analyses. Tests of differences in toxicity results (i.e., "sensitivity") were performed two ways 

 to satisfy two objectives. First, because these candidate tests were evaluated for future use in 

 the NS&T Program in which the sampling protocol involves collection of samples at three 

 unreplicated stations per site for chemical analyses, differences in mean results among sites 

 for end-points measured at all stations were of interest. Results of each toxicity test for each 

 site were analyzed by the non-parametric Kruskal-Wallis (K-W) test (oc=0.05) for differences 

 between sites and respective controls and among sites. For those tests where significant 

 differences were indicated, a non-parametric equivalent of the Student-Newman-Keuls 

 (S-N-K) multiple comparison test was performed to determine which sites could be 

 differentiated from each other, and a non-parametric equivalent to Dunnett's t-test was 

 performed for differentiating sites from the respective control(s). The data from each of the 

 three stations at each site were used as measures of within-site variance. Cumulative 

 (pooled) results from the five laboratory replicates were used in these determinations of 

 between-site differences. Second, to determine the relative sensitivities of each test to 

 individual samples, each composited sediment sample from each station was regarded as an 

 independent, individual sample and the differences between the samples and respective 

 controls were determined and tallied. In this case, the five replicates tested in the 

 laboratory per treatment were used as measures of within-treatment variability. In the 

 latter statistical tests, no attempts were made to determine geographic patterns in toxicity 

 response, but, rather, to determine sensitivity of each toxicity end-point to individual 

 samples. Differences in results between individual samples and respective controls were 

 tested with the K-W test and Dunnett's t-tests (Wilcoxon and Wilcox, 1964). 



In addition to the non-parametric tests, a few parametric tests were performed with the 

 data in an exploratory phase to further characterize patterns in the results. Since these 

 additional procedures used parametric methods, the results cannot be used in any inferential 

 sense, but, rather, provide additional qualitative descriptions of the results. 



The standard deviations (SDs) and coefficients of variation (CVs) for the five replicates 

 of each sample were calculated. Then, the averages of the CVs were determined for each of 

 the toxicity end-points to evaluate relative analytical "precision." The difference between 

 the maximum and minimum mean values observed in the samples was divided by the average 

 SD to estimate the discriminatory power, an index that was not biased by the data from the 

 controls. Correlations among toxicity end-points normalized to respective batch controls were 

 determined by Spearman's rank correlation (Spearman, 1904). Spearman's rank correlations 

 were also determined between toxicity and bulk sediment chemical data, in which the metals 

 data were normalized to percent fines and the organics data were normalized to TOC content. 



Benthic Community Analyses. The data from the benthos taxonomic analyses and the 

 sediment profiling survey were not completed for all five sediment sampling sites. 

 Therefore, they were not subjected to the same statistical treatments applied to the 

 sediment bioassay data. Methods described on pages 16 through 18 were used to classify 

 sites with the data that were collected. 



Measures of Effects in Fish. The biological data from the fish analyses were summarized 

 and listed as means and standard deviations for each site for each sampling episode. The 

 chemical data from the fish were treated similarly. The biological data were evaluated 

 with the Kolmogorov-Smirnov test to determine if they were normally distributed. 

 Differences among sampling sites were tested with the K-W test followed by the non- 

 parametric S-N-K test. Data from the analyses of nuclear pleomorphism in erythrocytes 

 were tested with chi-square. Differences in chemical concentrations among sites were tested 

 with one-way ANOVA. Correlations both among the biological measures and between 

 biological measures and liver contaminant concentrations were determined with Spearman's 

 rank correlation analysis. The range in response of each test among sites, the SD and the 

 CV among the samples at each site, and the maximum range divided by the SD were 

 calculated as indicators of methodological sensitivity. 



In addition to the non-parametric tests, a few parametric tests were performed with the 

 data in an exploratory phase to further characterize patterns in the results. Since these 



29 



