How to read a Highway Loss Data Institute loss bulletin

An HLDI table tells you how a group of vehicles performed on one insurance coverage, compared with the average passenger vehicle, which is always set at 100. A relative claim frequency of 140 means claims arrive 40% more often than average; a relative overall loss of 60 means insurers paid 40% less per insured vehicle-year than average. Every table covers a single coverage, such as collision or property damage liability, so a vehicle can be a 60 on one table and a 140 on another. The scores describe what insurers paid for a type of vehicle, not what you will be charged.

This guide takes one kind of table, the relative-loss table, and walks through its vocabulary, the 100 scale, the grouping by model year and class, the adjustments behind HLDI's bulletins, how driver-assistance results are reported, and the reasons you cannot turn a score into a price.

What the Highway Loss Data Institute publishes

The Highway Loss Data Institute (HLDI) is, in its own words, a nonprofit that gathers, processes and publishes insurance data on the losses associated with owning and operating motor vehicles. It is a sister organization to the Insurance Institute for Highway Safety (IIHS); IIHS and HLDI insurance loss data: what it shows about cars explains the relationship. Insurers send HLDI claims and exposure data. An April 2023 bulletin on animal strikes, for example, notes that 37 companies reported comprehensive coverage data, and the 25 that supplied animal-strike information represented 40% of the HLDI database at that time.

Readers of the public material will meet two formats:

• Loss facts. Short sheets that summarize results by vehicle class and size. The December 2025 collision sheet and the property damage liability sheet both state that they summarize a larger report available only to IIHS-HLDI member companies.
• Bulletins. Research papers on a specific question, such as a feature, a make or a seasonal pattern. The Mazda driver assistance bulletin from December 2025 is an example.

The sheets give you the relative-loss tables most readers are looking for; the bulletins show the method and the caveats. Reading both is the quickest way to understand either.

The vocabulary in every table

HLDI uses four terms consistently across publications.

• Insured vehicle year. The unit of exposure. One vehicle insured for a year is one insured vehicle year; so are two vehicles insured for six months each.
• Claim frequency. Claims divided by exposure, usually quoted per 100 or per 1,000 insured vehicle years.
• Claim severity. Total loss payments divided by the number of claims paid, in dollars per claim. HLDI warns that it is not a measure of crash speed or injury severity; it is a payment measure.
• Overall losses. Frequency times severity, in dollars per insured vehicle year. In the collision and property damage sheets, HLDI describes it as the average cost of insuring a vehicle for one year, excluding administrative costs.

That last phrase matters. Overall losses leave out the insurer's operating expenses, commissions, taxes and profit, so they are a component of a premium and nowhere near the whole. The relationship between frequency and severity is covered in claim frequency vs severity: which one actually moves premiums?.

Which coverage the table covers

Each table covers one coverage. HLDI's studies use these:

The coverage splits come from the Mazda bulletin and show why a single "vehicle score" does not exist: your state decides which of these coverages you buy and which table applies to you.

Reading the 100 scale

Every relative table sets the average for all passenger vehicles to 100. A score is a percentage of that average, so 60 means 40% below and 140 means 40% above.

The baselines on the two December 2025 sheets, which cover the 2022-24 model years, are:

Overall loss follows from the other two. Multiply relative frequency by relative severity and divide by 100, and you get relative overall loss, subject to rounding. Here are rows from the collision sheet that show it.

Check the arithmetic: 144 x 126 / 100 is about 181, and the sheet shows 182. 52 x 255 / 100 is about 133 against a printed 132. Rounding explains the gaps.

What 60 and 140 mean in dollars

Put the collision baseline to work. If 100 equals $604 per insured vehicle year, a score of 140 corresponds to about $846 and a score of 60 to about $362. The very large luxury car at 278 is about $1,679 in claim payments per vehicle-year, and the midsize wagon at 63 is about $380. These are claim dollars before expenses and profit, shown only to give the scale a sense of size. They are not premiums.

Look at the gap between 63 and 278. The collision sheet itself says very large luxury cars had more than four times the overall losses of midsize station wagons, the lowest category.

Frequency and severity can pull apart

The small sports car row is the clearest case. Its collision claims are far rarer than average (52) but the average claim costs more than two and a half times the norm (255), so the net score is still above average (132). The property damage liability sheet explains the low frequency side: sports cars tend to accumulate fewer miles than comparable vehicles of the same age. Fewer miles mean fewer chances to crash, regardless of the car's design.

