Discounts, telematics and rating factors
Driver assistance features and car insurance: what one make’s claims data show about claims and premiums
Insurer claim data on one make, Mazda 2015-23, associate bundles built around automatic emergency braking (AEB) with property damage liability claim frequency 13% to 39% lower than unequipped Mazdas. They measure losses paid, not premiums, and the average cost of a collision claim rose in five of the six bundles.
That split is the whole story in miniature. Property damage liability pays for the other driver’s car, so a braking system that prevents a rear-end crash removes a claim from that line. Collision pays for the equipped car itself, and its claim frequency fell only 2.6% to 17.3% across the bundles, while the claims that remained tended to cost more. The bulletin comes from the Highway Loss Data Institute (HLDI), the insurance-data group affiliated with IIHS. Whether a lower loss figure becomes a lower price depends on each insurer’s own rating plan, which the study does not describe.
Estimates and published averages, not a quote. Not insurance advice. State rules and insurer rules vary. Confirm requirements with your state insurance regulator or a licensed agent in your state.
What the Mazda bulletin measured
HLDI Bulletin Vol. 42, No. 30, published in December 2025 under the title “2015-23 Mazda advanced driver assistance systems,” continues a series of Mazda studies that HLDI began in 2011. Its unit of measurement is the insured vehicle year, one vehicle insured for one year. The exposure table lists 15 rows of Mazda series and drivetrains, from the Mazda3 and Mazda6 to the CX-3, CX-30, CX-5, CX-50 and CX-9, adding up to about 11.3 million insured vehicle years of collision exposure. Mazda gave HLDI the vehicle identification numbers (VINs) of cars with at least one driver assistance feature. Vehicles Mazda did not identify were assumed to have none of them and became the comparison group. Hybrids and electric vehicles were left out because Mazda did not supply their feature data.
The results apply to one manufacturer. Mazda’s front crash prevention uses a radar or laser sensor and, in some models, a forward-sensing camera, and its bundles combine features in Mazda’s own pattern. Another maker’s sensors, speed ranges and packaging can differ, so the Mazda results are not a general answer, and HLDI’s multi-make work gives a wider check. The owners are one group as well: the bulletin warns that people who pay for optional equipment may differ from those who do not in ways its statistical controls cannot capture.
The method is a regression that compares equipped and unequipped Mazdas while holding other factors fixed: model year and series, calendar year, garaging state, vehicle density, the rated driver’s age group, gender and marital status, the collision deductible range, and standard or nonstandard risk. Claim frequency was modeled as a Poisson count and claim severity, the average loss payment per claim, as a gamma distribution. Overall losses were derived from the two models. For bodily injury, medical payments and personal injury protection coverage, only frequency is reported. An effect is reported as exp(estimate) minus 1, times 100, so an estimate of -0.0265 becomes a 2.6% reduction. “Statistically significant” below means the 95% confidence interval excluded zero.
Mazda sold its features in packages, and a single sensor often serves several functions, so HLDI could not separate most of them one at a time. It grouped them into six mutually exclusive bundles, A to F, plus four stand-alone features. The grouping is the unit of the Mazda tables that follow.
| Bundle | Features in the bundle | Model years offered | Share of exposure (bulletin Table 2) |
|---|---|---|---|
| A | Front AEB and forward collision warning (FCW) | 2015-18 | 13% |
| B | A, plus adaptive cruise control (ACC), lane departure warning (LDW) and high beam assist | 2015-18 | 3% |
| C | B, plus lane departure prevention (steering assistance) | 2015-18 | 5% |
| D | C, with front AEB upgraded to detect pedestrians | From 2018 | 32% |
| E | D, plus Driver Attention Alert (fatigue and distraction prompt) | From 2021 | 5% |
| F | E, plus rear AEB | From 2021 | 2% |
One manufacturer only. The six shares add up to 60% of exposure. Benchmark, not a quote: the table describes equipment, not a price.
The model-year column limits any comparison between bundles. Bundles A, B and C exist only on 2015-18 cars, while D starts in 2018 and E and F in 2021. A step from C to D is therefore both a technology step, because the AEB gained pedestrian detection, and a step across model generations. HLDI itself describes the jump from C to D as “substantially greater benefits,” and notes that because so many newer vehicles carry pedestrian AEB, few unequipped cars remain to compare against, which produces wide confidence intervals for the injury coverages.
