168 – Inference
Toward a General Theory of Observational Causal Inference
C. Sterling Portwood
Center for Interdisciplinary Science
Neither frequentist nor Bayesian paradigms can be used to draw causal inferences because the word "cause" is not a part of their derivations. Nevertheless, the physical/experimental sciences have been overwhelmingly successful, by intuitively extending these classical paradigms with unstated, but simple and generally acceptable assumptions. The nonexperimental sciences have had the opposite experience, because their assumptions required for extension are voluminous, complicated, and anything but generally acceptable. Further, Young and Karr (Significance, 2011) showed empirically that 100% of the causal findings from 55 "randomly selected," observational extension studies were incorrect. I summarize a new inquiring paradigm for causal inference called causal statistics and layout the assumptions foundational to its derivation and application. The outcome of the derivation is a series of simultaneous expressions in Pearl's do-calculus form. Causal statistics infers causal connections in samples. Classical statistics transports these results to populations. They are complementary. Much work remains to fully develop this new causal inquiring paradigm, but this research points the way.