Family/Child Selection Biases. The possibility that families differ in their child care choices is a topic of interest in its own right (NICHD Early Child Care Research Network, 1997; Singer, Fuller, Keiley, and Wolf, 1998). It also is a critical issue for investigators who are interested in ascertaining the effects of child care on children (Howes and Olenick, 1986; Vandell, 1997). The problem is that ostensible “effects” of child care quality may be artifacts of family characteristics that are confounded with child care quality. As a result of this concern, it has become standard practice for researchers to incorporate family selection factors into their analyses. As is evident in Tables 2 and 3, almost all studies conducted in recent years have included controls for family characteristics.
As an example of this strategy, the NICHD Study of Early Child Care has utilized three criteria for identifying family variables that are then used as selection controls in analyses: (1) the family characteristic is significantly related to child care, (2) the family characteristic is related to the child outcome of interest, and (3) the family characteristic is not highly related to other family factors. The third criterion is applied to reduce collinearity among family characteristics.
At one level, concern about family selection bias is clearly merited. There is evidence, for example, that type and quality of child care are related to parents’ education and income (see Figure 1).
Parents who have higher incomes and more education are more likely to place their children in centers that have higher ECERS scores, lower child:adult ratios, and better-trained teachers (Blau, 1999c; Peisner-Feinberg and Burchinal, 1997). Children with more sensitive mothers are more likely to be placed in care arrangements that offer more positive caregiving experiences (NICHD Early Child Care Research Network, 1997). Children whose home environments are more cognitively stimulating and more emotionally supportive are more likely to be placed in child care settings that are stimulating and supportive (NICHD Early Child Care Research Network, in press-b). These family factors, if not controlled, may masquerade as child care effects.
At another level, however, selection effects do not appear to be as large as initially thought. In the Cost, Quality, and Outcomes Study, for example, the correlation between maternal education and the ECERS was .24; the correlation between family income and the ECERS was .09. In the NICHD Study of Early Child Care, correlations between maternal education and ORCE positive caregiving ratings were .11 at 6 months, .14 at 15 months, .22 at 24 months, and .19 at 36 months. Correlations between family income and ORCE positive caregiving were typically lower than these figures. These relatively modest associations between child care quality and family factors suggest that selection effects are not substantial, at least within the range of studies that have been conducted. In the future, selection effects may be greater as welfare reform is fully implemented and the numbers of children in child care increase.
Variability in Child Care Quality. The ability to detect child care quality effects also is dependent on obtaining sufficient variability in quality scores. Obviously, if there is no variation in quality, it is not possible to detect variations associated with quality. If quality is sampled within a truncated range, effects associated with quality are reduced and larger samples are needed to detect differences. One reason that the Swedish studies have not detected quality effects may be the restricted range of the quality scores that were sampled, coupled with relatively small sample sizes (Broberg, Wessels, Lamb, and Hwang, 1997; Lamb, Hwang, et al., 1988). These same issues are pertinent to child care research in the United States, when restricted ranges of quality are sampled and sample sizes are small.
Control for Prior Child Adjustment. A third challenge is determining when and how to control appropriately for prior child adjustment in examinations of child care effects. Some researchers have argued that stronger tests of child care quality require controls for prior child adjustment. Such controls could be used successfully in studies of after-school programs that controlled for children’s adjustment prior to entry into the programs (Vandell and Posner, 1999). Controls for prior child adjustment in studies of early child care quality are more difficult. Children typically begin child care during their first year of life, prior to the time that robust and reliable measures of child cognitive, language, and social adjustment can be administered. Using measures of child adjustment collected at some later period, after substantial child care experience has accrued, does not make sense because these measures may well be a reflection of the effects of quality to that point. By controlling for child adjustment scores that were already affected by quality, we may be eliminating (or at least minimizing) the very quality effects that are of interest. This potential confounding of child care quality and child adjustment scores means that fixed-effects models that control for prior (or concurrent) child adjustment must be applied with caution.
"report.pdf" (pdf, 132.7Kb)
"table1.pdf" (pdf, 43.75Kb)
"table2.pdf" (pdf, 43.32Kb)