larger sample sizes and, consequently, greater statistical
power, as each individual is used more than once. Tournament
designs, however, are not widely used because such
a design results in males having different agonistic experiences
and the effects of unmeasured differences in fighting
ability and ‘experience effects’ can be difficult to
disentangle. That is, a winner may keep winning because
he has a series of attributes that make him a better fighter
or because of a self-reinforcing ‘winner effect’ and vice
versa for losers.We show how this problem can be circumvented
through use of a generalized Bradley–Terry model.
Traditionally, the B–T model has been used to assign
ranks to individuals within a linear hierarchy based on
paired observations. The only two previous applications of
the B–T model to animal dominance or aggression used
the model to estimate a complete hierarchy, then tested
whether position in the hierarchy was associated with
individual traits in subsequent analyses (Appleby 1983;
Haley et al. 1994). However, individual-specific covariates
(in this case, male traits) can be incorporated directly into
a single structured model of B–T form, thereby circumventing
the problem of having to derive a linear hierarchy
first (Tufto et al. 1998; Firth 2005).We provide a case study
for the application of the structured B–T model for paired
comparisons and discuss its utility and the utility of tournament
designs for experimental studies of male contests