Official Statistics: Governing Definitions That Matter

Why Official Statistics Need Governed Definitions

When a national statistics office reports that “unemployment fell to 4.2%”, that single number carries enormous weight. Policy decisions, budget allocations, and political debates all hinge on it.

But what does “unemployed” actually mean?

• Must someone be actively seeking work?
• What counts as “actively seeking”?
• Are discouraged workers included?
• What about gig economy workers with irregular hours?

Different answers to these questions produce different unemployment rates — potentially by several percentage points. The definition IS the statistic.

This vignette shows how ontologyR helps statistical agencies:

1. Document definitions explicitly — no more buried footnotes
2. Version definitions over time — track methodological changes
3. Test definitions against reality — catch when proxies drift
4. Govern changes properly — require approval before publishing
5. Maintain international comparability — manage multiple scopes

---

Scenario: Labour Force Statistics

You’re the methodology team at a national statistics office. You need to manage definitions for the Labour Force Survey (LFS), ensuring:

• Consistency with ILO international standards
• Adaptation for national context
• Historical comparability
• Quality assurance before publication

Let’s build this system.

---

Phase 1: Setting Up the Survey Infrastructure

library(ontologyR)

# Connect to the statistics office ontology database
ont_connect("labour_force_ontology.duckdb")

# In reality, this would be your survey microdata warehouse
# Here we simulate LFS respondent data
DBI::dbWriteTable(ont_get_connection(), "lfs_respondents", tibble::tibble(
  respondent_id = paste0("R", 1:10000),
  age = sample(16:75, 10000, replace = TRUE),
  worked_last_week = sample(c(TRUE, FALSE), 10000, replace = TRUE, prob = c(0.6, 0.4)),
  hours_worked = ifelse(worked_last_week, sample(1:50, 10000, replace = TRUE), 0),
  has_job_attached = sample(c(TRUE, FALSE), 10000, replace = TRUE, prob = c(0.7, 0.3)),
  sought_work_4weeks = sample(c(TRUE, FALSE), 10000, replace = TRUE, prob = c(0.3, 0.7)),
  available_within_2weeks = sample(c(TRUE, FALSE), 10000, replace = TRUE, prob = c(0.8, 0.2)),
  reason_not_seeking = sample(c("discouraged", "family", "student", "retired", "other", NA),
                               10000, replace = TRUE),
  education_level = sample(c("primary", "secondary", "tertiary"), 10000, replace = TRUE),
  region = sample(c("Capital", "North", "South", "East", "West"), 10000, replace = TRUE),
  survey_wave = rep(c("2024Q1", "2024Q2", "2024Q3", "2024Q4"), each = 2500)
))

# Register source dataset
ont_register_dataset(
  dataset_id = "ds_lfs_respondents",
  dataset_name = "Labour Force Survey Respondents",
  physical_name = "lfs_respondents",
  dataset_type = "source",
  owner = "social_statistics_division",
  description = "Quarterly LFS microdata, respondent level"
)

# Register the object type
ont_register_object(
  object_type = "LFS_Respondent",
  table_name = "lfs_respondents",
  pk_column = "respondent_id",
  description = "Individual respondent in the Labour Force Survey",
  owner_domain = "labour_statistics"
)

---

Phase 2: Defining ILO-Compliant Concepts

The International Labour Organization (ILO) provides standard definitions. We’ll implement these as our baseline.

Employment Definition

# Define the employment concept
ont_define_concept(
  concept_id = "employed",
  object_type = "LFS_Respondent",
  description = "Person is employed according to ILO definition",
  owner_domain = "labour_statistics"
)

# ILO standard: Worked at least 1 hour in reference week, OR temporarily absent from job
ont_add_version(
  concept_id = "employed",
  scope = "ilo_standard",
  version = 1,
  sql_expr = "worked_last_week = TRUE OR (has_job_attached = TRUE AND worked_last_week = FALSE)",
  status = "active",
  rationale = "ILO 19th ICLS definition: worked 1+ hours OR temporarily absent with job attachment"
)

