LogFAQs > #934394405

red sox 777 posted...

In the experiment we don't know that the coins are slightly biased. That is just our model.

I think with the results, it is overwhelmingly likely that the coins are extremely biased. But in the same direction as we had predicted a small bias. So our model is wrong and yet more powerfully predictive than it claimed to be.

The model can only estimate a probability distribution; it cant say anything to distinguish between different processes that give the same probability distribution function (e.g. flipping 100 fair coins or flipping 100 double faced coins that each have a 50% chance of being double heads or double tails). But if you have two models with even slightly different distributions, they can be distinguished in the limit of infinite data.

---

Congrats to Advokaiser for winning the CBX Guru Challenge!