# The Price of Insuring Yourself: methodology and sources

Snapshot 2026-10-08. Companion to the post at gizmowarehouse.org/gizmo/self-insurance-cost. Code, pipeline, and the row-by-row labels: https://github.com/eichenbaumj/gizmowarehouse.org/tree/main/gizmos/self-insurance-cost

## The question

Do governments that carry their own liability risk (self-insure) spend more on claims and insurance than governments of similar size and type that buy coverage from a public-entity risk pool or a commercial insurer? The comparison is of spending as booked, not of pure loss, and it holds one state's tort law constant.

## Data

**New York State Comptroller, Financial Data for Local Governments.** Account-level bulk files for counties, cities, towns, and villages (also school and fire districts, used only as robustness and placebo arms), fiscal years 1995 to 2026, downloaded 2026-10-08. One row per government, account code, and year. Cost of risk = Unallocated Insurance (account 1910) + Judgments and Claims (1930) + Property Loss (1931), summed across every fund, deflated to 2024 dollars with CPI-U at each government's fiscal-year end, divided by population. Population: decennial counts (OSC/NHGIS 1990 to 2020) and Census Bureau subcounty estimates 2000 to 2024, interpolated between anchors. Secondary outcome: the same cost per dollar of total expenditure net of interfund transfers and debt principal. Workers' compensation (9040), the self-insurance administration line (1710), benefits and awards (1720), and excess insurance (1722) are carried separately; the fund family M, MS, S, CS holds workers'-comp and benefit claims for many entities and an operating-fund variant excludes it.

**Audited financial statements.** The general-liability arrangement of all 57 counties outside New York City, the 20 largest cities, and the 30 largest towns and 10 largest villages by spending (117 governments) was read from the GASB Statement 10 risk-management note in each government's most recent audited statements (fiscal 2024 for most; 2023 for Washington County and seven towns and villages; 2021 for Mount Vernon and Ithaca, the latest issued). Sources: government websites and the Federal Audit Clearinghouse, which serves the full audited statements of every New York county without login. Each label carries the document URL, page, and a verbatim sentence. Structures: self-insured; self-insured with excess coverage; pool; commercial; mixed; unclear. A stated retention at or below $100,000 is treated as a deductible and the entity as covered. Mixed and unclear entities are described but not compared.

**NYMIR subscriber list** (nymir.org, as of July 31, 2026; 1,016 subscribers) was parsed to flag pool membership. Coverage lines are not stated on the list and two counties on it self-insure liability per their notes, so for counties and cities the audited note governs.

**National Transit Database** (Federal Transit Administration, via the DOT open-data portal). Operating expenses by type, 2022 to 2024, object class 506 "Casualty and Liability Costs," which the FTA glossary defines as "the cost elements covering protection of the transit agency from loss through insurance programs, compensation of others for their losses due to acts for which the transit agency is liable, and recognition of the cost of corporate losses" (workers' compensation sits in fringe benefits). Directly operated service, full reporters only, joined to vehicle revenue miles from the service file. The structure of the largest agencies was read from their audited statements the same way.

## Comparison

Each government's cost of risk per resident is averaged over 2015 to 2024 (at least eight years required). Governments are matched within class (county, city, town, village) and population band (10,000 to 50,000; 50,000 to 150,000; 150,000 and up), and the ratio of self-insured to covered means is computed as a population-weighted average of within-band log differences. The 95 percent range comes from a bootstrap that resamples counties. A regression of log cost per resident on structure, log population and its square, police and capital shares of spending, class, and region is the twin. Sensitivities: ratio of medians; cost per dollar of spending; the 90th-percentile year; the 2015 to 2019 and 2010 to 2019 windows; an accrual version that adds the change in the judgments-and-claims-payable balance; a version that adds the law department's cost; the operating-fund variant. A decomposition of the raw gap into size, service mix, and place, with the residual reported as the most structure could explain.

Two further measures carry the piece's argument. Exposure: in the same regression, each additional 10 percentage points of spending on police goes with liability cost per resident about 1.43 times higher (range 1.03 to 1.99), while structure's coefficient is 0.99 and the fit is the same without it. Swing: for each government, the standard deviation of its yearly cost of risk divided by its mean over 2015 to 2024; the median is 0.54 for self-insured governments and 0.27 for covered ones, higher for the self-insured in every class, and the worst year (90th percentile) runs 1.45 times the mean for the self-insured against 1.26 for the covered. Subgroup ratios of level (governments with and without a police department) are reported in the models file and are not established in either direction.

## Entities held out, and why

Judgments and claims lines that carry self-funded benefit claims rather than torts (large, smooth, growing with medical costs): Cattaraugus, Chautauqua, Livingston counties; the City of Lockport; and a handful of towns and villages. Documented self-insurers whose claim payments do not appear on the Comptroller's judgments line: Suffolk (its note reports $68.4 million of self-insurance program payments in 2024 against $0.2 million on the line), Erie, Seneca, Otsego. Nassau is held out because its judgments line is large and smooth enough to trip the benefit-claims flag, which for Nassau is likely property-tax refunds. All of these appear in the downloads with the flag set.

## What the comparison cannot do

It cannot say what a large self-insured city would pay if it bought coverage, because no insurer writes the large ones first-dollar. It cannot separate selection from effect: a county that self-insures may be one with low expected losses, or one nobody would write. The four best-documented changes in a government's premium line turned out to be accounting reclassifications over an unchanged structure, so there is no within-government evidence here. The judgments line is cash paid, coded at each entity's discretion, and understates self-insurers' cost (defense lawyers, incurred-but-unpaid claims, claims booked elsewhere); the premiums line overstates covered entities' liability cost (property and auto lines, the insurer's expenses). The direction of the first bias favors the self-insured; the second favors the covered. A premium-only signature was tested as a way to label the 1,400 towns and villages and agreed with the documents only about half the time, so it is not used as a label anywhere in the piece.

## Sources

- New York State Comptroller, Financial Data for Local Governments: https://wwe1.osc.state.ny.us/localgov/findata/financial-data-for-local-governments.cfm
- Federal Audit Clearinghouse: https://app.fac.gov
- NYMIR, Subscriber List by County, July 31, 2026: https://nymir.org/wp-content/uploads/2026/07/SUBSCRIBER-LIST-BY-COUNTY-07.31.26.pdf; New York Department of Financial Services, Report on Examination of the New York Municipal Insurance Reciprocal as of December 31, 2020.
- Federal Transit Administration, National Transit Database: Operating Expenses by Type (j5uj-anzx) and Service by Mode (wwdp-t4re), data.transportation.gov; NTD Glossary.
- Joanna C. Schwartz, "How Governments Pay: Lawsuits, Budgets, and Police Reform," 63 UCLA Law Review 1144 (2016). John Rappaport, "How Private Insurers Regulate Public Police," 130 Harvard Law Review 1539 (2017). Tom S. Clark, "Municipal Liability Insurance as Police Oversight: Evidence from Vallejo, California," working paper dated April 23, 2026.
- Christian Science Monitor, "Cities pool their risks and insure themselves," June 20, 1986 (the National League of Cities' 650 percent figure). Governing, "Massive Legal Costs Weigh on Local Budgets," October 27, 2016. City of Chicago, Office of Financial Analysis, lawsuit cost study at the request of Alderman Cappleman, 2019.
- Matthiesen, Wickert & Lehrer, "Municipal/County/Local Governmental Immunity and Tort Liability in All 50 States," last updated February 14, 2022 (used for the state cap crosswalk in the transit file).