infant day-care, attachment, and the file drawer problem & lacasse, 1987, in clarke-stewart,...

16
Infant Day-Care, Attachment, and the "File Drawer Problem" Lori A. Roggman Utah State University Judith H. Langlois University of Texas, Austin Laura Hubbs-Tait Oklahom.a State University Loretta A. Rieser>Danner Pennsylvania State University, Ogontz Campus N, LORI A.; LANGLOIS, JUDITH H.; HUBBS-TAIT, LAURA; and RIESEB-DANNEH, LOHETTA A. Infant Day-Care, Attachment, and the "File Drawer Problem." CHILD DEVELOPMENT, 1994, 65, 1429-1443. The research literature on infant day-care and attachment may be biased by the unavailability of "Hie drawer" studies, (unpublished data showiriygjK)_statisticalliLsigiiificant^ef- fefitsj Replication studies, whether showing an effect or not, are essential to clarify the relation between day-care and attachment. {Tliis^ study of 105 12-month-olds is an attenigt_to_£egHcate Jour,siTnilarstudies summarized and comtinedi5y^BeIskyto show~3mtTriSnts in Jaitcai.e_8fffi..at lIslTfcrliBiiSgui^'WtacinilBnCJIn'the'pies'SrFsKiSy^^ to emerge consistently, altEbugTi the^Fe was a trend for more negative attachment outcomes to be associated with little or part-time day-care rather than with full-time day-care. In general, the results suggest that the specific measures, definitions of full- and part-time, and statistical techniques used in studies examining the relation between day-care and attachment are likely to affect the outcome of such studies. For any given research area, one cannot tell how many studies have been conducted but never reported. The extreme view of the "file drawer problem" is that journals are filled with the 5% of the studies that show Type I errors, while the file drawers are filled with the 95% ofthe studies that show nonsignificant results. [Rosenthal, 1979, p. 638] With this statement, Robert Rosenthal The status of research on the relation (1979) articulates a problem that is relevant between infant day-care and mother-infant to all areas of psychology and child develop- attachment has generated a great deal of con- ment but is of particular concern when in- troversy. The arguments and proponents of vestigations of a topic that has social policy each side of the. infant day-care controversy implications generate conflicting evidence, are, by now, well known and will be de- The present article is concerned with the scribed here only briefiy (for a full treat- "file drawer problem" and its implications ment, see Belsky, 1988, 1989, 1990; Glarke- for the interpretation of research on infant Stewart, 1988, 1989; Hoffman, 1990; Lamb day-care and attachment. & Stemberg, 1990; Lamb, Stemberg, & Pro- This research was supported in part by a grant from the University of Texas Research Institute to Judith H. Langlois and in part by a NIMH postdoctoral fellowship (MH08556-01) to Laura Hubbs-Tait while at the University of'Texas. We thank Kathy Matula and Alice Andrews for valuable assistance in data collection and George Holden for comments on a draft of this article. We also thank the more than 20 researchers in the fields of day-care and/or attachment who responded to our requests for information about "file drawer" studies. Portions of this research were presented at the meeting of the Southwestern Society for Research in Human Development, New Orleans, 1988. Gorrespondence and requests for reprints should be ad- dressed to Lori A. Roggman, Department of Family and Human Development, Utah State Uni- versity, Logan, UT 84322-2905. [Child Development, 1994,65,1429-1443. © 1994 by the Society for Research in Child Development, Inc. All rights reserved. 0009-3920/94/6505-0018$01.00]

Upload: nguyentuyen

Post on 29-May-2018

213 views

Category:

Documents


0 download

TRANSCRIPT

Infant Day-Care, Attachment, and the "FileDrawer Problem"

Lori A. RoggmanUtah State University

Judith H. LangloisUniversity of Texas, Austin

Laura Hubbs-TaitOklahom.a State University

Loretta A. Rieser>DannerPennsylvania State University, Ogontz Campus

N, LORI A.; LANGLOIS, JUDITH H.; HUBBS-TAIT, LAURA; and RIESEB-DANNEH, LOHETTA A.Infant Day-Care, Attachment, and the "File Drawer Problem." CHILD DEVELOPMENT, 1994, 65,1429-1443. The research literature on infant day-care and attachment may be biased by theunavailability of "Hie drawer" studies, (unpublished data showiriygjK)_statisticalliLsigiiificant ef-fefitsj Replication studies, whether showing an effect or not, are essential to clarify the relationbetween day-care and attachment. {Tliis study of 105 12-month-olds is an attenigt_to_£egHcateJour,siTnilarstudies summarized and comtinedi5y^BeIskyto show~3mtTriSnts in Jaitcai.e_8fffi..atlIslTfcrliBiiSgui^'WtacinilBnCJIn'the'pies'SrFsKiSy^^ to emergeconsistently, altEbugTi the Fe was a trend for more negative attachment outcomes to be associatedwith little or part-time day-care rather than with full-time day-care. In general, the results suggestthat the specific measures, definitions of full- and part-time, and statistical techniques used instudies examining the relation between day-care and attachment are likely to affect the outcomeof such studies.

For any given research area, one cannot tell how many studies have been conductedbut never reported. The extreme view of the "file drawer problem" is that journalsare filled with the 5% of the studies that show Type I errors, while the file drawersare filled with the 95% ofthe studies that show nonsignificant results. [Rosenthal,1979, p. 638]

With this statement, Robert Rosenthal The status of research on the relation(1979) articulates a problem that is relevant between infant day-care and mother-infantto all areas of psychology and child develop- attachment has generated a great deal of con-ment but is of particular concern when in- troversy. The arguments and proponents ofvestigations of a topic that has social policy each side of the. infant day-care controversyimplications generate conflicting evidence, are, by now, well known and will be de-The present article is concerned with the scribed here only briefiy (for a full treat-"file drawer problem" and its implications ment, see Belsky, 1988, 1989, 1990; Glarke-for the interpretation of research on infant Stewart, 1988, 1989; Hoffman, 1990; Lambday-care and attachment. & Stemberg, 1990; Lamb, Stemberg, & Pro-

This research was supported in part by a grant from the University of Texas ResearchInstitute to Judith H. Langlois and in part by a NIMH postdoctoral fellowship (MH08556-01) toLaura Hubbs-Tait while at the University of'Texas. We thank Kathy Matula and Alice Andrewsfor valuable assistance in data collection and George Holden for comments on a draft of thisarticle. We also thank the more than 20 researchers in the fields of day-care and/or attachmentwho responded to our requests for information about "file drawer" studies. Portions of thisresearch were presented at the meeting of the Southwestern Society for Research in HumanDevelopment, New Orleans, 1988. Gorrespondence and requests for reprints should be ad-dressed to Lori A. Roggman, Department of Family and Human Development, Utah State Uni-versity, Logan, UT 84322-2905.

