Data collected from partners in a dyadic instructional setting are, by definition, not statistically independent. As a consequence, conventional parametric statistical analyses of change and influence carry considerable risk of bias. In this article, we illustrate a strategy to overcome this obstacle: the longitudinal actor-partner interdependence model (APIM). Participants included 60 girls and 100 boys enrolled in public middle schools, who ranged in age from 10 to 14 at the outset. Students worked in pairs assigned by teachers. At the beginning and end of the instructional period, students completed surveys rating the degree to which the partner was a friend, confidence in one's own computing skills, and computer programming knowledge. APIM analyses revealed partner influence over the acquisition of computer programming skills among friends but not nonfriends. Students with higher initial levels of confidence in their own computing skills were more apt to be influenced by friends. This association was especially strong when confident partners were paired with friends who knew relatively more about computer programming.