An often implicit assumption of perceptual deterrence tests is that the elicited values pertaining to arrest risk reflect stable underlying beliefs. But researchers in other disciplines have found that reported expectations are highly susceptible to exogenous factors (e.g., anchors and question ordering), indicating that such values are somewhat arbitrary responses to probabilistic questions. At the same time, reported expectations are coherent within persons, such that respondents rank order them rationally. For deterrence, then, absolute values reported on arrest risks are likely not stable but individuals still rank order specific crimes in meaningful ways. We examine the interpretability of reported arrest risk for three possibilities: 1) Reported risks are stable probabilistic values; 2) reported risks are arbitrary and uninformative for deterrence research; or 3) reported risks display “coherent arbitrariness” with unstable values between individuals but stable rank ordering of crimes within individuals. Through the use of three random experiments of college students, our results indicate that elicited risk perceptions are arbitrary in that they are influenced by the presentation of anchors and question ordering. Nevertheless, the rank ordering of crimes within and across conditions is unaffected by the presentation of anchors, suggesting that reported risks are locally coherent within persons.