Reading the System Diagram

Opening Scene

Every method this series has covered — matching, instrumental variables, difference-in-differences, regression discontinuity — produces a causal estimate only if certain specific assumptions actually hold: that matching captured every relevant confounder, that an instrument truly affects the outcome only through treatment, that parallel trends genuinely would have held, that subjects can’t manipulate a threshold. A plumber’s final report isn’t complete without stating clearly which parts of the system were directly verified and which were inferred from indirect evidence. A causal analysis deserves exactly the same honesty.

In Plain English

Every causal inference method rests on identifying assumptions — specific, often untestable claims about how the world works, that must hold true for the method’s statistical machinery to produce a genuinely valid causal estimate rather than just a number. Stating these assumptions explicitly, rather than burying them in technical fine print, is what separates a rigorous causal analysis from a confident-sounding calculation that happens to produce a number.

The Old Way

Before this discipline of explicit assumption-statement was standard, causal claims were often presented without this crucial context:

A statistical result presented as a definitive causal finding, without any accompanying account of what assumptions were required for that interpretation to be valid.
A causal claim defended purely by pointing to a sophisticated-sounding method, without addressing whether that method’s specific required assumptions were actually plausible in this case.
Assumptions buried in technical appendices or footnotes, rarely engaged with by anyone actually acting on the headline finding.

Making assumptions explicit and central, not incidental, is itself a genuine methodological advance this series has built toward throughout.

What’s Changing (and Why AI Is the Reason)

The causal graph framework from Article 5 has made explicit assumption-statement considerably more systematic, giving analysts a structured way to lay out and communicate exactly what’s being assumed.
Growing methodological maturity across this entire field has shifted expectations: a causal claim without a stated, scrutinized set of assumptions is increasingly treated with real skepticism by careful, informed readers.
As AI systems increasingly generate causal-sounding claims from data, the discipline of demanding explicit, checkable assumptions has become a genuine defense against confidently wrong, technologically dressed-up conclusions — connecting directly back to the folk causal reasoning warning from Article 3.

The Metaphor, Fully Extended

Behind the Wall	Assumptions Concept
A final plumbing report stating what was directly verified versus inferred	A causal analysis stating what was directly tested versus assumed
Assumptions buried in fine print, easy to overlook	Assumptions buried in technical appendices, easy to overlook
A rigorous report that puts key assumptions front and center	A rigorous analysis that puts key identifying assumptions front and center
The difference between confident-sounding work and genuinely verified work	The difference between a confident-sounding number and a genuinely valid causal estimate

For Beginners: What to Actually Do

Practice identifying the specific identifying assumption behind any causal method you use — every single one covered in this series has at least one.
Get comfortable asking “what would have to be true for this method to give a valid answer?” as a standard, habitual question.
Learn to distinguish a testable assumption (like parallel trends, checkable with pre-treatment data) from a fundamentally untestable one (like an instrument’s exclusion restriction).

For Practitioners and Leaders: The Deeper Layer

Require explicit statement of identifying assumptions in any causal analysis presented for a major decision, not buried in an appendix.
Build a genuine culture of scrutinizing assumptions, not just headline results, when evaluating causal claims.
Recognize this discipline as this series’ central throughline: every method covered, from Article 6’s randomized trial through Article 16’s sensitivity analysis, is ultimately about making assumptions explicit and testing them as rigorously as possible.

Quick Recap

Every causal inference method rests on specific identifying assumptions that must hold for its estimate to be genuinely valid.
Stating these assumptions explicitly is what separates rigorous causal analysis from a confident-sounding calculation.
The causal graph framework provides a structured way to communicate these assumptions clearly.
Demanding explicit, checkable assumptions is a genuine defense against confidently wrong causal claims, however sophisticated the underlying method.

Where This Fits in the Series

Article 17 covered making every assumption explicit. Article 18 turns to a genuinely modern risk: AI agents making decisions based on causal claims nobody has actually verified this carefully.