Fischbacher et al. (2001), henceforth FGF, introduced an experimental design to measure conditional cooperation in public goods games. We collected data from 17 replication studies of FGF and observed that the criteria used to identify types are not always consistent. We refine FGF's definition of types to resolve ambiguous cases in FGF and its replications. Using our new classification scheme, we find in our combined data set with more than 7000 individual observations that FGF's original findings are by-and-large stable: conditional cooperation is the predominant pattern: free-riding is frequent, while non-minimal, unconditional cooperation is very rare.