You are here:
Combining correlated stressor variables to clarify stressor effects
Citation:
VAN SICKLE, J. Combining correlated stressor variables to clarify stressor effects. Presented at 56th Annual Meeting of the North American Benthological Society, Salt Lake City, UT, May 25 - 30, 2008.
Impact/Purpose:
The effects of different stressors on biota cannot be clearly compared when stressor variables are sufficiently correlated in aquatic survey data
Description:
The effects of different stressors on biota cannot be clearly compared when stressor variables are sufficiently correlated in aquatic survey data. The recent USA Wadeable Streams Assessment reported ecological conditions for macroinvertebrate IBI and for each of 4 stressor variables (total N, total P, habitat complexity, and riparian vegetation cover). In multiple regression, the 4 stressors "shared" 33% of their total explained variation in IBI, due to their confounding (correlation). As a result, the "unshared" portion of IBI variation attributable to each stressor (5-25% of total explained) gave an unreliable estimate of that stressor's overall effect. To potentially reduce confounding, we can instead estimate effects for interpretable combinations of similar, correlated stressor variables. For example, a "nutrients" combination (total P, total N) and a "habitat" combination (habitat complexity, riparian vegetation cover), separately accounted for 51 and 44% of explained variation in IBI, respectively, leaving only 5% shared. Model uncertainty aside, the modeled effect of "nutrients" on IBI was slightly, but clearly, greater than the "habitat" effect. Thus, the effects of a few, general classes of stressors (e.g., metals, nutrients, toxics, habitat) can sometimes be clearly separated and compared, unlike the effects of correlated, individual stressor variables