,
9 tweets,
3 min read
Read on Twitter
“Causality is non-quantitative, causal inference is quantitative” seems a nice solution to a definitional problem, but is it true? Not if you’re interested in advancing useful knowledge or in history it isn’t
1/n “Causal inference” (CI) is a notion with its origins in Plato, Hulme et al, and used formally by thinkers as distinct as G.E. Moore and Gilbert Ryle regarding empirical (and ultimately to-be-acted-on) inference
2/n A wide range of evidence goes into CI, some quantitative, some non-quantitative. Epidemiology is one discipline that has a long tradition of struggling with what approaches serve best as a basis for practical action ncbi.nlm.nih.gov/pubmed/23171877 academic.oup.com/aje/article/17…
3/n Not all evidence that would feed into a useful “triangulation” of support (and refutation) from as many useful independent sources is fully quantifiable nature.com/articles/d4158…
4/n CI is often strengthened by biological data that cannot be quantified in effect estimation, and even approaches interrogated in quantitative ways – such as “negative controls” - should not be used to purely adjust your quantitative estimate academic.oup.com/ije/article/47…
5/n The validity of Directed Acyclic Graphs (DAGs) depends on the “background knowledge” that go into constructing them. Where does this come from? Not purely the quantitative effects that pop out from analysing the prior DAG (or the infinite regress beckons)
6/n CI has, throughout its long history as a formal term, represented a much richer vein than effect estimation; as @neilpearce53x and @DeborahLawlor2 say “Causal inference: so much more than statistics” ncbi.nlm.nih.gov/pubmed/28204514
7/n As Kevin Hoover says on CI, it "is not a problem in statistical technique, it is a problem in the logic of empirical inference" cambridge.org/core/journals/…
8/n and quoting John Maynard Keynes 
