f7be3ebca4980b59fe3f665011115395-Supplemental-Datasets_and_Benchmarks_Track.pdf

Marginalisation Operation Suppose V can be split as V = Ṽ [ Ũ where we are interested in the causal structure over Ṽ and do not observe the variables Ũ. For all U 2 Ũ, add an edge Z! Z if the current graph contains Z! U! Z and then U Figure 5: Example of step one in the marginalisation, taken from Evans [22]. The explanation for the biases are given in Appendix E. In this section we analyse the datasets presented in Le Quy et al. [43] for the three biases we present For each bias we provide a justification of our decision. Therefore we drop them from the analysis. COMPAS have been well documented to exhibit all these biases and more [5].