Hacker News Reader

Comment by matt4077 | Hacker News Reader

If you're in a class, you will soon hear the overused phrase "The plural of anecdote is not data".

You can out-smartass them by replying "Actually, it is".

An anecdote is a single data point. It's usually dismissed for all sorts of reasons, many of them valid.

BUT: contrary to popular opinion, they can be interesting, or even significant, depending on your priors, the payoffs of different outcomes etc. It's become fashionable to repeat the lazy phrase above, similar to "correlation does not prove causation", or "a survey with n=<any number> is completely worthless".

Example: "I've never seen this tree before. I wonder if those fruits are edible" / "It's just an anecdote, but my father ate one of those, started throwing up 30min later, and died the same day".

View on HN

> If you're in a class, you will soon hear the overused phrase "The plural of anecdote is not data". > You can out-smartass them by replying "Actually, it is".

I don't think it is. The plural of anecdote is probably something more like "a lot of single points without any care taken to correct for variables."

To get real data there needs to be a lot than just more anecdotes. There needs to be careful measurement, followed by correction for a lot of other factors.

Here's one example: "I've never seen this tree before. I wonder if those fruits are edible"

"It's just an anecdote, but my father ate one of those, started throwing up 30min later, and died the same day."

"What did he do before that?"

"Oh, he swam the dirty river, then started playing with that striped snake over there. He was a brave man, what with his heart problems and the cancer."

The parent said "depending on your priors".

Someone drops dead right after eating a strange fruit. Maybe they had heart disease and coincidentally died right after eating said fruit, but it's far more likely that the fruit had something to do with it. This is a perfectly statistically valid assumption, given that the vast majority of plants are toxic.

Once you live in a bayesian universe, every anecdote is data. Every data point might be incorrect; that's a given. The "plural of anecdote" line is just a particularly rigid way of frequentist thinking.

I agree with you a bit. But I also think there's a danger in assuming every anecdote is a data point, which is best expressed here:

https://www.youtube.com/watch?v=tWr39Q9vBgo

(Short video with Richard Feynman)... tl;dv - it's very hard to know something and you have to be really careful that you've done the work and aren't just deluding yourself.

Or, to put it differently - every anecdote is really only a data point if you understand what exactly it's measuring and what has been excluded. If you haven't taken the care necessary, then it's very easy to draw false conclusions. People are really bad witnesses and really bad at seeing what is and isn't given in the information they think they have - particularly if the anecdote is time-shifted from the point it was taken.

I completely agree (and said in my post) that any single anecdote can come with all sorts of problems.

I just believe there are too many people repeating such criticism thinking it'll make them sound smart, when in reality, they wouldn't be able to function if they didn't trust single data points almost all of the time.

Most of these occasions don't appear to be "trials" to us, but they actually are. Examples:

- "I really enjoyed that restaurant and will go back there soon"

- "Don't take a taxi from the airport! The train is much cheaper and the taxi driver just ripped me off"

- "The Hitchhikers' Guide is a great book–I should read another one by Douglas Adams".

-...

All these are essentially statistical trials with n=1. What makes single experiences in these cases possibly relevant is (among other things) our believe in the low variability of what we're measuring: A restaurant's quality doesn't usually change dramatically from one day to the next etc...

Then, there are more complex examples, where learning from n=1 is slightly more dubious, but without alternatives: "Don't start a land war in Eurasia", or the other German saying: "It happened, therefore it can happen again". On a more personal level, we all have to learn an awful lot about human relationships from a very limited set of experiences.

Reply on news.ycombinator.com