4 93 REPLICATION AND COMPARISON NETWORKS fewer nodes, which applies to our study, power increases when networks are denser. The similarity of network structures was also investigated by correlating both the adjacency matrices of the networks and the rank of the node’s strength centrality across networks (Borsboom et al., 2017; Fried et al., 2018; Santos et al., 2018). If the correlation among network structures equals one, networks have a perfect linear relationship, meaning that the networks have essentially the same structure. A similar rationale holds for the correlation among centrality coefficients (Borsboom et al., 2017). 4.2.3.5 Network stability To check the robustness of the estimated networks and to examine the reliability of our inferences regarding strength centrality, the correlation stability coefficient (CScoefficient) was calculated based on a bootstrap procedure using 1,000 bootstrap draws for each network. This coefficient should not be below 0.25, and preferably above 0.5 to interpret centrality differences (Epskamp, Borsboom, et al., 2018). Networks were constructed, investigated, and compared using R-Packages qgraph (Version 1.6.1.; Epskamp et al., 2012), mgm (Version 1.2-5; Haslbeck & Waldorp, 2016), igraph (Version 1.2.2; Csárdi & Nepusz, 2006), bootnet (Version 1.2; Epskamp, Borsboom, et al., 2018) and NCT (Version 2.2.1; van Borkulo et al., 2019). Supplemental Material A contains the full R codes for the network construction, node centrality, community analyses, and network stability. 4.3 RESULTS 4.3.1 NETWORK CONSTRUCTION AND CENTRALITY OF DYNAMIC RISK FACTORS 4.3.1.1 Networks without recidivism In both the original (DSP) and the replication (BC) sample, within the network without recidivism all interrelationships between dynamic risk factors were positive (Figure 4.1a and 4.1c). General social rejection/loneliness, lack of concern for others, and poor cognitive problem-solving were the dynamic risk factors with the highest estimated strength centralities (i.e., having the highest influence within the network based on the number and magnitudes of edges) in these networks, along with sex as coping in the DSP network and impulsive acts in the BC network. Significant social influences and emotional identification with children were relatively weakly connected with other factors in networks constructed on both samples. Capacity for relationship stability appeared to have the smallest strength centrality in the BC sample, while in the DSP sample it was a middle-ranking dynamic risk factor (Figure 4.1c and 4.1d). The spinglass and the walktrap algorithms showed a distinct community (i.e., a cluster of meaningful-related dynamic risk factors) consisting of sexual preoccupation, sex as coping, deviant sexual interests, and emotional identification with children in the networks
RkJQdWJsaXNoZXIy MjY0ODMw