The Relation of Child Care to Cognitive and Language Development

(1)

Child Development, July/August 2000, Volume 71, Number 4, Pages 960–980

The Relation of Child Care to Cognitive and Language Development

National Institute of Child Health and Human Development Early Child Care Research Network

Children from 10 sites in the United States were followed from birth to age 3 to determine how experiences in child care relate to cognitive and language development (Ns varied between 595 and 856, depending on the as-sessment). Multiple assessments of family and child care environments and of cognitive and language compe-tence were collected. Analyses that adjusted for maternal vocabulary score, family income, child gender, ob-served quality of the home environment, and obob-served maternal cognitive stimulation indicated that the overall quality of child care, and language stimulation in particular, was consistently but modestly related to cognitive and language outcomes at ages 15, 24, and 36 months. The effect sizes for high (top quartile) versus low (bottom quartile) quality ranged from .18 to .48. After adjusting for child care quality, cumulative experi-ence in center-based care was associated with better outcomes than was participation in other types of care. The amount of time children spent in care was not related to outcomes. Children in exclusive maternal care did not differ systematically from children in child care. Tests for lagged relations of earlier child care experiences to later performance (adjusting for current child care) showed that language stimulation predicted subsequent cognitive and language performance 9 to 12 months later. Although children in center care at age 3 performed better than children in other types of care, earlier experience in child care homes was associated with better performance at age 3 than was experience in other types of care. The relations of child care variables to out-comes did not vary consistently as a function of family income, quality of home environment, child gender, or ethnic group.

INTRODUCTION

The last half of the twentieth century has witnessed a rapid increase in the percentage of women in the workforce and a corresponding rise in the number of children who receive routine care by someone other than their mothers. Children are entering child care at younger and younger ages. By 1990, in the United States, over half of the infants under one year old reg-ularly received care by someone other than a parent (U.S. Department of Labor, Bureau of Labor Statistics, 1994). The changing context of early caregiving and the increased number of infants in child care have raised questions among parents, researchers, and pol-icymakers about whether and under what conditions nonparental infant care poses risks or offers benefits for children’s development. These questions have fo-cused on the impact of the quality, type, amount, and timing of child care.

Preschool-age child care has been extensively stud-ied somewhat to the neglect of infant care (cf., Lamb, 1997; Scarr, 1998). Yet infant care is of special interest because experiences in the first year or two are partic-ularly formative: they establish the fundamentals of language and cognitive functioning. For example, one hallmark of language, vocabulary development, begins in the first 2 years of life, before children enter preschool. The relation between language input from caregivers and children’s early vocabulary acquisi-tion is well-documented (for reviews see Adamson,

1995; Messer, 1994). Moreover, the first 2 years mark an important period of brain development during which the density of short-range synaptic connec-tions reaches its peak. These connecconnec-tions are formed on the basis of available input from the environment (Elman et al., 1996). Hence, verbal and cognitive stim-ulation by caregivers in the first 2 years may have a pronounced impact on later language and cognitive competence. This leads us to ask how the quality, type, and amount of care relate to cognitive and lan-guage outcomes in the first 2 or 3 years of life.

QUALITY OF CARE

Interactions between children and caregivers have been linked conceptually and empirically to cognitive and language development (Bloom, 1991; Bornstein & Bruner, 1989; Feurestein, Klein, & Tannenbaum, 1991; Tomasello, 1992; Wertsch, 1985; Wood & Middle-ton, 1975). Quality of care has typically been indexed by process features of adult–child interaction that represent good caregiving, whether provided by a parent or someone else. These features include sensi-tivity and responsiveness to the child’s needs and sig-nals, positive affect, frequent verbal and social inter-action, and cognitive stimulation (e.g., Friedman &

(2)

NICHD Early Child Care Research Network 961

Cocking, 1986; Hart & Risley, 1995). The relation of child care quality in the infant and toddler years to cognitive and language development has been stud-ied in two ways: naturalistic, correlational investiga-tions of existing child care settings and experimental studies of planned interventions.

In the naturalistic investigations, high-quality care during the infant and toddler years is generally asso-ciated with better cognitive functioning, complex play, and language development when both are mea-sured at the same age (Burchinal, Roberts, Nabors, & Bryant, 1996; Galinsky, Howes, Kontos, & Shinn, 1994; Howes & Rubenstein, 1985; Howes, Smith, & Galin-sky, 1995; McCartney, 1984; Phillips, McCartney, & Scarr, 1987).

Although this finding is consistent across studies, there is some question about how to interpret these results. Some argue that although most of the effects are statistically significant, they are trivial in magni-tude (Lamb, 1997), which suggests the need to deter-mine effect sizes, particularly in studies with large samples. Others, like Scarr (1998), have asserted that even if there are effects of quality, they do not endure into later childhood and are therefore not sufficiently important to justify large public investments in high-quality care. Evidence for durability of effects is mixed. By way of example, some research supports Scarr’s argument. In a sample in Bermuda, positive relations between early quality and cognitive compe-tence in preschool years did not remain when chil-dren were ages 5–8 (Chin-Quee & Scarr, 1994). In stark contrast, investigators like Field (1991) and Rosenthal (1994) found that high-quality infant care is associated with academic and cognitive performance in the elementary school years. Likewise, a Swedish investigation found that quality of care in the first 3 years of life predicted mathematics performance at age 7 (Broberg, Wessels, Lamb, & Hwang, 1997).

In many studies investigating the relation between quality of care and cognitive and language outcome, quality has been defined by such global, structural in-dices as adult–child ratio and staff education. In sev-eral studies, however, investigators have articulated specific characteristics of child care that are associated with early cognitive and language development. Mc-Cartney (1984) found that children’s cognitive and language development was related to specific types of linguistic and cognitive experiences in child care. This finding accords with what is known about how family environments (specifically mothers) contrib-ute to language acquisition. Parents who talk more (Hart & Risley, 1995; Hoff-Ginsberg, 1991; Wells & Robinson, 1982), parents who talk about objects and events in the immediate environment (Harris, 1992;

Menyuk, Liebergott, & Schultz, 1995), and parents who engage in joint attention with their children as they label objects (Smith, Adamson, & Bakeman, 1988; Tomasello & Farrar, 1986; Tomasello & Todd, 1983) have children whose language development is more advanced. Although the causal direction ac-counting for these associations cannot be ascertained, there is no question in the language literature that language experiences play an important role in chil-dren’s acquisition of vocabulary and grammar. These experiences are therefore likely to be a central compo-nent of quality care.

The effects of child care quality have also been studied experimentally in investigations of planned early interventions for economically disadvantaged children or for those at risk for developmental prob-lems. The findings here are very consistent. Intensive, high-quality interventions begun in infancy have a positive effect on measures of intelligence and on school achievement throughout childhood and ado-lescence (Barnett, 1995; Brooks-Gunn et al., 1994; Burchi-nal, Campbell, Bryant, Wasik, & Ramey, 1997; Lamb, 1997; McLoyd, 1997; Ramey & Ramey, 1998). Further, early gains are associated with language ability in early childhood (Feagans, Fendt, & Farran, 1995; Rob-erts, Rabinowitch, Bryant, & Burchinal, 1989). Such programs not only enhance intellectual functioning but also contribute to infants’ responsiveness to their environments, which may further enable children to elicit stimulation from a range of environments (Burchi-nal, Campbell, Bryant, Wasik, & Ramey, 1997). Al-though some of these effects dissipate during the early school years, the impact of some high-quality infant and preschool programs on school performance and educational attainment continue into middle childhood, adolescence, and, in some instances, adult-hood (Campbell & Ramey, 1994; Currie & Thomas, 1995; Lazar & Darlington, 1982; Luster & McAdoo, 1996; McLoyd, 1997).

