Trust in Online Collaborative IS - Community-Built Databases: Research and Development - page 353

Fig. 14.5 CDR, SDR, DSR and RDR extracted after 2007 as functions of reputation computed
before 2007. The X-axis shows the reputation bins and the Y-axis shows the percentage of users in
the bins

Fig. 14.6 Transitions between high-quality and
low-quality states

Table 14.3 Correlation values for the three reputation models
Models RDR CDR SDR DSR

Model 1 (- 0.906, - 0.871) (0.434, 0.760) (0.757, 0.861) (0.999, 0.996)

Model 2 (- 0.927, - 0.939) (0.783, 0.852) (0.822, 0.833) (0.976, 0.975)

Model 3 (- 0.958, - 0.973) (0.779, 0.811) (0.791, 0.786) (0.944, 0.944)

of submissions (SDR), whereas in reality during 2007-2009 those users were respon-
sible for only 27% of submissions. To compare these two sets of diagrams (Figs. 14.4
and 14.5 ), we perform a Pearson correlation analysis. The results are described in
Table 14.3 , where each tuple shows the correlation between the two parameters
before and after 2007, respectively. For example, the entry ( 0.906, 0.871) sig-
nifies that the correlation between RDR and Model 1 reputation is 0.906 in

Fig. 14.4 , while it is 0.871 in Fig. 14.5 . These correlations are highly significant,

and the same is observed if one measures the correlation between the reputation value
themselves within or across models, and up to 2007 or up to 2009.
In combination, these results suggest that the reputation models are good at
predicting behavioral indices and reputation values at future times, not only for
extreme populations of very good or very bad users, but also across the entire
spectrum of reputation values.