Wildfire Smoke and Public Health in BC
Spatial Analysis in Exposure Assessment and Health Implications
Results II
Regression AnalysisNo significant regression is observed between all the exposure methods and the Mean daily counts of prescriptions. Note that the probability of Method 2 is close to 0.05 and may be considered has a weak association with Mean.
Method 1 and Method 2 have significant positive association with the Sum of cumulated prescription counts. The regression model can explain around 20% of the variance in changes of Sum. However, we need to examine the spatial autocorrelation in the later part before we can draw any conclusion.
Method 3, where exposure assessment is based on the prediction model, has the weakest association with both the mean and sum of prescription counts.
| Regression variables | Probability | Adjusted R-squared | Coefficient |
| Method 1 & Mean | 0.288037 | 0.001704 | 0.024177 |
| Method 2 & Mean | 0.065553 | 0.028706 | 0.041991 |
| Method 3 & Mean | 0.555298 | -0.007789 | -0.008007 |
| Method 1 & Sum | 0.000030* | 0.181468 | 0.000170 |
| Method 2 & Sum | 0.000001* | 0.250627 | 0.000199 |
| Method 3 & Sum | 0.916259 | -0.011913 | 0.000003 |
Residuals and Spatial Autocorrelation
Residuals of the significant regressions were mapped and the patterns for Method 1 & Sum and Method 2 & Sum are similar. By calculating the Moran's index, the clustered patterns were found in both of these two regressions. This result indicates that there is still some unexplained spatial variables that contribute to the association except for the PM2.5 exposure assessed by our methods.
By simply visualizing the residual patterns and the monitoring station locations, I suspected that the residual distributions is related to the distances between LHA's populations (in this case the DA centroids). A further analysis was done with this factor.
Among all DAs, the mean distance from the nearest monitor is 24km, ranging from 80m to 644km. The mean distance averaging from DAs within each LHA is presented here. To further investigate this possibility, another regression was made with variable "residuals of Method 2 & Sum" (this analysis was not performed with Method 1 & Sum since the residual patterns are similar for the two regressions) and "mean distance from nearest monitor". A significant association was found (probability=0.02 < 0.05). However, this factor, still, can only explain about 5% of the variance in residual patterns and more factors should be looked at.
Residuals of Method1 & Sum
(Click here for higher resolution map)
The results of spatial autocorrelation indicates a clustered pattern and there is less than 1% likelihood that it is the result of random chance. (Click here for the report)
Residuals of Method2 & Sum
(Click here for higher resolution map)
The results of spatial autocorrelation indicates a clustered pattern and there is less than 1% likelihood that it is the result of random chance. (Click here for the report)
Map of Mean distances from nearest monitors
(Click here for higher resolution map)
