Theoretical predictions of observed to expected ratios in RIVPACS-type predictive model assessments of stream biological condition

Abstract

AbstractRiver InVertebrate Prediction And Classification System (RIVPACS)-type predictive models assess the biological conditions of streams by comparing observed assemblage composition (O) with an expected assemblage composition derived from reference site observations (E). The ratio of the 2 values (O/E) can be interpreted as a measure of taxonomic completeness. Values of O/E that are near 1 at a test site suggest that the site is comparable to reference sites, whereas values that differ substantially from 1 suggest that the site is degraded. I analyzed the computations involved in estimating E mathematically. The analysis reveals that many predictive models produce values of O/E that are systematically <1, even when the models are applied to sites that have the same average assemblage composition as reference sites. The amount by which O/E values are <1 depends upon the number of reference sites used to build the model, the manner in which these reference sites are distributed among discrete clusters, ...