Causal Decision Theorists (Gibbard & Harper 1978, Lewis 1981, Joyce 1999) think that an act's expected valueis determined by what that act is likely to cause. Evidential Decision Theorists (Jeffrey 1965, Ahmed 2014) disagree and think that an act's expected value is determined by the evidence that act provides about the world. In this talk I argue that Causal Decision Theorists face a problem from decision instability: Causal Decision Theory may recommend an act that foreseeably looks worse than some available alternative. 

I restate Reed Richter's original (1984) challenge from decision instability and provide an updated version of that argument. I claim in particular that causalists are subject to money pumps in cases of decision instability. I then review one class of solutions: require that agents not just maximize expected utility but maximize a particular kind of expected utility (Weirich 1985, 1988; Rabinowicz 1989; Gustafsson 2011). The idea behind these solutions is that rationality requires agents not just to do the act they expect to do best, but to do the act they expect to do best and that meets some further 'stability' constraint. 

My claim is that these solutions do solve the paradox of decision instability, but at a price. In particular, such solutions give rise to a menu-dependent decision theory. This means the causalist faces a dilemma: accept the paradoxes of decision instability or give up on the idea that an act's expected utility is determined by what that act is likely to cause.