Claim frequency by coverage and bundle
Five coverages are reported, and each pays for a different loss. Collision pays for damage to the at-fault driver’s own vehicle. Property damage liability (PDL) pays for the damage an at-fault driver does to other people’s vehicles and property, and exists in every state except Michigan, where vehicle damage is paid on a no-fault basis. Bodily injury liability (BI) pays for injuries the at-fault driver inflicts on others, and medical payments (MedPay) pays for injuries to the insured driver and passengers. Both are analyzed in the 33 states with traditional tort systems. Personal injury protection (PIP) covers the occupants of insured vehicles regardless of fault, and is analyzed in the 17 no-fault injury states. The District of Columbia is excluded from the injury analysis. For how these coverages sit on a policy, the comprehensive and collision guide and the Michigan no-fault guide go further.
The table gives the bulletin’s own figures. Collision figures come from the full regression listing in Appendix C of the bulletin, which prints one decimal. The other cells come from the results text, which prints whole percentages and says in words whether a result was significant. Where the text gives a direction and a significance but no number, the cell says so. The bulletin shows some results only as unlabelled bars in its figures, and those cells say that no figure is given.
| Bundle | Collision | PDL | BI | MedPay | PIP |
|---|---|---|---|---|---|
| A: front AEB, FCW | -2.6% | -13% | -9% | Not significant; no figure given | Not significant; no figure given |
| B: A + ACC, LDW, high beam assist | -3.7% | -16% | -18% | Significant decrease; no figure given | Decrease, not significant; no figure given |
| C: B + lane departure prevention | -5.5% | -19% | -21% | Decrease, not significant; no figure given | Significant decrease; no figure given |
| D: C with pedestrian AEB | -16.2% | -30% | Decrease between 13% and 30%, not significant; no figure given | Decrease between 13% and 30%, not significant; no figure given | -13% (not significant) |
| E: D + Driver Attention Alert | -13.2% | -23% | Increase, not significant; no figure given | Decrease, not significant; no figure given | Decrease, not significant; no figure given |
| F: E + rear AEB | -17.3% | -39% | -21% (not significant) | -24% (not significant) | -6% (not significant) |
Mazda 2015-23 vehicles only, compared with unequipped Mazdas. Collision and PDL changes are statistically significant in all six bundles. A figure with no qualifier is statistically significant. For Bundle A the bulletin says BI claim frequency fell significantly but MedPay and PIP did not. Published averages and benchmarks, not a quote.
The PDL column is the cleanest. Property damage claim frequency fell in all six bundles and the change was significant in each, rising from 13% for the front AEB and forward collision warning pair to 39% for the full set with rear AEB. The progression is not perfectly smooth, because Bundle E, at 23%, sits below Bundle D at 30%. HLDI’s account is that Driver Attention Alert operates only under limited conditions, activating after roughly 20 minutes of driving between 41 and 86 mph and possibly not operating normally when lane markings are unclear or during lane changes, so it added no measurable benefit beyond Bundle D.
The pattern across coverages follows the crashes the systems were built for. In the bulletin’s words the front crash systems are designed “to primarily prevent front-to-rear crashes, which typically result in PDL and BI claims.” That is why the liability coverages, which pay for the car in front and the people in it, move far more than collision, which also pays for single-vehicle crashes, backing into posts and other events a forward radar does not address. Collision claim frequency fell by 2.6% for the basic pair and by 5.5% with lane departure prevention added. The step to pedestrian AEB took it to 16.2%.
The two -21% BI figures in the table mean different things. For Bundle C, a 21% fall in bodily injury claim frequency was statistically significant. For Bundle F, a 21% fall was not. Both are point estimates, and what separates them is the width of the interval around each. Bundles D, E and F rest on far fewer comparison vehicles, since a large share of newer Mazdas carry the advanced bundle, and Bundle F covers only 2% of exposure. A non-significant result does not show that the feature did nothing for injury claims; it shows the data cannot rule out zero. The collision column makes the same point with real numbers: Appendix C puts the 95% interval for Bundle D’s 16.2% reduction at roughly 25% to 7%, and for Bundle F’s 17.3% at roughly 26% to 8%. The point estimates are large and the ranges are wide.
