Perturbation graphs, invariant prediction and causal relations in psychology

Abstract

Networks (graphs) in psychology are often restricted to settings without interventions. Here we consider a framework borrowed from biology that involves multiple interventions from different contexts (observations and experiments) in a single analysis. The method is called perturbation graphs. In gene regulatory networks, the induced change in one gene is measured on all other genes in the analysis, thereby assessing possible causal relations. This is repeated for each gene in the analysis. A perturbation graph leads to the correct set of causes (not necessarily direct causes). Subsequent pruning of paths in the graph (called transitive reduction) should reveal direct causes. We show that transitive reduction will not in general lead to the correct underlying graph. There is however a close relation with another method, called invariant causal prediction. Invariant causal prediction can be considered as a generalisation of the perturbation graph method where including additional variables (and so conditioning on those variables) does reveal direct causes, and thereby replacing transitive reduction. We explain the basic ideas of perturbation graphs, transitive reduction and invariant causal prediction and investigate their connections. We conclude that perturbation graphs provide a promising new tool for experimental designs in psychology, and combined with invariant prediction make it possible to reveal direct causes instead of causal paths. As an illustration we apply the perturbation graphs and invariant causal prediction to a data set about attitudes on meat consumption.

Possible Futures for Network Psychometrics

“…However, the broader notion of conceptualizing constructs as networks invites the use of other modes of observation, which generate other kinds of data and require other kinds of models. An early extension of this type concerns the analysis of time series of momentary mood states (e.g., experience sampling and other forms of mobile assessment; (Bringmann et al, 2013), but other examples include results from activity monitoring (e.g., actigraphy data, location data), posts on social media (Kelley & Gillan, 2022;Golino et al, 2022), intervention data (Blanken et al, 2019;Waldorp et al, 2021), data on interactions with other individuals (Bodner et al, 2021), and data gathered at different levels of observation (e.g., registrations of brain activity, genetic data, social network data; Blanken et al, 2021).…”

“…In addition, extensions of the approaches of Ryan and Hamaker (2022) and Henry et al (2022) to related intervention situations such as persuasion (Zwicker et al, 2020) and education (Savi et al, 2019) would seem to be fruitful avenues to explore. Together with other recent approaches (Robinaugh et al, 2020;Lunansky et al, 2022;Blanken et al, 2019;Waldorp et al, 2021), a considerable set of methodologies to approach the analysis of interventions is now building up. I expect that the methodological study of these approaches as well as their substantive application will define an important research area in network psychometrics in the next decade.…”

Formalizing psychological interventions through network control theory

“…However, with the current network approach, the researcher will not be informed "how much" change in one component (e.g., rumination) induces "how much" change in others (e.g., sleep quality). Examples like this are numerous and include any study that contains an intervention such as controlled experiments with more than one condition (Waldorp et al, 2021). Related, the ability to predict the change in psychological behavior is a pivotal aspect of scientific discovery.…”

“…Here, we extend this approach to study the effects of psychological perturbations on a representative experiment aiming to modify the attitude towards eating meat (see Figure 1). Network models have previously been used to study attitudes and are shown to be psychometrically realistic formalizations (Dalege et al, 2016;Hoekstra et al, 2018;Waldorp et al, 2021). In this experiment, subjects are asked about their attitudes toward eating meat using an 11item questionnaire and are then subjected to psychological interventions in terms of hypothetical scenarios (see Methods for details).…”