Re-scaling social preference data: implications for modelling