# National adaptation: Same core, but restricted to working age (16-64)
ont_add_version(
  concept_id = "employed",
  scope = "national_headline",
  version = 1,
  sql_expr = "(worked_last_week = TRUE OR has_job_attached = TRUE) AND age BETWEEN 16 AND 64",
  status = "active",
  rationale = "National headline rate uses 16-64 working age population"
)

Unemployment Definition

This is where it gets interesting — unemployment has more nuance.

ont_define_concept(
  concept_id = "unemployed",
  object_type = "LFS_Respondent",
  description = "Person is unemployed according to standard definition",
  owner_domain = "labour_statistics"
)

# ILO standard unemployment (strict)
ont_add_version(
  concept_id = "unemployed",
  scope = "ilo_standard",
  version = 1,
  sql_expr = "
    worked_last_week = FALSE
    AND has_job_attached = FALSE
    AND sought_work_4weeks = TRUE
    AND available_within_2weeks = TRUE
    AND age >= 16
  ",
  status = "active",
  rationale = "ILO 19th ICLS: without work, seeking work, available for work"
)

# Extended unemployment including discouraged workers
ont_add_version(
  concept_id = "unemployed",
  scope = "extended_measure",
  version = 1,
  sql_expr = "
    worked_last_week = FALSE
    AND has_job_attached = FALSE
    AND (
      (sought_work_4weeks = TRUE AND available_within_2weeks = TRUE)
      OR reason_not_seeking = 'discouraged'
    )
    AND age >= 16
  ",
  status = "active",
  rationale = "Extended measure includes discouraged workers per Eurostat LFS guidelines"
)

Labour Force and Derived Rates

ont_define_concept(
  concept_id = "in_labour_force",
  object_type = "LFS_Respondent",
  description = "Person is in the labour force (employed or unemployed)",
  owner_domain = "labour_statistics"
)

ont_add_version(
  concept_id = "in_labour_force",
  scope = "ilo_standard",
  version = 1,
  sql_expr = "
    age >= 16 AND (
      worked_last_week = TRUE
      OR has_job_attached = TRUE
      OR (sought_work_4weeks = TRUE AND available_within_2weeks = TRUE)
    )
  ",
  status = "active",
  rationale = "Labour force = employed + unemployed"
)

# Underemployment (wants more hours)
ont_define_concept(
  concept_id = "underemployed",
  object_type = "LFS_Respondent",
  description = "Employed but working fewer hours than desired",
  owner_domain = "labour_statistics"
)

ont_add_version(
  concept_id = "underemployed",
  scope = "national_headline",
  version = 1,
  sql_expr = "
    worked_last_week = TRUE
    AND hours_worked < 35
    AND available_within_2weeks = TRUE
  ",
  status = "active",
  rationale = "Part-time workers available and wanting more hours"
)

---

Phase 3: Quality Assurance Through Auditing

Before publishing statistics, methodology teams should verify that definitions capture what they intend.

Sampling for Expert Review

# Sample respondents classified as unemployed for expert review
unemployed_sample <- ont_sample_for_audit(
  concept_id = "unemployed",
  scope = "ilo_standard",
  n = 50,
  concept_value = TRUE,
  seed = 42  # Reproducible sample
)

# Sample edge cases: not seeking but potentially discouraged
edge_cases <- ont_sample_for_audit(
  concept_id = "unemployed",
  scope = "ilo_standard",
  n = 30,
  where = "sought_work_4weeks = FALSE AND reason_not_seeking IS NOT NULL",
  concept_value = FALSE  # System says "not unemployed"
)

Recording Expert Judgments

Labour statisticians review individual cases against the full questionnaire.

# Expert reviews case R1234: System says unemployed, expert agrees
ont_record_audit(
  concept_id = "unemployed",
  scope = "ilo_standard",
  version = 1,
  object_key = "R1234",
  system_value = TRUE,
  reviewer_value = TRUE,
  reviewer_id = "senior_methodologist",
  notes = "Clear case: actively seeking, sent 5 applications last week"
)

# Expert reviews case R5678: System says unemployed, but expert disagrees!
ont_record_audit(
  concept_id = "unemployed",
  scope = "ilo_standard",
  version = 1,
  object_key = "R5678",
  system_value = TRUE,
  reviewer_value = FALSE,
  reviewer_id = "senior_methodologist",
  notes = "Respondent waiting to start job next month - should be employed (job attached)"
)

