Alternative to Von Neumann bias correction method

[tim1608]

Hi

I have discovered that the Von Neumann bias correction method only works when the bias is 100% stable, for example tossing the same loaded coin again and again.

Does anyone know of a bias correction method which can correct an unstable bias? Is this impossible?

Edit: Let's say I have a bucket of coins, each of which is differently loaded but they are all loaded on the tails side. How can the bias be removed if the coins are each tossed one-by-one?

 

[Stephen Tashi]

if the coins are each tossed one-by-one?

Are you going to label the coins and toss them in order till all are tossed and then toss them again in the same order?

 

[tim1608]

Hi Stephen

Thank you for your reply.

No. The hypothetical "bucket" in my original post has an infinite number of coins.

 

[Stephen Tashi]

To pose an interesting problem, I think you must say more about the population of coins and how they are selected. (For example, you haven't ruled out an infinite population of two-headed coins.)

 

[blue_raver22]

Thank you for bringing up this issue with the Von Neumann bias correction method. While this method may work well for stable biases, it may not be suitable for correcting unstable biases. In such cases, an alternative approach called the "Bayesian bias correction method" may be more effective.

The Bayesian method takes into account the prior probability of the bias being present, as well as the observed data, to correct for any potential bias. This approach has been shown to be effective in correcting for unstable biases in various scenarios, including coin tosses.

However, it is important to note that there is no one-size-fits-all solution for correcting biases. The effectiveness of any correction method will depend on the specific situation and data at hand. Therefore, it is always important to carefully consider the nature of the bias and the available data before selecting a correction method.

In the case of your bucket of coins with different but all loaded on the tails side, the Bayesian method may be a good option to consider. I would recommend further research and experimentation to determine the most appropriate approach for your specific scenario.