We develop a micro-founded framework to account for individuals' effort and cognitive noise which confound estimates of preferences based on observed behavior. Using a large-scale experimental dataset we find that observed decision noise responds to the costs and benefits of exerting effort on individual choice tasks as predicted by our model. We estimate that failure to properly account for decision errors due to (rational) inattention on a more complex, but commonly used, task design biases estimates of risk aversion by 50% for the median individual. Effort propensities recovered from preference elicitation tasks generalize to other settings and predict performance on an OECD-sponsored achievement test used to make international comparisons. Furthermore, accounting for endogenous effort allows us to empirically reconcile competing models of discrete choice.