Could this have anything to do with obesity?
I used to joke that English was a second language for me … my native language was really 8086 computer machine code. And I once had a significant other half-jokingly tell me, “Thank god you’ll never have to make a living based on your communication skills.”
Well, apparently, both of these jokes were not too far off the mark. I’ve had email and PM requests for clarification re yesterday’s post. Sorry if it was too confusing. Here’s some addendum to that post; please refer back to it to refresh yourself on what we’re talking about.
In the text below, the italics are questions and comments; the non-italicized material is mine.
> How come your last obesity map was 2007? Isn’t that cherry picking?
Oops! Sorry. No, I was just too lazy to get more images. Here’s the very latest image available, 2013. The trend is most definitely alive and well and continuing:
Note that the 2007 image I showed had over 26% obese as the highest category, whereas in 2010 there were 11 states with more than 30% obese; here in 2013 – three years later – it looks like 20 states and, also, we have a brand, new category – over 35%!
> I would have liked to see more on exactly how the findings from the studies in that book were manipulated to make the claim that 78/108 of the studies found that intentional weight loss increased mortality.
Campos’ claim mentioning that 108 number was not 108 studies, but 108 associations in a single study(!) – the Iowa Women’s Health Study. So, Campos wasn’t doing a meta-analysis of the published literature, and even if he were doing that, you can’t just tally them up, you have to weight them by study size. I have the Campos book requested from the library, so I will have a chance to delve into it further, but I doubt that he says anything more in the book as to what he did. Math is boring, incomprehensible to most, and just doesn’t make for a good read.
> It’s not exactly clear in what way he’s misrepresenting the data.
My guess is Campos found the raw data tables from the Iowa study and simply totaled up the % of deaths in each association category. When he found the proportion of categories with a higher death rate among the dieters to be greater than 50%, my guess is he said to himself, “Aha! I knew dieting was bad and this proves it!” But that a) is confirmation bias and b) is NOT the way to handle statistics correctly. Oh, if only statistics were that simple!
> I think you say it in that variables weren’t properly controlled for (maybe many of the people were trying to lose weight because they were already sick from obesity) and the lack of statistical significance from some of the results, but those points were kind of scattered.
Yeah, well what I didn’t emphasize enough was problem #1 with what Campos did was that he didn’t provide the P-value for 108 associations which would reflect both the number of people in each category and the number of categories. Not only that, there’s no consideration of how association categories might contain non-independent variables. And until you do that, you have no point of reference. That’s why it’s bogus. The more associations you add, the higher P gets for a given cohort size. And unless your P-value is low enough, your inference is worthless. So, just to be clear, as you increase the number of associations you are looking at, the requirements for making your case become increasingly more stringent. That’s why it raised a red flag for me – he was excited about 108 associations without mentioning P!
Problem #2 with it is that he doesn’t do an honest survey of ALL the research, as was done, for example, in the textbook I quoted. That practice is called cherry-picking.
resolvingthecontroversies 