Dearing, E., McCartney, K., & Taylor, B. A. (2009). Does higher quality early child care promote low-income children’s math and reading achievement in middle childhood? Child Development, 80(5), 1329–1349.
Doing well in school during the late-elementary grades is a strong predictor of economic independence during adulthood for children growing up in poverty (Feinstein & Bynner, 2004). Unfortunately, by the end of fifth grade, children from low-income homes and neighborhoods are twice as likely as students from more advantaged homes and neighborhoods to perform below “proficient” levels on measures of mathematics and reading (National Center for Education Statistics, 2007; Roberts & Bryant, 2011; Roberts, Mohammed, & Vaughn, 2010), creating a self-perpetuating cycle of poverty and low achievement for families and communities across multiple generations. The quality of early nonmaternal child care (e.g., day care, preschool) during the preschool years influences later achievement in the general population of preschool children. Experimental evidence also supports similar effects of quality, early, nonmaternal care for students from low-income homes and neighborhoods. However, the evidence-based practice models these studies used tend to be highly resource intensive, and no evidence supports their scalability for students from low-income homes and neighborhoods.
Dearing, McCartney, and Taylor (2009) used a large existing data set to address this gap in the literature, finding that lower income was less strongly predictive of underachievement in reading and in mathematics for children aged 53 months to 11 years who had been in higher-quality early child care compared to children who had not had this opportunity. Further, the authors found that the achievement of children from low-income homes and neighborhoods was associated with family characteristics that also predicted a child’s participation in higher-quality early care. This finding suggests that children from families who sought out quality early care may have outperformed other poor children, even in the absence of such care, for reasons that existed before the selection and onset of early care. This possibility is difficult to rule out, given the study’s use of extant, nonrandomized data.
The researchers used longitudinal data from the Study of Early Child Care and Youth Development, a large-scale survey project funded by the National Institute for Child Health and Human Development that covers birth through fifth grade, including data on (a) child care quality during infancy and early childhood, (b) family economics across the first 10 years of children’s lives, (c) early school-readiness skills, and (d) mathematics and reading achievement during middle childhood. The authors included a variety of child, parent, and family variables to estimate and attempt to control the selection bias associated with nonrandomized data (see recommendation 1 below). Bias in this study centered on the predictive influence of family characteristics on higher-quality early care and, independent of quality care, on the child’s subsequent achievement. In other words, although it is tempting to conclude that quality care “caused” improved achievement, it is possible, maybe even probable, that certain family characteristics are in fact “causing” both the quality of early care and later academic achievement.
The quality of early child care was measured on five occasions. It was found that each occasion rated as high quality was associated with slightly higher mathematics achievement in middle childhood (5% of a standard deviation in overall mathematics; 7% of a standard deviation in applied mathematics problems) and word reading (6% of a standard deviation). So, in the area of general mathematics, children found to be in high-quality child care for two occasions scored 10% of a standard deviation higher in middle childhood, and those who participated in three occasions of high-quality child care scored 15% of a standard deviation higher in middle childhood than children who did not participate in high-quality child care. For children who were never in high-quality child care, similar increases in later achievement would require about a 50% increase in income (from 200% of the poverty line to 300% of the poverty line).
Dearing et al. (2009) also report that higher-quality child care’s effect on later achievement for lower-income children was primarily through improved school readiness. In other words, results suggest that higher-quality care provided younger children with better school-readiness skills and that better school-readiness skills led to better achievement than might otherwise have been possible.
Finally, higher-quality child care significantly moderated the relationship between risk factors other than poverty (contextual risk factors) and later achievement in reading and mathematics—the association between risk and achievement varied by occasions for families in lower-quality child care. In higher-quality child care conditions, quality of care was associated with even greater “protection” from the other contextual risk factors for academic failure than in lower-quality child care conditions.
To interpret the findings of studies such as this one, it is important to understand differences between different types of statistical analyses and the different ways that variables can be modeled in these analyses.
Recommendation 1: Consider the differences between randomized controlled trials (the “gold standard” for using statistics to determine causality between variables) and propensity score modeling, which this study used.
When trying to determine the possible “causes” of certain outcomes, researchers design studies to reduce the possibility that their findings might be explained by variables they did not measure. Randomized studies assign participants to treatment conditions randomly, which maximizes the likelihood of beginning a study with comparable groups (and minimizes the chances that groups are different based on variables that are not being measured). This randomization means that differences in groups at the study’s conclusion can reasonably be assumed to be the result of differences in the variables of interest (in this study, quality of child care).
A randomized design in this case would have required assigning families to either a higher-quality child care group or a lower-quality child care group before following the children through their 11th year. There are several problems with this approach. The most important problem is the ethical concern involved in restricting a family’s access to higher-quality child care.
In this study, the analysts instead used propensity score models to approximate the benefits of randomized designs. Propensity score approaches avoid the ethical challenge of restricting access to higher-quality services. These models use patterns across a large number of child and family variables to estimate each family’s likelihood of being assigned to one or the other condition (higher-quality care and lower-quality care, in this case) and to statistically adjust for group differences at the beginning of the study accordingly.
However, propensity score models only measure, and thus control, observed variables, and randomized designs control for both measured and unmeasured differences. To the extent that unobserved or unobservable variables are correlated with a study’s outcome, propensity score models are limited in addressing selection bias, a particular concern given that the relationship between an unobservable variable and a given outcome cannot be known. In sum, propensity score models are useful in situations where randomization is not possible or advisable. However, as the researchers acknowledged, claims about causality in propensity score models should be examined carefully.
Recommendation 2: Consider the definitions of statistical mediation and moderation and how they can be used to investigate causal relationships between variables.
Mediation: This study represents an example of statistical mediation, with school readiness mediating the effect of higher-quality child care on later achievement. Statistical mediation is an important tool for “unpacking” the relationship between two variables, particularly when one of the variables is assignment to a treatment (or likelihood of assignment, in the case of propensity score models). In cases where an intervention appears to cause changes in an outcome (higher-quality care and later achievement, respectively, in this case), statistical mediation addresses questions about “how” the intervention may have affected the outcome and often provides insight on ways in which an intervention can be improved or enhanced. In this case, the data support the hypothesized mediation (that higher quality of child care is related to higher academic achievement through better school-readiness skills).
Moderation: Statistical moderation represents the interaction between two variables. In practical terms, the presence of moderation means that the independent variable has a different-sized effect at different levels of the moderating variable, so statistical moderation can inform questions about the groups for which or condition under which an intervention may be particularly effective. For example, knowing whether a variable like “gender” or “household income” is a moderating variable helps researchers understand whether their intervention is more or less effective for boys vs. girls or for students from higher- vs. lower-income households. In this study, risk factors and achievement were differently related (the association between them was less strong) in the lower-quality child care group than in the higher-quality child care group.
Feinstein, L., & Bynner, J. (2004). The importance of cognitive development in middle childhood for adult socioeconomic status, mental health, and problem behavior. Child Development, 75, 1329–1339.
National Center for Education Statistics. (2007). The condition of education 2007 (NCES 2007-064). Washington, DC: U.S. Government Printing Office.
Roberts, G., & Bryant, D. P. (2011). Early mathematics achievement trajectories: English-language learner and native English-speaker estimates using the Early Childhood Longitudinal Survey. Developmental Psychology, 47(4), 916–930.
Roberts, G., Mohammed, S., & Vaughn, S. (2010). Reading achievement across three language groups: Growth estimates for overall reading and reading sub-skills using the Early Childhood Longitudinal Survey. Journal of Educational Psychology, 102(3), 668–686.