Provenance rules

The canonical version of this standard lives in the repository, at method/provenance-rules.md. This page is the rendered copy.

The identity a number must carry, and when to refuse rather than report.

This is the standard behind series-identity, which resolves what a series actually is, and number-hygiene, which carries the identity record below as a required field set.

The failure being prevented

Not invented numbers. Invented numbers are a solved problem in the sense that everyone already knows to check them.

The failure is the right-looking number from the wrong series: a real figure, from a real source, correctly transcribed, that does not mean what the sentence around it claims. Nominal where real was intended. Not seasonally adjusted, compared against seasonally adjusted. A revised figure used to evaluate a decision taken when a different figure was on the table. An index whose base year is not the one assumed. A per-capita figure over the wrong denominator.

These pass every check a reader is likely to apply. The number cites a source and the source is correct. Only the identity is wrong.

The identity record

Every number that enters an analysis or a piece of writing carries:

Two of these are the ones that get dropped and matter most: transformation and vintage.

transformation matters because most reported numbers are not the published figure — they have been deflated, rebased, annualised, or divided by something. Each of those steps is a choice, and the choice changes the answer.

vintage matters because macroeconomic data is revised, sometimes by enough to change the sign of a quarter. A statement about what was known at a point in time requires the figure that was published at that time, not today’s.

Refusal rules

Refuse to report, rather than reporting with a hedge, when:

1. Adjustment status cannot be determined. Whether a series is seasonally adjusted is not inferable from its values. If it is not documented, the series cannot be compared to an adjusted one.
2. The deflator is unknown for a real series. “Constant prices” names a base year, not a price index. Different deflators give different answers, and the difference is not small for anything over a decade.
3. Units cannot be reconciled. Two figures that should be comparable and are not — because of currency, scale, coverage, or definition — must not be combined into a ratio or a difference.
4. The identity of a user-supplied figure cannot be established. Report it as the user’s figure. Do not adopt it into your own analysis as though it had been verified.
5. The number is recalled rather than retrieved. A specific statistic produced from a model’s memory is unreliable in precisely the way that looks confident. If it is checkable, check it; if it cannot be checked in the moment, say the figure needs verification instead of stating it.

A hedge is not a substitute for a refusal. The estimate travels; the hedge does not.

Precision

Report the precision the data supports and no more. Quarterly GDP growth to three decimal places asserts a precision that survives neither the revision process nor the sampling error.

Where a figure is uncertain, report the uncertainty: an interval, a standard error, a revision range, or the spread across alternative definitions. Where none of those can be computed, say so — that is itself information about the figure’s reliability.

Distinguish, in the words used:

• an outturn (published, subject to revision) from a forecast (a central projection) from a scenario (conditional on assumptions that are not predictions);
• a published figure from an own calculation based on a published figure.

Percentage points

A change from 4% to 6% is a rise of 2 percentage points, or 50 per cent. Reporting it as “a 2% rise” is wrong, common, and changes conclusions. Use percentage points for differences between percentages, always.

Why this is a standard and not a style guide

Every rule above exists because dropping it produces an error that is invisible in the output. A missing citation is visible. A missing vintage is not — the sentence reads identically whether or not the figure was the one available at the time.

That invisibility is why these are structural requirements rather than formatting preferences, and why the skills implement them as fields that must be present rather than as advice to be careful.

Related

• Econometric reporting standard — what an empirical result must contain before it is reportable.
• Reasoning standard — framework disclosure, contested versus settled questions, and the positive/normative line.

Licensed CC BY 4.0.