TYPE OF CARE

Despite the fact that most infants and very young children are cared for in child care homes (Hofferth, 1996), most research has been conducted in child care centers. Child care centers are more likely than child care home settings to have a planned curriculum and trained caregivers—features that might enhance cog-nitive development. On the other hand, the centers attended by children in the present study at age 6 months had higher child/adult ratios and lower levels of positive caregiving than did child care homes, rel-ative care, or in-home-care (NICHD Early Child Care Research Network, 1996). In general, children from

(3)

962 Child Development

low-income families with experience in center-based infant and toddler care have more advanced lan-guage development and perform better on measures of intelligence and achievement than do children in child care homes, but the differences may be a func-tion of program quality or family characteristics (Broberg et al., 1997; Caughy, DiPietro, & Strobino, 1994). In two studies, center care was related to lan-guage development for White children and for chil-dren of color, but was related more strongly for chilchil-dren of color (Burchinal, Ramey, Reid, & Jaccard, 1995; Helburn, 1995). In a third study (Clarke-Stewart, Gru-ber, & Fitzgerald, 1994), participation in center care was related to advanced cognitive performance in 2- and 3-year-olds from a range of family backgrounds, even when family socioeconomic status (SES) was con-trolled. Other investigations show no difference as a function of type of care (Brand & Welch, 1989; Mel-huish, Lloyd, Martin, & Mooney, 1990). Caughy et al. (1994) found that center care was related to better out-comes for children from low-income families whereas care in child care homes was associated with better per-formance for children from middle-income families.

TIMING AND AMOUNT OF CARE

One major debate about the effects of infant child care revolves around the possibility that early entry into extensive amounts of nonmaternal care is harmful to infant development, regardless of the quality of the care, because it disrupts the parent–child bond (e.g. Barglow, Vaughn, & Molitor, 1987). Because socio-emotional and cognitive/linguistic development are believed to be intimately linked in infancy and beyond (e.g., Kopp, 1997; Lazarus, 1991; Lewis & Michalson, 1983), the disruption of the parent–child relationship due to extensive child care and its impact on socio-emotional development raise the possibility that there will be secondary or mediated negative effects of early and extensive child care on cognitive devel-opment (e.g., Van Ijzendoorn, Dijkstra, & Bus, 1995).

Another concern related to extensive child care is that in settings with more than one child, adults spend less time with any given child, thereby reducing the amount of language and cognitive stimulation. Less overall stimulation might impede development as was shown in a study comparing language stimulation and language outcomes in singletons and twins of the same age (Tomasello, Mantle, & Kruger, 1986).

Most reviews of the literature have concluded that the evidence is mixed about whether the timing and amount of infant child care has positive, negative, or no relation to intellectual and language development (Ackerman-Ross & Khanna, 1989; Clarke-Stewart, 1986;

Hayes, Palmer, & Zaslow, 1990; Lamb, 1997). Some re-searchers have found a positive relation of the amount of infant care to school performance in middle childhood (Andersson, 1992; Field, 1991). In one lon-gitudinal study, there was an apparent sleeper effect. The amount of time children had spent in high-quality, center-based child care before age 3 was unrelated to verbal skill at 3 but was positively associated with verbal ability at age 7 (Broberg et al., 1997). Others, however, have reported poorer later performance for children who received extensive infant care (Baydar & Brooks-Gunn, 1991; Vandell & Corasaniti, 1990). Finally, in two studies, infant care experience was positively related to cognitive and language out-comes for girls but not for boys (Metcalf-Davenport, 1993; Mott, 1991).

VARIATIONS IN RELATIONS OF CARE TO OUTCOMES FOR CHILDREN OF DIFFERENT ENVIRONMENTS

The relation between infant care and intellectual de-velopment appears to depend partly on the family’s income and environment. Positive relations of infant care to intellectual development and achievement have been found more consistently for children from low-income families than for children from more ad-vantaged families (Lamb, 1997). In analyses of chil-dren in the National Longitudinal Survey of Youth sample, for example, the amount of infant care was positively related to reading skill at ages 5 and 6 for children from low-income families but negatively re-lated for children from middle-income families (Caughy et al., 1994; Desai, Chase-Lansdale, & Michael, 1989). This pattern is consistent with the “compensatory education” notion, that children from home environ-ments with limited opportunities for cognitive stimula-tion will obtain more benefits from high-quality care than will children from more advantaged family envi-ronments. However, some recent studies with preschool age children failed to find evidence for compensatory child care effects under conditions of lower parent edu-cation (Stipek, Feiler, Daniels, & Milburn, 1995), poverty, or developmental risk due to gender (Burchinal, Peisner-Feinberg, Bryant, & Clifford, 2000).

The mirror image of the “compensatory education” hypothesis, known as the “lost resources” hypothesis, has also been advanced. It has been suggested that for affluent children, child care environments often pro-vide less optimal stimulation, structure, and support than their family environments (Desai et al., 1989). Consequently, the prediction is that the cognitive and language outcomes for these children would be less optimal than if they were not in child care.

(4)

Finally, there are reasons to expect ethnic differ-ences in the relations between child care and develop-ment, even when the confound of ethnic group with income is taken into account. For minority children, family environments may be different from those of majority culture children (Garcia Coll et al., 1996).

One problem in testing these and other hypotheses about the role of child care in children’s development is the confound between child care experiences and other features of children’s lives. In most investiga-tions in the United States, including the present study, the quality, type, and amount of care experienced by a child are related to family economic and demographic characteristics, family structure, the quality of the home environment, parental behavior and attitudes, and characteristics of the child (Scarr & Eisenberg, 1993; NICHD Early Child Care Research Network, 1997b). Hence, in studies about the effects of child care on child development, detecting the unique con-tribution of child care requires that these child and family attributes be either experimentally or statisti-cally controlled.

PURPOSES OF THIS RESEARCH

In this report, we examine children’s cognitive devel-opment, acquisition of school readiness skills, lan-guage production, and lanlan-guage comprehension as a function of quality, type, and amount of child care during the first three years of life. The National Insti-tute of Child Health and Human Development (NICHD) Study of Early Child Care has a longitudi-nal, prospective design beginning at birth, and the sample is socioeconomically and geographically di-verse. Because the sample in this study was drawn from area hospitals at birth rather than from child care settings, we observed children in the entire range of child care settings used by parents. Child care char-acteristics were assessed on multiple occasions by using direct and intensive observation. Because lan-guage stimulation is theoretically important in these early years, we examined specific language interac-tions between each child and his or her caregivers as potential mediators of the effects of more global indi-cators of quality of care. The analysis controlled for such family characteristics as economic status and the quality of the home environment.

Five questions were addressed: First, do the cumu-lative quality, type, and amount of child care predict children’s cognitive and language skills during the first 3 years of life? Second, assuming some effects of cumulative quality, type, and amount of care on out-come, what are the magnitudes of these effects? Third, how do children raised almost exclusively by

their mothers compare with children who have expe-rienced different levels of quality of child care? Fourth, does child care in the first year or two of life have lasting associations with cognitive and language development at subsequent ages? Fifth, are the rela-tions of child care to cognitive and language outcomes different for children from different income levels, home environments, genders, or ethnic groups?

