untitled

Very often a field experiment breaches one or more rules of experimental design and

so no longer answers unambiguously the question being posed. Such an occurrence

has two causes: an unfortunate mistake or a necessary choice.

Mistakes

Very often there may be no logistical or technical justification for using an inappro-

priate design. Such a flaw is simply a mistake. One of the most common in ecolo-

gical and wildlife research is pseudo-replication (=subsampling), used under the

misapprehension that it constitutes treatment replication (Hurlbert 1984). In this case

site and treatment are confounded (see Section 16.4.1).

A second common mistake is the unbalanced design. Figure 16.4 illustrates an experi-

ment to evaluate the effect of grazing by sheep and rabbits on the density of a species

of grass. There are two factors (SHEEPand RABBITS), each with two levels (presence and

absence). “Presence” for sheep is taken as the standard stocking rate, and that for

rabbits the prevailing density. Variation of rabbit density across the area is taken care

of by the replication.

The four treatments may be symbolized by a code in which 1 indicates presence

and 0 indicates absence. Most of the practical details of setting up such a trial are

simple. A rabbit-proof fence around a quadrat excludes both sheep and rabbits (treat-

ment R0 S0). A sheep-proof fence excludes sheep but allows rabbits in (R1 S0). A

quadrat to measure the effect of sheep and rabbits together is simply an unfenced

square marked by four pegs (R1 S1). Thus, three of the four treatments are easily

arranged. They can be set up and then temporarily forgotten, the experimenter return-

ing after several months or even years to harvest the data.

The final treatment (R0 S1) cannot be managed in this way. No one has yet invented

a fence that acts as a barrier to rabbits while allowing sheep free access to the quadrat.

Hence R0 S1 must be handled differently. It requires a rabbit-proof fence around

the quadrat to exclude rabbits (as for R0 S0) but with sheep at standard stocking

density within the enclosure. That treatment cannot be set up and then left

untended. Sheep need water and husbandry. Hence, that treatment is often left out

278 Chapter 16

16.5.4Weak-

inference designs

Factor s: SHEEP (2 levels)

RABBITS (2 levels)

Response variable: Density of Themeda australisl plants

Design logic

No SHEEP SHEEP

No

RABBITS

RABBITS

2 reps

[Rabbit-proof

fence]

2 reps

[Rabbit-proof

fence with

sheep inside]

2 reps

[Sheep-proof

fence]

2 reps

[Open range]

Fig. 16.4Design logic

for a two-factor

experiment on the effect

of sheep and of rabbits

on the biomass of

pasture.

WECC16 18/08/2005 14:47 Page 278