We evaluated the associations for outliers; excluding potentially influential points from the analyses did not change the inferences.
No outliers (important influential points) were identified for any of the variables included in the final model.
Model diagnostics, including studentized residuals and Cook's distance values, were inspected for outliers and highly influential points and models were evaluated for coherence with known emission source patterns and for sensitivity to alternative emission source indicators.
On the basis of a statistical analysis of the residuals (classical, normalized, standardized, jackknife, predicted and recursive) and diagonal elements of a projection matrix, diagnostic plots for influential points indication are formed.
Within each cluster, we check model diagnostics and look for influential points.
Logistic regression models were checked for model fit using the Hosmer-Lemeshow test and were evaluated for influential points using delta-beta statistics.
We also conducted regression diagnostics for influential points using Cook's distance, but no substantial changes in estimates were observed before or after the influential points were excluded, and our results are therefore all based on the whole data set.
We identified influential points separately in models for the full cohort and the low-tobacco group and provide results for analyses that excluded these observations as well as results when we retained these data.
The final regression models to predict length of stay and total costs were checked for outliers and influential points [ 17].
To control for effects of influential points, we repeated all analyses excluding influential observations by a statistical procedure proposed by Belsey et al. (2005).
