Ratings are critical to the function and success of services in the emerging sharing economy. They are a means through which users develop trust in one another and in the services themselves. Ratings are designed to give users a proxy for the expected quality and risk of potential online transactions. We expect online ratings to reflect an objective measure of quality, but such evaluations in fact may be systematically distorted by many, complex social-psychological processes. Decoupling these subjective factors from rating systems to correct for biases and to provide neutral assessments of risk and quality has proved extremely challenging. We focus on one of the most prevalent factors in virtually every form of social exchange. Differences in resource ownership affect the balance of power in interpersonal interactions, likely impacting online ratings. We demonstrate how power imbalance affects mutual ratings using a massive dataset from CouchSurfing.org, an international online hospitality exchange network. Our methodology employs a deductive approach to knowledge discovery. Through a series of observational experiments, we find support for a sociological theory dating back to the 1960s, Power-Dependence Theory (PD), as a possible explanation. PD predicts that power-imbalanced relationships induce user behavior that attempts to balance power. We find support for status-giving as a likely mechanism driving the asymmetry of ratings between power-unequal users. Our findings underscore the need for ratings systems to account for the tendency of mutual ratings between users that hold differential resources to be asymmetrical, especially under conditions of resource scarcity.