The idea that more features mean more savings holds only partly. Bundle B added adaptive cruise control, lane departure warning and high beam assist to Bundle A and, in the bulletin’s words, showed “greater reductions in claim frequency across all coverage types” than Bundle A, and lane departure prevention in Bundle C pushed them a little further. But the same bulletin’s supplemental analysis found that bundles A, B and C were associated with increased frequency of high-severity collision claims even while they trimmed the low- and mid-severity ones, a pattern HLDI says may reflect early-generation systems being less effective at higher speeds. Bundles D, E and F showed reductions in low-, mid- and high-severity collision claim frequency too. The newer systems did something the older ones did not.
Severity: what each remaining claim costs
Claim severity is the average payment per claim. Lower frequency with higher severity is a trade, and the bulletin reports both sides for collision and PDL. Overall losses, the third measure, multiply frequency by severity, so they show the net effect on what insurers paid out.
| Bundle | Collision severity | PDL severity | Collision overall losses | PDL overall losses |
|---|---|---|---|---|
| A: front AEB, FCW | +4% (significant) | +4% (significant) | Nearly no change | -10% (significant) |
| B: A + ACC, LDW, high beam assist | +6% (significant) | +1% (not significant) | +2% (not significant) | -14% (significant) |
| C: B + lane departure prevention | +3% (significant) | -2% (not significant) | -3% (significant) | -21% (significant) |
| D: C with pedestrian AEB | Increase, not significant; no figure given | Decrease, not significant; no figure given | -14% (not significant) | -33% (significant) |
| E: D + Driver Attention Alert | -1% (not significant) | -5% (not significant) | -14% (not significant) | -27% (significant) |
| F: E + rear AEB | +7% (not significant) | +3% (not significant) | -11% (not significant) | -37% (significant) |
Mazda 2015-23 vehicles only. Percentages are as printed in the bulletin’s results text. Benchmark, not a quote: loss measures describe what insurers paid, not what drivers were charged.
Collision severity rose in five of the six bundles and fell, by 1%, in the sixth, though the rise was statistically significant only for A, B and C. PDL severity was mixed: up in A, B and F, by 4%, 1% and 3%, and down in C, D and E. The net effect under PDL is a significant fall in overall losses for every bundle, from 10% to 37%. Under collision the net is nearly flat for A and B, a significant 3% fall for C, and 11% to 14% for D, E and F, which are not statistically significant despite their size.
HLDI offers two explanations for the severity increases. The first is mechanical: “higher repair or replacement costs for the sensors and components that support these technologies” are “a likely contributor.” The second is a change in what is left. As systems prevent low-severity crashes, “the remaining claims may disproportionately consist of higher-cost incidents, thereby raising the average payment per claim.” Take away a thousand parking-lot nudges and the average of what remains goes up even if no single repair got dearer. Under PDL only the second explanation can apply, because PDL pays for the other vehicle, not the sensor-equipped one.
IIHS made the same argument in an August 11, 2026 article by its chief insurance operations officer, reporting a separate HLDI comparison of otherwise identical 2017-22 model year vehicles with and without a bundle of driver assistance features. The article does not name the make or give a bulletin number. It reports collision claim severity about 10% higher and PDL severity about 15% higher for equipped cars, with collision claim frequency about 10% lower, PDL frequency almost 40% lower, and overall losses about 5% lower under collision and almost 30% lower under PDL. IIHS reads the PDL severity rise as arithmetic: fewer low-speed fender benders leave a higher share of higher-speed crashes in the pool.
A background trend sits beneath all of this. The bulletin’s claim-size table shows the median payment on Mazda collision claims rising from $1,767 in 2014 to $4,750 in 2024, a factor of 2.7, and the median PDL payment rising from $1,980 to $3,675, up 86%. The severity differences attributed to driver assistance are single-digit percentages in the bulletin, small next to that drift. A premium that rose over the same decade need not trace to a safety feature.