METHOD Participants

Participants were recruited from hospitals located in or near Little Rock, AR; Irvine, CA; Lawrence, KS; Boston, MA; Philadelphia, PA; Pittsburgh, PA; Char-lottesville, VA; Morganton, NC; Seattle, WA; and Madison, WI. During selected 24-hour sampling peri-ods in 1991, 8,986 women giving birth were visited in the hospital. Of these, 5,265 met the eligibility criteria for the study and agreed to be contacted after their return home from the hospital.

In selecting participants, the following criteria were used to exclude cases from the pool of 8,986 po-tential participants born during the hospital recruit-ment periods: (1) mothers younger than 18 years of age at the time of the child’s birth (3.8% of potential participants); (2) families who did not anticipate re-maining in the catchment area of the study for at least the next 3 years (5.4%); (3) infants of multiple births, those with obvious disabilities, or those who remained in the hospital more than 7 days postpartum (6.8%); (4) mothers with medical problems or acknowledged substance abuse or who were placing their infants for adoption (4.3%); (5) mothers who did not speak En-glish (4.4%); (6) mothers who lived more than an hour from the laboratory site or who were enrolled in an-other study (9.2%); (7) man-others who lived in neigh-borhoods (generally high-rise projects) deemed by police too unsafe for visitation (1.5%); and (8) other exclusions (2.7%). Of the mothers who were eligible, 1.5% refused to be interviewed in the hospital and 3.4% asked not to be called when they returned home. When the infants were 1 month old, 1,364 families (58% of those contacted) with healthy newborns were enrolled in the study. Families were similarly distrib-uted across the 10 sites, with each site enrolling a min-imum of 120 families. When the children were 1 month old, the mean sample size per site was 136.4 (SD5 7.5) and the median sample size was 136. When the children were 36 months old, the mean sample size per site was 122.6 (SD5 4.8) and the median sample size was 122.5. At the time of recruitment, 53% of the recruited mothers were planning to work full-time in

(5)

the child’s first year of life, 23% were planning part-time employment, and 24% were planning no em-ployment. The families enrolled in the study included 24% ethnic-minority children (non-White or His-panic), 11% mothers without a high school education, and 14% single mothers. For a more detailed descrip-tion of the sample and its recruitment, see NICHD Early Child Care Research Network (in press). Table 1 provides information about the demographic charac-teristics of the families by child care usage over the first 3 years of life.

The sample was not designed to be nationally re-presentative. However, after the fact, we compared the demographic characteristics of our sample with those of people living in the same census tracts and nationally. We found that our sample was similar to families in the census tract records and to the nation as a whole on key demographic variables (household income and ethnicity).

All the analyses about the effects of child care pa-rameters on cognitive and language outcomes are based on data from children who were observed in child care and for whom we had complete data on predictor variables and data on at least one outcome variable (ns5 595 at 6 months, 595 at 15 months, 739

at 24 months, 856 at 36 months). The same analyses (but not including observed quality of care as a vari-able) were run on data from all the children partici-pating in the study, regardless of whether we could observe the setting to determine its quality. On vari-ables other than quality of care, these analyses yielded results that shadowed the ones we obtained for the more restricted sample of children observed in child care. Consequently, this paper focuses on the children observed in child care.

Any child who was spending 10 or more hr per week in nonmaternal care at 6, 15, 24, or 36 months was eligible for a child care observation. Of those eli-gible, 78.6% were observed at 6 months, 77.4% at 15 months; 85.8% at 24 months, and 90.3% at 36 months. The reasons that eligible children were not observed included caregiver refusal, child absence from child care, and recent changes in the child care setting (see NICHD Early Child Care Research Network, 1996). The observed sample differed from those eligible but not observed (all differences were significant at p, .05). The families in the observed sample had higher incomes and provided more stimulating home envi-ronments; the mothers had more education, higher Peabody Picture Vocabulary Test (PPVT) scores, and were rated as more stimulating in interactions with the child; and the observed children experienced more hours of child care and were more likely to be in a child care center or a child care home than to be in less formal child care.

The sample observed in child care includes a full range of families with diversity in ethnicity, maternal education level, income and family structure (see Ta-ble 1). In addition, the children were placed in diverse child care arrangements (see Table 2). The children who were observed in care entered child care at the mean age of 4.3 months (SD5 5.3). The mean age for entry into extensive child care—that is, for at least 30 hr per week—was 7.2 months (SD 5 8.1). By 6 months of age 61% of the children were in child care for at least 30 hr a week, and by 12 months of age, the comparable figure is 69%. Children who entered care early also experienced a higher quantity of care dur-ing the first 3 years of life. The correlation between en-try into care for any amount of time and quantity of care was moderate, r(846)52.49, p, .001. The cor-relation between entry into care for a minimum of 30 hr per week and quantity of care was high, r(749) 5 2.63, p, .001.

Procedure

Data were collected in three settings: the child’s home, a laboratory, and the child’s primary child care

Table 1 Demographic Characteristics of the Families by Child Care Usage during the Child’s First 3 Years

Observed in Child Care (n5 856) Not Observed in Child Care (n5 92) Exclusive Maternal Care (n5 210) Ethnicity White/non-Hispanic 678 (79%) 64 (70%) 171 (81%) Hispanic 48 (6%) 7 (8%) 11 (5%) African American 91 (11%) 18 (20%) 19 (9%) Other 39 (5%) 3 (3%) 9 (4%) Maternal education , High school 54 (6%) 12 (13%) 32 (15%) High school 177 (21%) 20 (22%) 39 (19%) Some college/voc 288 (34%) 32 (35%) 64 (31%) BA/BS1 337 (39%) 28 (30%) 75 (36%) Income (family income/

poverty threshold)a Low (, 1.8) 217 (26%) 39 (42%) 82 (39%) Average (1.8–4) 333 (39%) 34 (37%) 79 (38%) High av.-high (41) 300 (35%) 19 (21%) 47 (23%) Partner in householdb Always partnered 643 (78%) 60 (71%) 171 (86%) Sometimes partnered 131 (16%) 16 (18%) 12 (5%) Never partnered 52 (6%) 9 (11%) 17 (9%)

a_{In 1993 dollars, the cutoff of income to needs of 1.8 equals $26,573}

and the cutoff of 4.0 equals $59,052.

b_{Not all parents provided this information at each time point so}

(6)

arrangement (the arrangement in which the child spent the most time or, if the child spent equal time in two settings, the arrangement that was more formal). Mothers and children were visited in their homes when the children were 1, 6, 15, 24, and 36 months old; the primary child care environment was ob-served at 6, 15, 24, and 36 months of age; and children and their mothers were seen in the laboratory at 15, 24, and 36 months. At each assessment point, mothers responded to standardized interview questions about family demographics and other domains of family life not dealt with in this report. The families were telephoned at 3-month intervals between assessment points to update information about child care and family characteristics.

All data collectors were highly trained and certi-fied on data collection procedures. The performance of data collectors was monitored centrally to ensure uniform, high-quality data collection across the 10 sites. Training consisted of several phases: (1) Read-ing manuals for administerRead-ing procedures, which were written by the investigators of the study and

were tailored for the age of the children at the time of data collection and the context in which the data were collected; (2) Collecting initial familiarization pilot protocols that were sent to a centralized place for feedback; (3) Going to a centralized training session; (4) Official piloting with new participants; (5) Send-ing audio tapes or videotapes (as required) to experts for evaluation; (6) Certification—meeting prespeci-fied certification criteria that varied across measures. Details about all data collection procedures are docu-mented in manuals of operation for the study (NICHD Early Child Care Research Network, 1993).

