Really, Illinois?

[eve11]

Really?

REALLY?

(Not even one statistics class?)

ETA: Oh, hey, what do you know, I can pay $30 to find out whether this research is as full of crap as I think it is. Thanks, Elseveir.

Comments

[ladymercury-10.livejournal.com]

...I thought the U of I was supposed to have all the very clever mathematicians in our state.

[eve11]

They apparently don't have to take intro stats.

[ladymercury-10.livejournal.com]

On the one hand, I really wouldn't expect them to. On the other hand, it becomes very embarrassing when people assume stats is a subset of math and that they can therefore ask stats questions of unqualified mathematicians, or worse, when mathematicians decide to make stats proclamations.

[eve11]

We're not talking really advanced mathematical statistics here. We're talking about a really basic concept that should be understood by anyone who knows what "multiple regression" is. And by anyone who employs it.

Unless they're economists.

[ladymercury-10.livejournal.com]

Yeah, but I guess if they assume they don't have time for ANY stats because they have to power through several years of complex and real anaylsis or something to get to their dissertation stuff, and then by the time they decide they want to do stats they are Mathematicians and Assume They Know Stuff...it is still not a very good excuse, I don't know.

Why did they have a mathematician/computer scientist doing their stats in the first place?

[eve11]

This to me is just a prime example of blindly employing techniques because they want to be published, and not really understanding what is behind those techniques. Honestly, it may be that if I bothered to pay the money and read the article they actually attempt some kind of quasi-experiment. But the data sets that they mention.... I just absolutely cannot figure out how one might infer causation in a regression that uses that data. They are looking for interesting stuff in big data sets; I am betting they don't even really care about the results that they are purporting. Because if they did, they would have figured out how to collect data that truly can be used to answer that questions.

[kalypso-01.livejournal.com]

this....


http://sphotos-a.xx.fbcdn.net/hphotos-prn1/59776_10151475013117464_8252905_n.jpg

[eve11]

The thing is, I saw something similar last year in our HR newsletter. It's ridiculous. The claim may indeed be true, but hey, ever think about oh.... doing an experiment to test it?

This person isn't just wrong on the Internet. They are wrong and sharing it in something that espouses to be a peer reviewed journal. Cats and dogs, living together.

It at least has given me a nice slide for my presentation though.

[kalypso-01.livejournal.com]

i know - it is worse than just an idiotic comment troll. but the graphic is still funny :)

[eve11]

I'm going to have to go read about ice cubes to cool myself down. ;D

[kalypso-01.livejournal.com]

that is one of the funniest pages i've ever read on the internet.

[a-phoenixdragon.livejournal.com]

*Headdesk*

[abyssinia4077.livejournal.com]

*headdesk*

ETA: Oh, hey, what do you know, I can pay $30 to find out whether this research is as full of crap as I think it is. Thanks, Elseveir.

AUGH academic publishing. I'm on campus - if you want I could email it to you.

[eve11]

I'm on campus as well, but apparently my campus doesn't subscribe to that journal?

Don't worry about it; I will be able to request it from our librarian here if I want to read it.

[last01standing.livejournal.com]

So, my school apparently has access to this journal. Drop me your email and I'll forward you the PDF if you like.

[eve11]

Thanks; you can send it to my username at livejournal dot com. I will likely save it for when I feel like being outraged ;)

[last01standing.livejournal.com]

Done. Anytime (in the next three months before graduation at least) you need me to fetch a rage-inducing paper feel free to ask.

[eve11]

Thanks again :)

[giallarhorn.livejournal.com]

I don't understand why they thought it was a good idea for a CS and math prof to examine the data for correlations? Shouldn't that have been someone like a stat prof?

Plus, it never seems to account that BMI is a pretty horrid indicator of actual health, and definitely not the most sound source of data if you want to write an academic paper/study.

Also LOL at the comment.

[eve11]

They answered my comment! I asked, "how did you get to causal claims" and they said, "We cite other researchers and draw conclusions from them". spastic_visions sent me two of their citations and I skimmed them, and they make no causal claims.

Honestly, it is not hard to agree with the idea that a little bit more exercise per day can help you lose weight. But this is really, really not the data set to try and prove that on a national scale.

[lostrack621.livejournal.com]

Wow, just wow. I am no statistician (and I will rely on you to teach me what I need to know bwahahah) but that "study" makes ZERO sense to me.

