The Wonkery Way
How Workforce Wonkery turns workforce data into useful, transparent intelligence without hiding uncertainty or overstating what the evidence can support.
A consistent way to reason with workforce data.
Start with the question
Begin with the workforce decision or problem, not the dataset that happens to be available.
Read signals together
Use multiple measures when one dataset cannot answer the whole question.
Separate evidence from interpretation
Make it clear what the source shows and what Wonkery is inferring from it.
Show uncertainty and gaps
Say what is known, what is uncertain, and what evidence is still missing.
Make the math visible
Label calculations, weights, denominators, periods, and definitions so the analysis can be inspected.
End with the next question
Use evidence to identify what should be examined or validated next, not to manufacture certainty.
Cross-dataset synthesis rules
- Preserve source periods and definitions.
- For occupation analysis, declare the geography before interpreting the SOC. Jobs, openings, wages, training supply, WIOA activity, and other occupation evidence must not leak across geographies.
- Use one SOC-based occupation universe and the same calculation rules at every geography. If an external agency publishes a priority or high-demand list, retain it only as labeled source metadata with its own authority, source, period, and geography; it does not define the occupation universe or ranking.
- Use the current California workforce-region / Regional Planning Unit name from official state sources. Do not substitute an informal regional label; Monterey, San Benito, and Santa Cruz are the North Central Coast.
- Do not merge unmatched cohorts or reporting windows into a false common story.
- Separate expansion from churn by reading hiring with separations and turnover.
- Separate sector growth from job quality.
- Do not infer training shortage from openings minus completions.
- Label derived values as Wonkery calculations.
- Preserve source lineage when using a synthesis report.
Writing The Story
- Write for a smart general reader, not for a labor-market analyst.
- Lead with a concrete, specific headline about the local economy. Do not write a headline about the data itself, use vague framing, or manufacture novelty with words such as “now” unless the evidence shows a real change.
- Use plain language first and numbers second. Explain what is happening, which jobs or industries shape the place, and why it matters for people.
- Translate technical measures into everyday language in the narrative. Keep the technical detail available in the cards, source notes, and methodology.
- Preserve geography, source periods, uncertainty, and conflicting evidence. A cleaner story must never become a less accurate one.
- End with The big question: the practical question that should be watched, tested, or validated next.
Labor-shed rules
- Use Primary Jobs intentionally for the headline OnTheMap labor-shed view.
- Show county flows in both directions.
- Use enough destination depth so smaller strategic relationships are not hidden.
- Keep the base count visible beside percentages.
- Do not treat LODES earnings bands as living-wage standards.
- Do not treat broad industry as occupation.
Workforce-system rules
- When reporting alignment to an external priority or high-demand list, keep the list source, geography, version, and program year with the metric.
- Separate alignment from completion, employment, wages, living-wage attainment, and retention.
- Separate provider compliance from participant outcomes.
- Show concentration beneath headline alignment rates.
- Keep denominators explicit.
- Use privacy-safe aggregates and suppress small cells where needed.
Evidence confidence
- A: direct, current, authoritative evidence with strong geographic and definitional fit.
- B: strong evidence with limited timing, geography, or measurement caveats.
- C: directional or proxy evidence with important limitations.
- D: weak, stale, or poorly aligned evidence.
- U: unknown or insufficient evidence.
The grade applies to the evidence, not to a community, employer, program, occupation, or board.
Decision-brief rule: grade market opportunity, training supply, capacity/access, outcomes, and workforce-system status separately. Do not create a composite score. A current response should rest primarily on A/B evidence. C/D evidence can shape validation questions but should not independently drive a major investment. U evidence becomes an explicit verification task or a “What would change this finding?” condition.
Change rule: every decision-ready brief must state what new evidence could change the current response. This prevents a Wonkery finding from being treated as permanent.
One fact should have one authoritative record.
Market observations, training programs, outcomes, and decision findings are maintained once and then reused across Data products rather than copied into separate pages.