[Child Development, 1994,65,1429-1443. © 1994 by the Society for Research in Child Development, Inc.All rights reserved. 0009-3920/94/6505-0018$01.00]

1430 Child Development

dromidis, 1992; McCartney & Galanopoulos,1988; McCartney & Phillips, 1988; Richters& Zahn-Waxler, 1988; Thompson, 1988). Aspart of his 1988 review on day-care and at-tachment, Belsky summarized four studiesof the relation between infant day-care ex-perience and attachment classification inchildren from low-risk samples (Barglow,Vaughn, & Molitor, 1987; Belsky & Rovine,1988; Chase-Lansdale & Owen, 1987; Jacob-son & Wille, 1984). Because all four sampleswere low risk and all four studies used thestandardized Strange Situation (Ainsworth,Blehar, Waters, & Wall, 1978) to measure at-tachment, Belsky combined the data fromthese studies and showed that, across thefour samples, infant day-care experience wassignificantly related to insecure atta^^lynentsand higher avoidance ofthe mothe r^e con-cluded from these results that "some nonma-ternal care arrangement in the first year formore than 20 hours per week may be a riskfactor in the emergence of developmentaldifficulties" (Belsky, 1988, p. 235)n

Belsky's conclusion elicited severaltypes of criticism from a variety of research-ers. Some critics addressed the appropriate-ness of the Strange Situation for assessingattachment quality in a population for whomthe measure was not standardized or ques-tioned whether preexisting differences be-tween either parents or infants with varyinglevels of day-care experience may havecaused the results (e.g., Clarke-Stewart,1988, 1989; Lamb & Sternberg, 1990; Rich-ters & Zahn-Waxler, 1988; Thompson, 1988).Other critics emphasized that the differ-ences obtained in these four studies, al-though statistically significant, may not bevery meaningful because of their small mag-nitude (Clarke-Stewart, 1988, 1989; Thomp-son, 1988). Thompson (1988) investigatedthese small differences by comparing thepatterns of attachment found among day-care infants in the Belsky and Rovine (1988)and the Barglow et al. (1987) studies to thenormative patterns of attachment reportedby Ainsworth et al. (1978) and found no sig-nificant differences.

Our purpose here is not to enter into theexisting controversy but to illustrate anotherissue to be considered in evaluating the re-search literature on the impact of day-careon attachment security: the so-called filedrawer problem. As more inyestigators re-analyze data combined from several studiesto summarize general patterns of relationsbetween infant day-care experience and at-tachment classifications (e.g., Belsky, 1988;Clarke-Stewart, 1989; Lamb & Sternberg,

1990; Lamb et al., 1992; McCartney & Phil-lips, 1988; Thompson, 1988), it becomesmore important to consider the potential forbias inherent in the availability of studiesfor analysis. The lion's share ofthese studieshas come, understandably, from the pub-lished literature. These, after all, eire thestudies that are known and available. How-ever, such an "availability" bias removesfrom consideration those studies in which nosignificant differences in attachment wereobtained between day-care experiencegroups, studies that often remain unpub-lished and buried in file drawers because ofan understandable reluctance to publishstudies with null results.

The availability bias in summaries of at-tachment and day-care research is evidentfrom inconsistencies in which studies wereincluded in three summary analyses of day-care and attachment measured by theStrange Situation (Clarke-Stewart, 1989;Lamb & Sternberg, 1990; Lamb et al., 1992).Studies included in all three summary analy-ses were likely to be published studiesshowing significant effects of day-care on at-tachment, but studies omitted from one ormore summary analyses were likely to beunpublished studies with null results. Al-though a total of 20 studies were includedin one or more of the summary analyses,only eight were included in all three. Ofthese eight studies, all from the publishedliterature, five included both day-care andnon-day-care groups, each of which reportedat least one statistically significant relationbetween day-care and attachment (Barglowet al., 1987; Belsky & Rovine, 1988; Chase-Lansdale & Owen, 1987; Jacobson & Wille,1984; Thompson, Lamb, & Estes, 1982). Sixstudies were unpublished and apparentlyknown to some authors but not to others, be-cause they were not included in all threesummary analyses. Of those six studies, twoincluded only day-care infants, and of theremaining four, information is availableabout three: One has since been publishedand authors of two others provided a sum-mary of their results. In contrast to the pub-lished studies included in all three summaryanalyses, none of these three unpublishedstudies found a significant effect of day-careon attachment (Burehinal, Bryant, Lee, &Ramey, 1992, referenced as Burchinal &Bryant, 1986, in Clarke-Stewart, 1989; M. A.Easterbrooks, personal communication, Oc-tober 9, 1992, referenced as Easterbrooks& Harmon, 1985, in Clarke-Stewart, 1989;L. LaCasse, personal communication,September 10, 1992, referenced as

Lipsett & LaCasse, 1987, in Clarke-Stewart,1989).

The potential bias in published studiescomes from at least two sources: researchersand journals. Researchers are reluctant tosubmit findings of no significant differencesfor publication; instead, they "file" themaway (Creenwald, 1975; Kupfersmid, 1988).Journal reviewers and editors tend to pub-lish only studies that report "rejection ofthenull hypothesis" and tend not to publishreplications or failures to replicate (Kupfers-mid, 1988, p. 637; Rosnow & Rosenthal,1989). These sources of bias in the pub-lished literature, stemming from the atti-tudes of both researchers and journals re-garding null results, directly infiate theprobability of finding a significant relationbetween day-care and attachment in sum-mary reports and meta-analyses. Further-more, these sources of bias increase theprobability that such a relation may be onlya chance result, a Type I error (Creenwald,1975). If the studies available for review,summary, and meta-analysis include primar-ily those that are published, that is, primarilythose in which statistically significant effectsof day-care on attachment were found, theresulting problem is that "the impression ofsupport for the favored hypothesis will beartificially enhanced over what would beseen if the entire literature were integrated"and "misleading generalizations" will bepromulgated (Glass, McCaw, & Smith, 1981,p. (54).

Indeed, several studies that have failedto find significant differences in attachment,as assessed by the Strange Situation, be-tween day-care and non-day-care groupshave not been published (M. A. Easter-brooks, personal communication, October 9,1992; L. LaCasse, personal communication,September 10, 1992; Norberg, Mack, Beck,& Brazelton, 1989; O'Connor & Sigman,1988; Sutton & Fox, 1992). Additional un-published studies using the Q-sort method-ology (Waters & Deane, 1985) to assess at-tachment have also found no relationbetween day-care and attachment security(e.g., Caldera, 1991; Newman-Cedar, Plun-ket, Kessler, & Ward, 1988; Strayer, 1990).Sample sizes across these file drawer studiesranged from 41 to 102, and all of them in-cluded both day-care and non-day-caregroups. Four included continuous data onthe number of hours per week of day-care(Caldera, 1991; M. A. Easterbrooks, personalcommunication, October 9,1992; Norberg etal., 1989; Sutton & Fox, 1992), and three in-cluded data on the age of entry into day-care

Roggman et al. 1431

(Caldera, 1991; L. LaCasse, personal com-munication, September 10,1992; Norberg etal., 1989). Results were consistent acrossthese unpublished studies. They all reportno statistically significant relation betweensecurity of attachment and any ofthe follow-ing: day-care versus non-day-care groups,number of hours per week of day-care, orage of entry into day-care. In addition, atleast three dissertations and one master'sthesis testing the relation between securityof attachment and infant day-care have beencompleted in the last several years, none ofwhich found a significant effect (Everson,1982; Low, 1982; Moser, 1989; Phillips,1990). These studies also apparently remainin file drawers; none of them are indexed indata bases of published studies. Only oneunpublished (but very recent) study wasidentified that found a significant relationbetween day-care and attachment as mea-sured by the Q-sort in a sample of 23 infants(Wille, 1993). Although 13 unpublishedstudies are reviewed here, the total numberof file drawer studies is, of course, impossi-ble to estimate for the obvious and simplereason that they are filed away.

In general, the usefulness of file drawerstudies showing null results depends on thenature of the research question. At one ex-treme, for a relation between variables thatis not predicted by theory, has not been pre-viously reported, and has no social policyimplications, null results are not very usefulor even interesting. At the other extreme, fora relation of high magnitude that has beenwidely replicated, null results might onlysuggest something peculiar about the sam-ple or methods of a specific study. Null re-sults are of value when, as in the case ofattachment and day-care, they meet threecriteria: (a) the research question has theo-retical implications, (b) previous resultshave been equivocal, and (c) the policy im-plications are controversial.

