Effective early childhood intervention and child care policies should be based on an understanding of the effects of child care quality and type on child well-being. This article describes methods for securing unbiased estimates of these effects from nonexperimental data. It focuses on longitudinal studies like the one developed by the National Institute of Child Health and Human Development's Early Child Care Research Network. This article first describes bias problems that arise in analyses of non-experimental data and then explains strategies for controlling for biases arising from parental selection of child care. Next, it comments on attrition in longitudinal studies and outlines some strategies for addressing possible attrition bias. Finally, it discusses the need to translate "effect sizes" derived from these studies into the kinds of cost and benefit information needed by policy makers. (Contains 3 notes.)