L.3 Statistical Methods for Bioaccumulation

L.3 Statistical Methods for Bioaccumulation

Bioaccumulation tests are applied to determine whether exposure to dredged

material is likely to cause an elevation of contaminants in plant or animal tissues

compared with exposure to a reference. Bioaccumulation tests may be

conducted in the laboratory or in the field.

Situations may arise, particularly in the evaluation of plant or animal

contaminant uptake, where several sites, treatments, or dredged sediments are

simultaneously compared with a reference or control. If only one treatment is

compared to the reference, then the procedures described in Section L.2.1.1.1

(tests of assumptions followed by a t-test using a transformation or rankits if

necessary) for comparing two samples are used. If more than one treatment is

compared to the reference, then the procedures described below (tests of

assumptions followed by LSD, t-tests, or nonparametric equivalents) are used.

These analyses assume that individual sites are relatively large, and that a

decision concerning any particular site based on pathway testing results will be

made independently for each site.

Because contaminant concentration data are not easily expressed as propor-

tions, the arcsine transformation is not appropriate. The raw data are analyzed

first and, if necessary, a transformation may be employed. Contaminant concen-

tration data often follow a lognormal distribution so the logarithmic (either

natural or base 10) transformation is frequently used, but other transformations

such as square root are possible. As always, tests of assumptions must be rerun

on the data following transformation. If the transformed data violate the normal-

ity assumption, the data are converted to rankits (or ranks) and the assumptions

are retested.

L.3.1 Methods for multiple comparisons

Fisher's Least Significant Difference (LSD). Fisher's Least Significant

Difference (LSD) is appropriate for assessing differences in bioaccumulation

when more than two means are being compared. This a posteriori parametric

multiple comparison technique is discussed in many statistical texts, e.g., Steel

and Torrie (1980); SAS Institute, Inc. (1990c); Snedecor and Cochran (1989);

and Wilkinson (1990). The LSD controls the pairwise Type I error rate rather

than the experimentwise Type I error rate. This means that when the test

assumptions are met, the Type I error rate for each comparison is held to the

preset a even though the overall Type I error rate for all comparisons (i.e.,

experimentwise error rate) may be higher. A test that controls the pairwise error

rate is appropriate when decisions are to be made independently for each test site

regardless of how many sites are compared to the same reference. In situations

where rigorous control of experimentwise Type I error rate is important, e.g., if

decisions will not be made independently for each test site, Dunnett's test would

be preferred to the LSD test.

The LSD is usually performed in conjunction with analysis of variance

(ANOVA), and only if the data meet the assumptions of normality and equal

L27

Appendix L

Statistical Methods