406 – Sample Design - 2
Decomposing the Variance of Child Outcomes in Multistage Sample of Head Start Children
Barbara Carlson
Mathematica Policy Research
John Hall
Mathematica Policy Research
When designing a sample, estimates of expected precision are commonly made to help determine sample size. These calculations require specifying one or more of the following: type I error, power, population variance, design effects, finite population corrections, R2 for covariates, and the effect size. Sometimes, an earlier study can provide some of these, but often one must rely on educated guesses. Even when other studies are available, it is not straightforward to derive the various components of variance. In this paper, we decompose the variance and design effects for several key child outcomes from two rounds of the Head Start Family and Child Experiences Survey (FACES), in the hope that they can be used to help design samples for similar multistage samples of preschool-age children. This clustered sample involves selecting Head Start programs, then centers, classrooms, and children. Working backwards from the observed total variance for these outcomes, we first factor out the design effect due to unequal weighting, and then decompose the design effect due to clustering.