Environmental and demographic factors affecting childhood academic performance in Los Angeles County: A generalized linear elastic net regression model

Abstract

Modeling the influence of environmental exposures on early cognitive development requires summarizing dynamic spatial and temporal patterns. Spatial models often rely upon singular annual measures, precluding the understanding of how conditions and outcomes evolve across space and time. We modeled a large sample of environmentally and demographically diverse public elementary schools in Los Angeles County, California (N = 1,017) to explore how developing greenspace and air quality measures relate to standardized test performance. We show a time-discrete spatial representation of the data to identify extremes in test performance through a binomial elastic-net model. The regression model incorporates various aggregations of temporal-lag variables to detect the strength of relationships between the school's performance and preceding local environmental conditions. Our results show that through a binomial representation of extreme performance, the models captured how neighborhood factors lead to high or poor school-wide test performance with an average accuracy of 91% (88–94%). Furthermore, the model's longitudinal aspect revealed how cross-sectional measures of average exposure are reliable but may only partially reveal the environment's influence on health evaluations. Our findings support previous studies that showcase greenspace's mixed associations with development. Future incorporations of temporal lags and aggregation types into geographic assessments of environmental conditions will aid in uncovering health inequities.