Abstract:
|
Statistics makes an important impact on society by analyzing quantitative evidence related to public policy issues regarding socioeconomic well-being which must be based only on non-experimental data. Suppes' probabilistic causality theory establishes inequalities among probabilities of events. Instead of events, our ``cause' is a self-driven data generating process (DGP). We develop three criteria based on properties residuals of flipped kernel regression conditional expectation functions, Ef(X_j|X_i, X_k) and Ef(X_i|X_j, X_k). Our unanimity index aggregates measures of four orders of stochastic dominance and new asymmetric partial correlation coefficients, which yields decision rules for quantifying percent support for the competing causal paths X_i-->X_j, X_j--> X_i, X_i <--> X_j. A simulation supports our decision rules illustrated by many real-world examples, including the causes of US recession, policies for encouraging private investment in India and assessment of effective advertising media.
|