Figure 3

 A simple law between N (total number of social grooming partners) and m (mean number of social grooming days). These user behaviour data (black points) did not obey the null model (C=Nm; orange lines), where the assumed social grooming cost was independent of the strength of the relationship, but obeyed C=Nm^a (green and dash lines), where a=1.19 (a: Twitter), a=1.21 (b: 755 group chat), a=1.56 (c: 755 wall communication), a=1.10 (d: Ameba Pigg), a=1.07 (e: mobile phone) and a=1.24 (f: SMS), where user use-days u were 75th percentile, estimated by the regression models Equation (1), where adjusted R-squares were 0.990 (a: Twitter), 0.974 (b: 755 group chat), 0.959 (c: 755 wall communication), 0.997 (d: Ameba Pigg), 0.994 (e: mobile phone), 0.990 (f: SMS) (see Table 2 for details), and a>1 were significant excluding the mobile phone data set.