AURmetric
Methodology

How the Actual Rate is measured

A reproducible method for estimating unemployment and labor underutilization that captures the people the headline U-3 rate misses — discouraged and marginally attached workers, involuntary part-timers, and informal or undocumented labor — while counting people, not paychecks, and reporting ranges, not false precision.

National series built on public BLS, Census, and QCEW data. The informal component is a modeled estimate with explicit error bands.

Contents

  1. The unit: person, not job
  2. Labor-force accounting
  3. The U-1…U-8 ladder
  4. Data sources
  5. Estimating informal work
  6. Double-counting controls
  7. Uncertainty & reporting
  8. Relation to official U-6
The honest ceiling. No method can count unreported work — those transactions are intentionally invisible to administrative systems. AURmetric therefore produces a modeled estimate with explicit error bands, not a single precise number. Anyone presenting the informal-adjusted figures as an exact value is misrepresenting them.

1.The unit of measurement: the person, not the job

The foundational rule is one row per person per reference week, keyed by a stable pseudonymous identifier.

This rule is what prevents the classic error of inflating employment by counting paychecks.

2.Labor-force accounting identities

Every working-age resident (16+) is assigned to exactly one mutually exclusive status in the reference week:

StatusCodeDefinition
Employed — full-timeE_FTWorked ≥1 hour for pay; usually ≥35 hrs/wk
Employed — part-time, voluntaryE_PTV<35 hrs, does not want full-time
Employed — part-time, involuntaryE_PTI<35 hrs, wants and available for full-time
Employed — informal onlyE_INFWorked for pay, no administrative footprint
UnemployedUNo job, available, actively searched in prior 4 weeks
Marginally attachedMAWants work, searched in prior 12 mo but not last 4 wks
— of which discouragedMA_DSubset of MA not searching for job-market reasons
Not in labor forceNILFWants no job / unavailable

The identities below must hold exactly — enforced as automated tests:

Population_16+  = Employed + Unemployed + NotInLaborForce
Employed        = E_FT + E_PTV + E_PTI + E_INF
Employed_formal = E_FT + E_PTV + E_PTI            # excludes informal-only
LF_formal       = Employed_formal + Unemployed    # official basis (U1-U6)
LF_complete     = LF_formal + E_INF               # reveals hidden employment (U7-U8)

Formal vs. complete labor force. U-1…U-6 mirror official BLS definitions and use LF_formal, which excludes informal-only workers — the informal estimate never touches the official series. U-7/U-8 expand the denominator to LF_complete. Keeping these separate is what makes U-7 a genuine "reveal hidden employment" adjustment rather than a double count.

3.The output measures: U-1 through U-8

AURmetric reports the full BLS-style ladder plus two informal-adjusted measures, keeping the method comparable to official statistics while isolating the novel part.

# LF = LF_formal for U1-U6:
U1 = (unemployed >=15 weeks)                    / LF_formal
U2 = (job losers + completed temp jobs)         / LF_formal
U3 = U                                          / LF_formal        # headline
U4 = (U + discouraged)                          / (LF_formal + discouraged)
U5 = (U + all marginally attached)              / (LF_formal + MA)
U6 = (U + MA + E_PTI)                           / (LF_formal + MA)  # broadest official

# Informal-adjusted (complete labor force):
U7 = (U + MA + E_PTI)                           / (LF_complete + MA)
U8 = (U + MA + E_PTI + informal_underemployed)  / (LF_complete + MA)

Each measure is emitted as a point estimate plus a 90% interval. U-3…U-6 have narrow intervals (survey sampling error only); U-7/U-8 carry the wide informal-economy error and should never be quoted without the band.

4.Data sources and their roles

SourceCapturesBlind spots
Payroll / UI wage recordsFormal employees, earnings, hoursMisses informal, self-employed, undocumented
Tax filingsFormal + some self-reported incomeMisses fully unreported cash work
UI claimsRecent job losers certifyingMisses ineligible / exhausted claimants
Household survey (CPS-style)Self-reported status, search, availabilitySampling error; misreport of informal work
Anonymous supplemental surveyInformal & undocumented workNo individual linkage; estimate-only
National-accounts dataMacro size of informal economyIndirect; model-dependent

Linkage vs. estimation. Administrative sources are record-linked on the person ID to build an exact formal-employment spine. Survey sources feed population-level estimates added on top — precision where records exist, estimation where they cannot. The live public series on this site is the U-6 basis from the BLS Public Data API; see the Sources page for exact datasets and vintages.

5.Estimating the informal / undocumented component

This is the only modeled block and the sole source of the wide error band. It is estimated three independent ways and reconciled.

5.1 Direct anonymous survey

A large survey asking whether the respondent did paid work not reported to tax authorities. Anonymity and a no-linkage design raise truthful response, especially among undocumented residents. Yields an informal rate with a sampling interval, adjusted for nonresponse and social desirability.

5.2 Discrepancy / trace method

Compare reported earnings against observed consumption, currency demand, or output proxies at the aggregate level. The gap that formal income cannot explain implies informal activity.

5.3 Residual demographic method

Estimate total resident working-age population (census + demographic balancing, which counts residents regardless of status) and subtract those in administrative records. The unexplained residual, cross-checked against migration estimates, bounds the undocumented-worker pool.

5.4 Reconciliation — DerSimonian–Laird random effects

The three are treated as noisy estimates of one latent quantity. Fixed-effect inverse-variance weighting is the starting point; a between-method heterogeneity term τ² (from Cochran's Q) is added to every method's variance before reweighting. When the methods agree, τ² → 0 and it reduces to plain inverse-variance weighting; when they diverge, τ² grows and the 90% band widens — divergence becomes information, not noise to be averaged away. Implemented in accurate_unemployment/reconcile.py.

6.Double-counting controls

7.Uncertainty and reporting rules

8.Relation to the official U-6

This is U-6 plus an informal-economy extension, not a replacement for the BLS ladder. Reporting U-1…U-6 unchanged preserves comparability and lets users see exactly what the informal adjustment (U-7, U-8) adds. Any deviation from official definitions (age cutoff, reference week, search window) is documented so the numbers stay interpretable against published series. AURmetric is independent and is not affiliated with or endorsed by the U.S. Bureau of Labor Statistics.