Neurobehavioral effects of prenatal alcohol: Part II. Partial least squares analysis.

This paper, the second in a series of three, introduces Partial Least Squares (PLS) methods for assessing the effects of moderate levels of prenatal alcohol exposure on performance and behavior in young school-age children. Studies of human behavioral teratology pose statistical problems for which standard multiple regression methods are inadequate. Prenatal alcohol exposure, the teratogenic "dose," can be assessed only indirectly through a variety of measures of alcohol consumption. Similarly, the behavioral outcomes we examine--IQ, achievement, classroom behavior, and vigilance--are each measured indirectly in terms of multiple items or indicators. We find that a single latent variable, estimated as a linear combination of the measures of alcohol consumption, provides an appropriate measure of "dose" for summarizing the relationships between alcohol exposure and each of the four blocks of outcome variables. A pattern of alcohol consumption emphasizing binge behavior (i.e., reporting average consumption of multiple drinks per drinking occasion, or at least five drinks on any single occasion) in the period prior to recognition of pregnancy is significantly correlated with latent variables computed from each of the four outcome blocks: IQ, academic achievement, classroom behavior and attention/vigilance.