When one levels up and gets an increase in maximum hit points, one can either roll or take the average roll rounded up. Statistically, you’re better off with the non-random method.
For ability scores, the standard array is 15, 14, 13, 12, 10, 8. Using the point-buy system yields scores in this same basic range. But, according to AnyDice, the average rolls for 4d6-drop-one are 16, 14, 13, 12, 10, 9 — slightly better and, crucially, providing the likely possibility of an 18 to start (with a racial bonus of +2) and a 40% chance of starting with at least one ability score at 19 or 20.
It seems odd to me that the designers didn’t go the same route as with hit points: make the predictable choice be rounded up from the average. This answer delves into the math and makes the same basic point — and ends up wishing that the designers had picked a slightly higher standard array.
We can all speculate on the pros and cons of this; I’m curious instead about designer intent and official references and commentary.
IIRC, point buy models the most common outcomes of 4d6, drop lowest.
(Emphasis mine. IIRC stands for “If I remember correctly”. I didn’t know that myself. :))
His tweet seems to imply
That they relied on some other probability calculation tool instead of AnyDice
The designer intent was, indeed, to use the most common outcomes of 4d6, drop lowest.
(The short thread Mearls’ answer appears in discusses ways and whys of adding / not having a 16 in the standard array.)