6 September 2017This paper investigates how differences in military spending translate into probability of victory in a conflict. To do so the paper empirically estimates and compares the four main forms of contest success functions – the Tullock, the logit, the difference, and the relative difference form. In order to circumvent measurement issues and endogeneity biases associated with historical battle-related data, we advocate the use of virtual worlds. Data from virtual worlds is an innovative and promising tool in conflict research. It allows conflict analyses based on rich and objective empirical evidence on economic behavior in a warfare context; something which is difficult to achieve in real world settings. Thanks to collaboration with the developer of a virtual world, we are able to construct an original database of 19,229 battles that occurred during January 2011. The results show that the relative difference contest success function as proposed by Bevia and Corch6n (2015) outperforms other existing forms of contest success functions. Thus, the decisive factor to predict the outcome of a battle is the relative difference in forces devoted by rivals. Relative size matters. (C) 2018 Elsevier B.V. All rights reserved.