That is a reason to read all three columns rather than only the overall one. A vehicle with high frequency and low severity is claimed on often for modest amounts; one with low frequency and high severity is claimed on rarely but expensively. The first pattern may reflect who drives the vehicle and how much; the second may reflect what the vehicle costs to repair or replace. The tables show the pattern, not the cause. See car insurance myths: color, make, model year and horsepower for how these differences become myths.

Model-year and class grouping

Both sheets are labeled "2022-24 model years," and they sort vehicles into classes (two-door cars, four-door cars, station wagons, minivans, sports cars, luxury cars, pickups, SUVs, luxury SUVs and vans) and sizes from micro to very large. Three reading points follow.

1. The model-year range is part of the finding. A score describes the vehicles built in those model years as they performed in the period HLDI measured. A redesign, a new safety system or a change in who buys the vehicle can change the result. how a redesign changes a car's insurance rating group and your premium shows how redesigns affect rating.
2. A class is an average of many models. The "SUVs, midsize" row spans many nameplates with different owners and prices. A single model can sit well above or below its class.
3. Results from different sheets are not comparable. The baseline changes from one coverage and publication to the next. Compare scores only within a single sheet.

In its bulletins HLDI often combines a series and model year into a single variable so that comparisons are made within the same design. The bulletin comparing electric vehicles with their gasoline counterparts, for example, treats a series and a model year as one unit and then adds the powertrain as a separate factor.

What the bulletins hold constant

A raw relative score mixes the vehicle with the people who buy it and the places where it is driven. The bulletins try to separate them with regression analysis. In the Mazda study, the covariates were vehicle model year and series, calendar year, garaging state, vehicle density (registered vehicles per square mile), rated driver age group, gender and marital status, deductible range for collision, and the risk class. Claim frequency was modeled with a Poisson distribution and severity with a Gamma distribution, both with a logarithmic link.

You do not need the statistics to use the results, but three points help.

• The result is a percentage difference against a comparison group. HLDI converts the model's coefficient by taking its exponent, subtracting one and multiplying by 100. In the Mazda bulletin, a coefficient of -0.0265 becomes a 2.6% reduction in collision claims for vehicles with the first feature bundle compared with those without it.
• Unmeasured differences remain. The bulletin on electric vehicles added mileage from odometer records supplied by CARFAX. It found that the electric series and the Nissan Leaf were driven about 30% fewer miles a day than their conventional counterparts, and the claim frequency differences for most of those models shrank once mileage was added to the model. The Tesla Model S, driven about 10% more miles per day than comparable luxury cars, showed a larger difference after the adjustment, which HLDI said may reflect driving differences the database cannot capture. Mileage is not among the controls listed in the Mazda study, so a low-mileage vehicle can look safer than the vehicle itself deserves.
• The sheets do not repeat the controls. The loss facts I read describe what they cover but do not list the adjustments. Treat their class-level scores as summaries and go to a bulletin when you need to know what was held constant.

Reading the driver assistance results

Bulletins on safety features use a different table. Instead of a relative score on a 100 scale, they show the percentage change in claim frequency, severity and overall losses for vehicles with a feature compared with similar vehicles without it, with a 95% confidence interval for each estimate. A confidence interval is the range within which the true effect probably lies; if the range includes zero, HLDI calls the result not statistically significant.

The December 2025 Mazda bulletin analyzed six feature bundles for 2015-23 models. Two of them show the range:

Several features of this table are typical of ADAS bulletins.

• Reductions in frequency, increases in severity. Both bundles show claims becoming rarer but, on average, costlier. The bulletin offers two explanations: higher repair costs for the sensors and components, and a shift in what is left, because systems that prevent minor crashes remove cheap claims and leave a higher share of expensive ones. See driver-assistance sensors and the cost of repairs for the repair side.
• More features, bigger effects, with exceptions. In general, bundles with more features were associated with greater reductions in claim frequency. Driver Attention Alert added no measurable benefit beyond the bundle below it, and Traffic Sign Recognition was not associated with significant benefits under any coverage.
• Read the significance label, not just the number. Bundle F's injury-coverage reductions look larger than Bundle A's, but none of them was statistically significant. The bulletin notes for the newer Bundle D that few unequipped vehicles remained for comparison, which widens the confidence intervals on injury coverages.
• Limits HLDI states itself. There is no way to know whether a feature was switched on at the time of a crash. The data do not include detailed crash information, so all collisions are counted, whether or not a feature could have helped. And people who buy optional features may differ from those who do not in ways the controls miss.

