A study of the efficacy of all -day kindergarten: Parametric versus nonparametric approache

The goal of the present study is to explore the longitudinal effects of all-day kindergarten on the academic achievement of children after adjusting for two covariates: children's SES and their age upon entry to schooling. The study used the first four measures of the Early Childhood Longitudinal Study (ECLS) done by the National Center for Educational Statistics (LACES). Longitudinal multilevel analysis is the technique best suited to the present study, since it allows us to look closely at the educational contexts of children who are nested within two types of kindergarten program (half-day and all-day), and to take into account the dependencies of observations that are measured repeatedly.;An important methodological issue has, however, emerged during the analysis of the data. When a conventional multilevel analysis (HLM) is carried out where there is a violated distributional assumption, the standard errors can be biased and the conclusions of the study affected. Nonparametric maximum likelihood (NPML) is a technique that does not require the normality assumption required by HLM. Thus, in this study, both HLM and NPML are employed, so as to determine the effects of non-normality, and the results of the two approaches are compared with respect to sensitivity of the normality assumption.;This study did not find significant difference between parameter estimates obtained via HLM and NPML. There was, however, a small difference in the magnitude of the random components. Specifically, NPML gave higher variances of the intercept and the slope at Level 2, and a lower variance at Level 1.;With respect to educational policy issues, a beneficial effect of all-day kindergarten was not found. A basic finding is that all-day kindergarten children begin with significantly higher scores in math and reading, but improve their scores more slowly as compared to half-day kindergarten children.