Comment by xpe | Hacker News Reader

Comment by xpe | Hacker News Reader

Parent Full thread

Common writing, even from well-regarded sources, is often unnecessarily vague wrt statistics.

Claim: we would benefit by building a culture where basic probability ranges are expected and used.

Idea: Serious publications could:

A. Set basic guidelines for talking about probability estimates; something akin to what you see in government or business intelligence publications. For example, maps terms to ranges:

“highly likely”: > 90%

“likely”: 70% - 90%

“somewhat likely”: 50% - 70%

…

B. Update their editorial standards to not allow pure vagary around probabilities and impacts.

Could authors write more quantitatively? At the levels of the top 100 publications, I’d guess that 80% could, with the help of research and editing staff.

Would some readers be scared off? I’d guess than 80% of a college educated audience could handle it. The rest might complain, but could level up with a little peer pressure.

Would articles become more rigorous? Over time, I’d hope most would. Shift the expectations and make the vocabulary less forgiving to vaguery and fudgery, and I think the incentives point in the right direction.

I thought there was already common terminology for probability of side effects [0]

Very common = greater than 1 in 10

Common = 1 in 100 to 1 in 10

Uncommon [formerly 'less commonly' in BNF publications] = 1 in 1000 to 1 in 100

Rare = 1 in 10 000 to 1 in 1000

Very rare = less than 1 in 10 000

Frequency not known = frequency is not defined by product literature or the side-effect has been reported from post-marketing surveillance data

[0] https://bnf.nice.org.uk/medicines-guidance/adverse-reactions...

Perhaps, but… How many publications have any such guidelines? Make them public? How well do they follow them?

I work with readers and editors sometimes. They get swollen and puffy when they see numbers, though they will swallow their pain if it’s a scientific article. Otherwise they will complain loudly and the editors will use their red pens a lot.

With that said, I consider your idea has merit and should be implemented.

Thanks for sharing. To these folks I would ask: what about numbers is problematic? To the degree that it has to be with current reader perceptions, I understand and empathize.

But in terms of where we want to go, we can do better. Great writers care about precision, and wisely chosen numbers and ranges are able to provide that. It may be appropriate to situate them in footnotes or endnotes, depending on the context. But the numbers matter.

Explore the Associated Press "AP Stylebook" (pay-walled, sadly), and propose a mapping to them. It would take a few years to get accepted and more to propagate throughout the working press.

See also: https://en.m.wikipedia.org/wiki/Words_of_estimative_probabil...

Maybe it's because I'm tired but I don't see what that has to do with what I said. Did you reply to the wrong comment?

Perhaps :) Point taken. Re-reading, it looks like I skipped a transition sentence from the comment to what I wrote. (Or maybe the parent comment got edited? Not sure.)

> You can get cancer from not getting enough sleep. There are plenty of innocuous substances that look fine chemically but disrupt sleep. It doesn't have to be directly mutagenic to cause cancer.

I agree with the comment as written, though it strikes me that virtually no mainstream publications use writing that clarifies or quantifies these kinds of indirect relationships.

I see a lot of intellectual laziness or dumbing down. Two categorical errors happen often:

(A) writing “everything is connected”. This can be true in some sense while also being useless. The more interesting framing is to ask “to what degree” and under what modeling assumptions.

(B) Writing about various linkages without clarifying probability and impact dilution.

The connecting idea in my head was something like this: when most publications talk about cancer they tend to lack even a basic statistical language.

Backstory: Having hundreds of interactions with ChatGPT, which is trained on much mainstream writing, made this interaction pattern obvious and tiresome.

Another way of saying my complaint: It feels like otherwise intelligent people hit a barrier where they start speaking vaguely. They pretend like this is the best we can do. It ideally isn’t; we have extensive scientific and statistical studies. There is no reason why we as a culture have to accept such statistical phobia as quality writing.

Reply on news.ycombinator.com