How frequency and severity multiply
The overall-loss figures can be rebuilt from the other two. If frequency changes by f and severity by s, overall losses change by (1 + f) times (1 + s), minus 1. The table below does that with the bulletin’s printed percentages. Because each printed figure is rounded, the recomputed value can differ from the published one by about a point, so the table shows the range that rounding allows.
| Bundle and coverage | Frequency | Severity | Recomputed overall losses | Range allowed by rounding | Published |
|---|---|---|---|---|---|
| A, PDL | -13% | +4% | -9.5% | -10.5% to -8.6% | -10% |
| A, collision | -2.6% | +4% | +1.3% | +0.8% to +1.8% | Nearly no change |
| B, collision | -3.7% | +6% | +2.1% | +1.5% to +2.6% | +2% |
| B, PDL | -16% | +1% | -15.2% | -16.1% to -14.2% | -14% |
| C, collision | -5.5% | +3% | -2.7% | -3.2% to -2.1% | -3% |
| C, PDL | -19% | -2% | -20.6% | -21.5% to -19.7% | -21% |
| E, PDL | -23% | -5% | -26.9% | -27.7% to -26.0% | -27% |
| E, collision | -13.2% | -1% | -14.1% | -14.6% to -13.6% | -14% |
| F, collision | -17.3% | +7% | -11.5% | -12.0% to -11.0% | -11% |
| F, PDL | -39% | +3% | -37.2% | -38.0% to -36.3% | -37% |
Each published figure other than the verbal “nearly no change” equals a value in the allowed range once rounded to a whole number. Collision frequencies are the Appendix C figures with one decimal; all other inputs are whole numbers from the results text. Bundle D is omitted because the bulletin gives no severity figure for it.
The same product gives a break-even rule that is easy to carry to any headline: severity can rise by s percent without raising overall losses only if frequency falls by at least 1 minus 1/(1 + s). That works out to 3.8% for a 4% severity rise, 5.7% for 6%, 9.1% for 10% and 13.0% for 15%. Check the rule against Bundle B under collision. Severity rose 6%, which needs a 5.7% frequency drop to hold losses flat, and collision frequency fell only 3.7%. Overall collision losses therefore rose by about 2%. Bundle A under collision sits near the line: a 3.8% break-even against a 2.6% drop, which is why the bulletin calls the result nearly no change.
PDL clears the bar easily. The smallest PDL frequency drop, Bundle A’s 13%, would offset a severity rise of about 14.9%, and the largest PDL severity rise in the bulletin is 4%. The larger the frequency reduction, the more severity can rise before the saving disappears, and that is the sense in which the newer bundles are better than the older ones. In collision losses the early bundles produced almost no net change, and in PDL losses they produced reductions of 10% to 21%.
Stand-alone features and one that did not help
Four features could be evaluated alone. Blind spot detection with rear cross traffic alert shares radar sensors in the rear bumper and had to be analyzed as a pair. It is on 86% of the exposure, so it is close to the default Mazda in this dataset. Curve-adaptive headlights are on 34%, the heads-up display on 20%, and Traffic Sign Recognition on 3%, available only on 2021-23 models.
| Feature | Collision frequency | PDL frequency | BI frequency | Collision / PDL severity | Collision / PDL overall losses |
|---|---|---|---|---|---|
| Blind spot detection with rear cross traffic alert | -3.5% | Significant decrease; no figure given | -13% | +2% (significant) / +1% | -2% / -9% (both significant) |
| Curve-adaptive headlights | -2.3% | -5% | -8% | +3% (significant) / essentially unchanged | No meaningful change / -5% (significant) |
| Heads-up display | -2.7% | Significant decrease; no figure given | -9% | Essentially unchanged / +2% (significant) | -2% / -6% (PDL significant) |
| Traffic Sign Recognition | +1.1% (not significant) | +6% (significant) | +10% (not significant) | -1% / +1% | Unchanged / +7% (significant) |
Mazda 2015-23 vehicles only. Where a figure is not stated in the bulletin text, the cell says so. Collision frequency is from Appendix C.
Blind spot detection with rear cross traffic alert showed significant frequency reductions under every coverage, from 3.5% for collision (rounded to 4% in the bulletin text) to 13% for BI, which fits a feature that warns about vehicles beside and behind the car. Curve-adaptive headlights and the heads-up display both had significant reductions on all five coverages, small on collision. The headlights carry a severity cost: collision claims averaged 3% more, so collision overall losses did not move.
