The moment a traffic study becomes research on people
A tube count on an arterial is engineering. A study that recruits fifty commuters, puts a data logger in their cars and asks them about their trips afterward is human subjects research, and it needs an institutional review board before the first participant signs anything. Most transportation studies sit somewhere between those two, and the place a given study lands decides whether review takes a few days, a few months, or never comes up at all. The teams that lose a field season are almost always the ones who asked the question after the protocol was written.
The good news is that the test is short, the paperwork is predictable, and an IRB is not hostile to transportation work. Boards see driving simulators, intercept surveys and naturalistic driving data all the time. What they need is a clear description of who is involved, what you collect, how you protect it and what happens if something goes wrong. Give them that early and review becomes a scheduled task rather than a surprise.
One rule of thumb before the details: the IRB office at your institution, or the board your organization designates, makes the determination on any specific study. Your job is to bring it a clear question early enough that the answer can shape the design.
The two-part test
The federal policy for the protection of human subjects, usually called the Common Rule, uses two definitions. Both must be met before a project is human subjects research. Read them as two separate questions and answer each one in writing.
Is it research? The rule defines research as a systematic investigation designed to develop or contribute to generalizable knowledge. The word that does the work is generalizable. An operational evaluation of one signal retiming, done to decide whether to keep it, is usually not designed to produce knowledge that applies elsewhere. The same evaluation written up as a method other agencies should adopt, with a sampling plan built for that purpose, starts to look like research. Intent and design matter more than the final venue, though planning to publish is a signal reviewers notice.
Does it involve human subjects? A human subject is a living individual about whom an investigator obtains information through intervention or interaction with that person, or obtains identifiable private information. Intervention covers physical procedures and changes to someone's environment made for research purposes. Interaction covers communication: surveys, interviews, focus groups, recruitment calls. Identifiable private information is information a person reasonably expects is not public and that can be linked to them.
Notice what falls outside the second question. Counting vehicles, measuring speeds at a point with no link to a driver, analyzing aggregate crash rates by corridor: none of those obtain information about an individual. Notice also what falls inside. A dataset of trips with home and work locations is identifiable even if no name appears in it, because a person's trip endpoints identify them.
One more piece of context. The Department of Transportation codifies the Common Rule at 49 CFR Part 11. If DOT or one of its operating administrations funds or conducts the work, Part 11 applies, and its requirements track the Common Rule text other agencies use. State DOTs and university transportation centers often apply the same standard to work that is not federally funded, because their institutional policies say so. Ask before you assume a state-funded study is outside review.
Study designs and where they usually land
The table below is the first-pass sort we use when a study plan lands on the desk. It is a starting point, not a determination. The IRB, or the office your institution designates, makes the call.
| Study design | Usually human subjects? | Why |
|---|---|---|
| Automated vehicle counts, speed and classification | No | No information about an identifiable person is obtained |
| Crash records analysis, de-identified extract | Usually no | No interaction and no identifiable private information, if the extract is truly stripped |
| Crash records with names, plates or addresses | Often yes | Identifiable private information used for research; frequently eligible for an exemption |
| Travel or stated-preference survey | Yes | Interaction with individuals; often eligible for an exemption when risk is low |
| Driving simulator or test-track experiment | Yes | Intervention and interaction; usually needs full or expedited review, not an exemption |
| Naturalistic driving, in-vehicle video, GPS loggers | Yes | Continuous identifiable data on participants and often on bystanders |
Two designs cause most of the confusion. The first is roadside video. Video of public traffic with no recruitment and no attempt to identify anyone may fall outside the definition or into an exemption, but once you read plates, track a vehicle across cameras, or link footage to a registry, you are obtaining information about individuals. The second is the program evaluation that drifts. A pilot evaluation that starts as an internal operational review and later gets reframed for a journal should be checked again before any new data is collected.
Exempt does not mean unreviewed
The Common Rule lists categories of research that are exempt from most of its requirements. At a general level they cover low-risk work such as surveys and interviews where responses are not identifiable or disclosure would not harm the respondent, observation of public behavior under similar conditions, secondary use of identifiable information under defined conditions, and certain evaluations of public benefit or service programs conducted or supported by a federal agency. Each category carries conditions, and the conditions are where studies fail to qualify.
The practical point is that someone other than the investigator usually decides a study is exempt. Most institutions require an IRB office determination in writing, and many agencies want to see that letter in the project file. An exemption also does not remove your obligations to participants. You still owe them an honest description of the study, a real choice to decline, and protection of whatever you collect.
