This study applied nonparametric loess regression to describe the predictive association between birth weight and cognitive ability in childhood and generalized propensity score methods to control the confounding of multiple covariates that summarize prenatal differences.

The data from wave one of the National Longitudinal Study of Youth 1979-Children (NLSY-C) and the data from wave one of the Panel Study of Income Dynamics-Child Development Supplement (PSID-CDS) were analyzed in parallel and the results were compared.

Analysis of the two datasets proceeded in two stages. At the first stage, the association between birth weight and cognitive ability was examined first using loess regression models in which both birth weight and gestational age were the predictors, and then by simple loess regression models in which birth weight was the sole predictor while gestational age was restricted to the normal range, i.e., between 37 complete weeks (259 days) and 42 complete weeks (294 days) to help control its confounding effects. The association was examined separately for subsamples defined by race and sex. At the second stage, the distribution of birth weight was broken into low, medium, and high, and the challenge of controlling multiple covariates was addressed by stratifying the three groups on the balancing score estimated from generalized propensity score models that can accommodate more than two groups.

The shape of the association between birth weight and cognitive ability as depicted by loess regression was not consistent across the two datasets. Among the children in the NLSY-C who were born near term and with normal birth weights, the association had an upside-down U-shape. Among the children in the PSID-CDS who were born near term and with normal birth weights, the association was roughly a continuously increasing one. The shape of the association persisted, and was slightly strengthened when multiple covariates that account for prenatal differences were controlled by balancing score-based stratification.

Results from the two data sets demonstrated consistent racial and gender differences. The association was stronger for whites than for blacks. White males had slightly higher cognitive scores than white females, while black females had slightly higher scores than black males.