Hacker News Reader

Comment by pushrax | Hacker News Reader

> other fields where such principles aren't really applicable

Statistics and reasoning are almost always applicable, and that's the extent science is applied to most fields.

If you want people to act in an evidence-based way without invoking trust/faith, they need to understand statistics and be able to evaluate studies on their statistical merit. Unfortunately that's quite difficult and the educational systems we have do a poor job of teaching it.

Statistics is subtle. For example, "It turns out, most doctors don’t know actually know probability." https://blogs.cornell.edu/info2040/2014/11/12/doctors-dont-k...

View on HN

To build on that, most scientists* are also terrible at statistics.

Perhaps more controversially, I also don't think statistics in general is a particularly rigorous or well-founded field. Where there is any analysis at all in the field - that is, very rarely - it provides guarantees only with hefty assumptions and in the limit. I'm fine with that from a theoretical perspective, but practitioners of all stripes seem to assume that with real world phenomena, the data either satisfy the assumptions or, more defensibly, don't violate the assumptions by enough to invalidate the results.

I don't know about others, but I see real-world datasets all the time that wildly invalidate the assumptions of common regression analysis, to take just one example.

To reign this little rant back into a well-formed comment, perhaps my summary is "everything and everyone is wrong, almost all the time, and I'm not quite sure how to fix that."

* In my personal experience.

>To build on that, most scientists* are also terrible at statistics.

BTW, I think your use of the asterisk goes beyond what was intended. Its typically been used to omit things or add caveats that only matter in very specific situations. 'in my personal experience' could mean you know 5 scientists or 5 thousand so mentioning a rough range would certainly give more credibility to your comment over the asterisk. That said, I broadly agree with the rest of your comment, but this did bug me enough to comment I suppose.

>I don't know about others, but I see real-world datasets all the time that wildly invalidate the assumptions of common regression analysis, to take just one example.

I'm interested in knowing about them. Any major ones that come to mind?

>'in my personal experience' could mean you know 5 scientists or 5 thousand so mentioning a rough range would certainly give more credibility to your comment over the asterisk.

Agreed. Providing us with some statistics would help support their argument here.

It looks like the author of that article isn't too clear on the matter either, since there's a fundamental arithmetic error leading to a self-contradictory claim halfway through the article.

Indeed. The comments section shows the history; the author mindlessly corrected a non-error into an error. This looked like a decent summary at a glance but that issue is definitely distracting.

Reply on news.ycombinator.com