IIHS Comment

Status: Original and derived artifacts are available for this document.

Download: pdf

April 28, 2026
The Honorable Jonathan Morrison
Administrator
National Highway Traffic Safety Administration
1200 New Jersey Avenue SE
Washington, DC 20590
Agency Information Collection Activities; Notice and Request for Comment; Incident Reporting for
Automated Driving Systems (ADS) and Level 2 Advanced Driver Assistance Systems (ADAS).
Docket No. NHTSA-2026-0529.
Dear Administrator Morrison:
The Insurance Institute for Highway Safety (IIHS) appreciates the opportunity to comment on the National
Highway Traffic Safety Administration’s (NHTSA’s) data collection efforts on vehicles equipped with
driving automation involved in crashes under its Standing General Order (SGO). The SGO is currently the
only national-crash-reporting requirement for vehicles equipped with driving automation, especially SAE
Level 4 automation (L4 vehicles), and it provides a unified data system not subject to needless state-bystate variation. IIHS encourages NHTSA to begin a process that would more closely align the SGO data
with other crash data systems and require standardized reporting of vehicle miles traveled (VMT). These
recommendations were derived from an upcoming IIHS study that used the SGO data to compare the
crash experience of L4 vehicles on public roads with that of human drivers. Improving data collection on
crashes and exposure of driving automation used on public roads is an opportunity to more efficiently
conduct research studies and—per the original purpose of the SGO—identify defects.
Since the SGO collects data on crashes, it should be designed analogously to other crash databases
maintained by NHTSA including the Fatality Analysis Reporting System (FARS) and the Crash Report
Sampling System (CRSS), although it can be simpler than FARS and CRSS in terms of data structure
complexity and number of data elements. This would allow analyses of SGO data to be performed
quicker and more accurately, and a good rubric for that level of data quality is whether a researcher can
download the latest dataset, run a program, and see usable results. Currently, this is impossible with
SGO data because of the many duplicative records (often containing different information) and the large
number of very low-severity events that are not comparable with the crashes human drivers report to
police. The IIHS study excluded 25% of the unique SGO records investigated because the incidents did
not occur on a public road, involved the L4 vehicle being operated manually, or were not a crash in the
first place; 78% of the remaining crashes were deemed less severe than what humans typically report to
police and excluded from further analyses. The third-amended SGO set some minimum reporting criteria
for property damage crashes, but it is unclear if the resulting data are directly aligned with police-reported
crash data.
In making any major changes to this important data collection system, NHTSA should seek public
comment to gain additional insights from researchers and other stakeholders, and to ensure that any
such changes do not become overly burdensome to those required to report crashes. Overall reporting
burden is a function of both the complexity of reporting a crash and the number of crashes that must be
reported. Aligning reporting criteria with what human drivers report to police would substantially reduce
the number of reportable crashes. If less severe crashes are of interest for defect investigation, a
streamlined reporting procedure could be used for these.

The Honorable Jonathan Morrison
April 28, 2026
Page 2

Some specific recommendations based on the IIHS study are provided below. There are likely other ways
to improve data collection, content, and structure that could be identified through public comment.
•

Require standardized reporting of monthly VMT by location fleetwide.

•

Identify the primary record for each crash incident or, ideally, curate the database so that it
contains one record per crash incident with the most up-to-date information for all data elements.

•

Disallow redactions of entire narratives and reduce other redactions through clear guidance. For
example, it is unclear why road names are redacted when other crash databases include them.

•

Continue to include narratives in future versions of the database. While increasingly
unsustainable as a tool for studies of the full SGO database, narratives are still useful for more
targeted studies and for defect identification.

•

Ensure consistency of index variables (report_id, same_incident_id, same_vehicle_id) across
SGO database releases and ensure these are never missing.

•

Include a variable indicating whether the crash incident was submitted under SGO Request 1 or
Request 2, as the former more closely aligns with police-reporting thresholds than the latter.

•

Code types of crash damage (e.g., scratched bumper, dented body panel, structural intrusion,
damaged L4 sensor, etc.) to help infer severity and alignment with police-reported human data.

•

Provide guidance that damage reported in narratives should be more specific than “minor
damage.” In some cases, it seemed obvious that this meant a scratched bumper, and other times
it seemed like it meant something more serious like a broken rear windshield.

•

Code crash type at the vehicle level using the acc_type variable coding scheme in FARS and
CRSS. Ideally, this would be done at a reporting level rather than from narratives.

•

Code the specific SAE level of automation that was in use immediately before impact, not just the
within-30-seconds inclusion criteria. This would allow identification of cases in which a vehicle
was being driven manually at the time of impact. Moreover, the automation_system_engaged
variable should also identify the specific level because L3 and L4 have very different
functionalities (currently the variable groups L3–L5).

•

Distinguish in the driver/operator type variable between remote operation (directly manipulating
vehicle control) and remote assistance and define these carefully.

•

Use KABCO coding for injury severity to improve alignment with police-reported data. While SGO
levels appear similar, it is not clear whether they are based on the KABCO guidance.

In summary, IIHS supports NHTSA in its efforts to collect and maintain a national data system of crashes
involving driving automation technology on our nation’s roads. This is an important tool for evaluating the
real-world impact of these technologies, identifying defects, and comparing the safety of autonomous
vehicles with human-driven vehicles. As VMT increases and L4 vehicles are deployed in new cities, it is
important to enable monitoring and evaluation that is faster and more accurate through careful
improvement of this important data system.
Sincerely,

Eric Teoh
Director of Statistical Services