Is lecture attendance just a flip of a coin?

In the sciences, the content of sequential lectures is often planned under the assumption that the majority of students present in the class have attended the previous sessions. To some extent, students are expected to be able to recall and understand the previously delivered content. During a study into the effectiveness of a novel lecture engagement strategy (Interactive Lecture Demonstrations, or ILDs) in a first-year introductory electronics unit (Mazzolini, Daniel, & Edwards, 2012), student-generated codes were used to anonymously track individual students' responses to the ILD activities over successive lecture sessions. These codes offered an unintended proxy for attendance, with surprising results. To what extent can lecturers assume that the same set of students attend each lecture, and therefore construct a lecture sequence that builds lecture-by-lecture on previous work? To enable anonymous tracking of student responses over multiple lecture sessions, students were prompted with a set of 5 questions to generate an anonymous but unique identifying numerical code at the start of each session. For example, one of the prompt questions used was 'what is the last digit of your postcode?'. These codes, along with other student responses, were electronically submitted using audience response devices (i.e. clickers). By comparing which codes were reported in each of the sessions, patterns of attendance were established. Similar anonymously coded student data were collected from other units. Over the four lecture sessions for which data were collected in 2011, only six codes were reported in every session, implying that perhaps only six students attended all four sessions. The overall pattern of attendance for the students enrolled in the unit (N=148) can be coarsely approximated by a binomial probability distribution, with the probability of attendance being ~35%. Data collected from several other sources validate this initial coarse approximation. Analysis of student performance is reported elsewhere (Daniel, Mazzolini, Cadusch, & Edwards, 2012) but, unsurprisingly, students who attended all the instructional sessions showed higher learning gains, as determined by pre- and post-test scores, than students who hadn't attended any sessions. Lecture attendance is surprisingly inconsistent. The approximate fit to a binomial distribution means it is almost as if each student flips a slightly weighted coin in deciding whether or not to attend each lecture. It is often assumed that if the number of students attending lectures is approximately constant, then roughly the same group of students are attending each lecture. This assumption about the same group of students attending successive lectures underpins the common practice of building on the understanding of previously presented material in each new lecture. This assumption, and the purpose and efficacy of the lecture mode more generally, have to be questioned.