Results from generalised estimating equations† predicting maximum LCR‡ in US school districts (n=12 516 with ≥100 children§)
| Total | Point | On-road mobile | ||||
| b (95% CI) | P value | b (95% CI) | P value | b (95% CI) | P value | |
| Intercept | 3.577 (3.54 to 3.62) | *** | 0.154 (−0.06 to 0.37) | 1.818 (1.73 to 1.90) | *** | |
| Prop. children | 0.022 (0.01 to 0.03) | *** | 0.085 (0.02 to 0.15) | * | 0.010 (−0.01 to 0.03) | |
| Prop. children in poverty | 0.044 (0.02 to 0.07) | *** | 0.355 (0.19 to 0.52) | *** | 0.105 (0.06 to 0.15) | *** |
| Prop. children in poverty squared | −0.044 (−0.07 to −0.02) | *** | −0.378 (−0.61 to −0.15) | *** | −0.145 (−0.19 to −0.10) | *** |
| Prop. children that are foreign-born | 0.017 (0.01 to 0.02) | *** | 0.013 (−0.03 to 0.05) | 0.039 (0.02 to 0.06) | *** | |
| Prop. children that have a disability | 0.009 (>0.00 to 0.01) | *** | 0.090 (0.05 to 0.13) | *** | 0.014 (>0.00 to 0.02) | ** |
| Prop. children that are black | 0.041 (0.03 to 0.06) | *** | 0.221 (0.05 to 0.39) | ** | 0.076 (0.04 to 0.11) | *** |
| Prop. children that are Hispanic | −0.009 (−0.04 to 0.02) | −0.115 (−0.20 to −0.10) | ** | 0.086 (0.05 to 0.12) | *** | |
| Prop. children that are API | 0.021 (0.00 to 0.04) | * | −0.040 (−0.15 to 0.07) | 0.060 (0.04 to 0.08) | *** | |
| Prop. children that are American Indian | −0.022 (−0.03 to −0.01) | *** | −0.146 (−0.20 to −0.10) | *** | −0.074 (−0.09 to −0.06) | *** |
| Prop. children that are multiracial/other | 0.032 (0.02 to .05) | *** | 0.022 (−0.06 to 0.10) | 0.109 (0.09 to 0.13) | *** | |
| Prop. households with children that rent | −0.007 (−0.02 to >0.00) | −0.033 (−0.12 to 0.05) | −0.033 (−0.06 to <0.00) | * | ||
*P≤0.05, **P≤0.01, ***P≤0.001.
†Based on best fit, the models used slightly different specifications. ‘Total’ used Inverse Gaussian with log link, ‘point’ and ‘on-road mobile’ used Gamma with log link. All models use an exchangeable intracluster dependency correlation matrix and include children’s population density as a control variable. Independent variables were standardised before entering them into the model.
‡The LCR variable represents the highest census tract-level LCR estimate among all of the tracts that intersect or are completely contained within the school district boundary.
§School districts with less than 100 resident children as per the 2010 census were excluded to ensure reliable proportions.
LCR, lifetime cancer risk.