Collaborations become increasingly important among almost all scientific disciplines. Teams can be more productive and achieve more attention to their work. It is, however, unclear, whether teams also lead to higher research integrity. On the one hand, there may be a volunteer's dilemma in larger groups such that the responsibility diffuses who is controlling whom. The 'volunteer hypothesis' predicts that the more co-authors, the more scientific misconduct. On the other hand, larger groups may also achieve a higher level of social control. The 'control hypothesis' predicts that the more co-authors, the less scientific misconduct. Retractions are used as an operationalization of scientific misconduct. The data collection comprises of retracted articles from the Web of Science data set. In addition, control groups of non-retracted articles are constructed, using methods known from causal inference and bibliometrics. The analyses demonstrate that larger author groups have a lower retraction probability compared to smaller author groups. This suggests that teams prevent misconduct, most likely by their higher ability to social control. The results indicate that the development towards more and larger research collaborations may have positive macro-level consequences for the system of science.