Traffic Sign Recognition was the exception. It was the only system the bulletin found without a significant claim frequency reduction on any coverage, and its PDL frequency went up 6% and overall PDL losses rose 7%, both significant. HLDI offers no single cause, listing environmental conditions and obstructed signage, and cites an IIHS study (Cox, Reagan and Cicchino, 2025) finding that a large majority of Mazda drivers with the feature tended to disable the audible speed warning. That points at a real limit of studies like this one. HLDI cannot see whether a feature was switched on at the time of a crash, and the bulletin says that if many drivers turn features off, the reported reductions “may actually be underestimates.”
What other makes and rating programs show
One make cannot say how the technology performs generally, and HLDI has published two multi-make documents that put the Mazda figures in context. The first is Bulletin Vol. 40, No. 4, a compendium issued in April 2023 that combines HLDI’s evaluations of eight collision avoidance technologies across 11 makes, Mazda among them. The second is Bulletin Vol. 42, No. 38, issued in December 2025, which compares 2018-19 model year vehicles across 84 vehicle series by the rating their front crash prevention system earned in IIHS testing.
| Technology | Collision | PDL | BI | Source and scope |
|---|---|---|---|---|
| Forward collision warning | -3.1% | -9.0% | -17.3% | Compendium 40-04, six makes combined (Audi, BMW, GM, Honda Accord and Odyssey, Mercedes-Benz, Volvo); results and discussion text |
| Front AEB | -3.0% | -14.4% | -23.6% | Compendium 40-04, ten makes combined (Acura, Audi, BMW, GM, Honda, Mazda, Mercedes-Benz, Nissan, Subaru, Volvo); MedPay -4.2%, PIP -4.0%; the results text calls each significant |
| Lane departure warning | -0.3% (not significant) | -0.2% (not significant) | +6.2% (not significant) | Compendium 40-04, Audi, Mercedes-Benz and Mazda; BI from Appendix A |
| Rear AEB | -8.7% | -28.8% | -10.0% | Compendium 40-04, GM, Nissan Murano, Nissan Rogue and Subaru; PDL and collision called significant in the text, BI from Appendix A (95% CI -16.5 to -3.0) |
| Front crash prevention rated basic | +2% (significant increase) | -10% | -12% | Bulletin 42-38, 2018-19 models with optional systems; BI significance not stated |
| Front crash prevention rated advanced | Nominal decrease | -14% | -22% | Bulletin 42-38; PDL change significant |
| Front crash prevention rated superior | -7% | -19% | -28% | Bulletin 42-38; PDL change significant, PDL overall losses -22% |
Different studies, populations and years; the rows are not additive and do not describe any single car. Earlier HLDI studies of Mazda are among the inputs to the compendium rows, so the Mazda and multi-make figures are not independent. Published averages and benchmarks, not a quote.
Mazda’s basic pair sits close to the multi-make AEB averages on the damage coverages: a 2.6% collision reduction against 3.0%, and a 13% PDL reduction against 14.4%. On bodily injury the Mazda pair is well below the compendium, 9% against 23.6%. The two sets of figures are not independent, because the compendium, dated April 2023, draws on earlier HLDI studies of Mazda. The compendium also separates warning from action. Forward collision warning alone was associated with PDL claim frequency 9.0% lower and BI 17.3% lower, while systems that brake automatically were associated with 14.4% and 23.6% lower, and HLDI draws the conclusion that technologies which act in a crash-imminent situation do more than those that only warn the driver.
The compendium has two findings that reinforce the Mazda caution about lane departure and severity. Lane departure warning produced inconsistent results across makes and no significant average change, and HLDI notes that the technology is relevant to only about 3% of crashes reported to police. Forward collision warning without braking, and curve-adaptive headlights, were associated with significantly higher collision claim severity across makes, while front AEB showed no significant average effect on collision severity. In one Honda Accord comparison, severity rose with a radar unit mounted in the front bumper and was essentially unchanged with a camera mounted behind the windshield, though those cars also had LED headlights, which HLDI’s earlier parts analysis found added more to repair estimates than the radar did.
The 42-38 bulletin adds the rating dimension. Vehicles whose front crash prevention earned IIHS’s highest rating in the original test showed the largest reductions, and the basic-rated group showed a significant 2% increase in collision claim frequency. HLDI used each series’ lowest rating, which it describes as a worst-case choice, and cautions that the effects for basic and advanced ratings may be overestimated. The point for a driver is the same one the Mazda bundles make: two cars that both advertise automatic braking can differ in what the braking does.