I've recently read several studies about health risks related to sedentary office work (e.g., sitting at a desk for hours on end), but to my recollection, those studies actually used data related to people who had experienced heart attacks, etc....I'm not sure about the statistics in those studies, but I sure know that treadmill desks or standing desks seem to be becoming more popular!

Also, exercise balls as chairs (I use one much of the time - helps me keep proper posture and spine alignment). What I would really love is one of those "kneeling" chairs (for lack of better descriptor) where there's a cushion for your bum and then two cushions for your shins, so that you're actually sitting with your pelvis tilted forward (properly) and you're able to maintain proper spinal alignment, prevent vertebrae compression, and all kinds of other good health things. Unfortunately, that type of chair costs $100+. :(

/randomness. ;)

[eve11]

Yeah, there is also probably a risk along the lines of developing DVT or something like that? But the gist of the difficulties I have with this report is that it is essentially vacuous information. They stopped asking "Why" at entirely too early a point in time. Then they basically did the alt-text to that xkcd comic above: "Correlation doesn’t imply causation, but it does waggle its eyebrows suggestively and gesture furtively while mouthing ‘look over there.'"

[wiliqueen.livejournal.com]

I almost always blame this sort of thing on the journalism rather than the science.

Since this article's lone actual quote from an author of the study starts with "We're saying that" and includes several qualifiers, my very strong hunch is that it was in response to a question something like "So you're saying that all you need to do to lose weight is drive less?"

On the other hand, that last quote from the prof in charge makes it sound more than a little like he has a definite agenda, which makes me curious about what assumptions may have gone into the construction of the computer model. :-/

[eve11]

In many cases, I agree, it is journalists who misinterpret or want to sensationalize claims. But the scientists bear some of the responsibility in this case. The problem is that the specific causal claims they make in the paper cannot be supported from their data. They list caveats but then they state that there is in any way a meaningful interpretation of "reducing average travel" for people being correlated with a subsequent drop in BMI. The rest of this comment is for me more than you: I was thinking about it on the drive home today after I read through some of their citations.

For example they say "eliminating one mile of automobile travel per day for each licensed driver would be associated with a 0.21 kg/m^2 reduction in the national average adult BMI in six years." There is no way to truly determine if that is the case from the data they collected. It is also only one possible way to manipulate driving data to produce a new average. (Reducing driving time a lot for the extreme tails of the distribution will also reduce the national average driving time; would that be associated also with a subsequent decline in average BMI as well?) There is no support for the magnitude of their claim. In the past, I for one have had to drive further to park in a place where I would increase my activity levels by taking public transport and walking in.

That "would be associated" in their language is them trying to get around the inherent causal claim that they make by using the verb "eliminating". They are saying, "If we tweak the system to reduce road travel, then subsequently we will see a reduction in BMI in six years."

This is causal language, whether or not they say it is associational. I can paraphrase from Judea Pearl and Bryant Chen's recent tech report ("Regression and Causation: A Critical Examination of Econometrics Textbooks") to that effect:

E[Y|do(x)]: The expected value of Y given that we do x can also be interpreted as the expected value of Y in an idealized randomized experiment when the independent variable X is set to x. Clearly E[Y|do(x)] does not necessarily equal E[Y|X] (the expected value of Y given X, ie, the predictions produced from a regression analysis). For example, the expected amount of rain given that a barometer reads "low pressure" is different from the expected amount of rain given that we set the barometer reading to "low pressure".

I asked on the page how they supported causal claims and they said they cited other literature. I looked at two papers they cited as supporting literature. Neither were causal. None discussed model fit or showed any residual diagnostics from their models (and one had R^2 values of ~0.16 despite having "significant" p-values on the coefficients they cared about). Neither of the cited studies made causal claims, nor performed experiments. Neither reported population standard deviations along with their significantly different means; and one for example listed a difference of 23.7 BMI for average "good land use" persons vs 24.5 for average "bad land use" persons. This survey had a response rate of 27%.

Nor did the prior studies control for any kind of self-reported gym or workout time. One study recorded the number of children under 18 per person surveyed (which seems to me might be a good indicator of free time, desire for convenience, and life priorities, but did not include that information in the model and did not say why they did not. They did include marital status. Which one would be more indicative of possible underlying differences among suburban vs. more urbanized populations?