6 140 CHAPTER 6 for L1 penalized regularization (Costantini et al., 2015; Epskamp & Fried, 2017). This regularization technique which is used to control the Type I error rate has been shown to result in networks with high specificity and adequate sensitivity (van Borkulo et al., 2014). Assuming that research on networks of dynamic risk factors will typically include variables at various measurement levels, the appropriate PRMF model to estimate a network of dynamic risk factors relies on the use of mixed graphical models (mgm: Haslbeck & Waldorp, 2020). This is because mgm allows for the existence of not normally distributed data or data which is not measured on a continuous scale. Estimating and visualization of networks as described above can be done with respectively R-Packages graphicalVAR (Epskamp, 2020) and qgraph (Epskamp et al., 2012). However, given the limited sample sizes in research on adult males’ dynamic risk factors, there is a reasonable chance that the strength of connections between these dynamic risk factors may not be estimated accurately (Epskamp, Borsboom, & Fried, 2018). This probability increases exponentially when more dynamic risk factors are considered. The next section will describe the current approach to assess the accuracy of estimated networks. 6.3.2.1 Assessing accuracy of the estimated network The question which sample size is needed to accurately estimate network structures from psychological data is a subject of debate (e.g., Epskamp, Borsboom, & Fried, 2018; Fried et al., 2018; Williams & Rast, 2020). Contributing to this discussion, Epskamp and colleagues (2022) outlined several methods to gain insights into the accuracy of edge weights and the stability of centrality indices in the estimated network structure. These methods concern (a) the estimation of the accuracy of edge-weights, by drawing bootstrapped confidence intervals (CIs), (b) investigating the stability of (the order of) centrality indices, and (c) performing bootstrapped difference tests between edgeweights and centrality indices to test whether these differ significantly from each other. The estimation of the accuracy of edge-weights, by drawing bootstrapped confidence intervals (A) should always be performed, while the choice for method (B) and (C) depends on what is important to discover from the network (Epskamp, Borsboom, & Fried, 2018; Constantin et al., 2021). Studies on dynamic risk factors’ networks will often use methods (A) and (B) to gain insights into the accuracy of their network estimation because they will especially be interested in strength centralities of dynamic risk factors. 6.3.2.2 Towards establishing causality within networks of dynamic risk factors Networks of dynamic risk factors can be estimated from cross-sectional as well as longitudinal data. Cross-sectional data provides a contemporaneous network indicating to what extent dynamic risk factors are associated within the same window of measurement. These associations suggest at most potential causal relationships. After all, there is no certainty that variation in one dynamic risk factor causes change
RkJQdWJsaXNoZXIy MjY0ODMw