Selecting dominants and subordinates at conflict outcome can confound the effects of prior dominance or subordination experience.

Individuals with a previous experience of dominance are likely to be dominants in further encounters. To test this effect, individuals with a previous experience of dominance are used for the experiments. One way to obtain such individuals is to let opponents 'self-select': encounters between pairs of more or less equivalent opponents are staged and one selects ex post facto the dominant and subordinate from the ensuing conflict. This paper formally shows that the selection of dominant and subordinate animals modifies the dominance probability functions of the two corresponding sub-samples of animals. As a result, the propensity of previous winners to win again and of previous losers to repeat their loss can be attributed to this artefact rather than to prior social experience. This result has serious methodological implications. When one relies solely on selection to obtain winners and losers, equiprobability is no longer the appropriate null hypothesis against which prior social experience effects have to be tested. To clearly demonstrate the effect of dominance experience, one must show that prior winners defeat neutral opponents in at least two-thirds of all cases; reciprocally, to show that prior subordinate experience plays a role, prior losers must win in less than one-third of all fights against neutral opponents; finally, to conclude that a combined effect of the two kinds of prior experience is in operation, one must obtain that prior winners defeat prior losers in more than 83% of all planned conflicts. The present result does not imply that experience effects are not at work when the selection procedure is used, but that the procedure used to show their effects is inadequate because effects of experience on a subsequent encounter are confounded with those introduced by statistical selection.