# Batch import from expert review spreadsheet
audit_results <- readr::read_csv("expert_review_q4_2024.csv")
for (i in seq_len(nrow(audit_results))) {
  ont_record_audit(
    concept_id = audit_results$concept[i],
    scope = audit_results$scope[i],
    version = audit_results$version[i],
    object_key = audit_results$respondent_id[i],
    system_value = audit_results$system_classification[i],
    reviewer_value = audit_results$expert_classification[i],
    reviewer_id = audit_results$reviewer[i],
    notes = audit_results$notes[i]
  )
}

Checking Audit Results

# How well does our definition match expert judgment?
ont_audit_summary("unemployed", "ilo_standard", 1)
#> -- Audit Summary: unemployed@ilo_standard v1 --
#> i Total audits: 80
#> i Agreements: 72 (90%)
#> i Disagreements: 8 (10%)
#>
#> Most common disagreement reasons:
#>   - Job attachment edge cases: 5
#>   - Availability interpretation: 2
#>   - Seeking work definition: 1

---

Phase 4: Detecting and Managing Drift

Over time, definitions can drift from reality. New forms of work emerge, survey questions evolve, or classifier behavior changes.

Setting Up Drift Monitoring

# Check if unemployment definition is drifting
drift_result <- ont_detect_drift(
  concept_id = "unemployed",
  scope = "ilo_standard",
  threshold = 0.10,  # Alert if >10% disagreement
  min_audits = 50,
  window_days = 90
)

if (!drift_result$ok) {
  cli::cli_alert_warning("DRIFT DETECTED in unemployment definition!")
  cli::cli_alert_info("Disagreement rate: {drift_result$disagreement_rate}")
  cli::cli_alert_info("Consider methodology review before next publication")
}

# View overall drift status across all concepts
ont_drift_status()
#> -- Drift Status Summary --
#> v employed@ilo_standard v1: OK (5% disagreement, 120 audits)
#> v employed@national_headline v1: OK (6% disagreement, 85 audits)
#> ! unemployed@ilo_standard v1: WARNING (12% disagreement, 80 audits)
#> v in_labour_force@ilo_standard v1: OK (4% disagreement, 100 audits)

Investigating Drift

When drift is detected, investigate the patterns.

# Get disagreeing cases
disagreements <- ont_get_audits(
  concept_id = "unemployed",
  scope = "ilo_standard",
  version = 1
) |>
  dplyr::filter(system_value != reviewer_value)

# Analyze patterns
disagreements |>
  dplyr::left_join(lfs_respondents, by = c("object_key" = "respondent_id")) |>
  dplyr::group_by(has_job_attached, sought_work_4weeks) |>
  dplyr::summarise(
    n = n(),
    pct_system_true = mean(system_value)
  )
#> # A tibble: 4 x 4
#>   has_job_attached sought_work_4weeks     n pct_system_true
#>   <lgl>            <lgl>              <int>           <dbl>
#> 1 FALSE            TRUE                   3           1.0
#> 2 TRUE             FALSE                  5           0.0   # <-- Problem area!

# The issue: People with job attachment but not working/seeking
# Our SQL says "not unemployed", experts say "should check job status"

Resolving Drift

# Create improved definition based on investigation
ont_add_version(
  concept_id = "unemployed",
  scope = "ilo_standard",
  version = 2,
  sql_expr = "
    worked_last_week = FALSE
    AND (has_job_attached = FALSE OR (has_job_attached = TRUE AND hours_worked = 0))
    AND sought_work_4weeks = TRUE
    AND available_within_2weeks = TRUE
    AND age >= 16
  ",
  status = "draft",
  rationale = "v2: Clarified job attachment for zero-hours cases per audit findings"
)

# Compare versions
comparison <- ont_compare_versions("unemployed", "ilo_standard", v1 = 1, v2 = 2)
comparison$summary
#> # A tibble: 1 x 4
#>   total_objects v1_only v2_only both_true
#>           <int>   <int>   <int>     <int>
#> 1         10000      45     120       380

# v2 captures 120 additional unemployed (zero-hours job attachment cases)

---

Phase 5: Governance Gates for Publication

Before publishing official statistics, definitions must pass quality gates.

