An experiment with Babylon and LMNO

[Triple Zero]

okay. he didnt PM me yet btw.

also dont click the link, you! those numbers are mine. get your own.

[Cain]

This shouldn't be interesting. But I will have plenty to think about if it is.


Also: this experiment should be done at least twice, to ensure that the results can be reliably replicated. Not necessarily immediately after this one, but at some point.

I would suggest at least three times, and with a sceptic doing the guessing once.

Also, someone with statistics experience (NIGEEEEEEEEEEEEEEEL!) should look at the results, because there is a particular test (chi-squared?  I can't remember) which you can apply to tell you if deviations in answers fall within a reasonable range or if they indicate something else is going on.

[Roaring Biscuit!]

This shouldn't be interesting. But I will have plenty to think about if it is.


Also: this experiment should be done at least twice, to ensure that the results can be reliably replicated. Not necessarily immediately after this one, but at some point.

I would suggest at least three times, and with a sceptic doing the guessing once.

Also, someone with statistics experience (NIGEEEEEEEEEEEEEEEL!) should look at the results, because there is a particular test (chi-squared?  I can't remember) which you can apply to tell you if deviations in answers fall within a reasonable range or if they indicate something else is going on.

i'm not sure how applicable chi is to this situation, but then again im not sure I entirely remember how to do the test :P  I think it basically lets you calculate whether there is a significant difference between expected values and actual values, normally you'd use it on the whole data set, so, how many times BH guessed correctly against how many times he was expected to guess correctly or something.

there are probably other ways of testing the results for significance, like a Mann-Whitney U test, which could be used to calculate whether there was a siginificant difference between BH's guesses and a random skeptics guesses.  I think...

x

edd

[Cain]

That might be the one I was thinking of, then.  I only did statistics as a part of my biology and psychology courses and that was a long time ago, so I cannot really remember the names of various tests very well. 

I now realize I probably could have used Wikipedia to check.  Oh well.

[Jenne]

I have Thurney's predictions.

[LMNO]

Ok, I decided to start early.  Here are the die and coin:



I am putting this in an excel spreadsheet.

[Cain]

You can see where he's injected lead into the dice, if you look carefully.

[LMNO]

 

[Cain]

OK, I have the results. 

[LMNO]

Anyway, I have sent the results to Cain and Rat.

I have also flipped a coin, and recorded those results.

In addition, Ratatosk has sent me a binary number chosen at random, with the ones and zeros translated to Highs and Lows.


So, I was thinking: Any 3 or 4 I roll is automatically right.... the only real predictive ability comes between 1 & 2 and 5 & 6...


...So doesn't that just set the odds back to 50%?

Statistics majors, check the fuck in!

[Bebek Sincap Ratatosk]

Anyway, I have sent the results to Cain and Rat.

I have also flipped a coin, and recorded those results.

In addition, Ratatosk has sent me a binary number chosen at random, with the ones and zeros translated to Highs and Lows.


So, I was thinking: Any 3 or 4 I roll is automatically right.... the only real predictive ability comes between 1 & 2 and 5 & 6...


...So doesn't that just set the odds back to 50%?

Statistics majors, check the fuck in!

Are we only checking Low and High or are we crosschecking numbers as well?

[LMNO]

I think the terms of the experiment are that Babylon will predict either Low or High.

I still think that if the predictive ability will come from deciding low or high, and two of six numbers don't count, then there's still just a 50% chance of being right.

if 3 & 4 are considered both high and low, then they're always right.  We can discount them.  However many 3s and 4s I roll will automatically be recorded as "correct", which will artificially inflate the predictive ability.  

Sure, the "probability" that you'll either get a roll in the set of [1-4] or [3-6] is 2/3, but if we're testing the ability to predict the correct set, isn't there a 1 in 2 chance of predicting a roll that will actually be wrong?

[Bebek Sincap Ratatosk]

I think the terms of the experiment are that Babylon will predict either Low or High.

I still think that if the predictive ability will come from deciding low or high, and two of six numbers don't count, then there's still just a 50% chance of being right.

if 3 & 4 are considered both high and low, then they're always right.  We can discount them.  However many 3s and 4s I roll will automatically be recorded as "correct", which will artificially inflate the predictive ability.  

Sure, the "probability" that you'll either get a roll in the set of [1-4] or [3-6] is 2/3, but if we're testing the ability to predict the correct set, isn't there a 1 in 2 chance of predicting a roll that will actually be wrong?

I agree with you, I thought if it were weighted with the guessed number as well... then maybe that would help the numbers, but otherwise this looks like a 50/50 to me.

[Cainad (dec.)]

First... thing... in the... morning...
should not have clicked... too much math talk--GAH!


Oh, um, I have BDS's predictions.

[LMNO]

If you want to PM them to me, I can add it to the spreadsheet.