[CELS VII, held November 9-10, 2012 at Stanford, was a smashing success due in no small part to the work of chief organizer Dan Ho, as well as Dawn Chutkow (of SELS and Cornell) and Stanford's organizing committee. For previous installments in the CELS recap series, see CELS III, IV, V, and VI. For those few readers of this post who are data-skeptics and don’t want to read a play-by-play, resistance is obviously futile and you might as well give up. I hear that TV execs were at CELS scouting for a statistic geek reality show, so think of this as a taste of what’s coming.]
Unlike last year, I got to the conference early and even went to a methods panel. Skipping the intimidating “Spatial Statistics and the GIS” and the ominous “Bureau of Justice Statistics” panels, I sat in on “Internet Surveys” with Douglas Rivers, of Stanford/Hoover and YouGuv. To give you a sense of the stakes, half of the people in the room regularly use mTurk to run cheap e-surveys. The other half regularly write nasty comments in JELS reviewer forms about using mTurk. (Oddly, I’m in both categories, which would’ve created a funny weighting problem if I were asked my views.) The panel was devoted to the proposition “Internet surveys are much, much more accurate than you thought, and if you don’t believe me, check out some algebraic proof. And the election.” Two contrasting data points. First, as Rivers pointed out, all survey subjects are volunteers, and thus it’s a bit tough to distinguish internet convenience samples from some oddball scooped up by Gallup’s 9% survey response rate. Second, and less comfortingly, 10-15% of the adult population has a reading disability that makes self-administration of a survey prompt online more than a bit dicey. I say: as long as the disability isn’t biasing with respect to contract psychology or cultural cognition, let’s survey on the cheap!
Lunch next. Good note for presenters: avoid small pieces of spinach/swiss chard if you are about to present. No one will tell you that you’ve spinach on a front tooth. Not even people who are otherwise willing to inform you that your slides are too brightly colored. Speaking of which, the next panel I attended was Civil Justice I. Christy and I presented Clusters are Amazing. We tag-teamed, with me taking 9 minutes to present 5 slides and her taking 9 minutes to present the remaining 16 or so. That was just as well: no one really wanted to know how our work might apply more broadly anyway. We got through it just fine, although I still can’t figure out an intuitive way to describe spectral clustering. What about “magic black box” isn’t working for you?