Reading auto insurance studies by who paid for them

You can trust a study from a sponsor with a stake, but only as far as you can check it. The sponsor's interest usually shows up in which question gets asked, which data is chosen and how the finding is worded, far more often than in the arithmetic. So read a study the same way regardless of who paid: find the sponsor, the data, the sample and the measure, then see what a rival study of the same subject says.

This article builds that routine around four real examples: an insurer-supported research council, a consumer federation, the association of state regulators and an investigation by journalists. None of them is dismissed here, and none is accepted without its limits.

Why the sponsor matters, and why it does not decide

Auto insurance has few neutral observers. Insurers hold the claim and premium data. Regulators collect some of it, a few years late. Consumers experience prices but see only their own. Anyone who wants to measure the market has to borrow data from one of those holders or buy quotes, and every choice at that stage reflects an interest.

Sponsorship tends to act in five places:

• The question. An organization asks what it needs to know. An insurer-supported group might study fraud, claim costs or the share of drivers who are uninsured. A consumer group might study whether low-income or minority neighborhoods pay more.
• The data. Quote scrapes, closed-claim files, statistical-agent filings, regulator surveys and public records see different slices of the market.
• The comparison. "Higher than what?" The answer determines whether a gap looks large or small.
• The wording. A press release can lead with the biggest number in the study; the study may include a lot of caveats.
• What is left out. Studies seldom advertise what they did not examine.

None of that makes the findings false. A study with a visible sponsor, published method and transparent data is more useful than an anonymous one. The goal is not to find a sponsor-free source, which does not exist, but to read each source for what it can and cannot show.

Kinds of sponsors and what they tend to ask

Those categories overlap. A regulator's study can be commissioned because an advocate pushed for it, and a consumer group can rely on industry-collected claims data.

Example 1: the Insurance Research Council

The Insurance Research Council describes itself as "an independent, nonprofit research organization supported by leading property and casualty insurance companies and associations," founded in 1977. On its about page it also says it does not advocate public policy or lobby. Its member list on the same page includes the American Property Casualty Insurance Association, the National Association of Mutual Insurance Companies and several large insurers. It says its work analyzes data compiled by licensed statistical agents such as ISO and NISS, and that it runs periodic closed-claim studies on auto injuries.

What the sponsorship predicts. The council studies things insurers need to price: injury claim trends, attorney involvement, uninsured motorists. The about page itself says the industry relies on it for "critically needed property-casualty studies that are not ordinarily undertaken by other research organizations."

A concrete study. The council's uninsured motorist estimate reached 15.4% of motorists in 2023, according to the Triple-I's uninsured motorists page, which cites the IRC's 2025 study. The IRC derives it from claims: the ratio of claims by people injured by uninsured drivers to claims by people injured by insured drivers.

What to check. An estimate built from claim ratios depends on the idea that uninsured drivers are involved in injury crashes at about the same rate as insured ones, which may not hold. A state that counts uninsured vehicles through electronic verification will produce a different number. Our How uninsured driver rates are estimated: IRC vs state counts article lays out the comparison, and the Insurance Research Council studies on uninsured drivers and claims guide covers the council's other work. Also note that the series is revised between editions: do not splice tables from different editions.

Example 2: the Consumer Federation of America

The Consumer Federation of America describes itself as an association of nonprofit organizations founded in 1968 that works through research, advocacy and education. Its May 2025 testimony to a Massachusetts committee shows its dual role: the document states research findings and also urges lawmakers to pass a specific bill.

What the sponsorship predicts. Questions about price fairness and affordability for low- and moderate-income drivers, and policy proposals that follow.

A concrete study. In a November 2015 release, the organization reported that quotes from the five largest insurers averaged $1,060 in predominantly African American ZIP codes and $622 in predominantly white ZIP codes, a gap of 70%. The data came from Quadrant Information Services and described one driver: a 30-year-old woman with a perfect record who rents her home, works a clerical job and has fair credit. The release says the analysis covered basic liability coverage, not collision or comprehensive, and quotes the organization's director of financial services saying "We are not rushing to judgment about why this happens."

