A new paper published in the Journal of Economic Behavior and Organization (ungated version here; PDF) by Michele Belot (University of Edinburgh), Raymond Duch (Oxford) and Luis Miller (University of the Basque Country) looks to address this question. The authors conducted a series of classic economic experiments with three samples (one students only; one non-students only; and one comprising a mix of students and non-students), and compared the responses of the two groups (students and non-students).
The results suggest that there are substantial differences:
In all games, with the exception of the auction game, we find that students are significantly more likely to behave as selfish and rational individuals.
Starting with the dictator game, we find that 57 percent of the students donated nothing. In contrast, only 17 percent of our non-students donated nothing...
Turning to the binary trust game, we find that only 35 percent of our student trustors trust in comparison to 82 percent of non-student trustors...
We also find significant differences in the rates of reciprocation: 56% of the student population and 87% of the non-student population reciprocated...
Our third game involving other-regarding preferences is the public good game... students are considerably more likely to free-ride: in the first round of the public goods game, 24 percent of the students contributed nothing to the public good while only 9 percent of the non-students were strictly non-cooperative.
Turning to the beauty contest game... Thirty percent of our non-student subjects made choices consistent with at least level 1 iterative reasoning compared to 56 percent for the student subjects.Without going into detail about the experiments themselves (you can read about the dictator game here; the trust game here, the public good game here, and the beauty contest here), it demonstrates that students are more selfish (less other-regarding) and more rational decision-makers. The authors do note that "students are younger and smarter" than non-students, but the results persist even after controlling for cognitive ability, age and gender.
The results should give pause before we draw grand conclusions on the basis of laboratory experiment results. However, as the authors (and others) have suggested, laboratory experiments may still be useful for identifying qualitative effects. For example, they may tell us whether threat of punishment will induce less free-riding in the public goods game, but won't give robust information about how much less free-riding there will be.
[Update: Added in some links that I had forgotten to add first time!]