Hacker News new | ask | show | jobs
by fleitz 5171 days ago
Could be that it's a fluke given the small groups involved.
2 comments

Not likely, since statistical tests were performed to ensure that the results seen were not due to random chance. The sample size they used is actually pretty large for studies like this.

However, the study is complicated since rats are sensitive to abdominal gavages and it tends to reduce their lifespan. Similarly, there are studies that show chronic administration of olive oil increases life span.

So you have a situation where one control (water only) leads to decreased lifespan and another control (vehicle - olive oil only) tends to increase lifespan. Neither of these are ideal controls since they are known to affect the rat.

That said, they did fancy stats and presumably accounted for this fact.

Those tests for statistical significance aren't very strong in this case. It might be that the study in question only had a 5% chance of giving this result by chance, but given that this particular study was selected to our attention for news-worthiness the actual chance of it being a valid result are far less than 95%[1]. Also, a sizeable minority of published papers have statistical errors, even in good journals[2].

[1]http://xkcd.com/882/ [2]http://www.economist.com/node/2724226

You are right in general, but the study in question actually claims 99.9% significance, not 95%. That's pretty strong.
Ah, thank you for pointing that out. It changes things quite a bit.
An n of six per arm in a three-arm study is not "actually pretty large."
It's just a single paper by a single lab, and probably a pretty small lab. This isn't Merck running a clinical trial.
No amount of fancy accounting can ever save someone from the fact that they performed one experiment one time with a treatment group of 6. Extraordinary claims require at least a replicated experiment.
That's exactly the message my summary intended to convey :-)