A separate April 2026 HLDI bulletin projects how many cars have the technology. It predicts front automatic emergency braking on 37% of registered vehicles in 2025 and 59% in 2030, and says it typically takes decades after introduction for most vehicles on the road to have a given feature. That is why an effect per equipped vehicle does not immediately move overall industry losses. For older drivers, see driver assistance features for older drivers: what helps, what pays.

Why a score is not your premium

Several things stand between an HLDI score and the number on your bill.

1. Insurers price with their own data. HLDI publishes pooled results from reporting companies; an insurer sets rates from its own claims and filed rating plans. Vehicle ratings run through the insurer's vehicle symbol system, which may not match HLDI's groupings.
2. A premium is the sum of several coverages. Collision, comprehensive, liability and injury coverages are priced separately, each with its own factors. A car that is a 63 on collision and a 150 on property damage liability can be priced either way depending on the mix of coverages you buy and your limits.
3. Driver and place matter. Garaging state, driver age, record, credit where allowed, mileage and deductible all change the result. Two drivers in the same car can pay very different premiums.
4. Overall losses exclude expenses and profit. The insurer must add them to reach a rate.
5. Averages hide the tails. A class score is an average; your particular model or trim may differ, and Body style and insurance: sedans, SUVs, minivans and coupes covers how body style relates to price.
6. Time lags. The sheets describe past model years. Repair costs, parts prices and sensor content have changed quickly.

Used this way, the tables are a decent guide to which kinds of vehicles tend to cost insurers more, and a poor guide to what a quote will say.

A step-by-step way to read a table

1. Identify the coverage and model years. Is it collision, property damage liability, comprehensive or an injury coverage? Which model years? If your car is outside the range, the table is only a rough guide.
2. Find the baseline. Note what 100 stands for (claims per 100 vehicle-years, dollars per claim, dollars per vehicle-year).
3. Read frequency first. Is the vehicle crashing or being claimed on more or less than average?
4. Read severity next. Are claims cheap or expensive when they happen?
5. Check the overall score against the other two (frequency x severity / 100).
6. Consider who and where. Mileage, driver mix and region can explain a pattern. Look for a note about controls.
7. For feature studies, read the significance. Look for "statistically significant" and the confidence interval before trusting a percentage.
8. Translate cautiously. Use it to form questions for an agent, not predictions of a premium.

Questions to ask when you are comparing cars

• What is the rating symbol or group for each trim, and does it differ for collision and liability?
• Does the quote show a discount for the safety features on the car, and which ones are counted?
• How much does the quote change if I switch between the two models for the same coverages, limits and deductible?
• Are comprehensive and collision premiums separate in the quote, so I can see where the difference arises?

The car insurance quote comparison worksheet helps you line up quotes for two cars on the same terms.

Common questions

Is a lower score better?

For insurance cost, yes: a lower relative overall loss means insurers paid less per vehicle year. It says nothing about how safe the car is for its occupants. A vehicle with few collision claims might still do poorly in crash tests, and crash ratings measure something different; see vehicle safety ratings and insurance costs.

Can I look up my exact make and model?

The public loss facts are by class and size. HLDI says the larger reports behind them are available only to IIHS-HLDI member companies. Bulletins on specific makes and features are public. For your own car, the surest guide is to ask for quotes on the exact vehicle.

Why do sports cars show low frequency but high severity?

On the 2022-24 sheets, small sports cars had a relative collision frequency of 52 and a relative severity of 255. HLDI's property damage liability sheet notes that sports cars tend to accumulate fewer miles than comparable vehicles, which lowers frequency; the collision sheet shows the high severity but does not say why.

Do cars with driver assistance cost less to insure?

The research associates many features with fewer claims, especially front automatic braking, but it also finds that average claim size can rise. Whether a discount appears on your quote depends on the insurer, the state and how it identifies the feature. Ask which features it recognizes, and see how the car you drive affects your insurance rate for the broader picture. For the crash side of the picture, see Crash statistics and premiums: reading NHTSA FARS and IIHS data.

Key takeaways

• HLDI tables use a scale where 100 is the average passenger vehicle for one coverage: 140 means 40% above, 60 means 40% below.
• Overall loss is frequency times severity, before expenses and profit. In the December 2025 collision sheet, very large luxury cars had more than four times the overall losses of midsize station wagons.
• Compare scores only within a single sheet, coverage and model-year range, and read frequency and severity as well as the overall score.
• Driver assistance studies report percentage changes with confidence intervals; they often show fewer claims but higher average claim cost, and HLDI lists its own limits.
• A score describes pooled insurer payments for a vehicle type. It is not a premium prediction, because rates depend on the insurer, coverages, driver, place and expenses.