Stats question

fred_easey · 30 Oct 2006 at 14:37

Hi,

If I have an electronic dice and I want to test if it is truly random or not (or biassed towards a certain number), is there a way that I can test it? (eg some sort of stats test that looks at the number of outcomes compared to the expected number of outcomes for each number).

Anyone know how to do this?

Fred

ci_newman · 30 Oct 2006 at 16:05

Can't you just press the button 1000 times and record its output?

welshtom · 30 Oct 2006 at 17:17

i can tell you now its not truely random, computers cant generate random numbers, not true ones anyway

TheMagicPirate · 30 Oct 2006 at 17:23

welshtom said:
i can tell you now its not truely random, computers cant generate random numbers, not true ones anyway

^^ I had this exact same argument with a guy at work the other day.

Mr. Brightside · 30 Oct 2006 at 17:28

If it's anything like the electronic die we made in D&T at school a couple of years ago, it will be based on the length of time that the button is held down for (astable attached to a decade counter was how mine worked). So the outcomes will depend on the length of time the button is held for, and the previous value of the dice (if not reset before next seeding).

null

Gaverick · 30 Oct 2006 at 17:40

you can generate true random numbers on a computer using electrical noise if you really wanted to

VonClinkerhofen · 30 Oct 2006 at 17:44

welshtom said:
i can tell you now its not truely random, computers cant generate random numbers, not true ones anyway

what about carols random number generator on countdown? :confused:

Nash · 30 Oct 2006 at 18:25

Even if you recorded 1,000 results from it say, 800 were X and 200 were Y it could still be perfectly random. That's the beauty of randomness! I don't think there's a way to check it to be honest.

DAVEM · 30 Oct 2006 at 18:51

Yep,

Repeat the trial N times, and as N tends to infinity the distribution of the trial data should approximate the theoretical distribution. You could create a confident interval to measure how likely/unlikey your results are I guess.

David

sarutobi · 30 Oct 2006 at 18:54

DAVEM · 30 Oct 2006 at 18:56

theDave said:
idd

iid?
Indentically Independently Distributed?

Smithy · 30 Oct 2006 at 19:03

Only when threads like these come along do i remember why i hated the stats module in a-level maths.

Una · 30 Oct 2006 at 19:07

welshtom said:
i can tell you now its not truely random, computers cant generate random numbers, not true ones anyway

You can actually but not in the normal way (i.e rand()

), you need to sample a source of entropy outside of the computer. Such as radioactive decay, thermal noise, shot noise and clock drift, time between key stokes etc...

DAVEM · 30 Oct 2006 at 19:15

Una said:
You can actually but not in the normal way (i.e rand() ), you need to sample of source of entropy outside of the computer. Such as radioactive decay, thermal noise, shot noise and clock drift, time between key stokes etc...

In terms of defining a random variable it is quite mathematically involved, and at a graduate level looks at probability spaces. In terms of looking at 'real life' examples for a comparison set, it can often appear non random. Looking at would should be a random set, there is a possibility that the set may not actually appear random.

For example if you take what you think is a statistically random time series, then look at it's acf, then it shouldn't show acf lags that are above N^2/Sigma level, but I've seen a few examples that counter this.

Cheers

David