Dictator game experiments come in three avors: plain vanilla with strictly dichotomous separation of dictator and recipient roles, an interactive alternative whereby every subject acts in both roles, and a variant thereof with role uncertainty. We add information regarding which protocol was used to data from the leading meta-study by Engel (2011) (in addition to 296 corrections of reported means and standard errors), and investigate how these variations matter. It appears that interactive incentivization compared with the standard protocol, in addition to being relevant as a control for other effects, renders subjects' giving less generous but more efficient. Our results help organize findings and sources of confounds in the field.