3 min read

Seminars, bad and good

Rafael Irizarry recently proposed that the standard 50 min seminar be reduced to 25 min, with the extra time devoted to discussion. I commented there but should have just written at greater length here. I’ll do so now.

Bad seminars

The overriding characteristic of a bad seminar is a lack of empathy for the audience.

  • Skipping the motivation and jumping into a pile of theory.

  • Covering way too much material too quickly.

  • Talking about something that is boring and/or trivial.

  • Mucking about in details, particularly regarding simulations results.

  • Going over time, even by a minute.

Self-censorship is important. If you don’t have something new and interesting to say, talk about something old and interesting or don’t talk.

Good seminars

An interesting topic (particularly, interesting data with an interesting question) trumps almost all else. I can be entertained with drivel, but afterwards I’ll realize it was drivel.

Show the data first. Skip the outline slide and jump into the data. With the problem understood, you can then explain how you’ll spend the rest of our time.

Focus on one thing and explain it well. I don’t want to see everything you’ve done in the last year or two, especially if it means it’ll be too fast to be understood; I just want to understand one thing that you’ve been thinking about.

Know your audience. Know what you should explain and what you can leave unexplained. (If in doubt, explain it. Most people are happy to hear a nice summary of something they know well, and if you lose them, you’re not likely to ever get them back.) Know what people will find interesting and what they will want skipped over. This will be different for different audiences.

Skip the details of simulations results. Since you can modify the settings to get almost anything, I don’t really want to see detailed power curves (and definitely don’t show me tables); skip to the bottom line.

Show lots of pictures, a few words, and almost no formulas. (Rafa has said to use no Greek letters, and has also said to use only pictures. I say: at least find a way to dispense with the subscripts.)

Label your figures, and not with x[,1] and x[,2]. Don’t copy-and-paste from a paper (unless perhaps it’s to demonstrate how crappy the paper is; that’s entertainment!).

Give a succinct summary at the end, in case the audience has lost track.

25 minutes?

I’m warming to the 25 min concept, but I’m still cool. I can generally see how to edit others’ talks down to 25 min, but my own? I feel like I need that time. But perhaps that’s because

  • I work on such obscure problems that I have to spend 20 min on background.

  • I’m lazy.

  • I’ve been trained to talk for 50 min.

  • I just like to talk.

Still, it’s difficult or impossible to retain audience members’ attention for much more than 25 min, and I can edit even my own talks to 25 min if I make a few approximations.


If we’re going to cut seminars to 25 min, let’s not fill out the hour with questions and discussion. Let’s just go home early (or chat over coffee).

The discussion period of seminars is seldom so interesting that it deserves so much time. Of course there are exceptions, but usually it’s a few people asking lame purposeless questions.

Perhaps, though, the questions are lame only because we’re all tired and want to go home. After a refreshing 25 min talk, maybe we’ll be keen to delve deeper in an unstructured way.