I have done a little experiment fiddling with the data shown by other forum members above and estimated the amount of spins to have a high confident that a wheel is truly biased. What I have done is to divide the number of appearance of each number 0-36 by a factor of the total of 15069 spins of a wheel. I have also assumed that the distribution of the probabilities of each number is close to the real distribution for a large sample size (15000spins) and smaller sample size is most probable to have a probability close to the true distribution.
I have found out that for this particular wheel (probably weak to medium biased wheel), chance of the wheel being not biased is as follows,
- 7535 spins (chance of random is 1/19000)
2.5023 spins (chance of random only 1/26)
3.3014 spins (chance of random is 1/1.46 which is very random and shows nothing, despite the probabilities of each bias number are shown from the true distribution (15000 spins).)
The borderline of the minimum data that you need to take is around 5000spins before you can determine the wheel is biased. For me, I would start to play the wheel at a small unit at around 5000 spins, bigger unit as my data shows more confidence.
The above is the traditional method of modern wheels. However, today, although bias is strong for low profile wheels, but the bias might shift temporarily (not random fluctuations but physical factors that affect the bias of a wheel). It is more complicated than just collecting number but also to observe the bounciness of the ball, and other conditions.
I have found stronger wheels and am playing right now.