Quote:
Originally Posted by statmanhal
I am absolutely sure that the absolute difference will not fluctuate between positive and negative, I think
It may fluctuate between positive and negative, but it actually is more likely to remain either positive or negative for extended periods. It seems counterintuitive but a simplified example can help.
Suppose Peter and Paul play a game. They each flip a coin. If the coins match (both heads or both tails) Peter wins 1 point. If the coins are different Paul wins one point. Obviously since it is a fair game, the percentage of wins for each player will tend to equalize over many plays. Instead of that question, suppose we consider a different one — if they play some large number of games, what is the probability that Peter will be in the lead x% of the time during those plays? (Obviously the symmetrical question about Paul would have the same answer).
It is possible to calculate this value for any value of x ranging from zero to 100%. When we do so, an interesting and (IMO) counterintuitive result occurs - you get the maximum probability for either 0 or 100% and the minimum probability at 50%. It is more likely that one player or the other will lead the whole time than it is that they will go back and forth and each will lead about the same amount!
For those interested in the technical details, the calculation can make two assumptions about tie scores - either the player who led last is consistent to still be leading or the player who won to tie the score can be assumed to be the new leader. Either way the result holds. In technical terms the distribution is an arcsine distribution over the interval [0,1], which takes on maxima at the endpoints and a minimum at 1/2.
To make some kind of sense of this consider the first few games. Suppose (without loss of generality) that Peter wins game one. Then depending on your assumption about who the leader is on tie scores, either Peter will always lead after game 2 or either player is equally likely to lead. In the first case, we already have an asymmetry — the player who wins first will automatically lead for two rounds. In the second case, we find that if Peter also wins game 2, we get the same asymmetry - only Peter can lead after 3 games. Loosely speaking, these asymmetries add up over many games, making it more likely that one player or the other will lead most of the time.
Hopefully the parallel with long term poker winnings is obvious. Once you go positive over x hands, you either will go positive over the ensuing x hands, or negative over those hands. If you go negative, you are likely to be close to breakeven. If positive, you become even less likely to go negative (as your total winnings would be even greater after 2x hands than they were after x hands). Unlike the Peter vs Paul game, though a long term winner at poker isnÂ’t a fair game, and the probability of being positive after many hands is much greater than being negative.
Edit — of course the absolute difference, being defined as positive will not fluctuate, but it also is more likely that the signed difference will also not fluctuate, as described above.