Confounding and control in a multivariate system. An issue in causal attribution

It is widely agreed that, in establishing whether variable X causes variable Y, a third variable Z may confound the relation and thus hinder causal assessment. The solution developed within the ‘traditional’ framework is to control for any third variable, susceptible of confounding the relation between X and Y. This paper examines complex systems of variables, characterised by multiple causes and multiple effects. The paper advances the view that in such contexts confounding is a moot issue, under a suitable specification of the causal model. When networks of causal relations are considered, possible confounders are included in the appropriate causal paths from the causes to the outcome. The challenge for the model builder then amounts to developing a structural model that specifies the role of variables in each path, rather than just controlling for possible confounders.