Estimating the risks of multiple, covarying stressors in the National Lakes Assessment

The National Lakes Assessment (NLA) used relative and attributable risks to measure the apparent nationwide effects of excess N, reduced lakeshore habitat, and other stressors on planktonic assemblages in lakes. The risk measures, borrowed from human health research, use nontechnical language to compare the apparent effects of disparate lake stressors, thereby helping the public and policymakers identify stressors of greatest national concern. However, the investigators for the NLA and similar prior surveys of streams did not adjust their risk estimates of each stressor for possible confounding by other covarying stressors. As a result, the NLA point estimates overstate the risks of individual stressors and, thus, are unreliable for assessing their importance relative to that of closely related stressors. I used NLA data from 966 lakes to illustrate existing statistical methods of risk adjustment. Point estimates of adjusted relative risk for 7 stressor variables were 15 to 64% lower than unadjusted estimates. Adjusted attributable risks also were lower than unadjusted values, but for some stressors, they were very inconsistent across 3 adjustment methods. In addition, adjusted risk estimates used only part of the available data because each estimate was adjusted for many (6) covarying stressors. Closely related stressor variables (for example, N, P. and turbidity) can be combined into a bundle representing a broader type of stress (reduced water quality). For the NLA data, adjusted risk estimates for the stressor bundles (water quality and habitat) were more consistent across estimation methods and had lower relative uncertainty than estimates for their component stressors. In addition, closely related stressors with similar sources and modes of impact are more likely to be managed together rather than individually. For these reasons, I suggest evaluating stressor bundles in future aquatic surveys.