It wasn’t a bad dream. Last week the global press was actually debating whether it’s ok to associate with fat people based on a study, which was published (and heavily promoted) by the New England Journal of Medicine. For those of you with the good sense not to follow this kind of news, the report purported to show that obesity spread from friend to friend by fat friends changing each other’s norms of what weights are “acceptable.”
Sandy Scwarcz and Kate Harding, among others, did a beautiful job of debunking this study’s conclusions from different perspectives (the former from a sciency perspective, the latter from a position of pure common sense).
So it’s been said, and said better. But I want to add just a few more things I’ve noticed about what’s wrong with this study. There are so damn many things wrong, that it’s easy to miss a few.
Friends and Relatives
First, you couldn’t have picked a group of “friends” that were more likely to be genetically related if you made a concerted effort. Both the subjects and their “friends” were all children of the original Framingham Heart Study cohort, so their parents were all living in Framingham, Massachusetts sixty years ago. The “friendships” analyzed were narrowed to people who a) this group listed as contacts, b) were also in the Framingham heart study, and c) were not first degree relatives.
In other words, these “friends” were very likely to be second and third degree relatives – cousins, aunts and uncles – and my guess would be that the more likely they are to list each other as mutual contacts, the more likely they were to be related.
Genetics Doesn’t Work that Way. Really.
The researchers wave away the general possibility that genetic similarities could have influenced their results by making a fundamentally wrong assumption: that genetics is a single, static effect. Which is why they believe that adjusting for initial weight excludes the possibility of genetic similarity as a factor in weight change as the decades go by.
I can’t emphasize enough how critical that assumption is to their entire analysis. And how wrong it is. Different genetic effects influencing weight kick in at different points over the lifespan from infancy to old age. So, for example, two relatives with totally different BMIs in young adulthood could share genes that make them both prone to gain weight or distribute it in a certain way in middle age or prone towards frailty in old age.
So that’s a simply wrong understanding of the genetics of weight regulation that underlies their entire analysis (and has a lot to do with their not being obesity researchers).
An elaborate hypothesis covers a multitude of evils
The other obvious problem, which Kate Harding and Sandy zero in on, is that their complicated data analysis spat out downright bizarro results:
- At the start of the study, having a fat friend actually reduced your chances of being obese (They call it the “Laurel and Hardy effect”). In fact only weight change, and not fatness or thinness per se, had any positive association between “friends.” (To discover this interesting detail you had to look in a supplementary appendix which needed to be downloaded separately).
- People who listed each other as contacts (defined by the authors as “mutual friends”) were more likely than any other relationship to have a correlated weight change. But only among men. Female friends didn’t have a significant impact on each other’s weight change.
- Opposite sex friends and opposite sex siblings didn’t have a significant impact on each other’s weight change.
- Same-sex siblings, spouses and nonmutal friends all seemed to have some impact on each other’s weight change but the difference in impact between these groups doesn’t appear to be significant.
- Neighbors don’t impact each others’ weight change.
- Geographic distance has no importance. So the impact is the same if your “friend” lives with you, next door, or 3000 miles away. So apparently eating meals together or sharing recreation have no influence on weight change.
It’s a mess. My first impression – and I think the impression of common sense — is that all this is an artifact of the methodology (which is researchese for “you did it wrong.”)
Instead, what the authors did was come up with a flexibly vague, elaborate and speculative hypothesis to try and explain each of the details of the mess.
Here’s how it goes: apparently friends in 1971 had different-sized friends because fat friends like to be with thin friends, and vice versa. This changed in the intervening years because: “Whereas obesity has been stigmatized in the past, attitudes may be changing.” One of their two citations for that unlikely assertion? An article on the size acceptance movement. No, I’m not kidding.
So apparently over the next thirty years as one man falls prey to the nefarious influence of NAAFA (in Framingham Massachusetts, where I happen to know nobody’s ever heard of NAAFA – it ain’t no Northampton)….other male friends say to themselves “hey, I guess it’s OK to be fat.”
Opposite sex friends and siblings are immune to each others’ influence because they don’t see each other as role models. Apparently female friends don’t either. Spouses and siblings take fewer cues from each other than friends do. Shared behavior or shared environment have no role whatsoever because…well they just don’t.
It’s a lot to swallow. Or, Occam’s razor: They did it wrong.
Because It’s Just Not Directionality if it’s Not Significant
The authors make a LOT of the “directionality” of the effect, claiming friends who list you as their friend are less likely to impact your weight than friends whom you list. In other words, you’re more likely to take cues from people you think of as friends than people who think of you as a friend.
That direcionality, they argue, rules out the possibility of confounding variables, and is pretty much the sum total of their justification for arguing that this is about influencing attitudes, and has nothing to do with genetics or anything else.
And that would be interesting. If the “directionality” were statistically significant. But it’s not. The researchers don’t actually admit this, but you can basically eyeball it just by looking at the overlapping confidence intervals for the two types of friendships.
It’s hard to imagine that the entire thesis of an article in a peer-reviewed medical journal is based upon a result that is statistically meaningless. But that appears to be the case.