Jonathan Golob, you are my favorite addition to Slog EVER.

Correlation is not causation, thanks for reminding us of that. In other news, gay people often have gay friends and gay partners.

The other question is the quantities in these sub-questions. For example, let's say it was a study of 250 cases and 250 controls. But if we then sub-categorize them, to the point that, say "mutual friends" is only 4 respondents, how much meaning does the 200-400 percent increased probability of obesity impact have?

I would say pretty close to zero.

And many people in older communities, to follow Matthew's example @2, have gay roommates that are their partners, owning property together etc, but are not themselves gay.

The world is kind of like squishy jello - it's hard to nail it to the wall sometimes.

morbidly obese. LOL that term cracks me up.

being friends with lame-o fratboy types also causes the spread of lame-o-ism.

I'm no scientist, but I think your critique is off the mark. The researchers did actually take this into account (to some extent at least) by measuring weight gain, rather than static obesity. They are not looking for obese people who became friends, but friends who became obese (or, rather, what happened to you when your friend became obese over time). Or, in their words, "The use of a lagged independent variable for an alter's weight status controlled for homophily. The key variable of interest was an alter's obesity at time t+1. A significant coefficient for this variable would suggest either that an alter's weight affected an ego's weight or that an ego and an alter experienced contemporaneous events affecting both their weights." That doesn't completely pin down cause and effect but it does get closer to it than you give the authors credit for.

the study also followed 5,100 subjects, not 250 (and 12,000 total connected people), and I assume no claims were made about effects among statistically insignificant amounts of people. Peer review isn't a _total_ joke.

Valium b,

You are correct on that point; the authors indeed did look at the change in weight, so it cannot be as simple as "obese people have obese friends."

The authors, in their discussion, were far more responsible than the lay press has been. Take this section of the discussion:
"...changing the ego's norms about the acceptability of being overweight.. Unfortunately, our data do not permit a detailed examination."

The speculation made it into the lay press, but the caveat didn't.

I think the problem here is the word "suggests." It has a very different meaning for scientists, a sense that there is something worth studying here. To the press-release and lay-press writers, it seems to mean "proven to an extent that merits 150-point headlines above the fold." This is what makes me cranky.

yr totally right about the lay press--I just though the original authors were making a decent attempt to confront some of the complications you were mentioning. But what does the average headline writer (or reader) know of confidence intervals?

there is a 95% chance of correlation.

I believe this is incorrect.

A traditional confidence interval means that if you were to repeat this study 100 times and recompute confidence intervals each time, you should expect the true parameter (increase in risk of obesity, in this case) to fall within 95 of those intervals.

A Bayesian credible interval would have the interpretation you suggest, but the authors did not perform a Bayesian analysis.

actually...

this was my original interpretation - maybe fat people just like to be friends with fat people.

HOWEVER...

this is part of the framingham [sp?] study, in which a large number of subjects have been followed over the course of several decades, with weight being one of the outcomes measured at multiple points. in studies like this, where you observe changes over time, you can in fact deduce a cause-effect relationship. this would not be the case in a cross-sectional/single-point-in-time study, where one could only note a correlation between variables.

All you Americans do is talk about fat and you all just keep getting fatter. You’re all obsessed with calories and now you have made friends and family fattening, it´s their fault you are fat?

It´s the stupidest thing I have ever heard and it makes no sense.

So, danny, there's a 95% chance that this is one of the intervals which includes the true value, but not a 95% chance that the true value is within the interval.

Statistics is awesome.

@10. Fair enough -- I was certainly simplifying the stats into an ugly few words. How about "there is a 95% chance that the true relationship is somewhere on the horizontal black line" as a compromise?

@11. Indeed, this is a better study than just a single cross sectional study; if it wasn't longitudinal, I hope it wouldn't make it into the NEJM. Still, this is far from a randomized prospective study, or a clear demonstration of causality.

God, I love the slog commenters -- such excellent and thoughtful opinions.

I think @13 got it spot on. There's a 95% chance that this line belongs to a set of lines that contain the true value.

In a strict (and somewhat strawman...) frequentist interpretation, there is no partial probability that the true value lies on this particular line: it either does or it doesn't. Frequentist probabilities can only be interpreted by appealing to the frequency of an event, in this case, sampling populations.

Bayesian probabilities can be interpreted as a degree of belief, which would allow us to say there's a 95% chance the true value sits on this line---but that's a statement about our belief, not about the value itself.

Makes sense. Most normal sized people don't want to look at hang out or be seen with fat people, so one would assume they would self sort to each other. (Not to mention, they would have more in common with each other… “yum pie!”) Kind of like how poor people mostly know poor people and rich people mostly know rich people.

Lesson of the day: Just as the weather report seems to be wrong about half the time, when it comes to headlines about scientific studies, or even Slog posts about headlines, there is at least a 50% chance that the results will be be completely misrepresented.

As others have said, by using the half-century of prospectively collected repeated measures on a multi-generational cohort and looking at change in obesity, this is far closer than a "cause and effect" study than you give it credit for. No, it is not a controlled laboratory experiment, but I'm sure that you realize that such a thing isn't possible.

The NEJM made their article available (PDF) here.

A few notes that I haven't seen in the mainstream media:
- During the study, participants were asked to name a "close friend" who would have their address should the researchers have trouble contacting them. This was the basis for the "close friends" connections in the computer model.

- The increased chance of obesity between friends is actually between men. The researchers state that the increases between female-female and mixed-sex friends were "not significant". So it's only in male-male friendships.

According to the NY Times, researchers stated "that a person who became obese gained 17 pounds and the newly obese person’s friend gained five". Meanwhile, Jeffrey M. Friedman, head of the Laboratory of Molecular Genetics at Rockefeller University in New York, has said: “people can exert a level of control over their weight within a 10-, perhaps a 15-pound range" - which is roughly where we're at here.

I'm fairly skeptical of computer models, if only because I spend my work days finding bugs in them :) But one advantage of this being a model based on open-source software, and the availability of the Framingham data, is that other researchers can look at this and improve it over time. Or completely cook the books. Hell, it's software :)

URLs:

NEJM: http://content.nejm.org/cgi/content/full/357/4/370

NYTimes:
Catching Obesity: it's on the main page.

The Fat Epidemic: He Says It's An Illusion:
It's behind the TimeSelect firewall, but another copy is at http://math.hunter.cuny.edu/mbenders/Fat.txt

well you do not need a study like this just go to a gay bar.
Bears hang out with bears and are overweight and promote it.
Skinny twinks hang out with skinny twinks and strive for slimness through peer pressure.

I have no doubt that friends a family can influence one another in a variety of ways. But in the end, it comes down to this:

You are responsible for yourself. You are supposed to be a thinking, rational being, capable of making your own choices. If you choose to be influenced in a particular way by your friends and family, the responsibility is still on your own head.