Measures

Demographic, maternal, child, family environ-ment, and child care characteristics were selected as predictors of cognitive and language outcomes. The primary predictors included two measures of child care quality, two measures of child care type, and one measure of amount of child care. Also included were family and child characteristics because parents do

Table 2 Means and Standard Deviations for the Cumulative Predictor and for the Outcome Variables in the Observed Child Care Sample 6 Months (n5 595) 15 Months (n5 595) 24 Months (n5 739) 36 Months (n5 856) M SD M SD M SD M SD Cumulative predictors Background variables

Maternal PPVT-R (36 months only) 101.08 17.35 101.08 17.35 100.42 17.59 99.80 17.88

Income-to-needs ratio 4.22 3.46 4.22 3.23 4.05 3.20 3.84 3.24 Gender (male) .51 .51 .52 .51 HOME total 37.12 4.08 37.95 4.08 41.98 7.11 Maternal stimulation 2.64 .66 2.70 .61 2.74 .73 4.55 1.43 Child care Average hours/week 20.29 12.96 27.89 12.79 27.81 13.59 28.01 13.97

No. times center care .15 .36 .35 .68 .52 .95 .81 1.22

No. times child care home .28 .45 .58 .82 .85 1.12 1.04 1.39

No. times relative/in-home care .38 .49 .88 .86 1.13 1.15 1.34 1.40

Positive caregiving rating 14.75 2.94 14.63 2.89 13.96 2.87 19.50 3.29

Language stimulation .04 2.71 58.62 30.92 50.56 27.09

Outcomes

Bayley MDI 109.30 14.09

CDI Vocabulary Comprehension % 40.40 29.06

CDI Vocabulary Production % 39.31 27.67

Bayley MDI–Revised 93.42 13.98

CDI Sentence Complexity % 41.87 26.16

Bracken School Readiness Composite 42.91 26.61

Reynell Expressive Language % 45.59 26.49

Reynell Verbal Comprehension % 48.04 28.73

Note: Children who had complete data on predictor variables and data on at least one outcome variable collected at that age were in-cluded in these analyses.

(7)

not make choices regarding child care randomly. Their choices are determined by personal characteris-tics, family circumstances, and community condi-tions. Some of the same factors are related to chil-dren’s cognitive and language competence. Selected to reflect these concerns in our analyses were mea-sures of the mother’s vocabulary, family income, overall responsiveness and stimulation of the family environment in general, and more specifically, re-sponsiveness and stimulation by the mother. Gender was included because young girls typically perform better on cognitive and language measures than young boys (e.g., Fenson et al., 1994, or Halpern, 1992).

Table 2 provides means and standard deviations for cumulative predictors and for the outcomes for children who were observed in child care at any of the four assessment times. Table 3 provides comparable information about cumulative predictors and out-comes for children who were eligible to be observed in child care but who were not observed. Table 4 pro-vides information about the same variables for those children who were not in child care for a minimum of 10 hours per week at any assessment point. These are the children in the exclusive maternal care sample (see NICHD Early Child Care Research Network, 1997c, for more details about child care use in the first year of life). Because many of the predictor measures were collected longitudinally, cumulative scores were computed as the mean of that family’s scores up to and including the age at which the dependent vari-able was assessed.

Maternal and Family Characteristics

Maternal vocabulary was assessed by the Peabody Picture Vocabulary Test-Revised (PPVT-R; Dunn & Dunn, 1981), administered to the mothers when chil-dren were 36 months old. Family income was defined as an income-to-need ratio, computed as reported family income (exclusive of Aid to Families with De-pendent Children) divided by the federal poverty threshold for that family size for the year. Conse-quently, a ratio of 1 is the federal poverty threshold. In 1991, the year the children participating in our study were born, the poverty threshold (income to needs 5 1) for a family of four was an annual income of $13,812. Two measures of the family environment included the Total Home Observation for Measurement of the Environment (HOME) Score and the amount of ma-ternal cognitive stimulation in a videotaped observa-tion. The Home Observation for Measurement of the Environment (Caldwell & Bradley, 1984) was coded live to assess the overall quality of the physical and

social resources available to the child in the family context. The HOME consists of direct observation and a semistructured interview with the mother. The fo-cus is on the child as a recipient of inputs from objects, events, and interactions occurring in the family sur-roundings. The Infant/Toddler version was adminis-tered at 6 and 15 months and the Early Childhood version at 36 months. Both versions measure the qual-ity and quantqual-ity of stimulation and support available to a child in the home environment. Analyses used the total HOME. Because there was no 24-month HOME, the average of the 6- and 15-month scores was used for 24-month outcomes. Cronbach’s a was .77 for the 6 months score, .80 for the 15 months score, and .87 for the 36 months score.

A centrally orchestrated system was used to train HOME data collectors. Every four months, each ob-server coded videotaped home visits, and the coding was compared with expert coding. All observers were required to maintain a criterion of scoring like the master coder on 90% of the items (i.e., 41 out of 45 at 15 months, 50 out of 55 at 36 months).

A more targeted measure of maternal cognitive stimulation was obtained from a semi-structured mother–child interaction procedure conducted and videotaped at the family’s home at 6 and 15 months and in the laboratory at 24 and 36 months of age. At 6 months, mothers were instructed to use toys in two containers to play with their children. Some of the toys were provided by the experimenter, and others were the child’s toys that were selected by the mother. At 15, 24, and 36 months, toys were provided by the experimenter and were placed in three containers (Vandell, 1979). Mothers were told that 10 (at 6 months) or 15 min (at later assessments) were avail-able for play with the toys but they were not in-structed on how to divide the time. All tapes were coded at a central location by coders who were unac-quainted with the family or child care history. Mater-nal stimulation of cognitive development was rated for the number and quality of activities presumed to enhance perceptual, cognitive, linguistic, and physi-cal development on a 4-point sphysi-cale at 6, 15, and 24 months and on a 7-point scale at 36 months. Low scores indicate that mothers made little or no attempt to stimulate or teach the child, were totally unin-volved, or provided stimulation that was very poorly matched to the child’s developmental level or inter-est. High scores indicate that mothers consistently provided age-appropriate cognitive stimulation that was likely to lead to a higher level of mastery, under-standing, or sophistication. Intercoder reliability in coding maternal stimulation during mother–child play, calculated by using intraclass correlations

(8)

(Winer, 1971) was .81 at 6 months, .69 at 15 months, .72 at 24 months, and .78 at 36 months.

Child Care

Three dimensions of nonmaternal care were in-cluded in the analyses: observed quality, type, and quantity of care.

Quality of care. Measures of child care quality were coded live by using the Observational Record of the Caregiving Environment (ORCE; NICHD Early Child Care Research Network, 1996). The ORCE was de-signed specifically for this study to assess the quality of caregiver–child interaction experienced by an indi-vidual child. The ORCE measures environmental fac-tors that developmental psychologists in this culture believe constitute quality: care that is attentive and appropriately responsive, expresses positive affect and affection, is not excessively restrictive or intru-sive, and offers activities believed to promote chil-dren’s cognitive and social development.

Observations of child care quality were made

dur-ing two half-day periods scheduled within a 2-week interval. Typically, four 44-minute cycles spread over the two half-days were completed at 6, 15, 24, and 36 months. The first three cycles consisted of 10-min ob-servation periods during which recordings of prede-termined caregiver and child behaviors were made every 30 s. The three cycles were separated by two 2-min breaks during which notes were made for qual-itative ratings of caregiver behavior (ratings were made by using a 4-point scale). The last 10-min cycle was devoted exclusively to qualitative ratings. (The ORCE method is described in detail in NICHD Early Child Care Research Network, 1996.) Two scores from the ORCE were used for this study: a composite of the positive caregiving ratings and frequency of lan-guage stimulation.

