Reading employer health data

About 54 percent of Americans carried employment-based health coverage in 2023, making the workplace the country's largest coverage channel (Census Bureau, 2024). Reading that system's numbers well requires knowing which series measures what: the Census count of who is covered, the Bureau's benefit surveys of what employers offer, and the cost figures that sit inside plan-year disclosures.

This primer walks through each source and shows how to avoid the most common misreadings.

What does the Census data actually measure?

Coverage status, at a point in the prior calendar year. The Census Bureau's health insurance reports and Small Area Health Insurance Estimates count whether people had any coverage — employer, direct purchase, Medicare, Medicaid, and other types are counted separately — during the previous year (Census, 2024). The 2023 report put the uninsured share at roughly 8 percent for the full population.

The measures are annual, retrospective and mutually compatible by type. A person with employer coverage for six months and Medicaid for six appears as covered, with a mixed type — not as a gap.

Where do employer benefit numbers come from?

Two federal systems, doing different jobs. The National Compensation Survey measures the share of workers whose employers offer coverage and what the employer pays, published as the Employer Costs for Employee Compensation series (BLS). The department's Form 5500 filings document the existence and financing of large benefit plans themselves (DOL, EBSA). Neither system reports what a specific worker pays per paycheck; both are averages over defined populations.

Product marketing often quotes KFF-style premium averages, which are plan-year snapshots, not federal series. This article sticks to the federal sources so the citations resolve to documents.

Why do 'offered' and 'enrolled' differ so much?

Because the offer is the employer's decision and enrollment is the household's. Workers decline employer coverage for affordability reasons — the employee share of premiums — or because a spouse's plan is better, or because they are covered elsewhere. That is why employer-offer rates among private employers run well above the 54 percent of the whole population actually holding employment-based coverage (BLS NCS; Census, 2024).

Any claim that stitches the two together — 'employers offer X, therefore X million are covered' — is a category error. The offer counts workers; the coverage count is of all residents.

What is the employee share, and who says so?

The employee contribution is a plan-year fact, published in plan documents, not in the Census series. For a precise figure for a specific employer, the primary documents are the plan's summary plan description and its Form 5500 schedule of premiums (DOL/EBSA). Aggregated national figures come from annual employer surveys published by research organizations, which are useful but not federal statistics.

The reliable habit: attribute every premium number to its exact source and year, and refuse any figure that survives without one.

How does eligibility timing affect coverage counts?

Waiting periods and year-end separations create mechanical gaps. A worker hired in June who becomes benefits-eligible after a 90-day wait appears in Census data as covered for part of the year, and the point-in-time versus any-time distinction moves the national uninsured count by meaningful amounts (Census methodology notes). Seasonal and gig workers show the effect most sharply.

This is why 'the uninsured rate fell' headlines need the measure named. Any-time coverage and point-in-time coverage are different statistics about different people.

The dependent-coverage rule adds one more wrinkle worth knowing when reading household figures. Since the federal rule took full effect in 2011, plans covering dependent children generally must cover them to age 26, and Census tables show young adults' employment-based coverage moving accordingly (federal dependent-coverage rule, 2010; Census, 2024). A household coverage figure in any year after that reflects the rule, which means comparisons to earlier decades are not like-for-like.

Employer costs have their own measurement trap. The BLS compensation series reports health benefits as a cost per hour worked, averaged across all hours including overtime and part-time schedules (BLS ECEC). A per-hour average cannot be converted to a per-month premium without assumptions, and arguments that perform that conversion silently are the most common place benefits numbers go wrong in public debate.

For an individual worker, the documents beat every statistic. The summary of benefits and coverage, the summary plan description, and the plan's Form 5500 are the primary sources for what a specific plan covers, what it costs and how it is financed (DOL/EBSA disclosure requirements). Requests for those documents run on statutory timelines, and plans must provide them.

The statistical habit this primer leaves the reader with is attribution discipline. The 54 percent employment-based figure is a Census count of residents (Census, 2024). The offer and cost series are Bureau measures of employers (BLS NCS). The plan documents are the plan's own filings (DOL/EBSA). Each is real; none substitutes for the others; every misquote in the benefits debate is a quiet swap between them.

What should a reader do with a benefits statistic?

1. Name the population: all residents, all workers, or private-industry workers only.
2. Name the year and whether coverage means any part of the year or a point in time.
3. Name the source type: Census coverage count, BLS employer-cost series, or a plan document.
4. Reject any mix of the three without an explicit bridge.

The 54 percent employment-based figure (Census, 2024) and the roughly 8 percent uninsured share (Census, 2024) are the two anchors worth memorizing. Everything else in a benefits argument is a claim that needs its own citation.

Related: Shift work and the body clock · Algorithmic management at work.