Translator Disclaimer
23 July 2012 Deriving criteria-supporting benchmark values from empirical response relationships: comparison of statistical techniques and effect of log-transforming the nutrient variable
Author Affiliations +
Abstract

Empirical analyses of ecological changes accompanying nutrient enrichment provide one line of evidence for developing protective nutrient criteria. The purpose of my study was to elucidate 2 important aspects of such empirical analyses: 1) how statistical techniques compare regarding types of relationships fit and benchmark values obtained, and 2) how decisions regarding log-transformation of right-skewed nutrient data affect the results. I used data from Great Lakes coastal wetlands describing a suite of water-quality and biotic responses over a large nutrient gradient to conduct side-by-side evaluations of 5 statistical techniques (logistic regression, cumulative probability analysis, linear regression, piecewise regression, classification and regression tree analysis [CART]). With this somewhat noisy data set, differences in goodness-of-fit among techniques that modeled gradual changes vs ones that identified abrupt transitions were remarkably small, providing little evidence for superiority of one over another. However, differences among techniques in nutrient benchmark values and their ecological interpretation were substantial. Log10-transformation of nutrient data had little effect on residuals but shifted the benchmark values for all techniques except CART. Decisions concerning log-transformation ought to be based on implications for deriving criteria rather than perceived statistical assumptions. Multiple statistical techniques and response relationships provide relevant information and no transformation makes all relationships conform to the same pattern, so no cookbook recipe for analyses can be identified. Professional judgment is needed to convey empirical findings toward eventual criteria values regardless of the technique applied.

The Society for Freshwater Science
Anett S Trebitz "Deriving criteria-supporting benchmark values from empirical response relationships: comparison of statistical techniques and effect of log-transforming the nutrient variable," Freshwater Science 31(3), 986-1002, (23 July 2012). https://doi.org/10.1899/12-055.1
Received: 5 April 2012; Accepted: 1 May 2012; Published: 23 July 2012
JOURNAL ARTICLE
17 PAGES

This article is only available to subscribers.
It is not available for individual sale.
+ SAVE TO MY LIBRARY

SHARE
ARTICLE IMPACT
RIGHTS & PERMISSIONS
Get copyright permission
Back to Top