Why fewer claims is not a lower premium
The bulletin outputs changes in claim frequency, claim severity and overall losses, all derived from claims that insurers paid or reserved. No premium appears in any table. A reduction in overall PDL losses of 37% for Bundle F says what insurers paid on those cars, after adjusting for driver age, state and the other factors listed earlier. It does not say that any insurer charges 37% less, or anything less, for the same car.
Whether a loss difference reaches the price depends on the insurer’s rating plan. The NAIC’s Consumer’s Guide to Auto Insurance (2022 edition) opens its premium section with this: “Different insurance companies use different methods to rate their risk of insuring you and charge different premiums for similar coverage.” It lists the type of vehicle among the factors: “Generally, you’ll pay more for insurance, particularly for comprehensive and collision coverages, if your vehicle is newer or more expensive,” and adds that sports cars and high-performance vehicles “cost more to repair.” Its list of discounts to ask about names “protection devices such as airbags, anti-lock brakes and anti-theft devices.” That list names none of the crash-avoidance systems in the Mazda bulletin. It is a list of examples, so it neither includes nor excludes a discount for automatic braking at a given company. That is a question for the insurer, and the guide to reading a rate filing on SERFF shows where an insurer’s filed rating plan can be checked in the states that publish it.
Where the benefit falls matters too. The large gains are in PDL and BI, the coverages that pay for damage and injuries to other people. A driver whose car has a working braking system mainly shrinks what insurers pay out for the car in front and its occupants. Collision, the coverage on the driver’s own car, moved by a few percent in the early bundles, and overall collision losses rose by a nonsignificant 2% in Bundle B. That is a reading of the bulletin’s own coverage definitions, not a statement about any insurer’s pricing, but it is one reason a quote for a sensor-equipped car need not fall on every line.
The vehicle-level loss data that HLDI publishes shows what a model-by-model view looks like. The IIHS-HLDI page of insurance losses by make and model reports, for hundreds of passenger vehicles, the percentage above or below the average for all passenger vehicles. For collision, PDL and comprehensive the figure is overall losses, combining frequency and average payment, and for the injury coverages it is frequency only. The page adjusts for operator age, calendar year, density, gender, marital status, model year, risk and state, and for deductible under collision and comprehensive. It warns that a substantially redesigned model may not follow the experience of the earlier version with the same name. The page does not say how any insurer prices from these figures. What it shows is that the number summarizing a model is the net of the two forces in the Mazda tables, not frequency alone.
IIHS’s August 2026 article states the consumer case in terms of deductibles. It says that on average drivers of equipped vehicles “come out ahead” because “every crash avoided is a deductible you don’t need to pay.” That is a statement about out-of-pocket loss and not about premium, and it holds for a driver who files fewer claims. The same article concedes the other side of the ledger: an equipped vehicle in the same front-to-rear crash is “more costly to repair” because “the sensors are expensive and often require calibration after replacement.”
| Case | Invented base per 1,000 vehicles | Bulletin change applied | Equipped vehicles per 1,000 | Loss per vehicle year |
|---|---|---|---|---|
| PDL, Bundle A | 50 claims at $4,000 = $200,000 | Frequency -13%, severity +4% | 43.5 claims at $4,160 = $180,960 | $200.00 to $180.96 (-9.5%) |
| Collision, Bundle B | 60 claims at $5,000 = $300,000 | Frequency -3.7%, severity +6% | 57.8 claims at $5,300 = $306,234 | $300.00 to $306.23 (+2.1%) |
| PDL, Bundle F | 50 claims at $4,000 = $200,000 | Frequency -39%, severity +3% | 30.5 claims at $4,120 = $125,660 | $200.00 to $125.66 (-37.2%) |
The base figures are invented to show the arithmetic and are not HLDI data. The percentage changes are the bulletin’s. A loss figure is not a premium: an insurer’s filed rating plan stands between the two and is not modeled here.
Two of the three invented rows move in the direction of the bulletin’s headline, and one does not. That one, collision on a Bundle B car, is the case for asking before assuming. A driver who reads “driver assistance cuts claims” into a collision quote may be looking at the one line where the net effect, on this make and this bundle, was a small increase in losses. The bulletin does not say a price follows, in either direction.
