The Sleep Tax is built on four peer-reviewed sources, public Fortune 500 financial disclosures, and a transparent set of formulas. This page documents all of them, along with press-ready materials and contact information for journalists and researchers.
The Sleep Tax applies four peer-reviewed findings to the public financial disclosures of the Fortune 500. Every figure on this site resolves to a source and a formula. We do not hide assumptions.
We kept working notes at each step, from source verification and dataset cleanup through industry rate derivation, aggregate reconciliation, and US-wide extrapolation. They are available on request via the press contact below.
If you are a journalist, researcher, or analyst and something here does not reconcile, please email [email protected]. We will fix or annotate any finding that turns out to be incorrect.
We used the Gallup Panel's self-reported workforce sleep data as one of two independent absenteeism models. 6.2% of US workers report poor sleep quality, and Gallup's longitudinal workforce analysis attributes an additional 16.56 lost workdays per year to that cohort relative to good sleepers.
We used RAND Europe's multi-country economic analysis of workforce sleep duration. Workers who sleep under six hours per night (18% of the sample) lose six additional days per year; workers who sleep six to seven hours (27% of the sample) lose 3.7 days.
For the workplace-injury exposure component (available in the per-company reports), we applied the population-attributable risk from a 2014 meta-analysis of 27 studies covering 268,332 workers. Workers with sleep problems carry a 62% higher risk of workplace injury (RR 1.62, 95% CI 1.43 to 1.84), and 13% of all workplace injuries are attributable to sleep problems.
Where a dollar figure per injury is used, we applied the National Safety Council's published average cost of a medically consulted workplace injury.
The economic analysis is monetized for a reason: dollars mobilize capital and policy. But sleep loss is first a public health problem. These are the peer-reviewed sources behind the clinical claims made in the main report.
Recent longitudinal analysis: people with self-reported insomnia are 2.3× more likely to have depression than the general population. Zhang et al. report a separate 1.92× figure for all sleep disorders combined; we cite the insomnia-specific 2.3× finding here for accuracy.
Medium-to-large effect sizes for sleep problems increase the risk of depression, and this has been replicated across methodologies and populations. We cite Scott alongside Zhang to demonstrate the link is robust across study designs.
The commonly cited "4× increased odds of depression" figure comes from this review (Morin's text reads "four times more likely to develop new major depression over the next 3.5 years"), which in turn references Ford & Kamerow, 1996. We treat Zhang 2022 and Scott 2021 as the primary recent references; Morin is included here for continuity with older reporting that quotes the 4× number.
For each company we compute per-employee daily revenue, apply each absenteeism model, and multiply the implied lost workdays by daily revenue. The lower of the two model outputs becomes the conservative estimate, the higher becomes the upper bound.
1. revenuePerEmployeePerDay = annualRevenue / (employees * workingDays)
2. gallupLostDays = employees * 0.062 * 16.56
3. gallupRevenueImpact = gallupLostDays * revenuePerEmployeePerDay
4. randLostDays = employees * (0.18 * 6 + 0.27 * 3.7)
5. randRevenueImpact = randLostDays * revenuePerEmployeePerDay
6. conservativeEstimate = min(gallupRevenueImpact, randRevenueImpact)
7. higherEstimate = max(gallupRevenueImpact, randRevenueImpact)
Personnel cost burden follows the same arithmetic with personnel cost, computed as revenue times an industry-specific personnel cost ratio, in place of total revenue.
For the workplace injury component:
1. annualInjuries = employees * oshaTRIR / 100
2. totalInjuryCost = annualInjuries * $44,000
3. sleepAttributable = totalInjuryCost * 0.13
4. conservative = sleepAttributable * (1.43 / 1.62)
5. higher = sleepAttributable * (1.84 / 1.62)
Company financial data follows the 2025 Fortune 500 (fiscal year 2024 data), published by Fortune. Revenue and employee counts are sourced from each company's SEC 10-K filing via the EDGAR XBRL company-facts API, then cross-referenced against the Fortune 500 tabulation. Six rows were removed (subsidiaries, dissolved entities, taken-private companies); thirteen non-US-domiciled rows are flagged as excluded from the aggregate (their workforce is mostly outside the US); and five rows are flagged with a postFY2024Event note for transparency where the entity has changed materially since the FY2024 close. The aggregate covers 481 of the 494 retained Fortune 500 rows.
Employee counts reflect the figures Fortune publishes. Where a company reports different head counts in different filings (for example, separating full-time and contract workers in its 10-K), our dataset preserves Fortune's figure for consistency across the set, and notes the divergence internally.
Industry classification follows an eleven-category taxonomy: Retail, Technology, Healthcare, Pharmaceutical, Financials, Energy, Manufacturing, Transportation & Logistics, Consumer Goods, Telecommunications & Media, and Professional Services. This is our own rollup of Fortune's sector tags into broader, analyst-friendly buckets — we split Pharmaceutical out of Healthcare for finer industry-impact analysis. Four sub-category spotlights (Aviation, Banking, Defense and Aerospace, Healthcare providers) are layered on top for readers interested in specific workforce-safety or productivity angles.