The effect of day-care on attachment istheoretically interesting because attachmenttheory (Bowlby, 1969, 1973) has been inter-preted as suggesting that the repeated sepa-rations of mothers and infants for day-careare disruptive to the caregiying interactionsneeded for the forniation of secure attach-ment (Barglow et al., 1987; Jaeger & Wein-raub, 1990; Vaughn, Cove, & Egeland,1980). Although the summary studies citedabove offer at least some support for this in-terpretation, other recently published re-sults have been equivocal. In fact, a searchof the day-care and attachment literaturepublished since 1988 located four primary-

1432 Child Development

data studies of infant day-care and attach-ment measured with the Strange Situation:one study that reported a significant relationbetween day-care and attachment (Pierre-humbert, Frascarolo, Bettschart, Plancherel,& Melhuish, 1992) and three that did not(Burchinal et al., 1992; Howes & Hamilton,1992; Wille, 1992). Moreover, recently pub-lished studies of day-care and attachmentmeasured with the Q-sort methodology havereported no relation (Belsky & Rovine, 1990;Weinraub, Jaeger, & Hoffman, 1988), al-though one of these studies reported a sig-nificant relation between day-care and ameasure of attachment that combined be-havior scores, Q-sort scores, and an attach-ment scale (Belsky & Rovine, 1990). Impli-cations for social policy, based on researchon day-care and attachment, have been sug-gested in many sources and have includedarguments against social programs designedto improve the quality or availability of day-care for infants (e.g., Marshner, 1988; Shell,1988; Wallis, 1987). Results from method-ologically sound replication studies, then,whether their results show a significant ef-fect or not, are particularly important forreaching any conclusions about the effectsof day-care on mother-infant attachment thatmay be relevant for testing theories or for-mulating social policy.

For any single attempt at replication,the usefulness of the data will depend onthe extent to which its sample and methodsare comparable to those used in previousstudies that report the effect. The compara-bility of day-care and attachment studies de-pends on whether their measures of attach-ment and definitions of no, full-, andpart-time day-care are the same, whethertheir samples are similar, and whether theirdata are analyzed in the same ways. The is-sue of comparability is not a simple one.Measures, definitions, and analyses varywidely among studies of day-care and attach-ment. Even though the four studies summa-rized by Belsky (1988) were all based onlow-risk samples, and even though all usedthe standardized Strange Situation and de-fined day-care as a categorical measure, theirmeasures of attachment, definitions of full-and part-time day-care groups, and data anal-ysis techniques varied (Barglow et al., 1987;Belsky & Rovine, 1988; Chase-Lansdale &Owen, 1987; Jacobson & Wille, 1984). Fortu-nately, in our file drawer, we have data froma similar sample assessed in the Strange Sit-uation and representing a comparable rangeof full- and part-time day-care. Thus, we areable to replicate the measures, definitions.

and data analysis strategies used in each ofthe four studies.

In the four studies, the varying mea-sures of attachment included two groups ofinfants (secure, insecure), three groups of in-fants (secure, avoidant, resistant), and con-tinuous measures of attachment behaviors(proximity seeking, contact maintenance,avoidance, resistance). Across the four stud-ies, varying definitions of full- and part-timeday-care were used to divide subjects intotwo, three, or four groups based on the num-ber of hours per week in day-care, splittingthe groups at various points: 0, 4, 5, 20, and/or 37 hours per week. The various groupingsdefined "low" or no day-care as anywherefrom 0 to at most 20 hours per week and"high" day-care as anywhere from at least20 to over 37 hours per week, with somestudies making a distinction between part-time and full-time day-care and some not.This variation in the definitions of full- andpart-time care "in some cases appears to bebased on post hoc considerations rather thanon a priori predictions regarding the waysin which varying amounts of day-care mightaffect infant development" (Lamb & Stern-berg, 1990, p. 372). Finally, a wide varietyof statistical analysis strategies were appliedto these varying measures. Data analysistechniques used in the four studies includedchi-square tests, t tests, analyses of variance(simple and multivariate), and loglinearmodel testing. Some investigators includedpossible mediating variables in their analy-ses, others did not.

Because our data are based on the samemeasures used in the four studies, can beanalyzed in the same way, and include twoof the same potentially mediating variables(infant sex, parity), they provide a useful rep-lication and allow a comparison ofthe vary-ing measures, definitions, and analyses usedin studying day-care and attachment. Thepurpose of the present study was, then, toreplicate the four studies combined andsummarized by Belsky (1988) using (a) asimilar sample (low-risk, middle- to upper-income, intact families), (b) the same attach-ment measures, (c) the same definitions offull- and part-time care, and (d) the samedata analysis techniques. To accomplish thisgoal, we analyzed a data set from a com-pleted study of attachment for which familybackground data had been collected, includ-ing information about the timing and extentof all nonparental child care arrangements(Roggman, Langlois, & Hubbs-Tait, 1987).For these analyses, we used each of' the at-tachment measures, each of the definitions

of full- and part-time care, two ofthe mediat-ing variables (infant sex, parity), and each ofthe statistical analyses used in the four stud-ies cited by Belsky (1988).

Methodr^ Subjects.—Oiir sample included 105

Caucasian infants, 44% boys and 55% first-borns, from two-parent families of lower-

j^middle to upper-middle socioeconomicstatus (SES determination method byHollingshead & Redlich, 1958, based on thehigher of either parent's education and occu-pation). Families varied in caregiving ar-

prangements for their infants: 39% ofthe in-i fants received exclusively maternal care,

11% had a babysitter in their home, 27% at-I tended family day-care, and 24% attended(_a day-care center. The characteristics of the

sample are very similar to characteristics ofthe samples included in the four studiessummarized by Belsky (1988), as can be seenin Table 1. Each family was selected from afile of birth announcement records main-ta;ined by the Children's Research Labora-tory at the University of Texas. Familieswere initially contacted by a letter introduc-ing a study of infants' social relationshipsand then by a telephone call from a recruiterto further explain the procedures of thestudy, determine eligibility, and collect rou-tine family background information that in-cluded infonnation about day-care.^ Sub-jects were recruited for a study initiallydesigned to test the relations between at-tachment and other social behaviors (seeRoggman et al., 1987).^

^-"'^ Procedure.—Attachment status was1 evaluated when the infants were 12—13I months old (mean age = 12.25 months). In-( fants and their mothers were videotaped in

the standard Strange Situation (Ainsworth et

Roggman et al. 1433

al., 1978). Two coders,"' blind to day-care sta-tus of the infants,"* viewed the two reunionepisodes on each videotape and indepen-dently rated^ proximity seeking, intraclassr(208) = .91; contact maintenance, intraclassr(208) = .93; avoidance, intraclass r(208) =.81; and resistance, intraclass r(208) = .75;and detennined the classification of attach-ment for each subject, initial proportion ofagreement = .94, kappa = .82. Similar tothe four comparison studies in which 63%-78% of subjects were assessed as securelyattached, 78% of subjects in our sample wereassessed as securely attached, 13% as anx-ious/avoidant, and 9% as anxious/resistant.

