Sun et al. [10] considered a number of statistical strategies to examine multiple pollutants and their interactions using regression methods for high-dimensional covariates, such as least absolute shrinkage and selection operator (LASSO) [11], Bayesian model averaging (BMA) [12] or supervised principal component analysis (SPCA) [13]. This study showed that LASSO and other dimension reduction techniques worked well for estimating risk models when a large number of candidate pollutants exist. Elastic-net method [14] or the adaptive elastic-net method [15] were proposed to take into account the issue of multi-collinearity when highly correlated predictors are fit simultaneously.
