Collective Action and Intra-Group Conflict: An Experiment

ABSTRACTIn many situations, two rivals find themselves facing a common threat necessitating cooperation. Such alliances are uneasy as the parties anticipate eventually competing against each other in the future. Recent theoretical work shows that both standalone and joint contribution strategies, as well as a hybrid of the two, can emerge in equilibrium where the former strategy is characterized by one of the rivals contributing enough to eliminate the common threat, while the latter is characterized by rivals providing half the necessary effort to eliminate the common threat. Using a controlled laboratory experiment, we show that player behavior is best described by the hybrid strategy. However, none of these predictions closely describes the observed behavior, which is better described as following a proportionate rule.KEYWORDS: Blotto budgetsIntra-group conflictCollective actionAll-pay auctionLaboratory experimentsJEL CLASSIFICATION: C90D72D74 Disclosure statementNo potential conflict of interest was reported by the authors.Notes1. Evans-Cowley (Citation2005) provides a guide to using zoning ordinances to limit the impact of big box stores.2. See Sheremeta (Citation2018) for a general review of experimental results in group contests.3. Deck et al. (Citation2015) also considers endogenous alliance formation in a Tullock contest. In their setting, an opportunistic attacker observes the strategic decision to form an alliance or not of its potential targets. Theoretically, they show an alliance can increase the expected payoff of the defenders even as it increases the probability of a successful attack because it avoids an arms race between the potential targets. Behaviorally, Deck et al. (Citation2015) find that alliances do increase the expected payoff for defenders but do not increase the probability of a successful attack as defense investment does not drop as much as predicted when an alliance is formed.4. The ‘common threat’ is not a separate player in the game because it is not modeled as having a strategic decision.5. The total number of periods in the hour long sessions is based on the speed of play in a pilot session. Phase 1 involves twice as many periods as Phase 2 due to it being the primary focus of the experiment and to allow for learning in that environment.6. Given the distribution of allotments, the joint equilibrium is unique when b is less than 67.7. If b is in the interval (100, 200) then there is no Standalone equilibrium given the distribution of allotments.8. One observation in which a subject with an allotment of 91 contributed 91 is not shown in the figure due to the scale.9. Appendix B contains plots of individual level behavior in the first half of Phase 1, the second half of Phase 1, and Phase 2.10. One subject chose not to report being male or female. Of the others, 42% were male. For testing risk attitude and cognitive reflection, the sample was split as evenly as possible given the discrete nature of the responses.11. This strategy was identified empirically.12. Using session level observations accounts for the lack of independence between games within a session due to the random rematching protocol. We rely on the non-parametric sign test due to the limited number of sessions.13. The sign test p-values for comparisons between the observed success rate and that predicted when α=1,0.5,and0.373 are 0.0016, 0.0016, and 0.2059, respectively.14. It is worth noting that in Phase 2 only one subject contributes more than the amount necessary to meet the common goal and this is the same subject that frequently contributed more than the common goal in Phase 1.15. This strategy was identified empirically.Additional informationFundingThe work was supported by the University of Alabama.