How to read scientific journals: TV and obesity...maybe good science, maybe bad.

There are a lot of common mistakes in science. One is the tendancy to take "associated" phenomenon and claim they are "causal". Take this headline:

The No. 1 Way to Predict Obesity

How much TV do your kids watch? Based on that number, you can accurately predict if they'll be fat adults.

Researchers at New Zealand's University of Otago have found a direct correlation between the number of hours children and young teens watch television and the likelihood they'll be overweight or obese by age 26. Kids who are couch potatoes are likely to become obese or overweight adults. While this isn't the first study to link television with weight gain in children, it is the first to find that TV can be a better predictor of obesity than even diet or physical activity.

Before I go off on either the media spin or the scientists, let's differentiate their claims. The news article claims " If you want to reduce your child's chances of becoming obese, you should restrict their viewing time to no more than 2 hours a day, say the researchers. "

The title of the paper is: Watching television is associated with childhood obesity: but is it clinically important? (International Journal of Obesity, 2005) At least the scientists know how to phrase the question.

Obviously the medical condition of obesity has numerous factors. Genetics plays a large role along with an number of other 'lifestyle' factors including diet, exercise, etc... Could TV time be an indicator of obesity? Could it be a cause? Sure it could...but nothing in this entire study suggests anything other than an association. An association that is stated in the title. (Talk about begging the question...)

I watched a lot of television growing up, and still grew up to be nicknamed Lanky. But I'm no expert. Here's what the scientists say:

Although the aetiology of overweight and obesity involves a complex mix of environmental and genetic factors, this huge increase at the population level must be largely driven by changes in children's diet or levels of activity. Time spent watching television may be related to both.

Compare that to what I said 2 paragraphs ago. Ok, we're all on the same page. The scientists go on to claim that they found a correlation of 0.09-0.33 (correlation coefficient) even after regressing out sex, socio-economic data, and the father and mother's BMI. What on Earth does that mean? See this site for an illustration of corrlation coefficients. (Attention Political Readers! While I am definitely getting technical on your ass, I will still feed you a political fix. Check out this related page, especially if you've heard of Dan Drezner or want to see how government contracts vary with political donations)

There's sufficient evidence to show an association or a correlation between BMI and TV watching. However, I will again cite Juha Puranen "The correlation coefficient is a measure of linear association between variables - not a measure of the strength of causality." And if you read between the lines of the paper, the authors know they haven't established causality.

In summary, we provide further evidence of an association between television viewing and childhood BMI

...

Although the effect size appears small, the correlation between television viewing and BMI is stronger than reported corelations between BMI and diet or physical activity. We believe that watching television is an important contributing factor to the current epidemic of childhood obesity. [emphasis added]

When you read a scientific paper, and it uses the word "believe", it means, "this is what we personally think, but have not proven". It is also telling that they had access to the parents BMI. That was one of 3 variables worth regressing before, and yet no mention of whether or not that is more or less correlated than TV watching. (hint: you only regress things if they have a correlation). Overall, this paper is a B. It's not great, but it's not wrong. They should have written the title and left it at that.

On to the media article: Their title: "The No. 1 Way to Predict Obesity" was not actually claimed in the paper. Their suggestion "If you want to reduce your child's chances of becoming obese, you should restrict their viewing time to no more than 2 hours a day, say the researchers. " never appears in the paper (though they may have spoken with the authors). The claim: "These kids were also more likely to grow up to smoke and have high cholesterol" is unsupported by the data in the paper (same caveat). And finally, the assertion : "The number of hours they spent in front of the TV correlated almost exactly to their body mass index. " is profoundly wrong. It is correlated weakly at a high confidence.

Of course watching more TV isn't going to help one lose weight, but let's be realistic, obesity probably causes TV watching. It's a great activity if you're overweight, have low self-esteem, are anti-social, depressed or have any number of other problems (now correlate those with smoking...). Does it reinforce laziness? Probably. Does laziness correlate with obesity? Probably. But it is premature to claim this as a causal relationship. And I think it is irresponsible to suggest to parents that their kids won't be fat if they deprive them of TV. Far far better that you suggest to parents they get their kids some exercise. While on that note, here's a suggestion for parents of kids who play too many video games. It's also a fun drinking game.

Update:
I was conjecturing about the parental influence. (I should have used the word "believe"). Here's some evidence: Obese parents are the greatest influence for childhood obesity.