Data analysis.—To replicate the fourstudies summarized by Belsky (1988), datawere analyzed using three types of attach-ment measures, four breakdowns into full-and part-time day-care groups, and four dataanalysis techniques. Attachment was testedas a two-category measure (secure, anxious),a three-category measure (secure, anxious/avoidant, anxious/resistant), and as a set ofbehavioral measures (proximity seeking,contact maintenance, resistance, avoidance).Day-care groups were compared usinggroupings defined in the same way as in thefour studies: (a) two groups, defined as un-der 5 and over 20 hours/week (Barglow etal., 1987), (b) two groups, split at 20 hours/week (Belsky & Rovine, 1988), (c) twogroups, defined as 0 and over 37 hours/week(Chase-Lansdale & Owen, 1987), and (d)three groups, split at 4 and 20 hours/week(Jacobson & Wille, 1984).® Sample sizesvary, because day-care groupings used insome studies eliminated subjects in part-time day-care (i.e., Barglow et al., 1987;Chase-Lansdale & Owen, 1987). Data analy-sis techniques included categorical chi-square tests of attachment categories (usedin all four studies); loglinear models includ-

^ We use the term "day-care" to refer to any form of nonmaternal care, whether center-basedor not.

^ The N for this study differs from that reported by Roggman et al. (1987), a longitudinalinvestigation involving several visits to the laboratory. The subjects in the present study repre-sent all of those with complete data for the first visit only.

• One of the coders was trained in the scoring and coding of Strange Situation behavior atthe University of Minnesota, and she trained the additional coders in this study.

* The day"care data were not "exhumed" from the file drawer until after publication of theinitial attachment study (Roggman et al., 1987).

' Raters' resistance ratings were typically (89%) within one point but were not consistentlyhigher for one rater than the other, thereby reducing the predictability between raters and thecoriservative intraclass correlation which assumes random raters and unaveraged ratings.

^ The four full- and part-time day-care groups used by Belsky and Rovine (1988) were notused in this study for two reasons: (a) their results were identical for the four- and two-groupanalyses, and (b) dividing our sample using the same four groups would have resulted in verysmall cells when sex and parity were included as factors.

I, o o < N

3 ali

(2o

o2

13'V

3

ile,

n)

13

1CO

.2>

-1- be

68 2''

P<

> o S

idle

,

+ 2'

c"3)o

P.

. g - O.5" f Z(2o

ing categorical variables of day-care, attach-ment, sex, and parity (Belsky & Rovine,1988); analyses of variance (ANOVAs) ofavoidance and resistance behavior ratings(Barglow et al., 1987; Belsky & Rovine,1988); and multivariate analyses of variance(MANOVAs) of all the attachment behaviorraidngs (Chase-Lansdale & Owen, 1987).

Because the four studies either includedinfant sex and parity as a factor in their statis-tical analyses or restricted their samples tofirstborns, the present analyses either in-cluded sex and parity (firstbom vs. laterborn)as factors or were performed separately foreach sex and parity group. In addition, corre-lation analyses were conducted to assess therelation of the continuous variable of day-care hours per week with the attachment be-havior ratings.^ Because some researchershave suggested that the age at which infantsbegin to experience regular nonmatemalcare is important (Blanchard & Main, 1979;Goldberg & Easterbrooks, 1988), we testedage of entry into day-care in relation to theattachment measures. Type of day-care andfamily socioeconomic Status (Hollingshead& Redlich, 1958) were also tested in relationto attachment measures and quantity ofday-care.

ResultsThe relations between categories of at-

tachment, day-care, and other categoricalmeasures were analyzed by comparing a se-ries of loglinear models, following Belskyand Rovine (1988). To maximize statisticalpower, two series of three-way loglinearmodels of the day-care and attachment rela-tion were tested, first with sex and then withparity, rather than four-way models withsmaller cell sizes. These models were testedusing all four groupings of day-care as de-fined above in relation to both two attach-ment groups (secure, anxious) and three at-tachment groups (secure, anxious/avoidant,anxious/resistant).

Model comparisons testing the effect ofthree-way and two-way relations among day-care, attachment, and sex revealed no sig-nificant relations between day-care and at-

Roggman et al. 1435

tachment. Comparable tests including parityinstead of sex revealed only one significanteffect: a three-way relation among parity,day-care, and attachment when the three-group definition of day-care was used withgroups split at 4 and 20 hours/week (k la Ja-cobson & Wille, 1984). This effect was sig-nificant in relation to both two attachmentgroups (secure, anxious), x^(2) = 8.62, p =.01, and three attachment groups (secure,anxious/avoidant, anxious/resistant), x^(4) =11.66, p = .02. Thus, of the 32 tests of log-linear models, only two showed any signifi-cant relation among categories. Althoughprobably due to chance, this significant ef-fect was explored further in simple chi-square analyses described below (see n. 7below).

Chi-square tests for the entire sample,separate sex groups, and separate paritygroups were performed because they repli-cate analyses used in previous research byBarglow et al. (1987), Chase-Lansdale andOwen (1987), and Jacobson and Wille (1984)and allow for the exploration of the signifi-cant interaction revealed in the loglinearmodel comparisons. Again, all four group-ings of day-care were tested in relation toboth two and three attachment groups. Ofthe 40 separate chi-square tests of the rela-tion between day-care and attachment, onlythe tests for laterboms using the three-groupday-care definition, with groups split at 4and 20 hours/week (a la Jacobson & Wille,1984), were significant. The effect was sig-nificant for laterboms in relation to both twoattachment groups (secure, anxious), x^(2) =8.24, p = .02, and three attachment groups(secure, anxious/avoidant, anxious/resis-tant), x^(4) = 9.62, p = .047. Laterbomswere most likely to be secure when theyWere in day-care either more than 20 or lessthan 4 hours per week. In those respectiveday-care groups, 93% and 91% ofthe infantswere assessed as securely attached, but only55% of laterboms in part-time day-care wereassessed as securely attached with the re-maining laterboms in part-time day-careassessed as anxious/avoidant or anxious/resistant.

Day-care group differences in behavior

' Although the number of analyses increases the probability of a Type I error, the attemptto replicate previous findings required that we use all of the analytic techniques and definitionsused in previous studies reporting significant results. Nevertheless, all p values reported in thetables are calculated assuming single tests. Because increasing the probability of Type I errorsruns counter to the major argument of this article, that there is no relation between attachmentand day-care in this sample, inflating Type I errors is justifiable and, in this case, is the moreconservative approach.

1436 Child Development

ratings as measures of attachment weretested by MANOVAs, following Chase-Lansdale and Owen (1987), with either sexor parity as a second factor. Four behavioralratings in each ofthe two reunion episodes^were compared across groups divided byday-care and either sex or parity, using eachof the day-CEire groupings. The results ofthese eight MANOVA tests revealed no sig-nificant main effects of day-care and no sig-nificant interactions of day-care and eithersex or parity. To be thorough in our replica-tion ofthe data analysis methods used in thefour studies,^ we then conducted separateMANOVAs for each gender group (k laChase-Lansdale & Owen, 1987). In addition,we conducted separate MANOVAs for eachparity group. Of these 16 tests, there wereno significant day-care effects.

