Statistics Redux

Being as I lamented recently over the ease with which statistics can say, oh, anything you want, I thought I'd point out to everyone an example of rigorous results of a poll/survey/study. I'm not dissing Gallup or Knight-Ridder, or at least, not actively. I have no data, y'see?

Hope And Despair For The Obese

The article actually describes two studies, one having to do with self-rated 'quality of life' information from obese children, and the other related to the efficacy of a known drug in a new role -- weight loss. I'm going to talk about the numbers in the latter, and compare them to blanket statements like "40% of Americans believe the it's the earth, not the moon, that's made of green cheese."

Firstly, the obesity study is quite clearly peer-reviewed, as a publication in the Journal for American Medical Association.

Secondly the obesity study lists the group running the study, the lead researcher, the number of participants, their average weight, age range, and gender. The relevant article reveals that the diet of the participants was monitored by a dietitian, lowered by 500 calories (for the control and the experimental groups), and participants were encouraged to increase their activity levels.

These details are important to any scientist who wants to prove or disprove the -- admittedly preliminary -- finding that the drug in question may contribute to "signficant weight loss". Why? Because anyone who wants to duplicate these results needs to know the parameters of the experiment.

This is why it's impossible to take an off-the-cuff '40% of anything' statement at anything more than 1/20th of the value of the paper it's printed on. Because you don't know anything. But this obesity study? You know enough to replicate the experiment and support or contradict its results.

That's the difference between "science", and "making shit up", friends and neighbors.