Positive caregiving rating. A positive caregiving rat-ing composite was created by summrat-ing the ratrat-ings for five scales: sensitivity to nondistress, stimulation of cognitive development, positive regard, detach-ment (reversed), and flatness of affect (reversed). At 36 months, two additional scales were included:

fos-Table 3 Means and Standard Deviations for the Cumulative Predictor and for the Outcome Variables in the Unobserved Child Care Sample 6 Months (n5 174) 15 Months (n5 174) 24 Months (n5 122) 36 Months (n5 92) M SD M SD M SD M SD Cumulative predictors Background variables

No. times center care .05 .21 .10 .36 .19 .50 .25 .69

No. times child care home .39 .49 .75 .83 .98 1.16 1.02 1.39

No. times relative/in-home care .43 .63 .63 .66 .73 .89 .96 1.18

Positive caregiving rating Language stimulation Outcomes

Bayley MDI 108.58 13.91

(9)

tering exploration and intrusiveness (reversed). The composites had good internal consistency (Cron-bach’s as 5 .89 at 6 months, .88 at 15 months, .87 at 24 months, and .83 at 36 months).

Frequency of language stimulation. Because language stimulation is central to cognitive and language de-velopment, a subset of observed caregiving behaviors was identified as constituting language stimulation: asks questions of child, responds to child’s vocaliza-tions, and other (nonnegative) talk to child. Frequen-cies of each behavior were standardized and then summed to create composite scores at 15, 24, and 36 months. The internal consistency of these composites was high (Cronbach’s as 5 .88, .92, and .90 at 15, 24, and 36 months, respectively). This score was not com-puted at 6 months. At each age, frequency of lan-guage stimulation was positively correlated with pos-itive caregiving ratings (correlations ranged from .58 to .71, ps , .001).

Each ORCE observer was trained to reach criterion by using videotapes that had been coded by experts. The criterion was 60% straight match with the expert

coder. The criterion was 80% agreement with the ex-pert for grouped codes (for details see NICHD Early Child Care Research Network, 1996). Live interob-server reliability was also calculated three to four times at approximately 3-month intervals through-out each data collection period. Intraclass correla-tions among partners ranged from .89 to .99 (Winer, 1971).

Type of care. The primary care arrangement of each child was classified at each of the four child care as-sessments as one of three types: (1) child care center, (2) child care home (care in someone else’s home by a nonrelative or relative other than the child’s grand-parents), or (3) grandparent or in-home care (care in the child’s home, including care by father). For the purpose of data analysis, two variables of type of care were created. The first consisted of the number of times a child was observed in center care (numbers could be between 0 and 4). The second consisted of the number of times a child was observed in child care homes (numbers could be between 0 and 4).

Quantity of care. This construct was defined as the

Table 4 Means and Standard Deviations for the Cumulative Predictor and for the Outcome Variables in the Exclusive Maternal Care Sample 6 Months (n5 348) 15 Months (n5 348) 24 Months (n5 260) 36 Months (n5 210) M SD M SD M SD M SD Cumulative predictors Background variables

No. times center care 0 0 0 0 0 0 0 0

No. times child care home 0 0 0 0 0 0 0 0

No. times relative/in-home care 0 0 0 0 0 0 0 0

Positive caregiving rating Language stimulation Outcomes

Bayley MDI 107.75 13.64

(10)

average number of hours per week of regular, nonma-ternal care the child received up to the age point that was the focus of each analysis. Quantity of care was calculated on the basis of information obtained every 3 months by telephone or face-to-face interviews.

Cognitive and Language Outcomes

Cognitive performance was measured at 15 and 24 months by using the Bayley Scales of Infant Develop-ment (Bayley, 1969, 1993) and at 36 months by using the School Readiness subtest of the Bracken Scale of Basic Concepts (Bracken, 1984). Language measures were the MacArthur Communicative Development In-ventory (CDI) at 15 and 24 months (Fenson et al., 1994) and the Reynell Developmental Language Scales (RDLS) at 36 months (Reynell, 1991). Tables 2–4 dis-play the means and standard deviations for children in the observed child care sample, children eligible for observation who could not be observed, and children in exclusive maternal care. Table 5 presents zero-order correlations between outcome and predictor measures for children in the observed child care sample.

At 15 months, the Bayley Mental Development In-dex (MDI) was modestly to moderately correlated with the CDI Vocabulary Comprehended, r(687) 5 .30, p, .001, and Vocabulary Expressed, r(687)5 .45, p, .001, scores. The two CDI scores were more highly correlated, r(696)5 .57, p, .001.

At 24 months, the MDI was moderately correlated with the CDI Vocabulary Comprehended, r(751) 5 .49, p, .001, and Sentence Complexity, r(751)5 .50, p , .001, scores. The two CDI scores were more highly correlated, r(759)5 .68, p , .001.

At 36 months, the Bracken school readiness score was modestly to moderately correlated with the Rey-nell Expressive Vocabulary, r(812)5 .35, p , .001, and Receptive Vocabulary, r(828) 5 .65, p , .001, scores. The two Reynell scores were moderately correlated, r(832)5 .52, p , .001.

Cognitive measures. The Bayley Mental Develop-ment Index (MDI) is the most widely used measure of cognitive developmental status for children in the first 2 years of life. The original version of the Bayley (1969) was administered at 15 months. It assesses sen-sory perceptual acuity and discriminations; memory, learning, and problem solving; early verbal commu-nication; and the ability to form generalizations and classifications. The extensively revised Bayley (BSID-II; Bayley, 1993) was administered at 24 months. The revision covers the age range of birth to 42 months, thereby reducing the kind of ceiling-level problems for 24-month-olds that might have occurred with the original scale. About 10% of the children (126 out of a

total of 1,170) were given the standardization version of the revised Bayley rather than the final form of the test. Their scores were later transformed to be compa-rable to scores from the final version. The correlation between the 15 months Bayley scores and the 24 months Revised Bayley scores is r(796) 5 .49, p , .001. The Bracken School Readiness Composite consists of 51 items grouped into five categories: knowledge of color, letter identification, number/counting, com-parisons, and shape recognition (Bracken, 1984). The score analyzed was the percentile rank. The correla-tion between the Bracken scores and the 15 months Bayley scores is r(797) 5 .27, p , .001, and between the Bracken scores and the 24 months Revised Bayley scores is r(802)5 .50, p , .001.