Glass, calibration and the repair bill
The Mazda study covers collision, PDL, BI, MedPay and PIP. Comprehensive coverage, which pays for glass, is not among them, so the bulletin is silent on glass claims, where the cost of recalibrating a windshield camera enters. Other HLDI work covers that gap on other makes. Bulletin Vol. 36, No. 35, from December 2019, examined more than 270,000 glass claims on Honda and Subaru vehicles where front crash prevention was optional and could be identified from the VIN. The systems in those studies used windshield-mounted cameras. HLDI wrote that typical windshield replacement costs range from about $250 to $600, that claim amounts of $1,000 or more “likely included an FCP recalibration,” and that when a windshield on such a car is replaced, recalibration is generally required.
| Vehicles | Without front crash prevention | With front crash prevention |
|---|---|---|
| Honda models studied (Accord, Civic, Pilot) | Under 2% | Over 10% |
| Subaru Crosstrek | 7% | 22% |
| Other Subaru models studied | Around 1% or fewer | 1% (Legacy, Outback) to 8% (Impreza station wagon) |
Unadjusted percentages from the bulletin summary, selected models with optional systems. The Crosstrek figures may be inflated by an optional moonroof and heated wiper park area, as the bulletin notes. Earlier HLDI studies cited in the bulletin found glass claim severity 11% higher on Honda Accord and 16% higher on Subaru Legacy and Outback with the system. Not a statement about Mazda.
Windshield-mounted cameras are one place the cost enters. Automakers stipulate calibration anytime a sensor is removed and replaced or reinstalled, according to IIHS, and an IIHS survey of owners found problems after repair were most common when the repair involved a windshield or crash damage. The windshield replacement guide works through the survey figures, the state rules on glass claims and the deductible arithmetic for a job with a calibration line. Bumper-mounted sensors raise a different cost question, and the compendium’s Honda Accord comparison above is the example HLDI gives: forward-looking sensors in the front bumper may leave a car more exposed to repair cost.
IIHS’s 2026 article resists putting all of the blame on sensors. It cites Bureau of Labor Statistics data showing the average price of vehicle repairs up more than 40% since 2020, and lists other contributors: more sophisticated headlights, SUVs and pickups dominating the market over hatchbacks and sedans, features once reserved for luxury models, all-wheel or four-wheel drive on nearly two-thirds of new vehicles, and turbocharged, hybrid or electric powertrains on more than half. For a driver, the practical consequence is that a repair estimate on a newer car can include lines for calibration, parts and labor that have little to do with how many claims the car’s safety systems avoided. The EV insurance guide covers the same tension for electric cars, where repair cost and claim frequency also point in different directions, and the guide on choosing your repair shop covers the rules on where a sensor-equipped car gets fixed.
Using the findings on your own quote
No study cited here says what a particular insurer will do with a particular car. What the findings can do is sharpen questions. The list below is built from the bulletin’s own caveats and definitions, and each item is a question to put to an insurer or agent, not an instruction.
- Ask which driver assistance equipment the insurer has recorded for the car. The Mazda bulletin notes that the presence of optional features “is not discernible from the information encoded in the Vehicle Identification Numbers (VINs),” and HLDI had to obtain feature lists from the manufacturer. That is a reason to ask whether the equipment is on file for the car being quoted.
- Ask whether the company has a rating factor or discount for automatic emergency braking or other crash-avoidance equipment, and ask it to say in writing which coverages it applies to. The bulletin’s benefits concentrate in PDL and BI, so a discount that applies only to collision would match the data least.
- Ask how a glass claim is handled when the windshield holds a camera: whether recalibration is included in the covered repair, whether the glass deductible applies, and who performs the calibration. The car insurance claims guide helps weigh whether a given repair is worth a claim at all.
- Compare quotes on the same coverage and deductible across trims. If a trim with more driver assistance quotes lower on liability and higher on collision, the pattern is consistent with the bulletin, though one quote proves nothing, and the deductible guide shows how the collision deductible changes the comparison.
- If the car’s price has risen at renewal, ask for the factors in writing before assuming a safety feature is involved. The renewal increase guide lists what moves a renewal, and the high-premium guide does the same for a first quote.
None of these questions needs a finding that driver assistance lowers premiums. They work because the studies show what the equipment does to claims, and a price is an insurer’s separate decision. The wider set of ways to lower a bill, including documented discounts, is in the guide to lowering a car insurance premium, and the rest of the pillar sits under discounts and rating factors.
