L.1.3 Experimental design

L.1.3 Experimental design

Once the investigator has formulated the null hypotheses to be tested,

decided upon significance (a ) and power (1 - ) levels for hypothesis testing, and

determined the sample size necessary to achieve the desired power, the next step

is to design an experiment to test the hypotheses. Instructions for setting up and

conducting toxicity and bioaccumulation experiments are outlined in the CDF

pathway appendices, but it is important at this point to review the basic

principles of experimental design. These principles include replication,

randomization, interspersion, and controls (Hurlbert 1984).

Replication refers to the assignment of a treatment to more than one experi-

mental unit. The number of replicates, as stated earlier, is the sample size for

that treatment. Recall that an experimental unit or replicate is the test container

(e.g., beaker, pot, or aquarium), not an individual organism in the test container.

The number of organisms in the test container is important only in terms of

constituting an adequate measure of the endpoint being tested (e.g., providing

sufficient tissue to measure contaminant bioaccumulation). Replication of

treatments is necessary to control for random error in the conduct of the

experiment. The pathway appendices include guidelines for minimum number

of replicates for the various bioassays. However, we strongly recommend

determining sample size a priori using the power formulae in Sections L.2.1.1.1,

L.3.1, and L.3.2.2. In many cases, the number of replicates necessary for a

powerful statistical test will be greater than the minimum guidelines.

Randomization and interspersion refer to the actual placement of experimen-

tal units in the laboratory setup. A random numbers table, available in most

statistical texts, may be used to randomly assign treatments to the experimental

units. If the randomization does not achieve a reasonable interspersion of

treatments, e.g., if several experimental units of the same treatment are clumped

together, then a new randomization should be tried. Randomization and

interspersion are necessary to control for investigator bias, for initial or inherent

variability among experimental units, and for variability in environmental

conditions such as lighting, water flow, etc.

Replication, randomization, and interspersion all function to control extrane-

ous sources of variability in an experiment. In addition, control treatment(s) are

needed to control temporal or procedural variability. In the broadest sense, the

control treatment is simply the treatment against which the other treatments are

compared. This is the dilution water (or control water) in acute toxicity testing,

and the reference in bioaccumulation testing. Laboratory controls, such as a

clean sand exposure in bioaccumulation testing, may also be included. Labora-

tory controls, if needed, are used for quality assurance, and are not included in

the statistical analyses.

Testing in Tier III can in most cases be best accomplished using simple

experimental designs, either a completely randomized design or a randomized

complete blocks design. These designs are discussed in most general statistics

texts. In a completely randomized design, treatments are assigned to experi-

mental units randomly over the entire experimental setup. A randomized

Appendix L

Statistical Methods