Strip the vocabulary away and an intelligent transportation system is an inventory. When Wyoming’s transportation department described Interstate 80 to a legislative committee in December 2020, it called the route one of the most heavily instrumented rural corridors in the United States and then listed what that meant: 160 variable speed limit signs supported by 95 traffic sensors, 42 electronic message signs, 44 weather stations and 54 webcams.
They measure the road, tell drivers something, and put somebody in a room to watch the cameras: the honest answer to what these systems do. Everything else in the field is an argument about how much good those three activities accomplish, and the argument has been going on with remarkably poor data.
Why that particular corridor got instrumented
Wyoming’s case for spending on I-80 was freight and weather rather than commuter delay. The same presentation put truck traffic at 30 to 55 percent of the total stream on an annual basis and as much as 70 percent seasonally, moving more than 32 million tons of freight a year on the department’s assumption of 16 tons per truck. Blowing snow closes the route, and a closure on a corridor with no parallel route strands loaded trucks for hours.
An instrumented corridor of that kind answers a specific failure mode rather than serving as a general upgrade, and it was chosen alongside geometry and snow fence, both of which the department listed alongside the technology as parts of the same strategy. Corridors that carry that kind of freight load are surveyed in the most important freight corridors.
The inventory also reveals the design choice a driver never sees. There are 160 speed limit signs on that corridor and 95 sensors feeding them, which means the signs outnumber the measurement points by roughly five to three. Whatever a sign displays between two sensors is therefore inferred rather than observed, and the rules governing that inference are the least public part of the whole system. A driver reads a number on a gantry as a measurement. It is the output of a control algorithm working from readings taken somewhere up the road, and which algorithm a state chose is rarely published alongside the sign count. Undisclosed control logic sits under a good deal of what more on transportation technology records as deployed.
Incident management has the clearest theory and the worst measurement
The federal case for managing incidents has been stated the same way for two decades. An archived FHWA summary from April 2004 puts traffic incidents at about a third of all delay caused by congestion, with non-recurring causes including weather, work zones and special events responsible for nearly 60 percent of congestion delay, and states that roughly 20 percent of all incidents are secondary incidents caused by the first one. The same page records why responders care independently of delay: traffic-related events are the leading cause of on-duty death in law enforcement and the second leading cause among fire and rescue personnel.
That 20 percent figure has justified an enormous amount of public spending, and in June 2023 FHWA published the study that finally tried to measure it. The research team collected crash records from ten states, Arizona, Florida, Illinois, Maine, Nevada, Ohio, Tennessee, Utah, Wisconsin and Wyoming, covering roughly 5.46 million crashes. Just under 52,000 of those had been marked as secondary by the officer completing the report, which the study describes as about 1 percent of the total.
Then the team tried to verify them by matching each one to a plausible primary crash within a two-hour, two-kilometer window. Only 15,488 could be matched, about 30 percent of the original set. The report is careful about the ambiguity: some unmatched crashes were probably secondary to non-crash incidents such as a disabled vehicle, which would leave no primary crash record to match. It is equally careful about the limit of that explanation, judging it questionable that such cases account for as much as 70 percent of the total, and concluding that the analysis raises concerns about the veracity of the secondary crash data collected by law enforcement officers on crash reports.
The coding problem shows itself inside the matched set. Roughly 52.5 percent of linked pairs carried identical latitude and longitude, and about 30 percent were recorded as having occurred at the same time as their primary crash, which is a property of how forms are completed rather than of how vehicles collide. About 84 percent were within half a kilometer.
The study’s method section is unusually forthcoming about the work required to make ten states comparable at all. Because every state uses its own crash report form with its own data elements, the team had to build a uniform schema by hand and map each state’s fields into it, a process the report describes as largely manual and sometimes subjective, and one that forced some fields to be collapsed. Locations were then snapped to the federal linear referencing network for roadway attributes and matched against a commercial historical weather service for conditions. Even after that, 2.9 percent of the flagged crashes were judged to have erroneous spatial attributes.
Anyone building a national performance measure out of state crash reports is doing that work, whether or not they write it down. This report wrote it down.
So the field has a headline justification of 20 percent, state records that flag under 1 percent, and a verified subset amounting to about three tenths of that. Those are three different measurements of three different things, and none of them supports a performance target. Work zones sit in the same measurement trap, described in work-zone crashes.
The measurement gap is documented in the shape of the guidance
FHWA published a process document in September 2016 on establishing, implementing and institutionalizing a traffic incident management performance measurement program. A federal process document for institutionalizing a measure is evidence about the measure’s status: agencies were still being taught how to begin measuring incident clearance more than a decade after the federal summary above was written.
The pattern repeats in the field guidance. FHWA’s operations guide for safety and service patrols states that the agency recognizes those patrols as one of the most effective incident management strategies available, and the document is procedural throughout, telling operators how to work a scene rather than reporting what the strategy achieves. For a field guide the division of labor is reasonable. It also means the sector’s most confidently asserted intervention is asserted in a document that carries no evaluation, and a reader who wants the benefit-cost case has to go elsewhere to find it.
Traveler information is a designated number with no mandate
The FCC designated 511 as the national travel information telephone number on 21 July 2000. What it did not do, and what no federal agency did afterwards, was require anyone to run one. FHWA’s own account of the program records that implementation was left to state and local agencies with no federal mandate, coordinated through a voluntary deployment coalition working, in the program’s phrase, from the bottom up.
Twenty-six years of that arrangement produced exactly what the design implies. Coverage, content, funding and the definition of a reportable condition are state decisions, and the federal role remains guidance. The three-digit number is a genuine public good and it is also the clearest illustration in the field of what a designation without a mandate produces: a national brand on fifty separate products.
Everything here runs into the same denominator
These systems mostly work. Their benefits are events that did not occur, counted against exposure that nobody measures. Wyoming can tell a legislature exactly how many message signs it owns. No agency can say how many vehicle hours of exposure to a queue those signs prevented, because the queue that never formed leaves no record.
That asymmetry decides which investments survive budget cycles. A dark message sign generates complaints within an hour. A camera that stopped working is noticed by the operator who was watching it. A crash that did not happen because a speed limit dropped eight miles upstream produces nothing at all. The corridor systems that have been measured by removal rather than by model are examined in the future of smart highways. That absence is why the evaluations that do exist are so valuable, and why the field keeps recycling a figure from 2004.