untitled

done to the control, other than for the manipulation that is formally the focus of the

treatment. If vehicles are driven over the quail plots to distribute the grain they must

also be driven over the control areas.

Controls must obviously be appropriate, and often a good deal of thought is needed

to ensure that they are. We have previously dealt with the mistake of declaring “before

treatment” a control on “after treatment” (see Section 16.4) but there are more

subtle traps to keep in mind. If the treatment is an insecticide dissolved in a solvent,

then the control plots must be sprayed with the solvent minus the insecticide. If treated

birds are banded then the control birds must be banded. If animals are removed from

the field to the laboratory for treatment and then released, the control animals must

also be subjected to that disturbance. And so on.

There is no general answer to the question “How many replicates are necessary?”

other than the trite “at least two per treatment.” It depends upon the number of treat-

ments to be compared, the average variance among replicates within treatments, and

the magnitude of the differences one expects or is attempting to establish. These may

be estimated from a pilot experiment or from a previous experiment in the same area.

As a general rule, however, the fewer the treatments the more replicates needed

per treatment, but there is little to be gained from increasing replication beyond 30

degrees of freedom for the residual. Suppose the experiment had three factors with

ilevels in the first, jin the second, and kin the third. There are thus ijktreatments

and ijk(n−1) degrees of freedom in the residual, where nis the number of repli-

cates per treatment.

Most questions on the effect of this or that management treatment have a similar

logical structure, even though they deal with different animals in different conditions.

The most common questions lead to standard experimental designs.

One factor, two levels

Figure 16.1 represents the simplest design that will provide an answer that can be

trusted. It evaluates the operational null hypothesis that supplementary feeding with

EXPERIMENTAL MANAGEMENT 275

16.5.2Sample size

16.5.3Standard

experimental designs

Factor: WHEAT (2 levels)

Response variable: Density or rate of increase of quail

Design logic

WHEAT No WHEAT

Rep

Rep

Rep

Rep

Design layout

Site 1

WHEAT added

Site 2

No WHEAT

Site 3

No WHEAT

Site 4

WHEAT added

Fig. 16.1Minimum

one-factor experimental

design.

WECC16 18/08/2005 14:47 Page 275