Follow-up ANOVAs for each of the at-tachment behavior variables in each reunionepisode were then performed to replicatethe analyses of variance from previous stud-ies (Barglow et al., 1987; Belsky & Rovine,1988). Two-way ANOVAs were first con-ducted with infant sex as a second factor,following Belsky and Rovine (1988), andthen with parity as a second factor, followingBarglow et al. (1987). Ofthe 32 ANOVA testsof day-care group differences in attachmentbehavior ratings with sex as a second factor,six tests were significant, representing onlytwo behaviors, proximity seeking and resis-tance. Ofthe 32 ANOVA tests with parity asa second factor, six were significant, repre-senting the same two behaviors, proximityseeking and resistance. The specific behav-iors showing significant day-care effects canbeen seen in Table 2. Both sets of analysesindicate that proximity seeking in the firstreunion varied significantly only betweenthe two day-care groups defined as under 5and over 20 hours/week (a la Barglow et al.,1987). In addition, both sets of analyses indi-cate that resistance in the first reunion var-ied significantly between groups for everyday-care grouping. Resistance in the secondreunion varied significantly only betweengroups defined as 0 and over 37 hours/week(a la Chase-Lansdale & Owen, 1987) whensex was the second factor and only betweengroups defined as under 5 and over 20 hours/

week (a la Barglow et al., 1987) when paritywas the second factor.

The mean attachment behavior ratingsacross groups for each significant main effectof day-care are shown in Table 3. As can beseen from the group means, proximity seek-ing was higher among those infants in day-care less than 5 hours per week when com-pared with infants in day-care more than 20hours per week. For every definition of day-care grouping, ratings of resistance in thefirst reunion were lowest for the infants whospent the most time in day-care and highestfor those who were in day-care either part-time or not at all. There were no interactioneffects of sex and day-care, but there was asignificant interaction effect of parity andday-care when day-care groups were split at4 and 20 hours/week (a la Jacobson & Wille,1984), F(2, 99) = 7.6, p = .001. Explorationof simple effects within parity groups sug-gests that for firstborns resistance in the firstreunion was highest among infants in noday-care and for laterboms it was highestamong infants in part-time day-care.

Both the quantity of day-care, measuredas total hours per week, and the Onset of day-care, measured as age at entry, were testedin relation to measures of attachment. Quan-tity of day-care did not differ between at-tachment groups but was negatively corre-lated with behavioral ratings of resistance.That is, as the number of day-care hours perweek increased, ratings of resistance in thefirst reunion decreased, r(103) = — .24, p =.01. Onset of day-care, among infants in day-care at least part time, also did not differ be-tween attachment groups but was negativelycorrelated with resistance in the second re-union. That is, the younger the infant beganday-care, the lower the resistance rating,r(55) = .30, p = .02. To assess the combinedeffect of both early and extensive day-care,attachment classification and ratings of 33 in-fants who had received both early (before 6months) and extensive (20 or more hours/week) day-care were compared to the rest ofthe infants in the sample. Only one signifi-cant result was obtained: Ratings of resis-tance in the first reunion were lower for in-fants receiving early and extensive day-care(M = 1.5, SD = 0.8) than for the other in-

* Chase-Lansdale and Owen (1987) included two additional behaviors in their analyses(initial distance interaction, overall distance interaction) which were virtually identical betweenday-care groups.

^We did not include the separate t test analyses used by Barglow et al. (1987) to testday-care effects on avoidance, because they are statistically equivalent to the ANOVA tests ofthat behavior.

Roggman et al. 1437

TABLE 2

MAIN EFFECTS OF DAY-CARE ON ATTACHMENT BEHAVIORS IN TWO-WAY A N O V A S

SECOND FACTOR IN ANOVA

Sex Parity

DAY-CAJRE GROtrpiNG F p F p

Two groups, under 5 and over 20 hours/week:"Proximity seeking, reunion 1 5.15 .02 4.64 .03Resistance, reunion 1 6.85 .01 6.48 .01Resistance, reunion 2 N.S. 4.59 .03

Two groups, split at 20 hours/week:^Resistance, reunion 1 6.94 .01 6.54 .01

Two groups, 0 and over 38 hours/week:'^Resistance, reunion 1 9.77 .003 10.39 .002Resistance, reunion 2 3.85 .05 N.S.

Three groups, split at 4 and 20 hours/week:**Resistance, reunion 1 3.69 .03 3.58 .03

NOTE.—There were no significant interaction effects of day-care with either sex or parity.Groupings follow those used by other researchers as cited.

»Barglow, Vaughn, & Molitor (1987); df = 1, 82.^ Belsky & Rovine (1988); df = 1, 101." Chase-Lansdale & Owen (1987); df = 1, 66.<• Jacobson & Wille (1984); df = 2, 99.

TABLE 3

MEAN ATTACHMENT BEHAVIOR RATINGS FOR SIGNIFICANTLY DIFFERENTDAY-CARE CROUPS

CROUP MEANS (SDS)

DAY-CARE CROUPING Group 1 Group 2 Group 3

Two groups, under 5 and over 20 hours/week:"Proximity seeking, reunion 1 4.02 (1.9) 3.13 (1.9)Resistance, reunion 1 1.91 (.9) 1.43 (.7)Resistance, reunion 2 2.54(1.5) 2.00(1.1)

Two groups, split at 20 hours/week:""Resistance, reunion 1 1.94 (1.0) 1.43 (.7)

Two groups, 0 and over 38 hours/week:'^Resistance, reunion 1 1.93 (.9) 1.31 (.5)Resistance, reunion 2 2.61 (1.5) 1.90 (1.0)

Three groups, split at 4 and 20 hours/week:''Resistance, reunion 1 1.91 (.9) 2.00 (1.3) 1.43 (.7)

Firstborns 2.05(1.0) 1.20 (.4) 1.38 (.6)Laterborns 1.77 (.8) 2.73(1.4) 1.50 (.9)

NOTE.—Groupings follow those used by other researchers as cited.»Barglow, Vaughn, & Molitor (1987).'' Belsky & Rovine (1988).' Chase-Lansdale & Owen (1987).•• Jacobson & Wille (1984).

fants (M = 1.9, SD = 1.0), F(l, 103) = 4.48, qnly one attachment measure, proximityp = .037. seeking in the first reunion, F(2, 61) = 3.63,

p = .032, with those infants in center day-For those infants in day-care, we ana- care showing more proximity seeking (M =

lyzed type of day-care (in-home, family, or 4.2, SD = 2.0) than those in family day-carecenter day-care) in relation to measures of (M = 2.9, SD = 2.0) or those with in-homeattachment and the quantity and onset of babysitters (M = 2.6, SD = 2.1). When theday-care. Type of day-care was related to number of hours per week of day-care was

1438 Child Development

compared across groups divided by type ofday-care, there was a significant main effectfor type of day-care, F(2, 61) = 9.61, p =.0001. Infants in family day-care averagedmore hours of day-care per week (M = 34.9,SD = 10.1) than those in center day-care (M= 18.8, SD = 18.3) or those with in-homebabysitters (M = 20.1, SD = 12.7). The ageof infants when beginning day-care did notvary with the type of day-care.

Finally, we analyzed family socioeco-nomic status in relation to measures of at-tachment and the quantity, onset, and typeof day-care. None ofthe attachment or day-care measures were related to family socio-economic status for either the entire sampleor the day-care subsample.

Discussion,,-'' The data from four previous studies, suf-1 ficiently similar in sample and measurementI characteristics to be combined and summa-I rized (Belsky, 1988), showed that infant day-l care experience was significantly related toi insecure attachment. In the present study,'i/ive used the same attachment measures, thef same day-care group definitions, and the! same statistical techniques but did not find

a consistent relation between day-care andattachment, despite the many tests we con-ducted and despite the fact that we did not

\; ijontrol for chance efFects. Several research-ers who have used the Q-sort methodologyfor assessing security of attachment (Waters& Deane, 1985) have also reported no differ-ence in Q-sort security scores between in-fants who attend day-care and those who donot (Belsky & Rovine, 1990; Caldera, 1991;Newman-Cedar et al., 1988; Strayer, 1990;Weinraub et al., 1988). The current studyprovides evidence that the absence of a rela-tion between infant day-care and later at-tachment is not due to differences betweenthe Strange Situation and Q-sort methodolo-gies for assessing attachment but rather isgeneralizable across these two methodolo-gies. The few statistically significant effectswe did find were for specific group defini-tions and statistical analyses. There was noeffect robust enough to be significant across

,^all definitions, using all types of analyses.: These results suggest that obtaining signifi-

cant differences among day-care groups, atleast in our sample, depends on how full-and part-time care are defined, which attach-ment measures are tested, and how the dataare analyzed.