Language measures. Age-appropriate versions of the CDI were used (Fenson et al., 1994). The infant version of the CDI, administered when the children were 15 months old, is a vocabulary checklist. Mothers check each word that the child understands (vocabulary comprehension) and produces (vocabu-lary production). Percentile scores for each scale were calculated. (The vocabulary production scores for our sample are somewhat lower than those reported in Fenson et al. [1994]. This may be due to the fact that while the norming sample for the CDI is middle class, about a third of our sample consists of families that are either poor or near poor. On the basis of their re-search, Arriaga, Fenson, Cronan, and Pethick [1998] conclude that the CDI published norms may not be directly applicable to poverty samples. The same con-clusions can be reached on the basis of Feldman et al., [2000].) The toddler version of the CDI, administered when the children were 24 months old, includes a vo-cabulary production checklist and a section designed to assess syntactic and morphological development as well as nominal/pronominal style. Two percentile scores were generated: vocabulary production and sentence complexity. The correlation between the 15 months and the 24 months percentile scores of vocab-ulary produced was r(644)5 .56, p , .001. The corre-lation between the vocabulary produced percentile score and the 24 months sentence complexity percen-tile score was r(644)5 .42, p , .001. Cronbach’s as for the three vocabulary scales (infant comprehension, infant production, and toddler production) were .95, .96, and .96, respectively. In earlier reports, test–retest reliability was presented as .87 and .95 for comprehen-sion and production, respectively (Fenson et al., 1994). The Reynell Developmental Language Scale (RDLS) (Reynell, 1991), was administered at 36 months of age during the laboratory session. Composed of two 67-item scales, the scale assesses verbal comprehension and expressive language. For example, verbal

(11)

com-970 Child Development

prehension is assessed as a child is presented with a set of objects and asked, “Where’s the spoon?” Ex-pressive language is examined as an experimenter picks up an object and says, “What is this?” The cor-relation between the 15 months CDI and the Reynell verbal comprehension scores is r(692)5 .12, p , .001. The correlation between the 15 months CDI vocabu-lary-produced percentile score and the Reynell ex-pressive language percent score is r(680) 5 .14, p , .001. The correlation between the 24 months CDI vo-cabulary produced percentile score and the Reynell expressive language percent score is r(740)5 .31, p , .001. Alphas were .93 for verbal comprehension and .86 for expressive language.

RESULTS

Hierarchical Regressions

Do the cumulative quality, type, and amount of child care predict children’s cognitive and language skills during the first 3 years of life? The first analyses were designed to answer this question. Cumulative quality was defined as the average quality rating up to and including the age being considered for the analysis. Cumulative type ratings were calculated for center care and for care in child care homes. Each child received a score representing the number of as-sessment periods up to and including the age under consideration in which the type of care observed was in a child care center and a second score representing the number of assessment periods in which care was

observed in a child care home. Cumulative quantity of care was the average number of hours of care per week that the child had received up to and including the age being analyzed.

Hierarchical regressions were used to examine the association between child care experience, cogni-tive development, and language development. Two models were fit. Both models included controls for site, maternal PPVT, gender, HOME total, and ma-ternal stimulation as background variables. In Model 1, quality (positive caregiving rating), type (proportion of times observed in center and child care homes), and amount (average hours of care per week) were the child care variables tested. If the variance accounted for by the child care block of variables was significant, we examined the coeffi-cients associated with each child care variable. If the global measure of child care quality was significant in these analyses, Model 2 was tested. It added fre-quency of language stimulation to determine whether this aspect of the child care environment mediated the association between global positive caregiving and a child outcome. This measure was selected because it is correlated with the positive caregiving rating and it represents a moment-to-moment index of caregivers’ talk to the child. Evi-dence of mediation was obtained when the positive caregiving rating was significantly related to out-comes in Model 1, but markedly less related in Model 2 (Baron & Kenny, 1986).

The results of these analyses are displayed in Table 6. The child care block accounted for a small but

signif-Table 5 Simple Correlations Between Cumulative Predictors and Outcomes for the Observed Child Care Sample at 15, 24, and 36 Months

15 Months 24 Months 36 Months

Bayley MDI CDI Vocab. Comp. CDI Vocab. Prod. Revised Bayley MDI CDI Sentence Complex. CDI Vocab. Prod. Bracken School Readiness Reynell Verbal Comp. Reynell Express. Lang. Background variables Maternal PPVT-R .19*** .02 .06 .34*** .20*** .14*** .39*** .27*** .47*** Income-to-needs ratio .13*** .04 .07 .33*** .16*** .13*** .40*** .23*** .40*** Gender (male) 2.15*** .06 .12** 2.19*** 2.23*** 2.02 2.18*** 2.14*** 2.19*** HOME total .19*** .12** .16** .34*** .22*** .22*** .44*** .33*** .49*** Maternal stimulation .13*** .12** .10** .33*** .17*** .18*** .37*** .28*** .44*** Child care Average hours/week .00 2.07 .00 .06 2.02 2.03 .04* .06 .04

No. times center care .08 2.01 .07 .14*** .03 .04 .13*** .07* .14***

No. times child care home .01 2.13** 2.07 .02 .00 2.03 2.02 .03 .03

No. times relative/in-home care 2.09** .08 .00 2.07 2.03 2.02 .01 2.02 2.05 Positive caregiving rating .12** .17*** .13** .25*** .15*** .13** .23*** .16*** .25*** Language stimulation .15*** .20*** .19*** .23*** .18*** .20*** .17*** .08* .11** * p, .05; ** p, .01; *** p, .001.

(12)

icant amount of variance in all outcomes at each age (1.3% to 3.6%). When the variance accounted for by child care was calculated as a proportion of the total variance explained by all predictors, it ranged from .042 to .735.

Quality of Care

Both indices of the quality of child care were posi-tively related to most of the language and cognitive outcomes. In Model 1, cumulative positive caregiving was positively related to all child outcome measures except the 15-month Bayley MDI and the 24-month CDI vocabulary production scores. Children who ex-perienced more positive caregiving scored higher on both language measures at 15 months, the Bayley MDI and CDI sentence complexity at 24 months, and school readiness and both language measures at 36 months.

In Model 2, cumulative observed language stimu-lation was a significant predictor for all measures at ages 15 and 24 months and for the Reynell expressive language score at 36 months. A comparison of Models 1 and 2 shows that in all but one analysis, the relation of the positive caregiving rating to outcomes was nonsignificant once language stimulation was included. The comparison suggests that language stimulation accounted for the relations of the more global indicator of child care quality (positive care-giving) to cognitive and language scores, at least at 15 and 24 months, and to the expressive language score at 36 months.

Type of Care

The type of child care was related to children’s competence, especially at 24 and 36 months (see Table 6). The more times children were observed in child care centers (the maximum possible was four times, when the children were 6, 15, 24, or 36 months of age), the better they performed on all cognitive and lan-guage assessments taken at 24 and 36 months and the higher were their CDI expressive vocabulary scores at 15 months (Table 6, Model 2). The more times chil-dren were observed in child care homes, the higher were their 24-month Bayley scores and 36-month ver-bal comprehension scores.

Zero-order correlations among child care predictors indicated that hours in care and type of care (both num-ber of assessment periods in center care, and numnum-ber of assessment periods in child care homes) were nega-tively related to the two measures of child care qual-ity. More hours in care was associated with lower ob-served positive caregiving ratings, rs ranging from

2.11 to 2.19, ps , .01, and less frequent language stimulation, rs ranging from 2.08, ns to 2.09, p , .05. The more times the child was observed in center care, the lower the average positive caregiving rating, rs ranging from 2.23 to 2.30, ps , .01, and the lower the average frequency of language stimulation, rs rang-ing from 2.30 to 2.34, ps , .01. The number of obser-vations in a child care home was not significantly re-lated to average positive caregiving ratings, rs ranging from 2.03 to 2.10, ns, but was associated with lower average frequency of language stimulation, rs rang-ing from 2.10 to 2.13, ps , .01.

Amount of Care

The cumulative number of hours in child care did not contribute to the prediction of children’s cogni-tive or language development in any analysis. As shown in Table 6, there were no significant relations between average hours per week of child care and chil-dren’s cognitive and language performance. The same results were found in a similar analysis on the whole sample, including children who had received little or no child care. Average hours/week in care was not significantly related to any child outcome.

