Proefschrift

4 99 REPLICATION AND COMPARISON NETWORKS Statistical comparison between the original (DSP) and the replicated (BC) networks without recidivism, with sexual recidivism, and with violent recidivism (including sexual contact) using the NCT revealed statistically significant differences in network structure (respectively, p = .007, .034, and .032), but not in terms of global strength (respectively, p = .193, .255, and .142). Both the adjacency matrices and the strength centrality estimates of the compared networks were highly correlated. The Pearson correlation coefficients (PCC) of the adjacency matrices varied from .769 (DSP and BC without recidivism) to .803 (DSP and BC with sexual recidivism). The Spearman’s rank correlation of the strength centrality estimates varied from .719 (DSP and BC without recidivism) to .947 [DSP and BC with violent including (sexual contact) recidivism]. Table 4.3: Results of network comparison using NCT, and correlation of the adjacency matrices of the networks and strength centralities (van den Berg et al., 2022, p. 434) NCT Adjacency matrices Strength centralities Networks Structure Strength p-value p-value PCC p-value rho p-value Without recidivism .007 .193 0.769 < .001 0.719 = .007 With sexual recidivism .034 .255 0.803 < .001 0.921 < .001 With violent (including sexual contact) recidivism .032 .142 0.788 < .001 0.947 < .001 Note. Values in bold are statistically significant, p < .05. 4.3.3 NETWORK STABILITY We estimated a bootstrapped sampling distribution (n = 1,000) for each network on both the original (DSP) and replication (BC) data sets to check the accuracy of the estimated networks (i.e., its sensitivity to sampling variation) and to test the stability of our inferences about the network structures (especially the centrality measures). As mentioned in the method section, this coefficient should not be below 0.25, and preferably above 0.5 to interpret centrality differences (Epskamp, Borsboom, et al., 2018). Our centrality measure of interest, the CS-coefficient of strength, varied from .28 for the network including sexual recidivism to .44 for the networks without recidivism and with violent recidivism (including sexual contact) in the DSP sample and from .59 for the networks without recidivism and with violent recidivism (including sexual contact) to .67 for the network including sexual recidivism in the BC sample. Supplemental Material C contains the subset bootstrap and the CS coefficients of the strength centrality for all three networks estimated on the DSP and BC samples.

RkJQdWJsaXNoZXIy MjY0ODMw