A die with a higher expected value can still lose a pairwise comparison. Suppose die X has faces 2, 2, 4, 4, 9, 9, giving an expected value of 5, while die Y has faces 1, 1, 6, 6, 6, 6, giving an expected value of about 4.33. Die X has the higher average, yet die Y wins whenever it rolls a 6 against X's 2 or 4, which happens for 24 of the 36 face pairs, so P(Y beats X) is about 0.67. High faces concentrated on a few outcomes can beat a steadier, higher-average die.
Non-Transitive Dice: The Endless Dominance Spiral
What the Spiral Means for Probability and Choice
When the Higher Average Loses
2 / 5
Here is the moment the two questions pull apart. Take a die whose faces are mostly small, with a couple of big values, and another die that is steadier but slightly lower on average.
0:00 / 0:00