In summary, quality of care is positively associated with children’s acquisition of language skills, cogni-tive skills, and school readiness skills from ages 15– 36 months. Caregivers’ language stimulation appears to mediate the relations of overall positive caregiving to these child outcomes. Type of care also predicts child outcomes. With quality of care statistically controlled, experience in child care centers and, to a lesser extent, child care homes is positively associated with lan-guage and intellectual skills. By contrast, there is no evidence that cumulative amount of care is related to cognitive and language development.

Magnitude of Differences Associated with Child Care

The second and third questions posed in this re-search dealt with: the magnitude of the effects associ-ated with child care and how children raised almost ex-clusively by their mothers compare with children who experienced different levels of quality of care. Answers to these questions were derived from two analyses. Both were analyses of covariance (ANCOVA) that compared the cognitive and language performance of children for groups of children. The first analysis compared children who were observed in different levels of quality of care with one another and with children who received exclusive maternal care during the first 3 years of life. This ANCOVA included the

(13)

972Table 6Child DevelopmentStandardized Regression Coefficients and Adjusted R2_{s for Cumulative Child Care Variables as Predictors of Cognitive and}

Language Scores at 15, 24, and 36 Months for Children Observed in Child Care

Bayley— 15 months MDI CDI—15 months Vocabulary Production CDI—15 months Vocabulary Comprehension

Model 1 Model 2 Model 1 Model 2 Model 1 Model 2

Background variables Maternal PPVT-R .16* .17* .01 .02 2.04 2.03 Income-to-needs ratio .06 .05 2.02 2.03 2.04 2.04 Gender (male) 2.13* 2.13* .12* .12* .07 .07 HOME total .10* .09 .17* .14* .09 .07 Maternal stimulation .04 .03 .03 .02 .09 .08

Adjusted R2_{for background variables} _.129* _.129* _.037* _.036* _.013* _.013*

Child care

Average hours/week 2.01 2.03 .03 .00 2.01 2.03

No. times center care .04 .07 .06 .11* .03 .06

No. times child care home .06 .08 2.02 .02 2.10 2.07

Positive caregiving rating .08 2.02 .12* 2.01 .15* .05

Language stimulation .17* .25* .18*

Adjusted R2_{child care} _.000 _.013* _.005 _.032* _.024* _.036*

Revised Bayley— 24 months MDI CDI—24 months Vocabulary Production CDI—24 months Sentence Complexity

Background variables Maternal PPVT-R .17* .17* .04 .04 .11* .10* Income-to-needs ratio .11* .10* .03 .02 .05 .04 Gender (male) 2.15* 2.15* .04 .04 2.21* 2.21* HOME total .12* .11* .15* .14* .08 .07 Maternal stimulation .15* .15* .08 .06 .04 .03

Adjusted R2_{background variables} _.268* _.269* _.053* _.053* _.109* _.110*

Child care

Average hours/week 2.00 2.02 2.04 2.06 2.04 2.05

No. times center care .17* .20* .06 .10* .07 .10*

No. times child care home .08* .10* .01 .04 .03 .06

Positive caregiving rating .16* .08 .07 2.07 .12* .02

Language stimulation .14* .23* .16*

Adjusted R2_{child care} _.025* _.032* _.001 _.023* _.007* _.016*

Bracken—36 months School Readiness Reynell—36 months Expressive Language Reynell—36 months Verbal Comprehension

Background variables Maternal PPVT-R .16* .15* .10* .10* .24 .24 Income-to-needs ratio .14* .14* 2.01 2.01 .12* .12* Gender (male) 2.16* 2.16* 2.13 2.12 2.16* 2.16* HOME total .22* .21* .22* .21* .16* .16* Maternal stimulation .10* .09* .09* .09* .16* .16*

Adjusted R2_{background variables} _.316* _.316* _.164* _.164* _.387* _.388*

Child care

Average hours/week 2.02 2.03 .03 .02 2.02 2.03

No. times center care .12* .14* .09* .12* .14* .16*

No. times child care home .04 .05 .05 .07 .08* .09*

Positive caregiving rating .12* .07 .10* .03 .15* .12*

Language stimulation .08 .12* .06

Adjusted R2_{child care} _.012* _.014* _.008* _.014* _.021* _.021*

Note: Regression coefficients are from the full model. The adjusted R2_{s are reported from the hierarchical model as the contribution of that}

block as it enters the model. * p, .05.

(14)

same control variables and the measures of child care types and amount that were used in the cumulative regressions. The child care quality measure was trans-formed for these analyses. Four groups differing in levels of quality of care were created by performing a quartile split on whichever cumulative quality mea-sure was the best predictor of child outcomes at a given age. This was frequency of language stimula-tion at 15 and 24 months and positive caregiving at 36 months. A fifth group, consisting of children who re-ceived less than 10 hr/week of child care from birth through the assessment age of the dependent vari-able, was added. Effect sizes were computed as pair-wise comparisons of the means from the five-level “group” factor divided by the pooled standard devi-ation as estimated by the root mean squared error un-der that analysis model. The estimated means for each group appear in Table 7. The second analysis, also an ANCOVA, compared the cognitive and lan-guage performance of children who were observed in different levels of quality of care with one another and with children who received exclusive maternal care during the first 3 years of life. The second ANCOVA compared children who were and were not observed in center care and included the same control variables and measures of child care quality and amount that were used in the cumulative regressions.

Child Care Quality and Type: Effect Sizes

The largest discrepancy in quality was the top (High) versus the bottom (Low) quartile. Two effect

sizes were greater than .40. The effect size for the 36-month Reynell verbal comprehension battery was .48 and for the 24-month MDI it was .43. For the 36-month Reynell expressive language battery, four were between .30 and .39, two were between .20 and .29, and the lowest was .18. All but one fell between “small” (.20) and “medium” (.50) according to Cohen’s (1988) guidelines.

Comparisons of Quartile 1 (Low) with 3 (Medium High) and of 2 (Medium Low) with 4 (High) each rep-resent groups separated by two steps. Effect sizes were greater than .30 for three comparisons, between .20 and .29 for nine comparisons, and less than .20 for the remaining six comparisons. Most of the compari-sons of adjacent quartiles yielded effect sizes less than .20.

Comparisons of children observed in center care with those not observed in center care show only four effect sizes ranging between Cohen’s “small” and “medium” levels. The effect size was .43 for the 24 months revised Bayley, .28 for the 36 months Bracken, .21 for the Reynell Expressive Language, and .38 for the Reynell Verbal Comprehension measure.

Exclusive Maternal Care

In these analyses of covariance, children observed in the different levels of child care quality were also compared with children who were in the exclusive care of their mothers for their first 3 years. (Children were considered to be in exclusive maternal care if they had received 10 or fewer hr per week of

nonma-Table 7 Adjusted Means for Children in Exclusive Maternal Care and Children Observed in Four Levels of Quality of Care

15 Months 24 Months 36 Months

Bayley MDI CDI Vocab. Prod. CDI Vocab. Comp. Revised Bayley MDI CDI Vocab. Prod. CDI Sentence Complex. Bracken School Readiness Reynell Express. Lang. Reynell Verbal Comp. N 891 771 771 925 856 856 985 983 1008 R2 Total .16*** .08*** .08*** .32*** .10*** .15*** .35*** .21*** .41*** Group .005 .006 .011 .012** .018** .015** .011** .003 .012*** Adjusted means Low quality 107.3 35.4 35.6 90.5a _41.4a _37.8a _38.6a _42.3 _41.7a

Low/average quality 107.9 36.1 40.0 91.5a,b _43.5a,b _39.3a,b _39.5a,b _45.2 _46.9b

High/average quality 109.8 39.9 45.2 93.6b,c _48.3b,c _43.9b,c _43.4b,c _46.1 _47.1b

High quality 110.0 42.4 45.0 95.6c _50.1c _45.6c _47.2c _46.7 _52.7c

Exclusive maternal care 110.7 38.5 41.2 93.5a,b,c _37.2a _36.0a _43.6a,b,c _45.6 _50.9b,c

Range of cell sizes 140 ,n, 245 155 ,n, 200 135 ,n,225

Note: Means adjusted for maternal PPVT-R, income-to-needs ratio, child gender, HOME total, and maternal stimulation. Within a column of adjusted means, values with the same superscript are not significantly different from each other.

