Ch. 6— Alternatives to Animal Use in Research • 127 
the number of subjects needed. However, this 
also increases the chances of concluding that 
the experimental procedure produced an ef- 
fect when in fact the effect was due to chance 
alone. By convention, researchers generally 
accept the probability of a chance effect of 
5 percent or less as a statistically significant 
result. 
• Greater precision in the conduct of an exper- 
iment may reduce variability and increase 
power . For example, highly precise behavioral 
measurements coupled with the elimination 
or control of extraneous variables would re- 
duce the need for large numbers of subjects 
(198). 
• The use of different statistical analyses can 
increase the sensitivity and power of a proto- 
col (e.g., analysis of the data by parametric 
rather than nonparametric statistical tests) 
(198). 
• Alterations in experimental designs can in- 
crease power. Factorial designs (where two 
or more treatments are manipulated concur- 
rently), for example, are more powerful and 
can be used instead of testing the effects of 
different treatments in separate experiments. 
Not only does the use of factorial designs in- 
crease power, it requires fewer untreated, 
control subjects than multiple concurrent 
studies do. It is important to note, however, 
that in areas that have not been heavily re- 
searched there are inherent dangers to the 
use of factorial designs. For example, there 
may be no observed effect of treatments given 
in combination, as one treatment cancels the 
effect of another. Without sufficient back- 
ground information on the effects of the treat- 
ments administered individually, this finding 
would be erroneously interpreted. 
• Power is increased as the magnitude of the 
treatment effect is increased. Treatment ef- 
fects can be maximized by choosing widely 
spaced levels of the treatment variables or by 
including conditions that are thought to max- 
imize the appearance of the phenomenon 
under study (113). 
Within- Subjects Design.— Many experiments 
on animal behavior are conducted using a between - 
subjects design. That is, different groups of ani- 
mals are given different treatments, and the per- 
formances of the different groups are compared. 
However, individuals also vary in their behavior. 
Depending on the degree of variability, large num- 
bers of subjects may be needed in each group to 
obtain statistically significant results. Under cer- 
tain conditions, however, a within-subjects (or 
repeated measures) design can be used that re- 
quires only one group of animals instead of many. 
Under these conditions all members of the group 
serve in all treatment conditions. The advantage 
of this technique is that it minimizes variability 
by taking into account individual differences. The 
major drawback, however, is the possibility of con- 
taminating the data and nullifying the results: 
Treatments already received by a subject may in- 
fluence, and thereby confound, performance un- 
der subsequent treatments. Carry-over effects can 
be partially offset by counter-balancing, wherein 
the experimenter ensures an equal occurrence of 
each experimental treatment at each stage of the 
experiment; this balances any effect of prior test- 
ing equally over all treatment conditions (113). Al- 
though within-subjects designs are effective in re- 
ducing both variability and the number of subjects 
needed, the inherent danger of carry-over effects 
in many instances may invalidate the use of such 
designs. 
Random Block Design.— Randomized block de- 
sign consists of assigning subjects to groups based 
on evidence of their being similar to one another 
in one or more characteristics known to be related 
to the behavior under investigation. Two or more 
such blocks are formed and then each block is as- 
signed randomly to the treatment conditions. This 
design reduces variability by restricting the de- 
gree of individual differences within blocks, and 
thereby increases power (113). Although random- 
ized block designs are effective in lowering the 
number of animals needed in an experiment, they 
are not applicable to all areas of behavioral re- 
search. The technique requires substantial prior 
knowledge of the behavior being investigated and 
is therefore limited to intensively researched areas. 
Analysis of Covariance.— An analysis of covari- 
ance uses the same information as randomized 
block designs except that an estimate of variabil- 
ity is not needed beforehand. The covariance pro- 
