Psychologists note that social cognition often involves the creation, refinement, and use of models of one's interactive partners. The influence of categorical thinking on interpersonal expectations is commonly referred to as a stereotype. Using an algorithm that we created for stereotype learning, we investigate problems that can occur when the robot acquires its first models of people and learns its first stereotypes - the robot's early social development. We examine if the errors related to the creation of these initial models have a disproportionate impact on the robot's developing social skills, perhaps even reflecting some of the same challenges faced by humans . We hypothesized that errors in which the robot interacted with someone that did not represent the true nature of a category, an outlier, would impact the robot's performance on a social coordination task more if the error occurred earlier in the robot's social development rather than later. Results from simulation confirmed our hypothesis. The results of this work have potential implications for social robotics, autonomous agents, and possibly psychology.