Comparison of linear and log-linear general relative risk models for the association between lung cancer mortality and mean county-level residential radon concentrations (LBL) at enrollment (1982), follow-up (1982–1988), CPS-II cohort, United States. Comparison of linear and log-linear general relative risk models for the association between lung cancer mortality and mean county-level residential radon concentrations (LBL) at enrollment (1982), follow-up (1982–1988), CPS-II cohort, United States. According to the stratification criteria of Krewski and colleagues (10, 11) little difference in relative risk estimates obtained from either a linear (ERR = 1 + 0.00121 X; dotted line) or log-linear [Cox regression analysis; RR = exp(0.00119 X); solid line] model was observed. Relative risk estimates were also similar using the stratification criteria of Darby and colleagues [refs. 8, 9; ERR = 1 + 0.00108 X, RR = exp(0.00109 X)]. Michelle C. Turner et al. Cancer Epidemiol Biomarkers Prev 2011;20:438-448 ©2011 by American Association for Cancer Research