Isn't UCLA among the top universities in the world? How can they publish a research like this (asking a few questions to 35 people) and draw such formidable conclusions? The title sounds like they've discovered something new that applies to all humans...

It becomes quite evident in studies like this, but nearly all social science related studies are based on the exact same pattern. For instance the emotional intelligence study, which has been repeatedly cited and been reasonably influential, was based on putting people into groups, having them do tasks such as [literally] writing a shopping list, and then showing that the groups that did the tasks most efficiently were not strictly composed of those that had the highest IQ. So therefore it was implied that individual merit alone does not determine results of groups, but rather some 'emotional intelligence' does.

It's absolutely atrocious logic since B is in no way implied from A. For instance to show this then you would need to, at the minimum, put the people that all individually do the best at e.g. writing a shopping list, into a group, and then compare them against people that individually do worse but score well on the [again literally] 'Reading the Eyes in the Mind' test that was used to measure 'emotional intelligence'. Of course that was not done, and I don't think it's unreasonable to speculate because we know what the result would be - and that's not publishable. Yet bad science paired with an astounding headline and you are suddenly good to go.

Yeah, a bit of a rant but it's incredibly disappointing how awful the social sciences have become. And that decline pairs ironically with a massive increase in the reporting of their results, which become ever more grand and increasingly clickbaity, in the media.

Are you equating IQ with merit? If so, you need to broaden your perspective.

My opinion is not relevant one way or the other. I am describing the study.

That's actually how statistics works. You don't need 100,000+ people to make a readily generalizable conclusion.

If the differences are significant, they'll appear with as few as 35 people, carefully selected, and give you a confidence level of X that it applies to a population size of Y.

Whilst I don't know the details of this particular study, I do recall doing homework problems about correlation confidence.

In layman's terms, maybe this example will help:

Imagine a study where they found that 100% of people who were "sitting" had bent knees. Essentially no one could bring their hips and ankles closer together without bending their knees. With such a significant correlation, you could generalize from a population of 10 to one of trillions so long as the ten was randomly chosen, and there were no other correlative factors.

Aside from things like false legs and injury bringing your hips and ankles together like that is a physical impossibility; that doesn't provide a good analogue. People sit with straight legs too, but if you test by asking people to sit on a chair it will show everyone sits with bent knees, where in fact lots of people sit on the floor, beach, or bed (say) with straight legs. FWIW I sit on a chair with straight legs, on top of my computer tower.

If you take a sample of 35 people in China you can with a reasonable confidence level say all people are Chinese, or perhaps a very small proportion aren't, depends how lucky you get. It's wrong of course, despite Chinese being the majority (17% according to Wikipedia).

It's fine as a preliminary, but it doesn't show anything useful yet other than the need to perform further study IMO.

A report on this should say something like "viability study suggests sitting and brain changes are related but doesn't provide enough confidence as yet". Which it sorta does, from this section on:

>"This study does not prove that too much sitting causes thinner brain structures, but instead that more hours spent sitting are associated with thinner regions, researchers said."

But either ScienceDaily or UCLA have not lead with that information, matching the current style.

Sampling is studied as a way to save time without running a census for every experiment. Many universities use this technique. An intro to stats course could probably be found on Coursera that will go into this.

Self-reported data? That isn't verified? That's nearly always a red flag.

How about using a set of long haul truckers? In a before vs after sort of why? Their hours behind the wheel is documented (or should be). I would presume they're sitting behind the wheel :)

In other words, tiny sample, results likely don't generalize, may also be false positive.