I had a thought on this topic overnight and came to another conclusion.
If you use point-buy or standard array (where you allocate 15,14,13,12,10,8 as you wish), the chances of getting a 15+ are very good to the point of guaranteed. Now compare that to a random 3d6 roll:
* chances of rolling 15+ are 9.26%,
* chances of rolling 16+ are 4.63%,
* chances of rolling 17+ are 1.85%,
* chances of rolling 18 are 0.46%.
Some players choose 3d6 because they have the chance to roll 18 (there is a 2.75% chance of rolling an 18 among your six stats), but I wonder if they know exactly what their chances actually are? Of course, the roll chances follow a bell distribution curve, so the chances of rolling very low are equally small. 3d6 is a good way to generate an 'average' character because of the way the distribution falls, but it is not great for when you 'need' that particularly high stat.
Later editions of D&D began offering differing ways to roll, of course, 4d6 and dropping the lowest die result being the most common, but this still leaves you open to chance. I'm sure I'm not the only one who has rolled 4d6 and still ended up with 8 or less.
Then there are the 'go on, have another go' GMs and players. Either through pity, pressure or because they believe they will have a better game with higher stats, sometimes players will simply keep rolling until they get the result they want. It will happen eventually. This is, of course, the early BG/Icewind Dale approach because on a PC, nobody can see your shame (how annoying was it when rolling STR for a fighter and getting 18 but ending up with 18/03?). If this method is used, would it not simply be faster to let the player have an 18 straight off the bat and save all that meaningless rolling?
