Audience Size Projection
We have not finished analyzing the data. We would like to try to model speaker-reported audience size in terms of keywords. We identified certain speakers who attracted large audiences and will try to find ways to accommodate them better in future, but the topic may be as important as the speaker. Furthermore, although speakers vary from year to year, many of the same topics reappear year after year. Nonetheless, we do have some interesting findings about audiences.
The obvious
More prestigious session formats attract larger audiences. Table 10 shows how audience size varies by session type for lecture-style presentations. Presenters, panelists, and discussants in invited and topic-contributed sessions were twice as likely to report audiences of 50 or more than were presenters of regular contributed papers. Having a prestigious session is no guarantee of a large audience. Ten of these presenters reported audiences of fewer than 10 people. Conversely, regular contributed papers sometimes pull in large audiences. However, being in a prestigious session roughly increases your audience by two-thirds.
Measurement difficulties
Chairs and presenters have widely varying estimates of audience size. The gamma association measure between them is 0.56, and the Pearson’s correlation between chair estimates and the midpoints of the presenter interval estimates is 0.58. They vary more at the small end than at the large end. Where speakers saw an audience under 10, chairs saw an average audience of 34 with a minimum of five and a maximum of 110.
It is difficult to imagine how these estimates vary so widely. It seems theoretically possible that everyone deserts the room after a popular speaker, leaving the last speaker or discussant with a tiny audience, but I do not believe I have ever seen such a thing happen. When I ran a multilevel model on presenter estimates of audience size with session as a random effect, I got an intraclass correlation of 0.46. So presenters do indicate substantial variation in audience within a session, but it still seems hard to reconcile the presenter and chair estimates.
Table 11 shows that they do not even agree about when the room is filled beyond capacity. Presenters identified 66 overfill situations. Chairs identified just 44. The two agreed on just 27. If chairs and presenters disagree, should we bring in a third party to measure audience size? Perhaps we could have a network of graduate students tasked with measuring attendance. If we can’t get good measures of audience size, it seems unlikely that we will ever develop good models of audience size that could lead to less unhappiness about room-size assignment.
These tables indicate that presenters at JSM are generally well satisfied with the presentation experience. Those in more prestigious session types are more satisfied, as are those with prior presentation experience, those with audiences in the range of 20 to 200, those in coherent sessions, and those with good chairs. Poster presenters are less satisfied: even those with good audiences and who prefer giving posters. It is clear the regular contributed session has a loyal following. It is hard to see how the session type could be eliminated without seeing JSM attendance drop by hundreds. There are, however, many regular contributed sessions with very small audiences. Perhaps to save convention space, we could restrict regular contributed sessions to rooms with a capacity of 50.
Leave a Reply