Cyclist0108

https://arstechnica.com/science/2018...mon-diet-tips/

Worth a read, just for the puns.

canklecat

In an interview on NPR several weeks ago a researcher criticized data dredging, although I don't recall whether he pointed specifically to Wansink.

I'm wondering whether it's a mistake to disregard apparent patterns in data. Sure, it's a mistake to publish findings representing these as statistically significant or indicative of anything that might influence people's health choices.

But the fellow in the interview sounded a bit zealous, as if it were a sin against science to even consider data dredging to look for issues that might merit further study. He sounded a bit like those psychologists who say we're overwhelmed by consumer choices, so we're better off with fewer choices. No, thanks, I'm capable of choosing what I need, want or like without false monopolies being presented as some sort of neo-consumerist advocacy.

Cyclist0108

The problem is expectation bias vs. statistical noise.

Some apparent correlations really are meaningless.

OBoile

Agreed... but the comparison you showed here makes perfect sense. If you can't have a nice steak, what's the point of it all?

redlude97

I know this is a joke, but the graph doesn't go to zero. Plotted on a full scale axis the dip in beef consumption looks a lot less meaningful. Plenty of steaks would still be consumed at 58lbs

OBoile

Ah, but 58 lbs is the average. What if some people are consuming the same amount, but some cut it out entirely... and therefore have no reason to live?

Cyclist0108

What if the relative rate of suicide by handgun per age group is constant, but more people are living longer because they eat ethically? Then the rise in suicide absolute numbers is a statistical artifact, due to people living longer because they didn't consume cow carcass.

DrIsotope

And it only gets worse from here. They stopped eating beef, they starting eating organic, and bam! everybody's got autism.

Cyclist0108

Here is possibly the most frightening part of the article:

So, this guy is a Prof. of Marketing and he is publishing nutritional and dietary statistical studies in JAMA?

redlude97

Fake news

bruce19

So, the Autism "epidemic" has caused an uptick in organic food sales?

Cyclist0108

Statistics never lie.

himespau 

I just got done giving a test (and trying to explain the results) where a whole lot of my students didn't understand that, when faced with data that doesn't support your hypothesis, you can't just make up a new hypothesis that fits your data perfectly and then go and publish that as is. You still need to test your new hypothesis (usually with new/different controls and possibly a new design) if for no other reason than to show your findings are reproducible (but also because you probably didn't control for all the right things perfectly when your first hypothesis was different).

himespau 

Wow, I didn't see that. No wonder he was trolling large datasets for deviations. Isn't that the whole point of directed micromarketing or whatever the hell it is that they call that stuff?

Cyclist0108

Why not teach them that they can publish the refutation of the original hypothesis?

Seattle Forrest

This is absolutely wrong, what that chart is telling us is that organic food causes awareness of autism and increased diagnosability.

bruce19

That was supposed to be a joke, right? OTOH, organic food doesn't actually cause autism.

Seattle Forrest

Yeah, that was supposed to be a joke.

OBoile

Doubtful. The problem with your theory is the association with avoiding beef and living longer.

Cyclist0108

I've heard the word gullible is not in any dictionary.

Road Fan

In this thread, Internet Brands should be signing some kind of release with the "Journal of Irreproducible Results!"

HarborBandS

It sounds nuts at first glance, but a lot of modern marketing research is statistical analysis of large data sets.