Hacker News Reader

Comment by pfortuny | Hacker News Reader

The Type I and Type II errors is especially hurtful to me: I am totally unable to remember which is which and each time I read an article containing them I think: well, I am probably going to get it wrong again but I cannot bother to check the definition one more time... who cares?

View on HN

You know the story of the boy who cried wolf? The villagers committed several type I errors first, and then a type II error second.

(Yes, the terminology is stupid and should be changed. But as long as we're stuck with it, this helps.)

Wonderful mnemonic, thank you!

I don't understand why they need to be numbered at all. What's wrong with "False Positive" and "False Negative"? They even take less time to say.

I have trouble with those too: they're a composition of false/true with positive/negative to denote whether a finding of yes/no is right/wrong, but I can never remember whether false/true corresponds to yes/no or to right/wrong.

Just say things like "right yes" or "wrong no" and I'll understand. The only problem with that way of speaking is that it doesn't sound fancy.

> Just say things like "right yes" or "wrong no" and I'll understand.

Those have the same problem, don't they? I think you think you're replacing the positive/negative with right/wrong, but you're actually replacing the true/false. So, if you get a "wrong yes" result, is that "yes" supposed to be a good or bad thing? It's still ambiguous.

Maybe "true good" or "false bad" are better...

Then again, if both results are neither good or bad, then which is which? Maybe true/false positive/negative is indeed the less ambiguous. You just need to be clear on the question that's being answered. "Is there cancer?" "positive" means yes, "negative" means no.

We'd have to poll people to really know which terms are more easily remembered.

But here's how I'm thinking of it: a test says yes or no: is the thing present or not? And right/wrong is the correctness of the test. I can't see people easily getting confused and thinking "wrong yes" means that a test correctly said that the thing is not there. Anyone would think it means that the test incorrectly said the thing is there (among all the four possible meanings we're concerned about). OTOH false and negative could each more easily mean either wrong or absent.

But I may be wrong! This is only my best proposal.

> Just say things like "right yes" or "wrong no" and I'll understand.

OK. So s/false/right/ and s/right/true/ and s/negative/no/ and s/positive/yes/.

Does that work for you? Positive and negative are about the outcome of the test - did it report whatever or not? True and false are about if this result corresponds to reality or not.

It might just be me, but "wrong yes" sounds like it's claiming something very specific about the test. It's claiming that test said something was present and it wasn't. It's possible that this might not align with how the test works. Calling it a false positive would seem to make it clear that it's positive from the test's perspective.

But I may be wrong! This is clearly just opinion.

I'm not sure I get what distinction you're making. Are you saying that the standard terms don't commit to what we mean by a test's positive/negative, and this generality is more valuable than concreteness? It would be bad to imply too generally that we're talking about a test that says yes or no, a thing is present or absent?

I'm saying that in my opinion, the standard terms are pretty clear and yes or no create exactly the kind of confusion you're hoping to avoid.

Huh. Positive and negative are sort of notorious in medical tests for creating a confusion: positive is often bad!

There are actually three dichotomies here: was the test mistaken, what was its boolean value, and which boolean value means which thing in the world. Saying yes/no still requires a decision about the latter; though it seems clearer to me about this (both the need for interpretation and on suggesting a default), but it is subjective. And of course, any of these are better than Type I/II.

I do kind of insist that right/wrong or correct/mistaken clearly means the first question, where true/false is more ambiguous. To really resolve things maybe a better idea is like wrong/present and right/absent.

I've found that the pronunciation of English tends to dissociate the words "positive" and "negative" from their proper meaning, which has led to all sorts of misuse of the words. The following pronunciation isn't 'correct' English pronunciation, but try 'mispronouncing' it that way a few times (out loud) and see if that flips a switch in your head:

Po‑si–t—ive & Ne‑ga–t—ive.

po‑si–t—ing & ne‑ga–t—ing.

Po‑si–t—ed & Ne‑ga–t—ed.

Po‑si–t—ive‑ly & Ne‑ga–t—ive‑ly.

Gives a whole new meaning to the idea of 'positive' and 'negative' numbers, doesn't it?

"Sensitivity" and "specificity" are even worse:

http://languagelog.ldc.upenn.edu/nll/?p=14835

http://languagelog.ldc.upenn.edu/nll/?p=14888

In particular when one could just call them "false acceptance of null" and "false rejection of null", and abbreviate it to FAN and FRN, or so. How hard is that.

False {posi,nega}tive Type II? error

just 2 chars...........

Same here, and some definitions named by people's names. I totally understand the need to attribute and respect people's work, but a more descriptive name is almost always easier to remember and talk about.

This one at least has the justification that there often aren't any words. When you're hip-deep in topology, grabbing graph theory with your right hand and algebra with your left, you've left English way behind anyhow.

Now we just need to go back in time and convince Euler and Grothendieck to pretty-please either discover fewer things, or change their names every couple of years or something.

How about including given names with everything named after one of the many Bernoullis?

https://en.m.wikipedia.org/wiki/Bernoulli_family

Descriptive names, even when possible, translate very poorly to other languages.

I remembered them like this:

Type I: True -> False Type II: False -> True

Meaning something was True (H_0) and our results say it's False (Reject H_0), and vice versa.

The concepts and numbers and letters describing them are perfectly ordered.

  H0        H1
  alpha     beta
  Type I    Type II

Because error is negation of truth, you get H1 result with Type I error.

You say this as though it were helpful.

Too late to edit the above, but this reminded me of the infamous in-joke about category theory use in Haskell: "A monad is a monoid in the category of endofunctors. What's the problem?" [1]

[1] https://james-iry.blogspot.com/2009/05/brief-incomplete-and-...

They interviewed one of the Rust main guys for Linux Format a few years ago. I think he said the exact same thing lol.

On another note, I think I just met someone with both a power systems and functional programming background today. Small world :).

Reply on news.ycombinator.com