that, he would get a value of 1 for previous wins and
0.75 (0.5þ0.25) for previous losses, indicating the
relative advantage and disadvantage for male A resulting
from prior contest experience.
As the three types of experience effect are formulations
of the same phenomenon, we allowed only one type of
experience effect in any one model. In addition, we
assessed the following potential predictors of fighting
ability (x): SVL, condition, relative proportion of pink
flank patch, casque height, jaw length, head width and
tail length (the latter four variables corrected for body
size). To choose the most parsimonious model, we used
a standard stepwise procedure, examining reduction in
the model Akaike Information Criterion (AIC) as well as
the significance of variables. We used the criterion of
P < 0.1 for retaining variables in the model. For maximum-
likelihood fitting of models we used the statistical
package R (Ihaka & Gentleman 1996).