Configuring Publication Gates

# Create a custom gate for statistical publications
ont_create_gate(
  gate_id = "gate_statistical_quality",
  gate_name = "Statistical Quality Standards",
  gate_type = "audit_coverage",
  applies_to = "activation",
  conditions = list(
    min_audits = 50,
    min_agreement_rate = 0.90  # 90% agreement required
  ),
  severity = "blocking",
  description = "Definitions must achieve 90% expert agreement before publication use"
)

# Require methodology committee approval
ont_create_gate(
  gate_id = "gate_methodology_approval",
  gate_name = "Methodology Committee Sign-off",
  gate_type = "approval_required",
  applies_to = "activation",
  conditions = list(
    min_approvals = 2,
    approver_roles = c("chief_methodologist", "division_head")
  ),
  severity = "blocking",
  description = "Major definition changes require methodology committee approval"
)

Going Through the Approval Process

# Check if v2 unemployment can be activated
gates <- ont_check_all_gates("unemployed", "ilo_standard", 2, "activation")

gates$overall_passed
#> [1] FALSE

gates$blocking_failures
#> $gate_statistical_quality
#>   passed: FALSE
#>   details: audit_count = 25 (need 50)
#>
#> $gate_methodology_approval
#>   passed: FALSE
#>   details: approvals = 0 (need 2)

# Conduct more audits for v2
v2_sample <- ont_sample_for_audit("unemployed", "ilo_standard", n = 30, version = 2)
# ... expert review process ...

# Request methodology committee approval
ont_request_approval(
  concept_id = "unemployed",
  scope = "ilo_standard",
  version = 2,
  requested_action = "activate",
  requested_by = "methodology_team"
)
#> v Created approval request: REQ-20250116-abc123

# Committee members review and approve
ont_approve_request("REQ-20250116-abc123", "chief_methodologist",
                     "Reviewed v2 changes. Aligns with ILO guidance on zero-hours contracts.")
ont_approve_request("REQ-20250116-def456", "division_head",
                     "Approved. Impact analysis shows minimal effect on headline rate.")

# Now gates pass
gates <- ont_check_all_gates("unemployed", "ilo_standard", 2, "activation")
gates$overall_passed
#> [1] TRUE

# Activate for publication use
ont_activate_version("unemployed", "ilo_standard", 2, activated_by = "chief_methodologist")

---

Phase 6: Materializing Publication Tables

Official statistics need reproducible, point-in-time outputs.

# Materialize unemployment classifications for Q4 2024 publication
result <- ont_materialize(
  concept_id = "unemployed",
  scope = "ilo_standard",
  filter_expr = "survey_wave = '2024Q4'",
  output_table = "pub_unemployed_2024q4",
  triggered_by = "quarterly_publication",
  executed_by = "production_team"
)
#> v Materialized unemployed to pub_unemployed_2024q4
#> i 2,500 rows (Q4 respondents)

# Materialize all labour force concepts for the publication
for (concept in c("employed", "unemployed", "in_labour_force", "underemployed")) {
  ont_materialize(
    concept_id = concept,
    scope = "national_headline",
    filter_expr = "survey_wave = '2024Q4'",
    output_table = paste0("pub_", concept, "_2024q4"),
    triggered_by = "quarterly_publication"
  )
}

# Calculate and store headline rates
DBI::dbExecute(ont_get_connection(), "
  CREATE TABLE pub_headline_rates_2024q4 AS
  SELECT
    'Q4 2024' as period,
    COUNT(*) FILTER (WHERE e.concept_value) as employed,
    COUNT(*) FILTER (WHERE u.concept_value) as unemployed,
    COUNT(*) FILTER (WHERE lf.concept_value) as labour_force,
    ROUND(100.0 * COUNT(*) FILTER (WHERE u.concept_value) /
          NULLIF(COUNT(*) FILTER (WHERE lf.concept_value), 0), 1) as unemployment_rate
  FROM pub_employed_2024q4 e
  JOIN pub_unemployed_2024q4 u USING (respondent_id)
  JOIN pub_in_labour_force_2024q4 lf USING (respondent_id)
")

DBI::dbGetQuery(ont_get_connection(), "SELECT * FROM pub_headline_rates_2024q4")
#>    period employed unemployed labour_force unemployment_rate
#> 1 Q4 2024     1847        198         2045               9.7

---

Phase 7: Full Provenance for Reproducibility

When questions arise about published statistics, trace everything back.

