Pasta, BMI, and Simpson's Paradox

I try to be unbiased when I do these 150 second analyses, but full disclosure here: I am an unabashed pasta partisan. My favorite?

Midnight Pasta: Thank me later.

Midnight Pasta: Thank me later.

 Midnight pasta – so called because you can make it after a night on the town with only the ingredients you have lying around – spaghetti, olive oil, garlic, parsley, red pepper flakes, and some parmesan. So good.

So I was thrilled to see news reports like these telling me that eating pasta will help me lose weight:

What these articles are all talking about is this study, appearing in the journal Nutrition and Diabetes.

The researchers used data from two large Italian cohort studies totaling a bit over 20,000 individuals. The methodology varied by study a bit, but basically they were relating self-reported dietary intake to Body Mass Index.

And what they found was that people who ate more pasta had higher BMI's.

The good old "news reports report the opposite of the findings" gambit.

Wait… what? That is literally the opposite of what the news reports show. Are we really supposed to swallow this? Was this study funded by Big Semolina?

Actually, no, and the reasons why illustrate some of the unique challenges in dietary research.

I’ve said before that food frequency questionnaires are terrible and unreliable – but that’s not the most interesting problem. The interesting problem is that heavier people tend to eat more.

This isn't a problem if you're looking at total calories consumed, and multiple studies have confirmed that, broadly speaking, the more calories you eat, the heavier you are.

That's a might-nice R-squared value there, Lou.

That's a might-nice R-squared value there, Lou.

But when it comes to parsing the contribution of individual food sources or nutrients, the situation is much more difficult.

So, let's look at pasta. I've already shown you that more overweight people eat more pasta. But there is more to the story than that.  I’ve generated some fake data to illustrate the point.

Here’s a hypothetical set of data points relating pasta consumption to BMI. 

Keep your eyes on the dots...

Keep your eyes on the dots...

No surprise – people with higher BMI’s eat more pasta.

But now I’ll color code the dots by weight categories.  Yellow dots are the people who weigh the most, green people who weigh the least. Remember that weight is different than BMI. You can weigh a lot, and still have a normal BMI (provided you are very tall).

Same dots! What wizardry is this?

Same dots! What wizardry is this?

What you see is that, for any given weight range, those who eat pasta have a lower BMI than those who do not. Now that’s a spicy meatball. It’s also one of my favorite statistical quirks called “Simpson’s Paradox”.

Mmmm... data.

Mmmm... data.

There are a variety of statistical approaches to account for this, but basically they all boil down to assessing how much of your total energy intake comes from pasta. It’s closer to thinking of pasta as a percentage than pasta as an absolute amount. And it’s through a statistical manipulation like this that the researchers got their delicious results.

But what does it mean?  Hopefully, you can see that it is not quite so cut and dried (and boiled) as more pasta = more weight loss.  No, it’s closer to say that eating more pasta as a percentage of your total calories is associated with lower BMI status. In Italy. Where pasta is like a nice small appetizer with some veggies and olive oil.

So no extra pasta for you folks. But hey, replacing this:

With this:

Might just help this:

Sorry for the cheesy puns this week, folks.