(15)

ternal care from birth to the target assessment age). The overall difference among the five groups was tested first. Pairwise comparisons were tested only when the overall effect of group was significant. The results of these analyses with adjusted means for each group are shown in Table 7.

For the most part, children in full-time maternal care had scores similar to those of children in child care (see Table 7). The only exceptions were 24-month lan-guage scores (CDI vocabulary production and sen-tence complexity) and 36-month verbal comprehen-sion. Children in exclusive maternal care performed less well than did children in medium- and high-quality care on the former, and they performed better than children in low-quality care on the latter.

Developmental Patterns

The fourth question addressed in our research con-cerned whether child care in the first year or two of life would have lasting associations with cognitive and language development at subsequent ages. All results reported to this point pertain to cumulative experience on the basis of the assumption that child-rearing experiences and their relation to developmen-tal outcomes are incremendevelopmen-tal and additive. However, some theoretical positions (cf., Lamb, 1997) predict that experiences very early in life may have a particu-larly strong relation to subsequent cognitive and lan-guage development. Because there is some stability in children’s care situations, another possibility is that apparently cumulative relations are, in fact, primarily a function of current features of child care. Therefore, a set of analyses was performed to determine whether earlier child care experiences predicted child out-comes independently of concurrent care. The child care indices were divided into measures collected at the same age as the outcome measure (concurrent measures) and measures collected before that assess-ment point (lagged measures). Hierarchical regres-sions were computed to test the relations of concurrent child care characteristics to child outcomes (Model 3). In Model 4, lagged child care measures were entered to determine whether they made additional contribu-tions to explained variance that was independent of children’s current child care experiences. The results are shown in Table 8.

Quality

At 24 months, language stimulation in the concur-rent child care setting was positively related to all three outcome variables—Bayley MDI, vocabulary

production, and sentence complexity. The amount of language stimulation received at 15 months predicted additional variance in children’s scores on the two language measures.

At 36 months, concurrent positive caregiving pre-dicted the Reynell verbal comprehension score but not expressive language or school readiness. The amount of language stimulation in earlier child care experiences, however, was positively related to both the Reynell expressive language and verbal compre-hension scores.

Type of Care

At age 24 months, children who were currently cared for in child care centers and child care homes performed better on all outcomes than did children in other forms of care. By 36 months, children who were currently in child care centers performed better than children in other types of care, but being in a child care home at age 3 was not related to performance. Children who had been in child care homes during the first 2 years of their lives, however, performed bet-ter at age 3 than did children whose earlier experience had been in other types of care.

Variations in Relations of Child Care to Outcomes for Children from Different Environments

Finally, the fifth question addressed in our research was whether the relations of child care to cognitive and language outcomes are different for children from different income levels, ethnic groups, home environ-ments, or genders. Previous literature has suggested that children from families with low incomes or low levels of stimulation in the home are especially likely to benefit from high-quality care and that children from advantaged homes are potentially vulnerable to negative effects of low-quality care. Gender and eth-nic differences have also been found.

To address this question, the hierarchical regression analyses shown in Table 5 were modified slightly, and interaction terms were added. Family income and HOME were each categorized into three groups. Eth-nicity was entered as a dichotomous variable: White/ non-Hispanic or not White/non-Hispanic. Gender was dichotomous.

Interactions of these four variables with each char-acteristic of child care were tested. For example, the block of four interactions entered in the quality anal-yses was Income 3 Quality, HOME 3 Quality, Gen-der 3 Quality, Ethnicity 3 Quality. Similar sets of four interactions were examined to create the blocks for type and amount. The result was three analyses

(16)

(quality, type, and amount of care) for each of nine outcomes (three cognitive and language measures at each of three ages). The interaction block was signifi-cant in only three of the 27 tests, thereby yielding no apparent pattern. These analyses did not suggest that the relations of child care quality, type, or amount to cognitive and language outcomes differed by family income, home environment, gender, or ethnic group.

DISCUSSION

Information obtained as part of the NICHD Study of Early Child Care was used to explore the association between child care experiences and children’s

cogni-tive and language development. Results indicate that quality of child care was a reasonably consistent pre-dictor of children’s cognitive and language perfor-mance. One measure of quality consisting of ratings of responsive, sensitive caregiving was related to cog-nitive and language outcomes throughout the first 3 years of life. The other more specific measure of fre-quency of language stimulation appears to be a par-ticularly important component of such caregiving, es-pecially during the first 2 years of life. This finding is consistent with theoretical predictions as well as with earlier cross-sectional studies pinpointing the impor-tance of linguistic stimulation in child care for children’s language development (McCartney, 1984; Melhuish,

Table 8 Relations of Concurrent and Lagged Child Care Variables to Cognitive and Language Scores at 24 and 36 Months

Revised Bayley— 24 months MDI CDI—24 months Vocabulary Production CDI—24 months Verbal Comprehension Model 3 Model 4 Model 3 Model 4 Model 3 Model 4

Background variables R2 _.25*** _.25*** _.05*** _.05*** _.10*** _.10***

Concurrent child care DR2 _*** _*** _*** _*** _*** _*

Average hours/week b5 2.02 2.01 2.07 2.07 2.05 2.02

No. times in center .29 .22 .22 .23 .17 .13

No. times in home .15 .13 .15 .24 .15 .20

Positive caregiving .04 .05 2.09 2.10 2.03 2.04

Language stimulation .24 .20 .36 .29 .26 .18

Lagged child care DR2 _ns _* _*

Average hours/week b5 .01 .06

No. times in child care center .05 .09

No. times in child care home 2.12 2.08

Positive caregiving .02 .05 Language stimulation .16 .15 Total R2 _.31*** _.31*** _.11*** _.13*** _.14*** _.15*** Bracken—36 months School Readiness Reynell—36 months Expressive Language Reynell—36 months Verbal Comprehension Model 3 Model 4 Model 3 Model 4 Model 3 Model 4

Background variables R2 _.29*** _.29*** _.16*** _.16*** _.36*** _.36***

Concurrent child care DR2 _** _** _* _N_s _*** _***

Average hours/week b5 .00 .01 2.07 2.07 .01 .01

No. times in child care center .16 .15 .13 .10 .15 .13

No. times in child care home .03 2.03 .07 .02 .02 2.08

Positive caregiving .07 .05 .05 .05 .09 .09

Language stimulation .08 .08 2.02 2.09 2.03 2.11

Lagged child care DR2 _ns _** _***

Average hours/week b5 2.02 2.03

No. times in center .10 .08

No. times in child care home .12 .19

Positive caregiving .03 .02

Language stimulation .18 .21

Total R2 _.31*** _.31*** _.16*** _.18*** _.38*** _.41***

Note: Significant coefficients are bolded and underlined. * p, .05; ** p, .01; *** p, .001.