Most models of social preferences and bounded rationality that are effective in explaining efficiency-increasing departures from equilibrium behavior cannot easily account for similar deviations when they are efficiency-reducing. We show that the notion of sampling equilibrium, subject to a suitable stability refinement, can account for behavior in both efficiency-enhancing and efficiency-reducing conditions. In particular, in public goods games with dominant strategy equilibria, stable sampling equilibrium can involve the play of dominated strategies with positive probability both when such behavior increases aggregate payoffs (relative to the standard prediction) and when it reduces aggregate payoffs. The dominant strategy equilibrium prediction changes abruptly from zero contribution to full contribution as a parameter crosses a threshold, whereas the stable sampling equilibrium remains fully mixed throughout. This is consistent with the available experimental evidence.