This dissertation presents three essays in labor economics and risk. Chapter 1 examines how past effort can impact current effort, such as when effort is reduced following an interruption. I present a series of real-effort incentivized experiments in which both piece rates and leisure options were manipulated and find effort displays significant stickiness, even in the absence of switching costs. I demonstrate that this intertemporal evidence is indicative of effort “momentum”, rather than on-the-job learning, reciprocity, or income targeting. When employing an instrumental variables (IV) approach, approximately 50% of the effort increase persists for 5 minutes after incentives return to baseline. Thus if a worker suffers a complete interruption in productivity, it would take an average of 15 minutes to return to 90% of prior work effort. I further demonstrate that advanced knowledge does not significantly reduce this productivity loss. Chapter 2 examines how risk preferences differ over goods and in-kind monetary rewards. I study an incentivized experiment in which subjects allocate bundles of either Amazon.com goods or Amazon.com gift credit (which must be spent immediately) across uncertain states. Under a standard model of perfect information of prices and goods available, I demonstrate risk preferences across these treatments would be identical. In practice, I uncover substantial differences in risk preferences across goods and in-kind monetary rewards. With additional treatments, I find no evidence that these differences are driven by price or product uncertainty. Chapter 3 is joint work with David Dillenberger, Daniel Gottlieb, and Pietro Ortoleva. We study preferences over lotteries that pay a specific prize at uncertain dates. Expected Utility with convex discounting implies that individuals prefer receiving x in a random date with mean t over receiving x in t days for sure. Our experiment rejects this prediction. It suggests a link between preferences for payments at certain dates and standard risk aversion. Epstein-Zin (1989) preferences accommodate such behavior, and fit the data better than a model with probability weighting.