What does the evidence actually support?

What Does the Evidence Actually Support?

Evidence is often asked to carry more weight than it can bear. A correlation becomes a cause, a pilot becomes a market forecast, an expert view becomes a fact and a plausible story becomes the base case. High-quality analysis limits each conclusion to the strongest claim supported by the design, data and context that produced it.

Label the inferential step. Direct observations describe what was measured; estimates add sampling or model assumptions; causal conclusions require a credible counterfactual; forecasts add assumptions about future conditions; transfer claims assume one population behaves like another. Every step can be reasonable, but each widens uncertainty and must be visible to the decision-maker.

Assess evidence on four dimensions: provenance, measurement validity, representativeness and identification. Ask who generated the data, whether the metric captures the claimed concept, whom the sample excludes and what competing mechanism could produce the same result. Recency improves relevance but does not repair selection bias or a weak comparison group.

The 2025 AQuA Book makes assurance proportional to the financial, legal, operational and reputational risk of the intended use. It also separates verification from validation. This matters commercially: a dashboard may reproduce source data without error yet remain unfit for an investment decision because the coverage, horizon or causal interpretation is wrong.

Use a claim-evidence matrix to state the conclusion, evidence class, assumptions, confidence and boundary of use. Downgrade the wording when support is weak: �is associated with,� �suggests� and �would be consistent with� are not evasions when accurate. Precision in language protects option value by showing what is known, what is inferred and which next test could materially strengthen the case.