L.3.2.2 Comparison with an action level

A number of methods can be used to permit statistical comparisons of

censored data, including simple substitution, uniform distribution substitution,

maximum likelihood, and regression methods. Based on the results of a

simulation study conducted to identify which of 10 censored data methods work

best to maintain power and minimize Type I error rate in LSD comparisons when

n is small, Clarke (1998) recommended the use of nonparametric tests. A

constant lower than all reported values, such as zero, one-half DL, or negative

DL, is assigned to all nondetects and then the data are converted to rankits or

ranks prior to running a t-test or LSD test, or Dunn=s Test may be performed.

The power of any test will generally decline as the amount of censoring

increases; statistical analysis is not recommended when more than 60 to 80

percent of the data are nondetects. Deletion of nondetects is not recommended

as it results in excessive loss of information and power as amount of censoring

increases.

L.3.2.2 Comparison with an action level

In this comparison, the objective is to determine whether the mean bio-

accumulation of contaminants in plants or animals exposed to a dredged material

is significantly less than a specified action level or standard. If the mean tissue

concentration of one or more contaminants of concern is greater than or equal to

the applicable action level, then no statistical testing is required. If the mean

tissue concentrations of a contaminant of concern are less than the applicable

action level, then a confidence-interval approach is used to determine if these

means are significantly less than the action level. One-sided tests are appropriate

since there is concern only if bioaccumulation from the dredged material is not

significantly less than the action level. There are two different approaches to

conducting these tests, and both are acceptable.

The first approach is to calculate a value of t, much as in a t-test (this

approach is often called a one-sample t-test):

x - action level

(L-25)

s /n

where x, s2, and n refer to mean, variance, and number of replicates for

contaminant bioaccumulation from the dredged material.

If tests of equality of variances in the comparison of dredged materials with

the reference indicate that variances are equal for all treatments, then MSE from

the ANOVA is used as s2, and calculated t is compared to t0.95, with N - k degrees

of freedom. If the variances are not equal, then s2 for the individual treatment is

used, and calculated t compared with t0.95, with n - 1 degrees of freedom. If the

data were transformed to normalize the distributions or equalize variances, then

all calculations should be carried out on transformed values.

Another approach is to calculate the upper one-sided 95 percent confidence

limit (UCL), and compare it to the action level:

L36

Appendix L Selected Resource Documents