This means that in areas of higher radon risk, there is a neutral or slightly increased lung cancer mortality rate. This is an area of highest statistical significance (local R-squared 0.64 – 0.72), where the relationship between radon and lung cancer mortality is neutral or slightly negative. The results of our GWR indicate some statistical significance in Florida, where local R-squared values range from 0.49 to 0.46 across the entire state.Īnother area of interest from our radon coefficient surface is the region of convergence between Tennessee, Alabama, Georgia, South and North Carolina. It should be recognized that it is possible for indoor radon concentrations to exceed predicted values and air quality targets within each zone (EPA, 2017). Although Florida is classified as Radon Zone 3, which is predicted to have relatively low average indoor radon concentrations below 2 pCi/L, our model indicates that radon is a major influence on lung cancer mortality in this region. The radon coefficient surface generated by the GWR highlights a strong relationship between radon risk and lung cancer mortality in Florida. Figure 7.3: GWR coefficient surfaces highlight the strength and direction of the relationship between our explanatory variables and the 5-year lung cancer mortality rate The coefficient surfaces generated by the GWR highlight regional variation in the dependent variables’ explanatory power, providing insight into the associations between observed lung cancer mortality rates (Figure 7.2) and environmental and socioeconomic factors (Figure 7.3). This area, which is best described by our GWR model, overlaps the south-central Appalachian subregion which exhibits multiple significant hot and cold spots for lung cancer mortality rates and explanatory variables such as income, smoking prevalence, proportion non-white population etc.įigure 7.1: GWR output surface highlighting the range of local adjusted R-squared values over the eastern United States Figure 7.2: The observed 5-year lung cancer mortality rates over the eastern United States Medium-high local R-squared values (0.57 – 0.63) were identified in Wisconsin, Minnesota, and central Ohio. Highest local R-squared values were present on the western side of the region ( Figure 7.1 ), particularly in northern Georgia, northeastern Alabama, and Tennessee (0.64 – 0.72). The GWR had a higher adjusted R-squared value (0.62) than the OLS Models 1 (0.48) and 2 (0.54). GWR also allowed us to map variation in regression parameters across space. Geographically-weighted regression models assign a greater weight to observations which are nearer to each particular feature location than to those which are further away, removing any spatial dependencies which might have been present in the residuals of the OLS (global) linear regression model.
0 kommentar(er)
