Just some idle conversation here, voicing my confusion. I read about a computer generated coin flip years ago. I think they let it run billions of times or more. There was a high run of one side winning every flip of around 20-22 times. This was wildly improbable as the next highest run was under ten I believe.
Here is where I get confused. If the first twenty flips would have been all heads, the odds of the rest of the series being fifty-fifty remain the same. If there was a thousand flip run of all heads, the trial from there on out would still have fifty-fifty odds. It seems to me if the overall statistics are to remain fifty-fifty, then the tails side would have to do some catching up either with long runs of it's own or a series of shorter runs.
If the odds are fifty-fifty, after a billion flips shouldn't the results be within a very close percentage of fifty-fifty? The computers say that one side might be 10-15% ahead of the other. If so, I say the odds weren't fifty-fifty. Seems like if one side is well ahead the other side has to be favored over the following run if you can stay with it long enough.
This confusion is why I am much more inclined to wager than gamble. I hate to sound like Bert Kinister but my last wager was for a chicken dinner. A single backcut shot, a bad shot to bet against me on. To quote Bert, "Winner, Winner, chicken dinner!"
Hu
Interesting thought experiment! I would have to agree: that the real deal with coin tossing is eerily surprising.
Although the average number of heads has to "even out" to fifty-fifty (law of large numbers), the
total number of heads do not. In fact, for totals, the deviations from expectation grow at a rate proportional to half the square root of number of tosses.
So for example, for an idealize fair coin, after a million tosses there is (only) a 68% chance that the total number of heads is between 499500 and 500500, which is just +/- 500 deviation from expectation. (500 being half of the square root of a million)
After a trillion tosses, there there is only a 68% chance the total is within 499999500000 and 500000500000 which is +/- 500000 from expectation.
The size of the deviation grows, but just at a rate slower the the number of tosses, because "square root of n" grows more slowly than "n". This allows the law of large numbers (which is for the average, not the total) to still be true.
An anomaly in the beginning (such as 20 heads in a row) would be absorbed into these expected deviations
The notions above are due to the deMoivre-Laplace theorem (essentially the first "central limit theorem") proven in the 1700s. The deMoivre-Laplace theorem has a Wikipedia entry if you're interested.
William Feller's 1950 textbook devotes an entire chapter to fluctuations in coin tossing. So many interesting and non-intuitive things.