Cross-lagged network models do not prove causality: Results from simulations and analyses of symptoms of depression and anxiety

The cross-lagged panel network (CLPN) model is a version of the classic cross-lagged panel model (CLPM). The CLPN can be used to model a large number of cross-lagged effects between nodes (such as symptoms of ill health) across two waves of measurement. In spite of a robust debate about the pitfalls of using correlational data to infer causality, researchers using the CLPN often interpret findings in causal terms. We show with simulations that CLPN can indicate cross-lagged effects even if data have been generated without any direct effects between nodes/symptoms. Hence, findings from CLPN:s may be spurious and should not be interpreted in causal terms if used to model non-experimental data. We have previously proposed a method of triangulation to scrutinize findings from the CLPM. Triangulation can also be used to reduce the false-positive risk when using the CLPN. We show by reanalysis of data from a previous study how triangulation can be applied, finding that positive cross-lagged effects between some symptoms of depression and anxiety probably were spurious, rather than due to genuine increasing effects, as alternative models indicated simultaneous and paradoxical increasing and decreasing effects.