We used univariate regression to obtain crude odds ratios (OR) and to select the most important covariates (p < 0.20) from each of the four groups to be included in the multivariate logistic models. We kept variables in the final models if they attained statistical significance (p < 0.05) or if we considered them important potential confounders of the association between our indicator of exposure and perinatal mortality, regardless of their significance. We did not include birth weight, gestational age, and IUGR in our models because these conditions may lie in the causal pathway between traffic-related air pollution and perinatal mortality, so it would not be adequate to control for them.