Key points are not available for this paper at this time.
Abstract Research Findings: As policymakers expand access to preschool, the sociodemographic composition of preschool classrooms will become increasingly important. These efforts may create programs that increase the concentration of children from low-income families or, alternatively, foster the creation of socioeconomically diverse preschool classrooms. What effect the creation of such contexts would have on very young children remains unclear. Using multilevel methods and data on 2,966 children in 704 prekindergarten classrooms, this study explores the relationship between socioeconomic classroom composition and children's social and cognitive development. The results indicate positive associations between the mean socioeconomic status (SES) of the class and children's receptive language, expressive language, and mathematics learning, regardless of children's own sociodemographic backgrounds and the characteristics of their classrooms. However, the analyses indicate no association between the development of social competence and class mean SES. Practice or Policy: The links between classroom SES and language and mathematics development were comparable in size to those associated with instructional quality and even children's own SES. Neither structural nor instructional characteristics of prekindergarten classrooms explained these relationships, suggesting the possibility of direct peer effects. The findings indicate that the composition of children's classrooms should be considered an important aspect of preschool quality. ACKNOWLEDGMENTS The Multi-State Study of Pre-Kindergarten and State-Wide Early Education Program Study data were collected by the National Center for Early Development and Learning (NCEDL) using funds from the U.S. Department of Education, National Institute for Early Education Research, Pew Charitable Trusts, and Foundation for Child Development. Although we are very grateful to the scholars at NCEDL for generously sharing their data, the contents of this study do not necessarily represent the positions or policies of NCEDL or the funding agencies, and endorsement by these agencies should not be assumed. Notes 1The final model was as follows: where Y ij is the spring assessment score of child i in preschool j, β0j is the expected outcome (spring assessment score) for child i in preschool j when values of the Level 1 covariates are equal to the grand mean, β1j through β11j are Level 1 coefficients for child i in preschool j, r ij is the Level 1 error term for child i in preschool j (random effect), γ00 is the expected outcome (spring assessment score) for child i in preschool j when values of the Level 2 covariates are equal to zero, γ01 through γ09 are Level 2 coefficients for child i in preschool j, and u 0j is the Level 2 error term for child i in preschool j (random effect). 2The 60.28% response rate may have implications for these analyses given the potential biases it introduces into the classroom-level measures. Although the data do not include information on the backgrounds of nonresponders, it seems reasonable to conclude that lower SES families were less likely to respond to the survey. This arguably suggests that variability in both the classroom SES and classroom income measures is artificially reduced. As a result, the associated HLM estimates would be biased toward zero, which means that the results presented here may reflect conservative estimates. 3For missing data to be missing at random, their missingness must be explained by variables in the data set. In this instance, we conducted a logistic regression using a measure of whether children took the test in Spanish (1 = yes, 0 = no) as the dependent variable, with race/ethnicity, poverty status, mother's level of education, and whether the child was an ELL as independent variables. Together, these covariates correctly predicted whether a child took the test in Spanish in 92.9% of cases, indicating that the missing data satisfied the requirements for being missing at random (Graham, Citation2009). Note. SES = socioeconomic status; ELL = English language learner; IEP = individualized education plan; ECERS = Early Childhood Environment Rating Scale; BA = bachelor's degree; CDA = Child Development Associate credential. a Classroom SES is the average of two z-scored variables: class mean family income and class mean mothers' education. Low-SES classrooms have SES values that are less than 0.5 SD below the mean for all classrooms; high-SES classrooms have SES values that are more than 0.5 SD above the mean for all classrooms; middle-SES classrooms have SES values between the two. b Significance tests compare low- and high-SES classrooms to middle-SES classrooms. c Teacher has had a teaching certificate for less than 4 years. *p < .05. **p < .01. ***p < .001. 4It would also be worth exploring how the distribution of children by age might differ among classrooms and how any differences might relate to their learning. Unfortunately, we do not have class-level data regarding the distribution of children by age. We do know that the average age of children in the Multi-State/SWEEP studies, and its standard deviation, did not vary significantly by whether children were in low-SES, middle-SES, or high-SES classrooms. Note. SES = socioeconomic status; ELL = English language learner; IEP = individualized education plan. a All variables are centered on the grand mean. b SES is the average of two z-scored variables: child's family income and mother's education. c Coefficients are empirical Bayes estimates. d Comparison group is White. †p < .10. *p < .05. **p < .01. ***p < .001. Note. SES = socioeconomic status; ELL = English language learner; IEP = individualized education plan. a Outcome is the z-scored spring assessment score. b At the child level, all variables are centered on the grand mean. c SES is the average of two z-scored variables: child's family income and mother's education. d Coefficients are empirical Bayes estimates adjusted for all child- and class-level measures. e Comparison group is White. f At the class level, all continuous variables are centered on the grand mean, and all dummy variables are left uncentered. g Socioeconomic composition (i.e., classroom SES) is the average of two z-scored variables: class mean family income and class mean mothers' education. †p < .10. *p < .05. **p < .01. ***p < .001. Note. SES = socioeconomic status; ELL = English language learner; IEP = individualized education plan; BA = bachelor's degree. a Outcome is the z-scored spring assessment score. b At the child level, all variables are centered on the grand mean. c SES is the average of two z-scored variables: child's family income and mother's education. d Coefficients are empirical Bayes estimates adjusted for all child- and class-level measures. e Comparison group is White. f At the class level, all continuous variables are centered on the grand mean, and all dummy variables are left uncentered. g Socioeconomic composition (i.e., classroom SES) is the average of two z-scored variables: class mean family income and class mean mothers' education. h Comparison group is no BA. †p < .10. *p < .05. **p < .01. ***p < .001. 5When we compared the residual class-level variance in the fully unconditional model and final model for each of the four outcomes, we found that the models explained nearly all the variability between classrooms in spring receptive language scores (97.97%) and expressive language scores (92.21%) and slightly less in spring math scores (83.60%). The model was far less successful at explaining variance between classrooms in the spring assessment of social competence (48.17%), indicating that unmeasured variables are likely to relate significantly to this aspect of children's development.
Reid et al. (Wed,) studied this question.