I spent most of last week at a big biostatistics conference. Overall it was a great time - there were mixers and free coffee and old friends and new friends and famous keynote speakers and lots of talks. It was also exhausing, due to the huge number of people, the early-morning sessions, and all the talks. So many talks! One thing I noticed about the conference talks was that most speakers with 15-minute research talks (including me) struggled to stay within the time limit. While some of this is on the speakers’ shoulders, I think a 15-minute research talk at a huge conference is incredibly challenging, especially when the conference accepts contributed papers (up to 25 pages long!!!) as talk proposals. It’s really hard to effectively transform a 25-page research article into a 15-minute talk geared toward a fairly general audience.

Here are some features of research papers:

  • describes a project in detail
  • assumes the reader is familiar with the underlying concepts of existing research in the area, and if not, by virtue of being a paper, allows the reader time to go review said existing literature, or at least google unfamiliar terms
  • can introduce complex notation, since readers can refer back to the notation definitions at their leisure
  • usually contains sections Introduction, Methods, Results, Discussion, Conclusions, or slight variants
  • may contain proofs of theorems, regularity conditions that must hold in order for the method to work, supplementary material, tables of simulation results, and other boring but important things

But almost none of this translates well to very short talks:

  • short talks have to focus on the big picture instead of details, because the audience won’t have time to process all the details and if they try to process the details, they’ll likely miss the point of the talk
  • you can’t really assume an audience at a large conference knows anything, except the general topic of the conference which is usually very umbrella-like
  • complex notation is hard to introduce in such a short time, and the audience will forget the notation once the notation slide gets taken down
  • 15-minute talks probably can’t contain five fleshed-out sections
  • spending time on boring things like regularity conditions takes too much time away from overall concepts and main messages, so they really can’t be part of a short talk

So my idea for contributed talks is to bill them as lightning talks!

Instead of having folks submit 25-page proposals and giving them a 15-minute slot for a research talk, have people submit abstracts or proposals specifically for lightning talks. Proposals would not be accepted if the scope of the talk is too large or detailed for five minutes of speaking time. I think taking away the “this is a research talk!” mentality might help speakers prepare better for a short talk. Maybe thinking of the presentation as a quick overview of their work rather than a full-on research talk would help presenters think intentionally about the big-picture impact of their projects and how to communicate that.

There would be some rules for the lightning talks:

  • slides optional
  • only one slide can contain equations
  • no tables with all numeric entries
  • no proofs, unless the entire talk is a proof
  • talks must end with a one-sentence “take-home message”
  • other ideas?

Each lightning talk session could be 100 minutes long and could include 10 lightning talks, each of which would be strictly limited to five minutes. There would be five minutes for audience questions and presenter-switching between each talk. (Compare this to the current format of seven 15-minute talks in a 105-minute session, with no built-in time for questions). You would need lots of friendly-but-firm session chairs to keep the presenters on time. Only the contributed sessions would be these lightning talk rounds; the invited sessions could still be the longer 25-30 minute research talks. (I think that’s a much more reasonable time frame for a research presentation).

I’ve heard of lightning talks being done at places like PyCon, StrataConf, and Hacker School, but never at a statistics conference1. I’m pretty interested in seeing if the format would help speakers with short talks give clearer, more memorable presentations. If that were the effect, I think the whole conference would be more enjoyable, for audiences and speakers alike!

** 1. update, 3/24/14:** My statistician friends pointed out in the comments that JSM introduced SPEED sessions (including lightning talks) in 2013, and that they went over well! Hoping ENAR will follow suit soon :)