For Scientists, a Beer Test Shows Results as a Litmus Test | Hacker News Reader

For Scientists, a Beer Test Shows Results as a Litmus Test

nytimes.com

15 pointssanj9 comments

I'm not a statistician, but if I read the paper correctly ... there is barely a negative correlation in his 2006 study (if you remove the one significant anamoly). n=18 in his 2002 study and n=16 in 2006. He controlled for age and years of active publishing which may be correlated to consumption. In fact, he conveniently removed the 1st principal component in the 2002 study and not in the 2006 and covered it by an offhand remark of "almost quantitatively identical". If they were almost quantitatively identical why did he remove it from one study and not the other. Oh! Well, earlier in the paper he mentions that the results were only consistant between the two studies when he controlled for age. The median consumption of the 6 bohemians in his 2002 study is 200 litres/per year(!!!) and the remaining 12 Moravians drink 37.5 litres/per year (which is quite the bimodal distribution). The 2006 study were the same people as in 2002 "where available". So n=16 for the entire study. They are all researchers in "avian evolutionary biology and behavioural ecology" in the country with the highest per capita beer consumption rate in the world. Hardly a meaningful dataset in my eyes. Then again, maybe I'm biased in my analysis. Beer drinkers unite!

So, it documents a correlation without explaining it, so maybe there is no correlation. You cann't take two facts and say there is a correlation, may be it simply a coincidence.

Huh? Did you read past the headline? Listen:

The results were not, however, a matter of a few scientists having had too many brews to be able to stumble back to the lab. Publication did not simply drop off among the heaviest drinkers. Instead, scientific performance steadily declined with increasing beer consumption across the board, from scientists who primly sip at two or three beers over a year to the sort who average knocking back more than two a day.

That's not "two facts" - it's a direct correlation between two variables. One may not cause the other, but it certainly doesn't appear to be a complete coincidence

And that's why I got out of physics!

As far as the correlation goes - There's a clear correlation there, though it doesn't seem like they're trying to propose a causality. Though they kind of bill it as a causality at the beginning of the article, I don't think the actual study is suggesting that drinking more or less beer will directly affect how often you're published.

The most interesint paragaph is at the end:

"More important, as Dr. Grim pointed out, the study documents a correlation between beer drinking and scientific performance without explaining any correlation."

So, it documents a correlation without explaining it, so maybe there is no correlation. You cann't take two facts and say there is a correlation, may be it simply a coincidence. Maybe the difference is not the beer but the time spent in bars. Maybe the scientists that get less papers publisheds are depressed and go to drink beer to forget their bad performance.

How can people publish a fallacy like that and be so happy?

It’s good data and interesting ground for future research. As long as people don’t misinterpret it, I don’t see the harm.

>How can people publish a fallacy like that and be so happy?

Beer.

This seems apropos in light of demo day:

http://news.ycombinator.com/item?id=140628

I'm skeptical, but as a non-drinker, I can't say I'd be sad if avoiding drinks has some upside.

Interesting that it was a quantity measurement, not a quality measurement (which would be very hard to do, I imagine) Seems obvious that if you're out drinking beer you're not writing :)

Reply on news.ycombinator.com