What to check.

• Quotes are not premiums. A quote is the price offered to a hypothetical driver. It may not match what existing customers actually pay, and the profile may not resemble the typical driver in each ZIP code.
• One profile. Results for a 30-year-old clerical renter do not automatically extend to a 55-year-old homeowner.
• No loss data. Quotes alone cannot say whether higher prices track higher claim costs. That is the question the later journalists' and regulators' studies tried to answer with payout data.
• Date. The 2025 Massachusetts testimony relies on premium data from August 2020, which the testimony footnote states. Advocacy documents can use data several years old; always read the footnote.

What this study does well is open: it names the data vendor, the driver profile and the coverage, which lets another researcher repeat the exercise. For the full discussion of what ZIP-code investigations showed, see ZIP-code pricing and racial disparities: what investigations found.

Example 3: the NAIC

The NAIC is the association of state insurance commissioners. Its 2022/2023 Auto Insurance Database Report, adopted in December 2025, was developed by its Casualty Actuarial and Statistical (C) Task Force at the direction of the Property and Casualty Insurance (C) Committee. The introduction says the database was compiled to help states assess their markets, and names the statistical agents that supplied the data: the American Association of Insurance Services, ISO, the National Independent Statistical Service, Independent Statistical Service, Massachusetts Commonwealth Automobile Reinsurers and the Maryland Auto Insurance Fund, plus California and Texas data.

What the sponsorship predicts. Regulators want comparable state-by-state market data to monitor competition and cost. They are less likely to ask a question the data cannot answer.

What to check.

• Defined terms. The report defines premium, exposure, car year and other terms and says "formulas will vary depending on the application." Its worked examples show how a six-month policy is split across calendar years. Take definitions from here before you compare to someone else's table.
• Honest limits. It warns that, in making interstate comparisons, premiums reflect a complex set of state-specific factors and the data "by no means represent all such factors."
• Revisions. It says data may differ from earlier years' reports as statistical agents update their information.
• Age. A report adopted in December 2025 covers premiums through 2023 and losses through 2022. Regulators' data is rigorous and slow.

See What NAIC auto insurance data measures for a field guide to the tables.

Example 4: journalists and regulators on the same question

A fourth type shows how sponsorship interacts with scrutiny. In April 2017, ProPublica and Consumer Reports published an analysis of liability premiums and insurer payouts in California, Illinois, Texas and Missouri, the four states from which ProPublica said it could obtain ZIP-code-level payout data. It reported that some insurers charged minority neighborhoods up to 30% more than other areas with similar accident costs. The article published the criticism it received: the California Department of Insurance said an individual insurer's losses in a ZIP code may differ from the industry average and called the conclusion flawed, and the Illinois Department of Insurance called the methodology incomplete and oversimplified. The article also points readers to a methodology page and the data and code.

In November 2024, the District of Columbia's insurance regulator released its own market conduct review, which found that Black drivers paid on average 1.46 times as much as white drivers, a gap it put at $326 a year, but also found that Black and Hispanic drivers incurred claims with an even higher difference, and said further study was needed. That is a regulator reporting a finding that cuts both ways, which a quick reading of the headline would miss. We cover both reports in the article linked above.

Two lessons for sponsor-reading. First, a study with open data and a published response from the people it criticizes is more checkable than one without. Second, regulators are not neutral by definition either; they have their own political and statutory incentives, and they can disagree with each other.

A checklist for any new report

Run through these in order. Ten minutes with the full report beats an hour with the press release.