"TThe most consistent pattern of results

emerging from our diverse analyses of vari- iously defined variables suggested that indi- ications of anxious attachment were most 1likely among infants in little or part-time Jday-care. Resistance, which is measured bycontinuous behavioral ratings and consid-ered an expression of anxiety in the mother-infant attachment relationship, was lowestfor those in full-time day-care and those inearly extensive day-care. Anxious attach-ment, measured by categories, was mostlikely among part-time day-care infants, atleast for those who were iaterborns. Thus,part-time day-care infants showed more re-sistance in the Strange Situation, as revealedby ANOVAs of continuous measures, andwere also more likely to be classified as anx-iously attached, as revealed by loglinearanalyses of categorical measures.

Part-time day-care, therefore, may be re- ;lated to more anxiety in the mother-infant \relationship than either no day-care or full- 'time day-care. Perhaps the conflict and am-bivalence displayed by these infants re-sulted from coping with less regular workschedules. Or perhaps mothers working only10—20 hours a week expressed confiict aboutboth parenting and employment that wasthen reflected in infant attachment behavior.Alternatively, it may be that mothers work-ing such part-time hours depend on a varietyof child care arrangements on an "asneeded" basis, resulting in less consistencyfor both infant and mother. Goldberg and^Easterbrooks (1988) have reported that ma- jternal satisfaction is lower when there are |more child care arrangements needed. Such ^maternal dissatisfaction, coupled with the ipossible effects of disruptions in child care Iarrangements, may lead to increased mater- ;nal stress and conflict that, again, may be |refiected in infant attachment behavior. The "impact of such stress may be greater forlaterboms, because they already are likelyto receive less attention from their parentsthan do firstborns (Lewis & Kreitzberg,1979). It is also possible that groupings de-fining part-time day-care, particularly whendivided by parity groups, resulted in smallcell sizes that infiated the probability ofType I errors or reflected a peculiarity of thissample. An additional limitation is that resis-tance scores were estimated to have lowerreliability than the ratings of other attach-ment behaviors. It is noteworthy, however,that Rodning (1988) found a similar relationbetween part-time maternal employmentand insecurity in two separate samples. Nev-ertheless, these results were not necessarily

in the direction predicted by other research-ers or by attachment theory.

Regardless of the underlying reason,tliis pattern in which part-time day-care in-fsints showed the most anxiety suggests thatmisleading results could be produced if in-fants in part-time day-care had been omittedfrom sampling, eliminated from analyses, orcombined with full-time day-care infants, ashas been done in several studies. Indeed,when our subjects in part-time care wereeliminated from analyses, following Barglowet al. (1987), tbe results suggested that prox-imity seeking was lowest in the group withmore tban 20 hours per week in day-care andhighest in the group with less than 5 boursper week in day-care. As proximity seekingduring reunion is viewed as a positive at-tsichment bebavior, this suggests that infantswhose mothers work more tban part time areless likely to display secure attachment be-havior. Tbis result would be misleading ifadditional analyses did not provide clarifi-cation. Thus, directly conflicting patterns ofresults were obtained depending on howfull- and part-time care groups weredefined.

In addition to tbe variations in defini-tions, measures, and analyses, some of thedifferences in results among studies may beexplained by variations among tbe popula-tions sampled, even though all were low riskand lower-middle to upper-middle socioeco-nomic status. Locations ofthe studies variedfrom small communities to large metropoli-tan areas; samples varied from well-off tradi-tional families to professional dual-careercouples to families in which both parentswere working to maintain a middle-classlifestyle.

Recruitment of subjects may also ac-count for differences in results. Some stud-ies recruited families after motbers had re-turned to work while others recruitedfamilies before tbe infants were bom. Belsky(1988, 1989) believes that the results of day-care studies may be dependent on tbe timeat wbicb subjects were recruited, implyingthiat tbe effects of day-care may stem frompreexisting group differences. According toBelsky, studies recruiting subjects at the endof tbe first year, such as tbis one, may sys-tematically exclude tbose families mostlikely to find matemal employment stressfuland insecurity promoting. However, it is justas likely tbat families with no matemal em-ployment wbo find otber aspects of life tobe stressful and insecurity promoting would

Roggman et al. 1439

also be less likely to participate in researcbstudies. Furthermore, tbere are no compel-ling a priori reasons to assume that recruit-ment at the end of the first year is more bi-ased tban prenatal recruitment. Recruitingsubjects prenatally may provide a samplerepresentative largely of people who arewilling to agree, before their baby is evenbom, to participate in a longitudinal studytesting the "effects" of day-care. General re-lations between day-care and attachmentwill only emerge by examining a variety ofsamples recruited at a variety of time points.It currently does not seem prudent to ex-clude samples based on time of recruitmentwithout compelling empirical evidence tbatone method of recruitment is biased and an-other method is not.

Anotber variable tbat may account for^differences in results among studies of day- \care and attacbment is tbe quality of day- \care provided for tbe infants tested. Tbetypes of day-care varied widely across thesestudies, from exclusively in-home care (Bar-glow et al., 1987) to a mix of in-home care,family day-care, and day-care centers such --''as that found in tbe present study. Althoughwe did not specifically assess tbe quality ofday-care, the type of day-care, which mayrepresent one aspect of quality, was not re-lated to category of attachment, althoughproximity seeking in the first reunion washigher for infants cared for in day-care cen-ters than tbose in otber types of day-care.Tbe quantity of day-care, related to lowerresistance, was greater for family day-carecompared to center and in-home day-care,but tbis pattern of relations among type ofday-care, quantity of day-care, and attach-ment behavior does not explain tbe lack ofoverall relations between the quantity ofday-care and attachment. Furthermore,there is no reason to suspect that the generalquality of available day-care was any differ-ent in the community in which this studywas conducted tban in the communities inwhich any of the replicated studies wereconducted. Therefore, it is not likely tbat theabsence of a negative relation between day-care and attacbment in our study is due toa higber quality of day-care. Moreover, thesubjects in tbe Barglow et al. (1987) studyreceived exclusively in-home care tbat wasassumed to be of tbe highest quality becauseof tbe small group size and ratio (one care-giver for tbe infant arid any siblings in a sin-gle family), yet that study did find a sig-nificant relation between day-care andattachment.

1440 Child Development

Although we did not find many signifi-cant relations between day-care and attacb-ment, the lack of findings does not seem tobe due to lack of power, because our samplewas approximately tbe same size as otberstudies finding significant relations. More-over, a series of power analyses (assuming amedium effect size) indicated adequatepower to detect a statistically significant ef-fect if one exists (Coben, 1988). Reducedpower estimates resulted only when thesample size was severely reduced for tbechi-square analyses involving separate sexand parity groups. In addition, our analyseswere powerful enough to detect other sig-nificant correlates of day-care. **

Tbis study was not originally designedto examine tbe issue of infant day-care andits relation to attacbment (see Roggman etal., 1987). Tbe advantage of sucb a post bocanalysis is tbat tbere is no possibility of ex-perimenter bias, because researchers in-volved in tbe project were blind to day-carestatus of tbe subjects during the stages ofdata collection. Tbe usual disadvantage ofsucb "accidental" studies is tbat tbe mea-sures employed are not appropriate for test-ing tbe hypothesis of interest. Such is nottbe case here, however, because we bave ex-actly tbe same kind of data as tbe four stud-ies we attempted to replicate. By using tbevarying measures, definitions, and analysesfrom tbose studies, we uncovered varyingresults, most indicating little relation be-tween day-care and attacbment or perhapseven a positive relation between full-timeday-care and attachment.

