Considerable research has focused on understanding variation in reproductive skew in cooperative animal societies, but the pace of theoretical development has far outstripped empirical testing of the models. One major class of model suggests that dominant individuals can use the threat of eviction to deter subordinate reproduction (the 'restraint' model), but this idea remains untested. Here, we use long-term behavioural and genetic data to test the assumptions of the restraint model in banded mongooses (Mungos mungo), a species in which subordinates breed regularly and evictions are common. We found that dominant females suffer reproductive costs when subordinates breed, and respond to these costs by evicting breeding subordinates from the group en masse, in agreement with the assumptions of the model. We found no evidence, however, that subordinate females exercise reproductive restraint to avoid being evicted in the first place. This means that the pattern of reproduction is not the result of a reproductive 'transaction' to avert the threat of eviction. We present a simple game theoretical analysis that suggests that eviction threats may often be ineffective to induce pre-emptive restraint among multiple subordinates and predicts that threats of eviction (or departure) will be much more effective in dyadic relationships and linear hierarchies. Transactional models may be more applicable to these systems. Greater focus on testing the assumptions rather than predictions of skew models can lead to a better understanding of how animals control each other's reproduction, and the extent to which behaviour is shaped by overt acts versus hidden threats.