Five checks before trusting a safety-tech headline
A summary of this study can easily quote the largest number, 39%, and drop the rest. Four checks apply to any claim study. First, find the make and the model years: the 39% is a Mazda figure for a bundle available from the 2021 model year, covering about 2% of the exposure. Second, find the coverage: 39% is the change in PDL claim frequency, and the same bundle’s collision frequency fell 17.3%. Third, find whether it is frequency, severity or overall losses, because a 7% rise in collision severity belongs to the same bundle. Fourth, find the significance: the BI figure of 21% for Bundle F is a point estimate whose interval includes zero.
A fifth check applies to anything about premiums: look for a price in the source. HLDI bulletins report losses. If a headline says drivers “save,” the saving is either a loss figure, an avoided deductible, or an insurer’s own rate statement, and each is a different kind of claim. Published averages and benchmarks, not a quote. Not insurance advice. This site does not sell insurance.
Questions readers ask
Do driver assistance features lower car insurance rates?
The studies measure claims, not rates. In HLDI’s December 2025 Mazda bulletin, bundles with automatic emergency braking were associated with property damage liability claim frequency 13% to 39% lower, but the bulletin reports no premium. Whether a company gives a discount or rating credit is set in its own filed rating plan, so the answer for a given car is a question for the insurer.
How much does automatic emergency braking reduce insurance claims?
On Mazda 2015-23 models, the basic bundle of front AEB and forward collision warning was associated with a 13% fall in PDL claim frequency, 9% in BI and 2.6% in collision. HLDI’s 2023 compendium, which combined ten makes for front AEB, found it associated with a 14.4% PDL reduction, a 23.6% BI reduction and a 3.0% collision reduction.
Why do cars with safety sensors cost more to repair if they have fewer crashes?
Both are true in the HLDI data. Sensors and components raise repair cost, and systems that prevent low-speed crashes leave a higher share of costlier ones, so average cost per claim rises while claim count falls. In the Mazda bulletin collision severity rose in five of six bundles and overall PDL losses were still 10% to 37% lower.
Does the Mazda study apply to other makes?
Not directly. It covers one manufacturer’s systems and owners. HLDI’s 2023 compendium combines 11 makes and its December 2025 bulletin 42-38 covers 84 vehicle series, and both show reductions in claim frequency under PDL and BI for front crash prevention, with sizes that differ by system and rating.
Does lane departure warning reduce insurance claims?
HLDI’s 2023 compendium found lane departure warning associated with changes of 0.2% in PDL and 0.3% in collision claim frequency, neither statistically significant. The Mazda bundles that added lane departure prevention showed somewhat larger reductions than those without it, but the bulletin measures bundles, not the feature alone.
Does windshield replacement cost more on a car with a camera?
HLDI found high-dollar glass claims far more common with it. In a 2019 study of Honda and Subaru vehicles, under 2% of glass claims on Hondas without front crash prevention were $1,000 or more, against over 10% with it, and HLDI said typical replacement costs range from about $250 to $600. The Mazda bulletin did not study glass claims.
Sources, checked 2026-11-03
- Highway Loss Data Institute, Bulletin Vol. 42, No. 30: 2015-23 Mazda advanced driver assistance systems (December 2025), checked 2026-11-03.
- Highway Loss Data Institute, Bulletin Vol. 42, No. 38: IIHS front crash prevention ratings and insurance losses (December 2025), checked 2026-11-03.
- Highway Loss Data Institute, Bulletin Vol. 40, No. 4: Compendium of HLDI collision avoidance research (April 2023), checked 2026-11-03.
- Highway Loss Data Institute, Bulletin Vol. 36, No. 35: Glass claims in Honda and Subaru vehicles with and without front crash prevention (December 2019), checked 2026-11-03.
- IIHS, How crash avoidance tech simultaneously raises and slashes repair costs, by Matt Moore (August 11, 2026), checked 2026-11-03.
- IIHS, Crash avoidance features improve safety but complicate repairs (February 15, 2023), checked 2026-11-03.
- IIHS-HLDI, Insurance losses by make and model, checked 2026-11-03.
- NAIC, A Consumer’s Guide to Auto Insurance (2022 edition), How insurers determine your premium, checked 2026-11-03.