A multilevel non-hierarchical study of birth weight and socioeconomic status.

Background: It is unclear whether the socioeconomic status (SES) of the community of residence has a substantial association with infant birth weight. We used multilevel models to examine associations of birth weight with family- and community-level SES in the Cape Cod Family Health Study. Data were collected retrospectively on births to women between 1969 and 1983 living on Cape Cod, Massachusetts. The sample included siblings born in different residences with differing community-level SES.

Methods: We used cross-classified models to account for multiple levels of correlation in a non-hierarchical data structure. We accounted for clustering at family- and community-levels. Models included extensive individual- and family-level covariates. SES variables of interest were maternal education; paternal occupation; percent adults living in poverty; percent adults with a four year college degree; community mean family income; and percent adult unemployment.

Results: Residual correlation was detected at the family- but not the community-level. Substantial effects sizes were observed for family-level SES while smaller magnitudes were observed for community-level SES. Overall, higher SES corresponded to increased birth weight though neither family- nor community-level variables had significant associations with the outcome. In a model applied to a reduced sample that included a single child per family, enforcing a hierarchical data structure, paternal occupation was found to have a significant association with birth weight (p = 0.033). Larger effect sizes for community SES appeared in models applied to the full sample that contained limited covariates, such as those typically found on birth certificates.

Conclusions: Cross-classified models allowed us to include more than one child per family even when families moved between births. There was evidence of mild associations between family SES and birth weight. Stronger associations between paternal occupation and birth weight were observed in models applied to reduced samples with hierarchical data structures, illustrating consequences of excluding observations from the cross-classified analysis. Models with limited covariates showed associations of birth weight with community SES. In models adjusting for a complete set of individual- and family-level covariates, community SES was not as important.