4 91 REPLICATION AND COMPARISON NETWORKS old (Hanson et al., 2007). For the DSP sample, men who exclusively offended against individuals older than 13 scored a 0 on emotional identification with children by default on the STABLE-2007. This resulted in a modified score on this item for about 7.7% of our total sample. More specifically, 57 offenders with a score of 1 and four offenders with a score of 2 on emotional identification with children on the STABLE-2000 received a score of 0 on this item in the STABLE-2007. For the replication study, we approximated STABLE-2007 scores for all participants using the 2000 version of the items when it was the only STABLE-version available. Because we lacked data on victim age, we were not able to transform the STABLE-2000 score on emotional identification with children for men who exclusively offended against individuals aged 14 years and older. Based on frequencies, we expect that few offenders without victims younger than 14 would have scored more than a 0 on this item. 4.2.3 NETWORK ANALYSIS To estimate network structures, nodes centrality measures, identification of communities, and network stability, we followed this original methodology (van den Berg et al., 2020). 4.2.3.1 Network structure To meet the statistical assumptions, mixed graphical models (mgm) were used to estimate network structures, as both data sets included count and categorical data (Haslbeck & Waldorp, 2016). Furthermore, the networks were constructed using regularized network estimation to control for Type I errors (van Borkulo et al., 2014). As a result, connections found in the sample network are likely to be present in the population as well, while connections that are absent may either be absent in the population network or too weak to be picked up by the regularization technique. Extended Bayesian information criterium (EBIC) for L1 penalized regularization was used to control for spurious connections and to estimate sparser network models (Costantini et al., 2015). To compare the networks, we used the layouts of the networks without recidivism and with sexual recidivism based on the DSP sample as a template for the other networks and set the maximum value (1.00) and cutoff (0.08) for possible connections for all networks. Green edges indicate positive statistical associations. The stronger an association, the more saturated and wider the edge. 4.2.3.2 Node centrality metrics In terms of node centrality metrics, this replication study focused on strength centrality. Strength centrality indicates the likelihood that activation of a dynamic risk factor will be followed by activation of other dynamic risk factors (McNally, 2016). In a weighted network, strength centrality reflects the importance of a node based on its number and magnitudes of edges compared to those of the other nodes (Opsahl et al., 2010).
RkJQdWJsaXNoZXIy MjY0ODMw