We are not claiming tbat tbe results ofthe four previous studies themselves (Bar-glow et al., 1987; Belsky & Rovine, 1988;Cbase-Lansdale & Owen, 1987; Jacobson &Wille, 1984) or tbe study combining datafrom tbese studies (Belsky, 1988) are flawedor erroneous. We do claim, however, that ad-ditional studies, sucb as tbe one reportedhere, are useful and necessary for exploringtbe overall relation between day-care andmother-infant attachment. Gottfried andGottfried (1988), after concluding from stud-ies reported in their edited volume tbat ma-ternal employment does not impair cbil-

dren's development, recommend tbatdecisions about maternal employment"should be based on scientific knowledge"(p. 284). Sucb knowledge requires a reviewof all metbodologically sound studies of day-care and attacbment, whether they find sta-tistically significant effects of day-care ornot. Our search for unpublisbed studies onday-care and attacbment indicated tbat nullresults are very likely to be a cbaracteristicof studies tbat remain in file drawers. In-deed, studies are likely to be in file drawersfor just tbat reason: Tbere were no signifi-cant eflects to report. If reports of null re-sults from methodologically sound studieswere published, the addition of even a rela-tively small number of file drawer studiescould cbange the results of summary analy-ses, and tbe conclusions one migbt reacbabout the relation between day-care and at-tacbment would be quite different from tbeconclusions reached by examining only tbeexisting publisbed literature.

Recently, Child Development pub-lisbed a study reporting no group differ-ences in aggressive bebavior between at-tachment groups (Fagot & Kavanagb, 1990).Tbis report of null results, albeit unusual tofind in a major journal, is as important to ourunderstanding of tbe etiology of aggressionand its relation to attachment as tbe findingof a difference may bave been. Certainly, thesame is true for our understanding of tbe ef-fects of day-care on attacbment. Assumingthat tbe studies are equivalent in power anddesign (we are not advocating tbat all studieswitb null results be published), metbodo-logically adequate studies tbat concernpublic policy, even tbose witb null results,should be reported in tbe literature and"regarded as a researcb outcome that is asacceptable as any otber" (Greenwald, 1975,p. 16).

Several day-care researchers (Glarke-Stewart, 1989; Ricbters & Zahn-Waxler,1988; Thompson, 1988) bave concluded tbatall tbat is known for certain about tbe truerelation between day-care and attacbment isthat tbis relation is currently unknown.Based on our data, we would concur witbsucb an assessment. A comprehensive longi-

^ Indeed, results of exploratory analyses indicating significant relations of day-care M ithphysical attractiveness and birth history (physically less attractive infants are likely to be inday-care more hours per week, as are boys delivered by Caesarean section) suggest that thepower of our design was certainly adequate to detect effects of day-care on attachment had therebeen any.

Roggman et al. 1441

tudinal multi-site research project to test theeffects of day-care on randomly selectedsamples of infants, currently funded by theNational Institute of Child Health and Hu-man Development, may provide data thatwill clarify how the mother-infant relation-ship is affected by daily separations (Fried-man, 1992). Meanwhile, unless studies find-ing no relation between attachment andday-care as well as those finding a relationare published, the true relation will neverbe known. Until the true relation is known,inferences about the public policy implica-tions of research on attachment and infantday-care will be both premature and fiawed.Rosenthal (1979; Rosnow & Rosenthal,1989) and others (e.g., Greenwald, 1975;Kupersmidt, 1988; Lane & Dunlap, 1978)who have examined the bias in research lit-eratures resulting from the significance cri-terion adopted by the field would surelyagree.

References

Ainsworth, M. D. S., Blehar, M., Waters, E., &Wall, S. (1978). Patterns of attachment. NewYork: Wiley.

Bairglow, P., Vaughn, B. E., & Molitor, N. (1987).Effects of maternal absence due to employ-ment on the quality of infant-mother attach-ment in a low-risk sample. Child Develop-ment, 58, 945-954.

Belsky, J. (1988). The "effects" of infant day care"- reconsidered. Early Childhood Research

Quarterly, 3, 235-272.Belsky, J. (1989). Infant-parent attachment and

day care: In defense of the Strange Situation.In J. S. Lande & S. Scarr (Eds.), Caring forchildren: Challenge to America. Hillsdale,NJ: Erlhatim.

Belsky, J. (1990). Parental and nonparental childcare and children's socioemotional develop-ment: A decade in review. Journal of Mar-riage and the Family, 52, 885-903.

Belsky, J., & Rovine, M. (1988). Nonmatemal carein the first year of life and infant-parent at-tachment security. Child Development, 59,157-167.

Belsky, J., & Rovine, M. (1990). Q-sort securityand first-year nonmaternal care. New Direc-tions for Child Development, 49, 7-22.

Blanchard, M., & Main, M. (1979). Avoidance ofthe attachment figure and social-emotionaladjustment in day-care infants. Develop-mental Psychology, 15, 445-446.

Bowlhy, J. (1969). Attachment and loss: Vol. 1.Attachment. New York: Basic Books.

Bowlhy, J. (1973). Attachment and loss: Vol. 2.Separation. New York: Basic Books.

Burchinal, M. R., Bryant, D. M., Lee, M. W., &Ramey, C. T. (1992). Early day care, infant-mother attachment, and maternal respon-siveness in the infant's first year. Early Child-hood Research Quarterly, 7, 383-396.

Caldera, Y. M. (1991, April). Infant daycare andmaternal characteristics as predictors of at-tachment and com,pliance in toddlers. Paperpresented at the meeting of the Society forResearch in Child Development, Seattle.

Chase-Lansdale, P. L., & Owen, M. T. (1987).Maternal employment in a family context:Effects on infant-mother and infant-father at-tachments. Child Development, 58, 1505-1512.

Clarke-Stewart, K. A. (1988). "The 'effects' of in-fant day care reconsidered" reconsidered.

1— Early Childhood Research Quarterly, 3,293-318.

Clarke-Stewart, K. A. (1989). Infant day care: Ma-ligned or malignant? American Psychologist,44, 266-273.

Everson, M. D. (1982). The impact of day care onthe attachment hehavior of 12- to 24-month-olds. Dissertation Abstracts International,43, 021.

Cohen, J. (1988). Statistical power analysis forthe behavioral sciences (2d ed.). Hillsdale,NJ: Erlhaum.

Fagot, B., & Kavanagh, K. (1990). The predictionof antisocial hehavior from avoidant attach-

' ment classifications. Child Development, 61,864-873.

Friedman, S. L. (1992). The National Institute ofChild Health and Human Development(NICHD) study of early child care: A compre-hensive longitudinal study of young chil-dren's lives. Bethesda, MD: National Insti-tute of Child Health and HumanDevelopment, Human Learning and Behav-ior Branch. (ERIC Document ReproductionService No. ED 353 087)

Class, C. V., McCaw, B., & Smith, M. L. (1981).Meta-analysis in social research. BeverlyHills, CA: Sage.

