Deep Causal Graphs for Causal Inference, Black-Box Explainability and Fairness

Causal Estimation is usually tackled as a two-step process: identification, to transform a causal query into a statistical estimand, and modelling, to compute this estimand by using data. This reliance on the derived statistical estimand makes these methods ad hoc, used to answer one and only one query. We present an alternative framework called Deep Causal Graphs: with a single model, it answers any identifiable causal query without compromising on performance, thanks to the use of Normalizing Causal Flows, and outputs complex counterfactual distributions instead of single-point estimations of their expected value. We conclude with applications of the framework to Machine Learning Explainability and Fairness.

Centrality-oriented causality. A sudy of EU agricultural subsidies and digital development in Poland

Results of a convincing causal statistical inference related to socio-economic phenomena are treated as an especially desired background for conducting various socio-economic programmes or government interventions. Unfortunately, quite often real socio-economic issues do not fulfil restrictive assumptions of procedures of causal analysis proposed in the literature. This paper indicates certain empirical challenges and conceptual opportunities related to applications of procedures of data depth concept into a process of causal inference as to socio-economic phenomena. We show how to apply statistical functional depths in order to indicate factual and counterfactual distributions commonly used within procedures of causal inference. Thus, a modification of Rubin causality concept is proposed, i.e., a centrality-oriented causality concept. The presented framework is especially useful in the context of conducting causal inference based on official statistics, i.e., on the already existing databases. Methodological considerations related to extremal depth, modified band depth, Fraiman-Muniz depth, and multivariate Wilcoxon sum rank statistic are illustrated by means of example related to a study of an impact of EU direct agricultural subsidies on digital development in Poland in the period 2012-2018.

Estimation and Inference in an Ecological Inference Model

We interpret an ecological inference model as a treatment effects model in which the outcomes of interest and the conditional covariates come from separate datasets. In this setting, the counterfactual distributions and policy parameters of interest are only partially identified under a selection on observables assumption. In this paper, we provide estimation and inference procedures for structural prediction and counterfactual analysis in such models. We also illustrate the procedures with an application to US presidential elections.