2.4.2. Core metric screeningWe conducted core metric selection using a stepwise processrecommended by Barbour et al. (1999),Hering et al. (2006) andStoddard et al. (2008). First, a range test was used to filtercandidate metrics. This test would eliminate poor metrics forassemblages with fewer taxa. Furthermore, the coefficient ofvariation (CV) for reference sites was used to select better metricsthat had a small CV (CV<1). Second, we tested the normality ofthe candidate metrics with normal probability plots and with aKolmogorov–Smirnov test of normality. If metrics had a normaldistribution, we would use parametric tests. Otherwise, we usednon-parametric tests. Third, box and whisker plots tests wereperformed for each candidate metric, and reference values werecompared to impaired values for discrimination power analysis.The discrimination power of each metric was judged according tothe degree of inter-quartile overlap in the box-plots. Metrics withsignificantly different values (p<0.05) and a higher discriminationpower (interquartile range, IQ_2) (Barbour et al., 1996) werefurther screened. Lastly, we evaluated metrics for redundancyusing the Spearman correlation analysis and metrics wereconsidered redundant if the Spearman correlation coefficientwas >0.70 and the p value was <0.05. Redundant metrics werethen eliminated from further analyses (Stoddard et al., 2008;Whittier et al., 2007). Only one metric per category should remain
according to Stoddard et al. (2008).
การแปล กรุณารอสักครู่..