1. Who paid, and who wrote it? Look for a funding or membership statement. If there is none, check the organization's website and public filings. Do not assume a nonprofit label means neutral, or an industry label means wrong.
2. What question was asked? State it in one sentence. If the headline answers a different question than the report's title, be careful.
3. Where did the data come from? Statistical agent, regulator filing, purchased quotes, survey, closed claims, public records. Each has known gaps.
4. What is the sample? How many insurers, states, years, drivers? Is it one profile, or a representative set?
5. What exactly is measured? Premium, expenditure, quote, claim frequency, claim severity, loss ratio? Check the units: per car year, per policy, per household.
6. What is the comparison? Against what, and why that group? "Higher than white ZIP codes" and "higher than the loss data predicts" are different statements.
7. What period? Is the data from the same year as the headline? Has the market changed since?
8. What do the authors say they cannot show? Good reports have a limitations section. If the press release lacks one and the report has it, quote the report.
9. Can someone else reproduce it? Is the method, code or data available? Studies that publish them invite correction.
10. What do the other camps say? Find at least one study from a different type of sponsor on the same topic, and see where the numbers agree.

A worked example with invented numbers

Suppose a headline reads: "Study finds drivers in ZIP code 99999 pay 40% more than the state average." Here is how the checklist runs. All figures are hypothetical.

• Sponsor. A consumer group, based on the byline. Question: do residents of low-income areas pay more?
• Data. The report says it bought quotes for one driver profile, a 35-year-old with a clean record and minimum liability limits, from six insurers. That is a quote study, not a study of what people actually pay.
• Numbers. The state-average quote is $700 and ZIP 99999 averages $980. That is exactly 40% more ($280 / $700). Arithmetic is fine.
• Comparison. The state average includes rural ZIP codes. The report also lists a neighboring urban ZIP at $950. Against that neighbor, the gap is about 3% ($30 / $950). The 40% gap is mostly an urban-versus-state-average gap.
• Missing measure. There is no loss data. Without it you cannot say whether $980 is a fair reflection of claim costs in the ZIP.
• Other camps. You look for the state insurance department's published territory loss data and for an insurer-side explanation of its territory factor. If the loss data shows ZIP 99999 has claim costs close to its urban neighbor's, the advocate's point about the state average is fair but incomplete; if it shows claim costs far below, the point is stronger.

Result: the headline is arithmetically right and analytically thin. You would cite it as "a quote analysis found," with the profile and the comparison group, and you would not say that insurers overcharge the ZIP.

Where regulators and the public can look next

For a claim about your own state, go beyond national studies. The state insurance department publishes rate filings, complaint data and sometimes territory-level loss statistics. Our how to read car insurance rate studies and average-cost claims guide covers reading rate studies and average-cost claims. The state car insurance requirements lookup tool can start you at the right regulator. For a dispute about prices or a possible violation of rating law, the state insurance department or a licensed attorney can advise on specifics.

Common questions

Is a nonprofit study more reliable than an industry study?

Not automatically. A nonprofit can have strong advocacy aims, and an industry-supported group can publish careful, open methods. Judge the method and the transparency of the data first. The sponsor tells you which questions to expect, not whether the answer is right.

Should I ignore any study where the sponsor benefits from the result?

No. If you discarded every study with an interested sponsor, you would have almost nothing to read, because insurers, regulators and advocates are all interested. Use the checklist, and give extra weight to claims that both an insurer-side and a consumer-side source accept.

What if two credible studies reach opposite conclusions?

Compare their units, years and comparison groups. Frequently one measures quotes and the other claims, or one covers liability and the other all coverages. Opposite headlines are often answers to different questions. If they cannot be reconciled, say so rather than choosing one.

How do I find out who funds an organization?

Look at the organization's about page, member list and annual report. Many nonprofits also file public tax returns (Form 990) that show their finances and officers, so check what yours discloses. Our Insurance Information Institute statistics and how to use them article applies the same routine to the Insurance Information Institute.

Key takeaways

• A sponsor's interest shapes the question, data and wording of a study more than its arithmetic. Check the sponsor, then check the method.
• Quote studies show what insurers offer a hypothetical driver, not what customers pay, and say little about claim costs without loss data.
• Regulator and statistical-agent data is the most comparable but also the oldest; read the data years before comparing with a fresh press release.
• Look for open methods, stated limits and published responses. A study that invites checking deserves more weight.
• Read at least one study from a different type of sponsor on the same subject before forming a view.