About that rat and High Fructose Corn Syrup study.

[elfs]

Ars Technica has an awesome takedown of the methodology, leading to the conclusion that the study is "suggestive" but hardly definitive. I have to agree with the reporter that the photo of the lab students looking "sciency" was indeed awesome, however.

I'm reminded of this comic, only backward.

Comments

[ladyerin.livejournal.com]

I have to agree with the reporter that the photo of the lab students looking "sciency" was indeed awesome, however.

I'm pretty sure they are really making jello shots in those tubes. ;-)

[sirfox.livejournal.com]

speaking from firsthand experience... Labwork just doesn't translate well into interesting photo-ops. 99% of the stuff being worked with is either white or clear/colorless, and there's something about a lab environment where you may as well stage a photo, because anything authentic enough to be believable is also boring as sin.

http://www.adlyfe.com/adlyfe/news_012306.html The person on the left was actually employed by a different company, but was judged the most photogenic of the people on hand, so they asked her to pose with Cindy and Alan. I think they're really laughing about the fact that the "Data" they're "discussing" is really a shipping receipt.

[gromm.livejournal.com]

These days, nearly all science is "suggestive" instead of definitive. And the news media always portrays it as some kind of breakthrough.

[http://users.livejournal.com/_candide_/]

Oh, what are you people talking about.

Having a PhD in physics, I can say the following with absolute certainty from personal experience as well as training:

Science is never definitive.

^_^

Science is the process of taking observational phenomena, looking for commonalities in it, creating a mathematical model to unify those observed phenomena, then whacking the h311 out of that model to see if you can break it.

If the model survives, you deem it good … and keep whacking on it. The longer you whack on it, the bigger a "x% correct" grade you give to it.


That reticence even holds for individual research papers. Presented with a pattern in the data that your average person would declare "certain," anyone worth their doctorate will look at and say, "Hmmm… looks suggestive of causation. We should research it more to make sure it's not just correlation."
Oh! That's what I wanted to mention. In physics (and likely chemistry, biology, and all of the fields of engineering), only correlations above 95% are considered high enough to be causative. Anything less is only vaugely suggestive of a possible causation, the further below 95%, the vauger.

[gromm.livejournal.com]

Oh sure, and don't I know it! But there have been lots of real breakthroughs in science, most of them about a hundred years old or so. Consider the following:

- The development of the theory of general relativity.
- The Polio Vaccine
- Penicillin
- Sulfa drugs
- Proving that perpetual motion is impossible.
- The theory of evolution

When there's lots of low-hanging fruit around, it's relatively easy for science to make big gains. :) Those sorts of breakthroughs were the result of a single person's years of labour (except Thermodynamics anyway). These days, it takes large teams of scientists decades to make that kind of progress.

Also, results lower than 95% are pretty much inevitable in biology and psychology, due to the fact that we're all unique snowflakes and all. That's why we breed lab mice to be very close genetically - so that we can actually prove something.

Also, I suspect that a lot of labs are under pressure to produce press releases, which is why marginal gains make it into the news so often.

[http://users.livejournal.com/_candide_/]

Well, Penicillin was more serendipity. ^_^

Perpetual motion had to be impossible, or Conservation of Energy went out the window. Since everything in physics from Newton onward kinda needs Conservation of Energy (and Momentum), it was a given early on that perpetual motion wouldn't work. (Unless there's no friction, which is a different ball of wax. Oh, and Special Relativity pulled 3 conservation laws into a single Conservation of Matter-Energy/Energy-Momentum.)

Everything else fits into my, "Look at nature. Try to unify things into a common model. Then whack the bajeezus out of the model to try and break it." ^_^

My, "95% or better," statement more applies to statistical data. You can take a dataset containing nothing of points on a circle and fit a line to it that has 70% correlation to that data. This is just one reason why the bar is pretty high when correlating statistical data to a model. When the model is nonlinear, the idea of a "percent correlation" gets tricky; sometimes 95% can be too low.

And, really, percent-correlation is just a convenient number that says, "How much does this data look like noise?" The closer to 0%, the more the data looks like it's purely (Gaussian) random.