Guideline 12: Discriminatory Ability 



The ability of the indicator to discriminate differences among sites along a known condition gradient should 

 be critically examined. This analysis should incorporate all error components relevant to the program 

 objectives, and separate extraneous variability to reveal the true environmental signal in the indicator 

 data. 



Sites from 1993 and 1994 were classified as degraded or undegraded based on our a priori criteria for 

 dissolved oxygen, sediment chemistry, and sediment toxicity that were used to choose test sites in the 

 development of the index. Of the 31 sites sampled in 1 993 and 1 994, only 1 95 could be classified as either 

 degraded or undegraded and these were used in the first validation step. In a Monte Carlo exercise, we 

 randomly chose 50 subsets of the 1 95 sites where each subset consisted of 50 degraded and 50 undegraded 

 sites. Correct classification occurred when the benthic index was either < 3 at degraded sites or > 5 at 

 undegraded sites. Misclassification occurred when the benthic index was < 3 at undegraded sites (false 

 negative) or > 5 at degraded sites (false positive). Using the 50 trials, we determined the percent of sites that 

 were correctly classified as degraded and undegraded by the benthic index. The benthic index correctly 

 classified 66-82% of degraded sites (x = 74%; SE = 0.5) and 70-84% of undegraded sites (x = 77%; SE = 

 0.4). The high degree of variability in the benthic communities in the Gulf of Mexico region influenced the 

 classification success. Although we attempted to minimize this variability during the development phase, we 

 may have sacrificed a level of precision in favor of a generalized index that is applicable across a wide 

 geographic area with an inherently large spatial variation. We also investigated the kappa coefficient (k) to 

 measure the degree of agreement (Stokes et al. 1995) between classification of a site by the benthic index 

 versus classification by sediment contaminants, toxicity, and dissolved oxygen. The average kappa coefficient 

 for the 50 trials was 0.509 where k > 0.4 indicates moderate agreement and the null hypothesis that there 

 was no agreement {H^: k = 0) was rejected at the a = 0.05 level of significance. 



An important consideration for the benthic index was that it not be significantly correlated with any natural 

 habitat factors like salinity or sediment type. This was addressed during the development of the index by 

 adjusting any benthic parameters that were correlated with salinity, or sediment type. One of the components 

 of the benthic index, proportion of expected diversity, represents a salinity-adjusted variable because diversity 

 is highly correlated with salinity in estuarine waters. Figures 3-6 and 3-7 show the relationship between the 

 benthic index and salinity and percent silt-clay content of sediments. The benthic index was still significantly 

 correlated with salinity and percent silt-clay but the R^ for both of these correlations was <1 5%. We determined 

 that these relationships were insignificant from an ecological perspective with statistical significance primarily 

 driven by the large number of samples (n = 338). 



Anthropogenic impacts may be correlated with salinity and silt-clay as well; therefore, residual correlations 

 between the benthic index and salinity or silt-clay may not indicate a lack of discriminatory power in the index. 

 This is important because the benthic index was designed to be an indicator of environmental condition that 

 is representative of the degree of sediment contamination and hypoxia experienced at a site, regardless of 

 the inherent salinity and sediment characteristics. 



3-20 



