Hacker News Reader

Comment by dholowiski | Hacker News Reader

Can somebody who knows math comment - There is a 5 % difference on a sample of 700. Not huge, but that seems statistically significant? Please post the source code so we can verify, and try this ourselves...

View on HN

Assuming each trial is independent and the experiment is a Bernoulli experiment (e.g. akin to flipping a coin 700 times), the probability that you will get a result within 5% of the expected mean (50%) is 81.4%. Remember, though, that the probability of getting EXACTLY 50% is only 3%. The probability of getting exactly +/- 5% is about 2.5%.

So, you have about the same chance of getting that result as the theoretical. That by itself wouldn't mean anything, though - we need more information on the experiment. The best impromptu way to tell if it was significant would be to look at the distribution of the success rates of individual people.

@Parapsych - What kind of data are you logging and would you be willing to release that?

Currently down to a 0.5% difference with samples of a bit over 3000. Seems pretty insignificant to me.

At 8700 samples, it's 0.5% in favor of predicting non-erotic. Yup, seems pretty insignificant now.

likely to happen once in 50 experiments or so.

I think we need more samples to gauge significance.

Why do you need to try this offline? I'm not opposed to the idea, just curious.

Not offline, just on our own servers, so we can be sure it's random. But I didn't realize it was written in Javascript, so anyone can inspect the code.

I don't like that it shows you your percentage as you go. Aren't people more likely to quit when their Porn finding skills are higher?

Also, maybe i've been spoiled, but some of the 'erotic' pictures don't seem that erotic.

However... it seems like the #'s are evening out now, both around 51%

I think with a bit of re-design this could make for a great, and useful experiment.

Whether the picture appears on curtain 1 or 2 is predetermined, and the server provides proof in the form of a hash. You can see that on the bottom under your stats.

Yeah, I had a tough time finding erotic pictures. If you have a better idea of where I can find some to hot-link, also for gay / female audiences, lmk.

What are you talking about, of course I have no idea where to find those kind of pictures... although, and this is pure speculation, a Google images search with safe search turned off might return some interesting results.

> likely to happen once in 50 experiments or so.

How many hypotheses did he investigate? I seem to remember reading that he said he had investigated a fairly large number of them before finding this one.

If he investigated many different ways that people might manifest ESP only to find negative results in all but the porn study, then the odds change a bit. I mean, suppose you have 50 studies and a 1/50 chance of something like this--you'd have a 63.6% of ending up with a study like this one.

Reply on news.ycombinator.com