User input to recommendation systems such as Netflix provide an excellent opportunity to study human choice and preferences. We present a probabilistic model that captures two processes that underlie human input to recommendation systems; the process by which individuals choose items to rate, and the process by which they select a rating for those items. Using movie rating data collected by Netflix, we demonstrate that this model can generate accurate predictions about missing movie ratings. Furthermore, we show that the implicit information that users reveal through their choice processes can be used to improve prediction accuracy even in the total absence of explicit ratings.