Ask for the determination in writing even when the answer seems obvious. A one-page letter saying a crash records analysis is not human subjects research is worth more than a confident email, because the letter is what an auditor, a journal or a future program officer will ask for.
Federalwide Assurance and IRB registration are two different things
An institution engaged in federally supported human subjects research files a Federalwide Assurance with the Office for Human Research Protections at the Department of Health and Human Services. The assurance is the institution's written commitment to follow the rule, and it names a signatory official and a human protections administrator. An IRB, separately, registers with the same office. An organization with no board of its own can hold an assurance and designate an outside registered IRB to review its studies. Confirm which applies to your organization before the proposal goes in, because a funder may ask for the assurance number at award.
When an independent IRB makes sense
Universities have their own boards. Consulting firms, planning offices, nonprofits and many state agencies do not. For them the choice is between relying on a partner institution's IRB or contracting with an independent IRB, a registered board that reviews studies for organizations without one.
Relying on a partner's board works when a university partner is already engaged in the research. The two institutions sign a reliance agreement that names one board as the IRB of record, and the other institution accepts its review. This avoids reviewing the same protocol twice. The catch is timing: the reliance agreement itself goes through the university's research office and can take as long as the review.
An independent IRB is a fee-for-service board. It often turns submissions around on a predictable calendar, publishes its meeting dates, and is used to reviewing studies from organizations that are not academic. Choose one that has reviewed social and behavioral studies, not only clinical trials, because the templates and expectations differ. Ask for a sample determination letter and the fee schedule, and check that its registration is current.
Whichever route you choose, name it in the proposal and the work plan. An evaluator reading a study that recruits drivers wants to see who reviews it, not a promise to figure that out later.
What an IRB will ask for
Every board has its own forms, but the substance is the same. Assemble these before you open the submission portal and the first round of comments will be about the study, not about missing pieces.
- A protocol that states the research question, the design, the population, recruitment, procedures, data collected, analysis and the end of participation.
- Recruitment materials exactly as participants will see them: flyers, emails, intercept scripts, social posts, screener questions.
- Consent documents or a request to waive written consent, with the reason the waiver fits.
- Every instrument: survey questions, interview guides, simulator scenarios, the list of data fields a logger records.
- A risk section that names physical, privacy and legal risks plainly and says how each is reduced.
- A data management plan: where data lives, who can see it, how identifiers are separated, and when raw data is destroyed.
- Training records for everyone who will interact with participants or touch identifiable data.
- Compensation details: amount, schedule, and what a participant receives if they withdraw partway.
- Funding and sponsor information, including the agreement number once it exists.
- Conflict of interest disclosures for the investigators.
The risk section is where transportation protocols are weakest. Writing "minimal risk" is not an analysis. A simulator study carries a known risk of simulator sickness, and the protocol should say how you screen for it and when a session stops. A naturalistic driving study carries a risk that recorded data shows a participant breaking a traffic law, and the protocol should say who can see that footage and whether it could ever be released.
Consent when the participant is behind the wheel
Consent is a process, not a signature. The document is the record of a conversation in which the person understood what you are asking, what you will collect, what could go wrong, and that saying no costs them nothing. Transportation studies raise four specific consent questions.
Who is a participant? In an instrumented vehicle study, the enrolled driver consents. Passengers and other household drivers who use the car are also recorded. Decide how you handle them: separate consent, a rule that only the enrolled driver uses the car, or recording that pauses when an unenrolled driver is detected. Write the choice down.
What about people outside the vehicle? Forward-facing video captures pedestrians, cyclists and other drivers who never agreed to anything. Boards generally accept this when the footage is used for the research question, is not used to identify those people, and is protected and eventually destroyed. Say exactly that in the protocol.
Is the incentive coercive? Compensation should pay for time and inconvenience, not pressure people into a study they would otherwise decline. Prorate it so a participant who withdraws in week three is paid for three weeks.
When can written consent be waived? For a short anonymous intercept survey, a signed form can create the only record linking a person to the study. Boards can waive documentation of consent in situations like that, or approve an information sheet in place of a signature. Ask for the waiver with the reason; do not assume it.
Video and GPS traces are identifiable by default
Treat any trajectory or video data as identifiable until you have done something specific to change that. Removing names from a trip dataset does almost nothing, because a person's trip endpoints are their identity. A GPS trace that starts at the same house every morning and ends at the same office identifies its owner to anyone with a map, and a few points at unusual times can be enough.
The protective steps are well known, and each one costs some analytic value, so choose them against the research question.
- Trim trip ends. Drop the first and last few hundred meters of every trip, or snap endpoints to a zone, so home and work cannot be read off the trace.
- Aggregate space and time. