How uninsured driver rates are estimated: IRC vs state counts

Nobody counts uninsured drivers directly, so every published rate is an estimate built from something else. The Insurance Research Council (IRC) infers a rate from insurance claims. State verification databases count registered vehicles that lack a matching policy. Surveys ask people what they carry. Each measures a different thing, which is why a state can be "12 percent uninsured" in one source and "under 10 percent" in another without anyone being wrong.

The practical rule is to treat a rate as a measurement with a unit, a source and a year, not as a fact about the world. This article explains how each method works, where the estimates diverge, and how to read a figure before using it in an argument.

The IRC method: counting claims instead of cars

The IRC's national figure is the one most often quoted. In its study published in 2025, the share of motorists it estimated to be uninsured was 15.4 percent in 2023. The Insurance Information Institute's summary of the study lists the trend as 12.4 percent in 2017, 12.1 in 2018, 11.6 in 2019, 14.3 in 2020, 14.6 in 2021, 15.2 in 2022 and 15.4 in 2023.

The method, as the summary describes it, uses insurance claims rather than driver surveys. The estimate is a ratio of uninsured motorist (UM) claim frequency to bodily injury (BI) claim frequency. In effect it compares the claims made by people injured by uninsured drivers with the claims made by people injured by insured drivers. The logic is simple: if, in a state, more injured people are collecting under their own uninsured motorist coverage relative to the number collecting from an at-fault driver's bodily injury liability policy, then more at-fault drivers there must have lacked insurance.

Three consequences follow from that design.

1. It measures injury crashes, not parked cars. The rate describes the uninsured share of at-fault drivers who hurt someone, which tilts toward drivers who crash more often. It is not a count of how many registered vehicles are uninsured on a given day.
2. It depends on claim behavior. A UM claim exists only if the injured person had UM coverage and filed. Changes in coverage take-up, in how readily people file, or in how insurers classify claims can move the ratio without any change in how many drivers are insured.
3. It is a statewide ratio built on insurer data. The III page notes that state ranks are calculated from unrounded data, so the displayed percentages may not line up perfectly with rank order. The summary describes insurer claims data as the foundation; the full methodology is in the IRC report itself, which was not available for this article, so check its definitions before relying on the detail.

What UM claims include

The National Association of Insurance Commissioners (NAIC) describes UM coverage as paying an insured who is injured by an at-fault motorist who has no bodily injury liability insurance, who is the owner or operator of a hit-and-run vehicle, or whose insurer denies coverage or becomes insolvent (NAIC 2022/2023 Auto Insurance Database Report). That definition of the coverage matters for how to read a claims-based rate. A hit and run claim is paid under UM coverage even though the at-fault driver is unknown and may well have been insured. Whether a given study counts those claims inside the uninsured rate is a definitional choice the report spells out. When you read a claims-based number, find out what sits in the numerator.

The III page adds a footnote for Florida: its compulsory laws cover personal injury protection and property damage but not third-party bodily injury coverage. For a ratio built on bodily injury claims, that difference in state law is one reason Florida's figure and a figure from a state that requires BI liability are not perfectly comparable. Florida's 2023 estimate of 20.6 percent is on the same table, flagged with that asterisk.

State verification counts: vehicles, not crash involvements

Most states now check coverage electronically. The mechanics, described in our guide to state insurance verification systems, are that insurers report policies and the state matches them to registered vehicles by VIN and sometimes by plate. A vehicle with no matching policy is flagged.

That produces a different kind of number: the share of registered vehicles with no match on the day, week or month the file was built. It counts vehicles, not drivers or crashes. It can be produced for every county without waiting for claims to settle. And it includes anything the match misses.

The Oklahoma District Attorneys Council says an insured owner can receive a notice "because of how carriers or brokers report policies," which the carrier then corrects in the state system. A 2019 Insurance Journal report quoted the program's prosecutor saying nearly 5 percent of flagged vehicles had been flagged in error, with causes including plates not re-registered after a sale and out-of-state or commercial coverage not required to be in the database. A count built from raw flags without correcting for these errors overstates the uninsured share. A count built from vehicles that fail the database and then fail follow-up letters understates the lapses that were cured.

Uninsured drivers, uninsured vehicles and lapses are different things

The word "uninsured" hides three separate concepts.

A state's registered-vehicle count can include a car that is parked all year, a car between policies for three weeks, a car sold but still registered, and a car that really is driven daily without a policy. Only the last is what most people mean when they say "uninsured driver." A crash-based estimate weights by driving: a car driven every day by a young driver shows up in injury claims far more often than a parked one.

Our companion article on who goes uninsured and why: cost, lapses and enforcement gaps covers why people go without coverage. For lapses specifically, see lapse in coverage and the guide on what happens if your car insurance lapses.

What happens when two methods meet: New Mexico

A rare side-by-side comparison exists in a study by New Mexico State University's Insurance and Financial Services Center for the state's Insurance Division, which compared the IRC's accident-based figures with the state Motor Vehicle Division's insurance database (Uninsured Motorists in New Mexico). It covered 2000 to 2007, with database figures through December 2008, so it is old; it is useful for how it explains the gap, not for current numbers.

The researchers expected the IRC's accident-based numbers to run higher than the database numbers, and they did: they estimated the chance of being struck by an uninsured vehicle in New Mexico at about 29 percent in 2007, while the state database put uninsured registered vehicles slightly below 10 percent in December 2008. They also expected the two to move in the same direction, and they did not: the database showed a steep decline in the first three years after implementation and a flattening decline afterward, while the IRC series dipped in 2004 and then, by 2007, was at or above pre-database levels. Their list of possible explanations is a good checklist for any such comparison:

• Many uninsured cars involved in accidents are also unregistered, so they appear in the claims-based method but not in the registration database.
• The state's figures concern vehicles registered in New Mexico only and omit uninsured out-of-state cars.
• Younger drivers cover disproportionately many accident-involved miles, and they are more likely to be uninsured, which biases an accident-based ratio upward compared with a count of all cars.
• The IRC study at the time drew on nine insurers representing about half of private passenger auto liability premiums, so the data are a sample.
• New Mexico's figures included uninsured motorist property damage and underinsured coverage, which broadens what a "UM claim" is.

The authors' conclusion was that, because of the difference between uninsured vehicles and uninsured drivers and the differences in method, neither study provides a "true" measure of uninsured motorists. They suggested a best approximation lay between the two, in a range they put at 10 to 29 percent, because the number of unregistered vehicles was hard to estimate. They also noted that they could find no national database of uninsured registered vehicles, and used a 2004 California Department of Insurance estimate of 14.4 percent uninsured registered vehicles as an admittedly crude proxy.

Why numbers move from year to year

A few mechanisms can change a published rate without a matching change in behavior.

• Edition differences. The IRC revises its series with each study. A trade-press report on an IRC study put the 2022 uninsured share at about 14 percent, while the 2025 edition's table gives 15.2 percent for 2022. Compare years only inside a single edition.
• Claims timing. Injury claims settle over years. A claims ratio for the latest year can shift as more claims close.
• Event shocks. The jump from 11.6 percent in 2019 to 14.3 percent in 2020 coincided with the pandemic, when driving patterns and claims changed sharply. The NAIC notes on its uninsured motorists page that the rate has steadily increased since the pandemic began in 2020.
• Enforcement and data upgrades. When a state improves its verification system, the number of flagged vehicles can rise at first (better detection) and later fall (people comply). A 2018 KGOU report quoted Oklahoma officials giving a rate of about 10 percent, down from roughly 26 percent in 2012, while the IRC's 2025 table shows 12.0 percent for Oklahoma in 2023. Those are different measures at different times, so the gap is not evidence of an increase.
• Reporting practices. In a California analysis for the state insurance department, collisions involving at least one uninsured motorist rose from 12.2 percent of collisions in 2014 to 18.0 percent in 2017, and the authors said the rise may reflect, among other things, improvements and changes in how the California Highway Patrol tracks uninsured motorists (Sacramento State report). A trend in a count can reflect better recording rather than more of the thing.

Limits of survey and claim-based methods

Each method has weaknesses that the other does not share.

Surveys ask drivers whether they have insurance. People may not answer honestly about an illegal gap, uninsured people may be harder to reach, and the sample may not represent the whole state. The IRC's claims method avoids self-reporting, which is part of why it has become the standard national series.

Claims-based ratios need claims. In a state where many drivers decline UM coverage, or where people hurt by uninsured drivers rarely file, the numerator is smaller. They also lag, because the data come from claims that must be reported and developed. And they say nothing about uninsured cars that did not hurt anyone.

Database counts are fast and local, but they depend on the quality of insurer reporting and on a match that can fail for innocent reasons. They also rarely capture drivers using vehicles they do not own, and, as the New Mexico study noted, they omit unregistered vehicles.

None of these is wrong. Each is a tool suited to a different question. If your question is "how likely is the person who hits me to be uninsured," a claims-based figure is closer. If it is "how many cars on the registry have no policy," a database count is closer.

How to read a rate before you use it

Use the following checklist when a statistic about uninsured drivers appears in an article, a bill analysis or an argument.

1. Name the unit. Is it drivers, vehicles, crash involvements or claims?
2. Name the source and method. Claims ratio, database match, survey, or a roll-up of several?
3. Check the year of the data and the year of publication. A 2025 study of 2023 data describes 2023, not today.
4. Check the edition. Do not compare numbers from different editions of a study.
5. Check what is included. Do UM claims in the numerator include hit-and-run and insolvent-insurer claims? Does the state require BI liability?
6. Check the state's rule. Some states require UM coverage, some require insurers to offer it, and some let the buyer reject it, which affects UM claim volume. The NAIC database report says 23 states require UM coverage and 19 more require it unless rejected in writing.
7. Compare like with like. Two states are comparable only if the same source measured both in the same year.
8. Look for the margin. A one-point difference between two states is often within what these estimates can resolve. The rank order of the top and bottom states is more reliable than the exact spacing.

A worked example with invented numbers

Say a state has 1,000,000 registered vehicles. On the day its database is matched, 80,000 vehicles have no policy on file, which looks like an 8 percent rate. A follow-up reveals the following about the 80,000:

• 20,000 are insured but reported wrong, with a VIN typo or a carrier that did not file.
• 15,000 are between policies and are covered again within a month.
• 10,000 are parked, sold or unregistered in practice.
• 35,000 are in use without coverage.

The registry-based uninsured figure is 8 percent before follow-up, but only 3.5 percent of vehicles are truly in use without insurance. Now suppose those 35,000 vehicles are disproportionately driven by young and high-risk drivers who crash twice as often as the average insured driver. A claims-based estimate could easily put the share of at-fault injury crashes involving an uninsured driver near 7 percent. Three honest numbers, 8, 3.5 and 7, all describe the same state. These figures are invented to show the arithmetic, not to describe any real state.

Common questions

Which figure is the right one for my state?

It depends on your question. For national comparisons and trends, the IRC series is the standard. For enforcement or registration questions, your state's database count is the relevant one. For how likely you are to be hit by an uninsured driver, use the claims-based figure and pair it with the guidance in uninsured drivers by state.

Why does Mississippi rank so high and Maine so low?

The 2023 IRC table shows Mississippi at 28.2 percent and Maine at 5.7 percent. The table does not explain the gap. Causes such as income, enforcement, rural driving, premiums and claim behavior are all plausible, and the estimate itself is affected by how many people carry UM coverage. See the state pages for Mississippi and Maine, and read the ranking with the margin in mind.

Does a high uninsured rate mean my premium is higher?

Not in a simple way. UM coverage is priced from UM claims, and how those costs reach your bill depends on your state and your coverages. See Uninsured drivers and your premium: how UM costs are priced into rates.

Can I trust a number that does not name its source?

Treat it as unverified. Ask which study, what year, what measure and what edition. The habit applies to all rate studies; our guide how to read car insurance rate studies and average-cost claims covers it more broadly.

Key takeaways

• The IRC's national rate comes from claims: it compares UM claim frequency with BI claim frequency, and it was 15.4 percent for 2023 in the 2025 study.
• State database counts measure registered vehicles with no matching policy, which can include data errors, short lapses and cars that are rarely driven.
• Uninsured drivers, uninsured vehicles and lapses are three different measurements, and none is a "true" rate on its own.
• Rates move because of method, edition, claim timing and reporting as well as behavior, so compare only like with like.
• Before citing a rate, name its unit, source, year, edition and definition.