http://en.wikipedia.org/wiki/Stanford_prison_experiment
Bad science.
http://skeptoid.com/episodes/4102
Thats one of the worst rebuttals I've ever seen.
Firstly, the issue of "selection bias" making the study unreliable is completely dud.
The idea that the kind of people who would volunteer for an experiment are any more likely to be sadistic than the kind of people who would actually volunteer for being an actual prison guard.
Of course a small and narrow sample will not yield as accurate results as an experiment with a larger sample. Due to modern legal issues its hard to recreate this experiment, but in the times I have seen it replicated it has generally supported the original results.
The Milgram study only used 40 people. Hardly enough people to make any strong conclusion on the general populace, they could have easily got 30 complete psychopaths by chance. but its still a very important study, and it being performed on a small scale led to many reproductions that reinforced the original studies findings.
The objection that most of the prison guards didn't engage in sadistic behavior is also dud. Its made quite clear in the study that only 1/3rd exhibited sadistic tendencies.
Also the objection that one guard was acting what he saw from a movie is benign, it was real people, and real sadism, so whatever motives it still supports that a significant amount of people will abuse power in a prison type situation.
Does Stanford Prison experiment prove all people are evil bastards waiting to happen? No. The results don't support that.
Does the Stanford Prison experiment provide concrete proof that X percent of people will act in a certain way X percent of the time in situation X?
No. Neither does the Milgram study, but they're both important studies in providing limited, but important indications on how the nature of authority positions can effect human action.