In this chapter we outline and discuss statistical approaches to the analysis of contest data, with an emphasis on testing key predictions and assumptions of the theoretical models described in Chapters 2 and 3. We use examples from an array of animal taxa, including cnidarians, arthropods and chordates, to illustrate these approaches and also the commonal-ity of many key aspects of contest interactions despite the differing life histories and morphologies (includ-ing weaponry) of these organisms. We first deal with the analysis of contest outcomes, a useful approach for determining which traits contribute to an indi-vidual's resource holding potential (RHP). Here we outline alternative statistical approaches that treat the outcome as either an explanatory (independent) variable or as the response (dependent) variable. In both cases, we treat a single contest as one 'experimen-tal unit' and consider ways in which multiple measures taken from the same experimental unit should be accounted for in the analysis. Thus, we introduce paired and repeated measures approaches for contest data and also the calculation of composite measures. We then discuss more complex mixed models, which are particularly useful for dealing with multi-party contests when multiple individuals from the same group occur in more than one observation. Having established what factors influence RHP, one might then ask questions about the roles of information-gathering and decision-making during contests. These questions are prompted by the theoretical models of dyadic contests discussed in Chapters 1 and 2, and we consider the advantages and limitations of using analysis of contest duration to distinguish between 'mutual-' and 'self-assessment' type contests. An additional tool that we can use to address this question is the analysis of escalation and de-escalation patterns , and we thus shift the focus to within-contest behavioural changes. As well as discussing changes in rates of agonistic behaviour we introduce 'sequence analysis' , which is a tool for analysing transitions between different behaviours. Thus, it is possible to investigate changes in agonistic behaviour in both quantitative and qualitative terms. We then turn to the wider social setting within which many contests take place and consider the range of methods currently used for analysis of dominance hierarchies. We then return to the idea of information and describe an approach for analysing the potential for bluffing during a contest. This question of signal honesty is one that pertains not only to contest behaviour but also to the wide range of contexts where signals are used to advertise some aspect of individual quality. Finally, we suggest some key points that should be considered when designing an experiment on contest behaviour. 4.2 Introduction Understanding contests from an evolutionary perspective is complicated by the fact that they involve at least two individuals and the benefits that accrue to one contestant will be influenced both by its own behaviour and by that of its opponent. When contests take place between opposing groups of individuals , the situation is complicated further because outcomes will be determined not only by the size of each rival group, but also by the behavioural interactions of the individuals within each group. For these reasons simple optimality models are often insufficient to account for the full complexity of animal contests. Therefore, approaches based on game-theoretic modelling , which deals with decisions made in the presence Animal Contests, ed. I.C.W. Hardy and M. Briffa.