Personnel cost ratios are derived from the BEA Industry Economic Accounts (Table 25, Composition of Gross Output by Industry, FY2023), with cross-checks against BLS Industry Productivity Statistics for manufacturing sub-sectors and bottom-up extraction from a sample of Fortune 500 10-K filings. For two industries with revenue pass-through (Healthcare insurers, Retail trade margin) we adjust the BEA labor share to reflect F500-weighted composition; full per-industry derivation is in the audit log. Where a single company is an outlier within its industry (for example, Nvidia within Technology), the aggregate calculation may over- or under-state that company's specific exposure. Individual company reports use the industry ratio unless otherwise stated.
The headline US figure extends our bottom-up Fortune 500 analysis to the full US civilian workforce of approximately 158 million workers using the BLS Current Employment Statistics 2024 annual averages.
Method. For each of our eleven industry buckets, we compute the per-employee-per-year impact rate implied by the Fortune 500 data (separately for both the Gallup and RAND absenteeism models, and for both the revenue-at-risk and personnel-cost denominators). We then multiply that rate by BLS 2024 US industry employment to produce the industry's US-wide contribution.
Sectors outside the F500 footprint. About half of the US workforce is employed in sectors where we do not have per-company financial data: government, education, leisure and hospitality, construction, wholesale (general), administrative services, real estate, and others. For these sectors, we apply the F500-weighted-average per-employee rate as a uniform assumption. This is a simplifying assumption that will be refined in future updates.
Two known sources of bias in the extrapolation. First, Fortune 500 employee counts as reported on 10-K cover pages include international workers; we estimate about two-thirds of the F500 workforce we sample is US-based on average, with significant variation by company. The per-employee impact rates we derive are therefore global rates, not US-only. Second, F500 industry composition differs from the broader US sector composition — F500 healthcare is insurer-heavy where the broader sector is provider-heavy, F500 retail is large-format chains where the broader sector includes many small establishments. These compositional differences could bias individual industry extrapolations up or down.
Revenue vs. GDP, and the RAND cross-check. Our denominator is revenue (and, for the low end of the range, personnel cost). RAND Europe's 2016 analysis used GDP. Because gross revenue double-counts intermediate inputs while GDP does not, our revenue-based figures are larger than a GDP-based equivalent would be. This is intentional: companies measure absenteeism exposure against revenue, and we preserve that framing. As a sanity check, RAND's $411 billion 2016 figure (2.19% of US GDP) scales to approximately $640 billion at the same share of 2024 GDP, which lands within 1% of the midpoint of our range. RAND's GDP-based estimate sits inside our wage-to-revenue range.
The Sleep Tax presents estimated ranges, not audited numbers. The underlying research describes population-level effects, not firm-specific attribution. We treat the Gallup and RAND findings as independent upper and lower estimates of workforce sleep impact applied to each company's workforce, and we publish both.
Absenteeism effects are modeled as lost workdays translated into revenue or payroll using each company's public financials. This translation assumes that lost workdays have an economic value equal to the average daily revenue or payroll per employee. It is a standard productivity-loss accounting method, not a causal measurement.
We do not estimate presenteeism (showing up for work but performing at reduced capacity) in the headline figures. The research literature suggests presenteeism is larger than absenteeism, which means our estimates are conservative against the broader workforce sleep literature.
Uniform sleep prevalence across industries. The Gallup absenteeism rate (6%) and RAND short-sleep distributions (18% under 6 hours, 27% at 6–7 hours) are applied uniformly to every company in our model. Real industry-specific rates vary, but the variations approximately cancel at the headline level.
Working days held constant at 250. All per-employee per-workday arithmetic uses 250 working days per year (the standard US business-year denominator). This matches the framing used in the Gallup and RAND source studies and keeps the math comparable to those primary references. Companies with non-standard work patterns (continuous shift operations, four-day weeks, etc.) may have slightly different real-world denominators.
RAND days-lost combine absenteeism and presenteeism. RAND Europe's 2016 source study computed productivity loss using a model that combines outright workplace absence (absenteeism) and reduced-capacity attendance (presenteeism) into a single “days lost” figure. We use that combined figure for the upper end of our range; our framing emphasises absenteeism specifically because it is the more conservative, more easily explained subset. The full RAND number is internally consistent with our methodology, just slightly broader than what we describe in the headline.
State numbers reflect HQ attribution, not employee distribution. Each Fortune 500 company's Sleep Tax is attributed entirely to its corporate headquarters state. This is the cleanest mapping available because 10-K filings disclose total employees but not state-by-state employee breakdowns. As a result, smaller states with mega-companies headquartered there (Arkansas with Walmart, Rhode Island with CVS, Washington with Microsoft and Amazon) appear over-weighted relative to their workforce share. State-level numbers should be read as “F500 companies headquartered in state X” rather than “sleep loss within state X borders.” The headline US-wide figure does not depend on this attribution.
Journalists and researchers may cite, quote, and reproduce the findings and visualizations on this site with attribution to The Sleep Tax, published by Rest, and a link to getrest.app/cost-of-sleep where possible. Press-ready charts, the full dataset, and a one-page summary are available on request.
Press inquiries: [email protected]
Research and methodology questions: [email protected]
About Rest. Rest is an AI-powered sleep coaching app grounded in Cognitive Behavioral Therapy for Insomnia (CBT-I). Founded out of Y Combinator's W18 batch. Clinical advisors at UCSF. Rest publishes The Sleep Tax annually as part of its work to make the economic cost of poor sleep legible to employers and policymakers.