Peer Instruction observed in the wild

\u27Flipped\u27 or \u27inverted classroom\u27 approaches have been shown to enhance the effectiveness of teaching contact time. We have adopted this pedagogy in our introductory physics classes, in which we teach a mixed cohort of around 300 major and non-major students at a large, research-intensive university. A crucial element of class activity is the Peer Instruction methodology, in which students discuss and defend their conceptions of topics, with levels of student understanding gauged using clicker questions. We have studied student behavior during these Peer Instruction sessions using \u27smart pen\u27 technology, which allows us to match audio-recorded student conversations and pen strokes with their clicker votes in an unobtrusive way. We find that the large majority of student conversations result in students settling on the correct answer to the post-discussion clicker questions, and that the success of these conversations does not depend on the technical fluency of the students\u27 vocabulary. The smart pen recordings also allow us to assess any problems or ambiguities in the clicker questions themselves, thus affording the opportunity for informed revision of pathological clicker questions. Over two iterations of the class we have successfully modified a number of clicker questions in light of this feedback and find them to have correspondingly increased learning gains. Thus, this authentically-situated \u27in the wild\u27 study of student behavior has benefits both in terms of increased theoretical understanding of the instructional approach and also in direct refinement of the classroom resources, i.e. more learning with less confusion