121 Idiographic personality networks 5 within situations, but the behavioral pattern is highly variable as a result of encountering different situations over time (Cervone, 2005; Mischel & Shoda, 1995). Recent idiographic research using machine-learning has shown considerable between-person differences in which situational features are personally relevant and the degree to which situations were predictive of behavior (Beck & Jackson, 2022). Thus, when estimating idiographic personality networks, researchers need to identify which situational features and internal components need to be modeled, and whether that differs between individuals. As a preliminary step, Bringmann et al. (2022) recently suggested conditions for selecting the network’s components. They pointed out that, in theory, nodes of psychological networks should be separately identifiable (i.e., able to be assessed separately) and independently malleable (i.e., outside influences should be able to have an effect on a node without it affecting any other nodes). Instead of trying to theorize non-stationarity away, it also possible to embrace non-stationarity as a feature of the personality system. Most analytical advances, however, characterize it as a to-be-overcome challenge (Ryan et al., 2023). It is worth noting that statistical timeseries analyses advance quickly, now allowing for some non-stationarity in the form of gradual mean-shifts (Bringmann et al., 2018) or a-priori specifications of the number of stable states the system has (e.g., Haslbeck & Ryan, 2021). Most of these models still assume that interactions among the variables are linear. There are, however, methodologists that move away from the linearity assumptions of statistical models and develop nonlinear analytical toolboxes. Nonlinear dynamics can for example be modeled with recurrence networks, in which the nodes represent time points, the edges connect recurring values, and the weights of the distance in time between two recurring values (Hasselman & Bosman, 2020). Although this descriptive method does not assume stationarity, it is considerably harder to intuitively interpret these networks. In terms of intuitiveness, linear models have an advantage, which is perhaps why they are more popular. However, our study demonstrates that this intuitiveness may be misleading. We therefore encourage future research to employ alternative network models (e.g., Bringmann et al., 2018; Hasselman and Bosman, 2020; Haslbeck & Ryan, 2021) when estimating idiographic personality structures. Importantly, this study showed that the GVAR model (Epskamp et al., 2018) did not accurately grasp stable personality within our sample. Also (more) advanced linear network analyses should not automatically be assumed to yield valid aggregated network estimates. Stationarity should always be examined.
RkJQdWJsaXNoZXIy MTk4NDMw