Hacker News Reader

Comment by ultrasaurus | Hacker News Reader

One thing that constantly saddens me about statistics is that a large amount of energy is expended using is almost correctly to "prove" something that was already the gut feel. Even unbiased practitioners can be lead astray [1] but standards on how not to intentionally lie with statistics are very useful.

[1] http://euri.ca/2012/youre-probably-polluting-your-statistics...

View on HN

There's no way to tell whether or not that "gut feel" is accurate without proof. Often it's right, but occasionally it's very, very wrong (cancer risk and Bayes theory provides a good illustration: http://betterexplained.com/articles/an-intuitive-and-short-e...). Consequently it's still worthwhile proving things even when they're seemly obvious.

"Consequently it's still worthwhile proving things even when they're seemly obvious."

Not true. It's all about risk vs. payoff. Some things are low risk enough we can go by gut, others we need more evidence. It's all about tuning for false positives and negatives.

EDIT: Added quote of what I was responding to.

He said he's saddened by people using statistics "almost" correctly to prove what was already "gut feel". I can't quite tell whether his real concern is what you thought he was saying (i.e., wasted effort), or whether it's the "almost" (but not quite correctly) part, i.e., that researchers use statistics to wrongly prove the gut feel. If it's the latter, then I think a big part of the problem is that people don't understand statistics well enough, not that they intentionally misuse it.

I think his point was that people seem to "prove" common sense statistically all of the time - but when doing so make a lot of thoughtless assumptions about representativeness, significance, definitions, etc. stemming from the unspoken assumption of a particular outcome being inevitable.

Or maybe I'm projecting?

One of the functions of the prior in Bayesian analysis is to incorporate this "gut feel" into your calculations. Given that you have a strong prior belief and weak data (ie not much data) your belief will strongly influence the posterior. As you collect more data your belief will be increasingly overridden by the reality.

Reply on news.ycombinator.com