Figure 3

Causal discovery using different experimental conditions. A linear causal model is used to describe the causal structure of the measured variables. Different experimental conditions (environments) possibly “shift” the values of some of the variables (i.e. \({c}_{j}^{i}\) can be zero for some i, j) while the causal structure of the variables remains invariant. BACKSHIFT exploits this invariance and can identify the causal structure given observations from at least three different environments.