# "Where did the 9.7% unemployment rate come from?"
prov <- ont_get_provenance("DS-pub_unemployed_2024q4")

# What definition was used?
prov$concept
#> $concept_id
#> [1] "unemployed"
#>
#> $scope
#> [1] "ilo_standard"
#>
#> $version
#> [1] 2
#>
#> $sql_expr
#> [1] "worked_last_week = FALSE AND ..."

# When was it generated?
prov$last_run$ended_at
#> [1] "2025-01-15 09:30:00"

# Who approved this definition?
ont_get_gate_history("unemployed", "ilo_standard", 2)
#> Shows approval chain with timestamps and approvers

# What was the audit quality?
ont_audit_summary("unemployed", "ilo_standard", 2)
#> Shows 92% expert agreement

# Full lineage
ont_get_upstream("DS-pub_unemployed_2024q4")
#> DS-lfs_respondents -> DS-pub_unemployed_2024q4

---

Phase 8: Managing International Comparability

Statistical agencies often need multiple versions for different purposes.

# Same concept, different scopes
ont_list_versions("unemployed")
#> # A tibble: 4 x 5
#>   concept_id scope             version status     rationale
#>   <chr>      <chr>               <int> <chr>      <chr>
#> 1 unemployed ilo_standard            2 active     ILO 19th ICLS with zero-hours fix
#> 2 unemployed national_headline       1 active     16-64 age restriction
#> 3 unemployed extended_measure        1 active     Includes discouraged workers
#> 4 unemployed eurostat_lfs            1 active     EU-LFS compatible definition

# Compare scopes for the same quarter
scopes <- c("ilo_standard", "national_headline", "extended_measure")

comparison_results <- lapply(scopes, function(s) {
  ont_evaluate("unemployed", s) |>
    dplyr::summarise(
      scope = s,
      unemployed_count = sum(concept_value),
      rate = round(100 * mean(concept_value), 1)
    )
}) |> dplyr::bind_rows()

comparison_results
#> # A tibble: 3 x 3
#>   scope             unemployed_count  rate
#>   <chr>                        <int> <dbl>
#> 1 ilo_standard                   823   8.2
#> 2 national_headline              756   9.7  # Different base (16-64 only)
#> 3 extended_measure               981  10.3  # Includes discouraged

---

Key Benefits for Statistical Agencies

Challenge	ontologyR Solution
“What does ‘unemployed’ mean exactly?”	Explicit SQL definitions with versioning
“Why did the methodology change?”	Rationale field + governance log
“Is our definition still accurate?”	Audit sampling + drift detection
“Who approved this for publication?”	Approval workflow + gate checks
“Can we reproduce last year’s figures?”	Materialized outputs + provenance
“How do we compare internationally?”	Multiple scopes per concept

---

Recommendations for Statistical Offices

1. Start with International Standards

Use ILO, UN, or Eurostat definitions as your ilo_standard or international scope. This ensures comparability and provides authoritative rationale.

2. Document National Adaptations Explicitly

Don’t bury adaptations in footnotes. Create separate scopes (national_headline, national_detailed) with clear rationale for differences.

3. Build Auditing Into the Production Cycle

Schedule regular expert review samples. Catch drift before it affects publications.

4. Require Approval for Headline Statistics

Configure gates that require methodology committee sign-off for definitions used in headline indicators.

5. Maintain Full Provenance

When parliamentary questions arrive asking “where did this number come from?”, you should be able to trace every step.

---

Next Steps

• See vignette("what-is-an-ontology") for conceptual foundations
• See vignette("governance-gates") for detailed RBAC and gate configuration
• See vignette("transforms-and-lineage") for building statistical production pipelines

Remember: In official statistics, the definition is the statistic. Govern it accordingly.