From No Limit Hold 'em: Theory and Practice by David Sklansky and Ed Miller.
Concept No. 25: The button is the true bread and butter position in no limit. In many games you can play an extremely wide range of hands from the button, even for a raise.
Sklansky and Miller say that if effective stack sizes are at least 200 times the big blind, then, as long as you have at least one opponent who plays too loose after the flop, it is correct to limp on the button with over 50% of hands ("probably any two suited cards, any big offsuit cards, any ace, and any offsuit connector down to at least five-four"), and possibly with 100%. They also suggest that you call with over 30% of hands "if the raise represents only a few percent of the stacks (e.g., no more than maybe $50 with $1000 stacks)."
This is another one that is tough to analyze precisely, but in general, I think this advice is encouraging the reader to play too loosely. Unless you are much, much better than your opponents, this advice is probably -EV. I would be especially wary of the advice to call a raise to $50 with KTo in a $2-5 blind game, even if the stacks are over $1000. First of all, this advice completely neglects to consider whether the raise is coming from early or late position, which can have a drastic effect on your opponent's range. Even if the player is in late position and is rather aggressive, though, I would not be happy calling with KTo in this spot. You need to flop a straight or have the flop to contain KT or TT in order to be confident with your hand in there is a lot of action. Otherwise, a hand like this is going to have no implied odds, or worse, negative implied odds.
I do think the general idea that the button is a powerful position is correct, and indeed there are a lot more hands that can be played profitably from here than any other position (except the BB with no raise). However, in my humble opinion, S+M take the idea a bit to far here. If I were an online player, or if I took careful track of my results during poker sessions, such data might be helpful in coming to a well-informed conclusion. As it stands, all I (and also S+M, I suspect) can really do is guess.