Public Data products use the same published evidence layer, so a factual update can flow across tools without silently creating competing versions.
Material factual changes retain source, date, and review context so a current finding can be understood in relation to the evidence that supported it.
Source changes can update factual evidence, but an interpretive decision finding does not change automatically just because a source changed.
The labor market is the public analysis geography.
Workforce Wonkery publishes Data profiles for California and official metropolitan statistical areas / core-based statistical areas. These are the public geographies readers select and compare.
Every source keeps the geography it actually measures. County, workforce-area, workforce-region, or other component records may remain in source lineage behind the scenes without becoming separate public profiles.
A WDB, workforce region, county, or CSA is never substituted for an MSA simply because a source is easier to obtain at that geography. If the evidence cannot support the labor market, the public value stays unavailable.
Additive measures may be combined from non-overlapping components that exactly form the MSA. Rates are recomputed from their underlying counts. Medians, wages, projections, and percentages are not averaged to manufacture a market estimate.
Current economic health stays separate from occupational outlook.
Monthly labor force, employment, unemployment, and unemployment rate are stored separately from occupation projections, openings, and wages.
MSA and California comparisons use the same not-seasonally-adjusted EDD series when displayed together.
When official non-overlapping components exactly form an MSA, labor force, employment, and unemployment may be summed and the unemployment rate recomputed. The component geographies do not become public profiles.
For headline labor-market profiles, if an exact MSA measure cannot be constructed without changing the meaning of the source, Workforce Wonkery leaves it unavailable rather than inventing an estimate. The Occupation Explorer is a narrow exception: when exact MSA occupational evidence is unavailable, it may show the nearest accurate published geography only when the substitute is clearly labeled and the note explains what geography is being shown and why.
New labor market should mean new evidence, not new rules.
Occupation findings are keyed to the selected labor market and occupation. A finding from one MSA never becomes the default for another.
The Occupation Explorer prefers exact MSA projection geography. When exact MSA occupational evidence is unavailable, it may use the nearest accurate published geography as a clearly labeled substitute. Broader workforce-region, county, or neighboring-market evidence is never relabeled as MSA demand.
A second labor market does not get custom weights or a weaker evidence threshold simply because its data are harder to obtain.
A labor market may have a full profile but only a labeled nearest-accurate occupation view. If no substitute has acceptable geographic and definitional fit, the occupation value stays unavailable.
A living brief should say when the evidence changed.
Every curated decision brief stores the evidence signature that was last reviewed for its finding.
If market values, program status, capacity, ETPL, or verified outcomes change, the review process can flag Finding may need review rather than silently carrying forward the old response.
A recent source may still be a proxy or a poor geographic fit. Confidence and freshness remain separate concepts.
A response is a current interpretation of the evidence. Every brief must retain its decision-changing conditions and last-reviewed date.
What readers should always be able to see.
- What question the analysis is trying to answer.
- Where the evidence came from.
- What geography and time period apply.
- What Wonkery calculated or inferred.
- What remains uncertain or unavailable.
- What question should be validated next.
Technical safeguards underneath the design.
Source, release, geography, period, definition, and refresh logic remain visible.
Participant-level workforce records are never published. Small cells are suppressed or combined when needed.
Method changes, model weights, and major interpretation updates should be documented rather than silently replaced.
Program existence, eligibility, capacity, access, and outcomes are different questions.
Every training or occupation Decision Brief must complete the same evidence check before a training response is treated as publication-ready.
Verify a current program or pathway from an authoritative provider, regulator, or administering-agency source. A directory hit is discovery, not proof.
For college, adult-education, and other training, verify current program-level CalJOBS status. For registered apprenticeship, distinguish automatic eligibility for ETPL listing from the sponsor’s actual CalJOBS listing or opt-in status.
DAS registration is its own evidence field. Registration does not prove current recruiting, available employer sponsorship, indenture, dispatch, or additional capacity.
Verify whether the program is actually accepting applications or enrolling now. An existing catalog page or registered program is not proof of current intake.
