« DIC stuff | Main | Causal inference and regression, or, chapters 9, 10, and 23 »
December 7, 2007
Statistical consulting mini-symposium next month
Posted by Andrew at December 7, 2007 4:33 PM
Trackback Pings
TrackBack URL for this entry:
http://www.stat.columbia.edu/~cook/movabletype/mt-tb.cgi/1328
Comments
Posted by: Peter at December 7, 2007 6:30 PM.
Posted by: Andrew
at December 7, 2007 8:25 PM.
"Some Recent Progress in Simple Statistical Methods"
Suspect the strong bias against simple methods in applied statistics journals is still there - but is that changing somewhat?
Keith
Posted by: Keith O'Rourke at December 7, 2007 10:43 PM.
to Keith:
My favorite professor in grad school said that he noticed a trend. People who knew very little statistics used simple methods, because they knew nothing else. But they used them incorrectly. People who knew a little more statistics used fancy methods, but used them incorrectly, when simple methods would do. And, he said, people who knew a lot about statistics often used simple methods, but did it right
Posted by: Peter at December 8, 2007 10:44 AM.
Keith,
Right now I think the best journals for simple statistical methods are Technometrics, Statistics in Medicine, and the American Statistician. If there's a bias against simple methods in other journals, I'm perhaps partly to blame, as I'm on the editorial boards of several of them.
One of the difficulties is that, although I like simple statistical methods, and I like _my_ new simple statistical methods (that's what I'll speak on next month), I don't always like the new methods that _other people_ develop. In grad school we joked that every issue of the American Statistician featured a new proof that the median is between the mean and the mode, or some new criterion for outliers. That's the kind of simple method I can do without.
More generally, I think what is needed is bridging between simple and complex methods, so that when they differ, we understand how this happened. My impression is that this used to be a standard feature in statistical analysis (for example, the "partial correlation coefficient") but has gone away. We do some of this bridging in our recent Jasa article on police stop-and-frisk, and we discussed it a bit in our recent book on regression and multilevel models, but I think there's room for more research on the topic.
Posted by: Andrew
at December 8, 2007 11:21 AM.