sage Rosenthal, R (1995) Writmg meta-analytic reviews Psy- Minde, K , Corter, C , Goldberg, S , & Jeffers, D (1990) Ma- diological Bulletin, 118, 183-192 ternal preference between premature twins up to age Rosenthal, R , & Rosnow, R L (1985) Contrast analysis Fo- four Journal of the Amenicm Atademi/ of Child and Adole^- citsed compansons in the analysis of vanance New York u'nt P^/duntn/ 29 367-374 Cambridge Umversity Press Mivake, K , Chen, S J & Campos, J J (1985) Infant tem- Rosenthal, R, & Rubin, D B (1982) Companng effect sizes perament, mother's mode of interaction, and attachment of mdependent studies Psychological Bulletin, 92, 500- m Japan An intenm report in I Bretherton & E Waters 504 (Eds ), Growing pomts m attachment theory and re- Sagi, A , van IJzendoorn, M H , Aviezer, O , Donnell, F , & search (pp 276-297) Monograph^ of the Soaeti/ for Re Mavseless, O (1994) Sleeping out of the home in a kib- sem/i m Child Development 50(1-2 Senal No 209) [2, butz communal arrangement It makes a difference for 4, 7 9] * mfant mother attachment Child Development, 65, 992- Müllen B (1989) Adoatned BASIC meta anali/bii, Hillsdale, 1004 NJ Erlbaum Sameroff, A J , & Chandler, M J (1975) Reproduchve nsk Nakagawa, M , Lamb, M E , & Miyake, K (1992) Anteced- and the contmuum of caretaking casualty In F D. Horo- ents and correlates of the Strange Situation behavior of witz (Ed ), Review ofchdd development research (Vol 4, pp Japanese infants Journal ofCro^ Cnltural Pbi/dwlogy 23, 187-244) Chicago Umversity of Chicago Piess 300-310 [l 4, 7, 8] * Schneider Rosen, K , & Rothbaum, F (1993) Quahty of pa- NICHD Earlv Child Care Network (1996, April) Infant rental caregivmg and secunty of attachment Develop- chdd cate and attadiment ^ecunti/ Resiilts of the NICHD mental Psychology, 29, 358-367 [7, 8] * sfiirfv ofcarhf diild care Svmposium presented at the In- Schwarzer, R (1989) Meta-analysis programs Unpubhshed ternational Conference on Infant Studies, Pro\ idence, manual, Umversity of Berlin, Department of Psy- RI [7]* chology Pederson, D R , & Moran, G (1996) Expressions of the at- Sechrest, L , & Yeaton, W H (1982) Magnitudes of expen- tachment relahonship outside of the Strange Situation mental effects in social science research Evaluation Re- Child Dei elopment, 67 915-967 [1]* view, 6, 579-600 Pederson, D R , Moran, G , Sitko, C , Campbell, K , Ghes- Seifer, R , Schiller, M , Sameroff, A J , Resnick, S , & Rior- quire, K, & Acton, H (1990) Maternal sensitivity and dan, K (1996) Attachment, maternal sensitivity, and rn- the secunty of mfant-mother attachment A Q-sort fant temperament dunng the first year of hfe Develop- studv Ould Development, 67 1974-1983 [1] * mental Psychology, 32, 12-25 [1] * Persson-Blennow, I, Binett, B , & McNeil, T F (1988) Off- Shaw, D S , & Vondra, J I (1995) Infant secunty and ma- spring of women with nonorganic psychosis Anteced- ternal predictors of early behavior problems A longitu- ents of anxious attachment to the mother at one year of dmal study of low-mcome families Journal of Abnormal age Atta Psychiatrien Skandinavien 78, 66-71 [2, 3, 8] * Child Psychology, 23, 335-357 [2] * Plomm, R (1994) The Emanuel Miller Memorial Lecture Smith, P B , & Pederson, D R (1988) Maternal sensitivity De Wolff and van IJzendoorn 591

and patterns of infant-mother attachment Ould Develop- Dutch families A contnbution to the cross-cultural ment, 59, 1097-1101 [4, 6] * studv of attachment Intel national Jouinal of Behavioial SPSS, Ine (1990) SPSS categona, Chicago SPSS, Ine Development, 13 333-344 [1] * Sroufe, L A (1988) The role of mfant-caregi\ er attachment van IJzendoorn, M H (1995) Adult attachment representa- in development In] Belsky&T Nezworski (Eds ), Chn- tions, parental responsiveness, and infant attachment A ical impliLatwm of attachment (pp 18-40) Hillsdale, NJ meta-analysis on the predictiv e vahdity of the Adult At- Erlbaum tachment Interview Pb\/Jiological Bulletin, 117 387-403 Stifter, C A , Couleham, C M , & Fish, M (1993) Lmkmg van IJzendoorn, M H , Juffer, F, & Duvvesteyn, M G C employment to attachment The mediating effects of (1995) Breaking the intergenerational cycle of insecure maternal Separation anxiety and interactive beha\ior attachment A re\ lew of the effects of attachment-based Clnld Development, 64, 1451-1560 [l, 4] * intervenhonb on maternal sensitivity and infant secu- Stroebe, W , & Diehl, M (1991) You can't beat good expen- ntv Journal of Clnld Ρ^ι/üiology and P^\/clnatn/ 36 225- ments with correlational evidence Müllen, Johnson, 248 and Salas' meta-analytic mismterpretations Ba^ic and \ an IJzendoorn, M H, Kranenburg, M J, Zwart- Applied Socml Psijchology, 12, 25-32 Woudstra, H A , Van Busschbach, A M , & Lamber- Szajnberg, N M , Skrmjanc, J, & Moore, A (1989) Atfect mon, M W E (1991) Parental attachment and chil- attunement, attachment, temperament, and zvgositv A dren's socio-emotional development Some frndings on twm study Journal of tlie American Acadenn/ of Clnld and the vahdity of the Adult Attachment Interview in the Adolescent Psychiatry, 28, 249-253 Netherlands Intet national Journal of Behavioral Develop- Teti, D M , Nakagawa, M , Das, R , & Wirth, O (1991) Se- ment, 14, 375-394 [1] * curity of attachment betWeen preschoolers and their \ an IJzendoorn, M H , & Kroonenberg, P M (1988) Cross- mothers Relations among social interaction, parenting cultural patterns of attachment A meta-analysis of the stress, and mothers' sorts of attachment Q-set Develop Strange Situation Clnld Development, 59, 147-156 mental Psychology, 27, 440-447 [7] * v an IJzendoorn, M H , Vereijken, C M J L, & Riksen- Thompson, R A (in press) Early socio-personahtv dev el- Walraven, J M A (in press) 1s the Attachment Q-sort opment In W Oamon (Ed), Handbook of ünld p^i/chologi/ a \alid measure of attachment secuntv in young chil- Vol 3 Social, emotional, and pei^onahti/devehpment New dren? In B E Vaughn & E Waters (Eds ), Patteim ofbe- York Wiley cuie bnt,c behavwi Q-so/f pet ψι'ι f;ws on attachment and Thompson, R A , & Lamb, M E (1983) Secuntv ot attach- Laiegnnng Hillsdale, NJ Erlbaum ment and stianger sociabihty in infancy Devclopmental Vaughn, B E , & Waters, E (1990) Attachment behavior Psychology, 19, 184-191 at home and in the laboratory Q-sort observations and Thompson, R A , Lamb, M E , & Estes, D (1982) Stabihtv Strange Situation classifications of one-year-olds Clnld of infant-mother attachment and its relation to changing Development, 61, 1965-1973 life circumstances in an unselected rmddle-class sample Verkes,R J, Van der Kloot, W A ,& Van der Meij, J (1989) Child Development, 53, 144-148 The perceived structure of 176 pam descnptive words Ungerer, J A , Dolby, R , Waters, B , Barnett, B , Kelk, N , Pam, 38 219-229 Lewin, V , & Blaszczynski, A (1990) The early develop- Waters, E , & Deane, K E (1985) Defming and assessing ment of empathy Self-regulation and individual differ- individual differences in attachment relationships Q- ences in the first year Motivation and Emotion, 14, 93- methodologv and the organization of behavior m in- 106 [8]* fancy and early childhood In I Bretherton & E Waters Vandell, D L , Owen, M T, Wilson, K S , & Henderson, (Eds ), Growmg points m attachment theory and re- V K (1988) Social development in infant twms Peer search (pp 39-40) Monogiaphs of the Society for Re^eaich and mother-child relationships Clnld Development, 59, m Chüd Development 50(1-2, Senal No 209) 168-177 Wille, D E (1991) Relation of preterm birth with quality van den Boom, D C (1988) Neonatal irrüabdity and the de- of infant-mother attachment at one year Infant Behavior velopment of attachment Observation and Intervention Un- and Development, 14, 227-240 [6, 7, 8] * pubhshed doctoral dissertation, Umversity of Leiden, Zaslow, M J , Rabmovich, B A , Suwalsky, J T D , & Klein, The Netherlands [1] * R P (1988) The role of social context in the prediction Van der Kloot, W A , & Van Herk, H (1991) Mulhdimen- of secure and insecure/avoidant infant-mother attach- sional scalmg of sortmg data A companson of three ment Journal of Applied Deivlopmental Psychologe, 9, 287- procedures Multwauate Behaviotal Research, 26,563-581 299 [3, 7, 8] * van IJzendoorn, M H (1990) Attachment in Surinam-