Z's dual role - his own behaviour affected the way in which events unfolded, thus questioning the validity of the findings
However, there was good internal validity - participant selection (stable and random allocation) helped to rule out individual differences in prisoner/guard allocation