Association for Public Policy Analysis & Management

Surveys of student perceptions produce multiple measures of classroom quality. Our aim of this chapter is to decide which of these measures are most useful in predicting student learning. Conventional statistical methods can produce misleading results because the measures of classroom quality are quite highly correlated. We therefore introduce a new method, the Multilevel Variable Selection Model, and we apply this method to the Tripod survey of student perceptions, which provides seven indicators of classroom effectiveness based on twenty-eight items. We find that classrooms that are well-controlled and intellectually challenging produce comparatively large learning gains. Our new methods can readily be extended to study the combined contribution of student perception data, classroom observation data, and other measures to student learning gains.