Coldherg, W. A., & Easterhrooks, M. A. (1988).Maternal employment when children are tod-dlers and kindergartners. In A. E. Cottfried &A. W. Gottfried (Eds.), Maternal em.ploymentand children's development: Longitudinal re-search (pp. 121-154). New York: Plenum.

Cottfried, A. E., & Cottfried, A. W. (1988). Mater-nal employment and children's development:An integration of longitudinal findings withimplications for social policy. In A. E. Cott-fried & A. W. Cottfried (Eds.), Maternal em-ployment and children's development: Longi-

1442 Child Development

tudinal research (pp. 269-288). New York:Plenum.

Creenwald, A. (1975). Consequences of prejudiceagainst the null hypothesis. PsychologicalBulletin, 82, 1-20.

Hoffman, L. W. (1990). Bias and social responsi-bility in the study of maternal employment.In C. B. Fisher & W. W. Tryon (Eds.), I. E.Sigel (Series Ed.), Annual advances in ap-plied developmental psychology: Vol. 4. Eth-ics in applied developmental psychology:Emerging issues in an emerging field (pp.253-271). Norwood, NJ: Ahlex.

Hollingshead, A., & Redlich, F. C. (1958). Socialclass and mental illness: A community study.New York: Wiley.

Howes, C, & Hamilton, C. E. (1992). Children'srelationships with child care teachers: Stahil-ity and concordance with parental attach-ments. Child Development, 63, 867-878.

Jacohson, J. L., & Wille, D. E. (1984). Influenceof attachment and separation experience onseparation distress at 18 months. Develop-mental Psychology, 20, 477-484.

Jaeger, E., & Weinrauh, M. (1990). Early nonma-ternal care and infant attachment: In searchof process. New Directions for Child Devel-opment, 49, 71-90.

Kupersmidt, J. (1988). Improving what is puh-lished: A model in search of an editor. Ameri-can Psychologist, 43, 635-642.

Lamh, M. E., & Sternherg, K. J. (1990). Do wereally know how day care affects children?Journal of Applied Developmental Psychol-ogy, 11, 351-379.

Lamh, M. E., Sternherg, K. J., & Prodromidis, M.(1992). Nonmaternal care and the security ofinfant-mother attachment: A reanalysis of thedata. Infant Behavior and Development, 15,71-83.

Lane, D. M., & Dunlap, W. P. (1978). Estimatingeffect size: Bias resulting from the signifi-cance criterion in editorial decisions. BritishJournal of Mathematical and Statistical Psy-chology, 31, 107-112.

Lewis, M., & Kreitzherg, V. S. (1979). Effects ofhirth-order and spacing on mother-infant in-teractions. Developmental Psychology, 15,617-625.

Low, N. E. (1982). Family day care in early in-fancy: Effects on the infant-mother attach-ment relationship. Dissertation Abstracts In-ternational, 43, 05B.

Marshner, C. C. (1988, May 13). Is day care goodfor kids? National Review, 40, p. 30.

McCartney, K., & Calanopoulos, A. (1988). Childcare and attachment: A new frontier the sec-ond time around. American Journal of Ortho-psychiatry, 58, 16-24.

McCartney, K., & Phillips, D. (1988). Motherhood

and child care. In B. Birns & D. F. Hay (Eds.),The different faces of motherhood (pp. 157-183). New York: Plenum.

Moser, M. H. (1989). The effects of infant temper-ament, maternal stress, and maternal employ-ment on maternal attachment to the infant andinfant attachment to the mother. DissertationAbstracts International, 50, 06A.

Newman-Cedar, M., Plunket, S. W., Kessler, P. B.,& Ward, M. J. (1988, April). Maternal employ-ment and security of attachment. Paper pre-sented at the International Conference on In-fant Studies, Washington, DC.

Norherg, K., Mack, S. A., Beck, D. L., & Brazelton,B. (1989, April). How much, how soon, andwho: The effect of daycare on attachment.Paper presented at the meeting of the Societyfor Research in Child Development, KansasCity, MO.

O'Connor, M. J., & Sigman, M. (1988, April). Ma-ternal work status in relation to infant-mother attachment. Paper presented at theInternational Conference on Infant Studies,Washington, DC.

Phillips, S. L. (1990). Effects of family day careexperience, home environment, maternal lifestress and attitudes toward dual roles forwomen on infant-mother attachment in intactfamilies. Masters Abstracts International, 30,02.

Pierrehumhert, B., Frascarolo, F., Bettschart, W.,Plancherel, B., & Melhuish, A. (1992). A lon-

_,^ gitudinal study of infants' social-emotionaldevelopment and the implications of extra-parental care. Journal of Reproductive andInfant Psychology, 9, 91-103.

Richters, J. E., & Zahn-Waxler, C. (1988). The in-fant day care controversy: Current status andfuture directions. Early Childhood ResearchQuarterly, 3, 319-336.

Rodning, C. J. (1988, April). Parttime workingmothers: A closer look. Paper presented at theInternational Conference on Infant Studies,Washington, DC.

Roggman, L. A., Langlois, J. H., & Huhhs-Tait, L.(1987). Mothers, infants, and toys: Social playcorrelates of attachment. Infant Behavior andDevelopment, 10, 233-237.

Rosenthal, R. (1979). The "file drawer prohlem"and tolerance for null results. PsychologicalBulletin, 86, 638-641.

Rosnow, R. L., & Rosenthal, R. (1989). Statisticalprocedures and the justification of knowledgein psychological science. American Psycholo-gist, 44, 1276-1284.

Shell, E. R. (1988, August). Bahes in day care. At-lantic, 262, pp. 73-74.

Strayer, F. (1990, April). Daycare infiuence on thequality and stability of Q-sort attachmentclassifications. Paper presented at the Inter-

Roggman et al. 1443

national Conference on Infant Studies, Mon-treal.

Sutton, D. B., & Fox, N. A. (1992). An affectiveprofile of infants in extensive day care in rela-tion to behavior in the Strange Situation. Un-published manuscript.

Tltiompson, R. A. (1988). The effects of infant daycare through the prism of attachment theory:A critical appraisal. Early Childhood Re-search Quarterly, 3, 273-282.

Tliiompson, R. A., Lamb, M. E., & Estes, D. (1982).Stability of infant-mother attachment and itsrelationship to changing life circumstances inan unselected middle-class sample. Child De-velopment, 53, 144-148.

Vaughn, B. E., Gove, F. L., & Egelsuid, B. (1980).The relationship between out-of-home careand the quality of infant-mother attachmentin an economically disadvantaged population.Child Development, 51, 1203-1214.

Waters, E., & Deane, K. E. (1985). Defining andassessing individual differences in attach-

ment relationships: Q-methodology and theorganization of behavior in infancy and earlychildhood. In I. Bretherton & E. Waters(Eds.), Growing points of attachment theoryand research (pp. 41—65). Monographs oftheSociety for Research in Child Development,50(1-2, Serial No. 209).

Walhs, G. (1987, June 22). Is day care bad for ba-bies? Time, 129, p. 63.

Weinraub, M., Jaeger, E., & Hoffinan, L. (1988).Predicting infant outcome in families ofemployed and nonemployed mothers.Early Childhood Research Quarterly, 3,361-378.

Wille, D. E. (1992). Maternal employment: Im-pact on maternal behavior. Family Relations,41, 273-277.

Wille, D. E. (1993, March). Parental characteris-tics and infant security of attachment. Paperpresented at the meeting of the Society forResearch in Ghild Development, New Or-leans.