A structural equation model postulates causal relationships among a set of interacting variables. Each such model is naturally represented by a directed graph whose edges describe which of the other variables each variable causally depends on. Prior work has shown that under different model assumptions, this causal structure can be identified from observational data. We present two different approaches for calculating a confidence interval for the causal effect in two- and three-dimensional linear structural equation models where the error terms follow a centered normal distribution with equal variance. Both these methods are based on the inversion of hypothesis tests and account for the uncertainty of both the causal structure as well as the size of the causal effect. Whereas one of these methods relies on a classical approach of likelihood-ratio type, the other one is based on the framework of Universal Inference introduced in recent work by Wasserman et al. (2020). We compare both these methods with regards to their statistical power. In particular, we provide a local power analysis of the latter testing procedure in a general framework.      «

A structural equation model postulates causal relationships among a set of interacting variables. Each such model is naturally represented by a directed graph whose edges describe which of the other variables each variable causally depends on. Prior work has shown that under different model assumptions, this causal structure can be identified from observational data. We present two different approaches for calculating a confidence interval for the causal effect in two- and three-dimensional lin...     »