Keep seats, cohorts, applicant volume, clinical sites, worksite sponsorship, employer placements, indentures, and dispatch capacity separate. Do not infer capacity from program existence.
Examine schedule, language, travel, cost, prerequisites, childcare, clinical/worksite access, application timing, and other barriers separately from capacity.
Completion, certification, licensure, employment, earnings, retention, advancement, and living-wage attainment retain their own source, cohort, denominator, and reporting period.
Keep training geography and labor-market geography explicit. A provider serving a county, workforce region, or neighboring market does not create exact MSA demand evidence.
No single outcome system is expected to cover every training program.
Outcome evidence follows the system that actually holds the strongest comparable information for that provider type. The goal is defensible evidence, not a universal match rate.
Use DataVista first. Community-college records use the strongest program or TOP-level view available. Adult-education records use CAEP measures such as employment after exit, annual earnings, credentials, and transitions when the published reporting level aligns.
Use California DAS registration and completion dashboards first. Program completions and five-year completion rates are valid apprenticeship outcomes. They are not silently converted into employment or earnings measures.
Use occupation-specific official sources when they answer a different outcome question. For example, RN licensure pass rates can supplement training outcomes but do not substitute for employment or earnings.
Use provider-published results only when the exact program, cohort, denominator, period, and metric can be reproduced. Marketing claims without those elements do not become outcome evidence.
Use as supplemental evidence when provider identity, program lineage, occupation, and location support a defensible one-to-one match. There is no target federal match rate and no occupation-family matching backlog.
If none of the appropriate systems support the program at the required level, the outcome stays undocumented. “Not evaluated” and “no verified federal match” are not quality judgments.
Choose the industry that best explains the labor market, not simply the largest one.
Each MSA can have up to three Industry Deep Dives. The selection starts with comparable 2-digit NAICS evidence, narrows to five data-selected industries for deeper review, and then adds current local context before anything is published.
How much of the market’s employment is in the industry?
Is the industry more concentrated here than it is across California?
How much is annual-average employment changing, whether it is growing or shrinking?
How does average annual industry pay compare with the market’s overall pay level?
An industry must have at least 250 jobs and at least 0.5% of covered market employment. The higher threshold controls. This keeps a very small, high-concentration sector from winning on location quotient alone.
Industry employment share is ranked against other eligible sectors in the same MSA. The score is relative to the structure of that market.
The MSA industry share is divided by the California industry share to create a location quotient, then ranked across eligible local sectors.
Year-over-year change is recalculated from annual-average employment totals for the current and prior year. The score uses the magnitude of that change, so a sharp decline can be as decision-relevant as rapid growth. The direction remains visible. County growth rates are never averaged.
Average annual industry pay is compared with average annual pay across the MSA, then ranked against other eligible industries.
Complete candidates use disclosed QCEW ownership pieces throughout. Partial candidates contain one or more suppressed ownership pieces; the disclosed values can support screening and analyst review, but a partial-confidence candidate cannot become the published Industry Deep Dive unless independent evidence confirms the choice is robust.
The composite score narrows the research. It does not make the publishing decision. Ties favor specialization, then scale, then employment, but analyst validation still comes next.
Screening evidence is kept separate. Census resident-industry data can help check whether the candidate set looks plausible, but it is not substituted for employer-side QCEW employment, wages, concentration, or employment change. Qualitative reporting is then used to explain current change, not to manufacture another numeric score.
The Wonkery Way collects less on purpose.
A field belongs in the maintained data layer only when it is reproducible, reasonably maintainable, and likely to change a workforce decision. Interesting is not enough.
Core labor-market measures, geography, and region-level occupation evidence that can be reproduced consistently across California.
Decision-relevant local developments from credible journalism, official announcements, WARN notices, providers, employers, and public agencies.
Training supply, capacity, admissions, detailed program cost, outcomes, current eligibility, employer validation, worker experience, and feasibility when a live decision requires them.