Hacker News Reader

Comment by ffffruit | Hacker News Reader

There are statistical ways to ensure that the results obtained from a study are both a) not due to chance and b) statistically significant to be applicable to larger sample sizes. Hopefully, both were applied and evaluated during the peer review process when a study gets published. I could potentially extract the actual data from the studies but alas, traditional media consider it a sin to link to the original papers and of course, a simple 'recent studies' is sufficient for us common people.

Edit: spelling, clarity

View on HN

The data is not being questioned, the conclusion is.

I was answering to the "Correlation is not causation" statement in the comment. Epidemiological studies usually have to satisfy most of the "Bradford Hill" criteria in order to claim a certain association.

More information: http://www.drabruzzi.com/hills_criteria_of_causation.htm

Would you be willing to make a wager that their p-value < 0.001?

You need to perform an intervention to test for correlation vs. causation, and that would involve getting lots of normal children, and traumatising half of them. Those studies are considered a bit retro in the scientific community, for some reason.

I do not think they are considered retro. Interventional studies are widely considered the "golden" standard but they are not perfect and, in many cases, not feasible. Like you point out, in many cases, it would not be ethical to perform one ; for example, making people smoke on purpose and then try out smoking cessation interventions on them. "Luckily" though, people usually self-select their selves into these categories.

Judea Pearl disagrees.

Links about Judea Pearl:

http://www.danielpearl.org/about_us/Judea_Pearl.html

http://ftp.cs.ucla.edu/pub/stat_ser/R320.pdf

The usual link to Peter Norvig (director of research at Google) on experimental research, which is good food for thought:

http://norvig.com/experiment-design.html

He disagrees with the idea that traumatising children is retro, or with the causation vs causality? It's been a few years since I read the book, but I had his do() operator in mind when I wrote the comment.

Edit: Okay - I clearly have to clarify my comment. Intervention studies are not retro. Studies in which you deliberately traumatise children, however, went out with Ceacescu. I was making a joke (albeit apparently a poor one).

Reply on news.ycombinator.com