Investigating whether the 5e standard array (15, 14, 13, 12, 10, 8) is better or worse than rolling (4d6k3)
For this I randomly generated a million sets of 6 stat rolls, each of which was formed from 4 samples of the U(0,1) distribution, ordered them, and then plotted their various distributions.
In conclusion, in the long term the rolled stats will beat the standard array, hands down.