Friday, March 16, 2007

Confounding

My favorite statistical concept is confounding. According to the wonderful Wikipedia, a confounding variable is "an extraneous variable in a statistical or research model that affects the dependent variables in question but has either not been considered or has not been controlled for."

It's always been my opinion that, in the social sciences at least, most connections that are drawn between conditions are plagued with confounding.

Take the graph above - from the National Center on Addiction and Substance Abuse at Columbia University report released yesterday: Wasting the Best and the Brightest: Substance Abuse at America’s Colleges and Universities. Doesn't it seem likely that those individuals who are more likely to join Greek houses are also going to be, relative to the average student, potentially more social? After all, joining a Greek house is an active choice made by that student. And realistically, isn't drinking a part of being social in college? Those are clearly summary statements, but I'm just speaking in averages.

I'm not condoning binge drinking, and I do believe that the group mentality that Greek systems foster can lead to absolutely inappropriate behavior, but I just don't like statistics that seem plagued with confounding.

Confounding comes up everywhere you look. Once you're aware of the concept it seems easy enough to poke holes in every argument. Next time you hear someone propose an inappropriate summary connection, ask them if they've considered all the confounding variables. Likely they'll look confounded.

1 Comments:

At 7:23 PM, March 16, 2007, Blogger Anne said...

If you read the report closely, though, they do note:

"It is important to note, however, that although binge-drinking rates have remained relatively stable over the past decade across various demographic groups, including gender, race/ethnicity, age and most types of college residences (e.g., substance-free residence halls, off campus housing), the main exception has been among residents of sorority and fraternity houses where binge drinking has declined from 83.4 percent in 1993 to 75.4 percent in 2001."

 

Post a Comment

<< Home