The Contested Baseline: Decision-Making When Official Statistics Lose the Benefit of the Doubt
By Diana Whitfield, Chief Economist
The machinery behind official economic statistics is visibly straining. The US Government Accountability Office reports that response rates to the establishment survey behind the monthly US jobs report fell from 62 to 42 per cent between October 2015 and September 2025, and that three statistical advisory committees were eliminated in 2025 — removing the Bureau of Labor Statistics' regular external technical scrutiny.
Yet the outputs remain statistically normal on every formal test. That asymmetry frames this quarter's decision question: not whether official statistics can be trusted, but what stops an organisation over-correcting — swapping audited, self-revising series for unaudited private alternatives. In our reading, over-correction is now the greater risk.
The Inputs Are Degrading, Measurably
The decay is broader than one survey: the GAO finds household-survey response rates fell from 87 to 69 per cent over the same decade, warns the Bureau "faces risks to data quality", and notes a planned 2027 online-collection fix may slip on funding constraints. The eliminated committees, including the Federal Economic Statistics Advisory Committee, were the formal channel for external methodological scrutiny; the GAO's headline recommendation is a plan to close the gap.
Trust itself now carries a price. Analysis published by CEPR estimates that the August 2025 dismissal of the Bureau's Commissioner amid data-manipulation allegations drove a policy-uncertainty spike consistent with roughly US$20 billion in lost US GDP — part of a cross-country pattern of political pressure, falling response rates and harder-to-measure economies.
The Outputs Refuse to Break
If degraded inputs were corrupting outputs, the annual benchmark revision — the survey's public reconciliation against the Quarterly Census of Employment and Wages, a near-census of unemployment-insurance tax records — is where it would show. The Federal Reserve Bank of Cleveland ran that test on 1980–2025 revisions, with checks back to 1965, and finds no structural break: most sit inside the Bureau's ±0.5 per cent acceptable range, and even the record 2025 markdown (-0.54 per cent) was, in its words, "just marginally outside" the normal range. Revisions also predict future ones — information, not scandal.
The most direct test came on 28 August 2026, when the Bureau reported a preliminary benchmark revision of -79,000 (-0.1 per cent) to March 2026 nonfarm employment — an order of magnitude smaller than 2024's -818,000 preliminary or 2025's record -911,000. The first benchmark after the series was politicised was the smallest in years. A series that never revised against administrative truth would be the red flag; large revisions are the audit working in public. Treating a preliminary print as final truth is the cousin error of conflating trade value with trade volume.
Governing Triangulation
The disciplined response is neither faith nor abandonment but triangulation governance: official series as spine, revisions read as signal, cross-checks against indicators that publish their methods.
| Indicator | What has degraded (inputs) | What remains stable (outputs) | How to use it now |
|---|---|---|---|
| Monthly payrolls (establishment survey) | Response rate 62% → 42%, 2015–2025 (GAO); advisory committees gone | No structural break since 1980 (Cleveland Fed); 2026 preliminary benchmark revision -0.1% (BLS) | The spine; widen the error band, keep the level |
| Benchmark revisions | Wider recent survey-to-census gaps: -818,000 (2024), record -911,000 (2025) | 2026 preliminary -79,000 (-0.1%), inside the ±0.5% norm; predictive of future revisions | Audit trail and leading signal, not manipulation in either direction |
| Private high-frequency indicators | Most never published response rates, revision policies or scored forecasts | Scoreable only where methods are public | Cross-check direction and breadth; never a substitute for the official level |
The same posture we applied to logistics indices under chokepoint stress applies here: an index score is a hypothesis, not a verdict — the jobs report included, and our own index.
Holding Our Own Index to the Same Standard
The August 2026 edition of the Logistics Index prints the World reading at 51.7, down 0.2 points: expansion, momentum easing. Breadth matters more — eight of ten sub-indexes remain in expansion but only three are rising, with Warehouse Capacity, below 50 and down 1.3 points, the main drag. And the scoreboard: July forecast 52.0 for August; the print was 51.7 — a 0.3-point miss, logged, not buried. An indicator publishing none of this is not more trustworthy than the official series it claims to replace — merely less falsifiable. Its stated limits — GDP-weighted diffusion, 42 economies, not seasonally adjusted, short history — are why we favour three-month trajectory and breadth over single prints.
What Could Change the Verdict
- The early-2027 final benchmark. The preliminary -79,000 could widen when the final lands (2024's moved from -818,000 to -598,000); a materially larger final would weaken the outputs-stable leg.
- Cumulative input decay. A 42 per cent response rate with no advisory committees raises variance even without bias; whether the band holds as rates pass the low 40s is open.
- Institutionalised distrust. Trust strain carries macroeconomic cost independent of data quality; further committee eliminations, funding cuts or a contested appointment could turn an inputs problem into an outputs problem — the one scenario where over-correction stops being the error.
What This Means for Decision-Makers
- Audit FY2027 planning assumptions for single-source dependency now. The test is not whether the number is right, but what the decision does if the input revises by ±0.5 per cent — the historically normal band.
- Adopt triangulation governance, not indicator replacement. Official series as spine; two independent cross-checks, one administrative, one high-frequency private; divergence is a flagged event with an owner, not licence to pick the friendlier number.
- Score every input, official or private. Demand published methodology, weights, a corrections log and a forecast record from any series you rely on.
- Treat revisions as information with sign and size. In our reading, the 2025 markdown signalled weaker hiring months before the narrative caught up: read as scandal, noise; read as data, signal.
Evidence and data cut-off: 12 September 2026. Third-party figures are as published by the cited sources; Logistics Index readings are the World Research Institute's own. See our Research Standards.
The World Research Institute provides macroeconomic monitoring, data-governance advisory, and evidence-led market intelligence. Contact our team to commission tailored research.