## Through a Glass, Darkly

I must start reading Dr. Helen more.

She’s right to point out that physicians as a whole are not particularly interested in statistics – even medical statistics.

Funny, talk to most doctors and they will tell you that only MDs can prescribe because they “know all that calculus, stats and stuff.” Really? I’ve never seen a doctor do any calculations to write a prescription. Now, I’ve learned that many of them don’t know how to interpret a piece of research thoroughly. That really breeds confidence.

And she should know, for she’s seen enough of them.

She quotes from the Yale Alumni Magazine (you must search for Don’t Know Much About Biostatistics by Rhea Hirshman):

Almost every medical school student takes a course or two in biostatistics to learn how to understand research data. But Donna Windish, an assistant professor at the School of Medicine, has shown that the information often doesn’t stick. “A significant percentage of physicians-in-training do not understand the statistics they encounter in the medical literature,” she says.

That’s pretty ugly.

I used to teach undergraduate physics. If I had to do it over again, there’s one thing I would change big-time, and that’s how “labs” were done. In my undergraduate days, I remember working (um… wrestling with) incredibly antiquated tubes and rheostats and odd meters of all sorts. There was a reason for that, above and beyond the fact that most schools were loath to modernize their equipment, leaving students in the early ’70’s to use state of the art instruments circa 1930. The reason is that we were actually *doing* cutting edge physics – from 1890!

The change I would make is this: I would tell the students to prove, oh, V(oltage) = i*r**2 (that is, voltage is the product of current times the square of the resistance. That’s wrong. It’s not (it’s V=i*r, actually). But they must prove me wrong (the reward, of course, is a good grade). To do that, they would need to design an experiment, one that would allow them to nail down that relationship. The equipment they need would be the same; a way to measure voltage, a way to measure current, a way to measure resistance.

But that’s not the science. After designing the experiment, they also need to collect the data, and then prove to me it’s “real” (by examining why it doesn’t fit their prediction exactly). When they finally figure out that V does not equal i*r**2 (but fits the curve V=i*r much better), then we’re in a position to talk about possible errors and how they creep in there.

Sorry – all that is introduction. Dr. Helen’s post got me thinking about my weight, and how, when I plot it, it’s a noisy, straight line. Every time I try to weigh myself regularly, (every day or so), I get inspired to go on a diet and lose a few pounds. Inevitably, after a couple of weeks or so, the graph makes a nice straight line, sloping in a downward direction. Being a geek, I compute the least-squares fit to a straight line and measure the residuals and immediately I’m impressed with the direct relationship, something like (2000 minus calories-consumed)/3500 cal per lb = lbs lost per day.

This formula predicts that by next May, sometime in the middle of the month, I’ll weigh 25 lbs.

Cool. I’ve inspired every anorexic in the country!

Actually, what’s happened is that I’ve imposed a curve on raw data, and then committed several cardinal sins of science. First, I put a “curve” to the data, instead of letting the data show me what it is. I based all my assumptions, therefore, on the curve, not on the data. Then, I extrapolated waaaayyyy into the future, well beyond the data.

Not only the medical profession, but most every reporter should be made aware of this. Anyone who calls himself a scientist and hasn’t seen first hand the pitfalls of using statistics badly should be whapped on the knuckles by the nearest nun with a ruler and given remedial courses.

**Explore posts in the same categories:**Science

November 19, 2007 at 9:50 pm

Here’s a direct link to the alumni magazine article: http://www.yalealumnimagazine.com/issues/current/findings.html#4

November 21, 2007 at 5:20 pm

Thanks, Anon!