Ergative case is said to mark transitive subjects, and it is widely assumed that this is true under the ordinary definition of transitive; however, Bittner and Hale (1996) propose that ergative languages fall into two types, neither of which is based on the ordinary notion of transitivity. In one, a direct object is not necessary for ergative case: any verb with an external argument counts as transitive, following Hale and Keyser 1993 (e.g., Warlpiri). In the other, a direct object is necessary, but not sufficient: the subject gets ergative case only if the object moves out of the VP (e.g., Inuit). This article argues that Niuean, Dyirbal, and Nez Perce are also of this object shift type. A search yielded no language where ergative case is clearly governed only by ordinary transitivity; languages that do fit the stereotype have only ergative agreement. A formal account of the correlation between object shift and ergative case is proposed, under which ergative case can be used as a ''last resort,'' as one of three ways to avoid the locality violation that object shift creates.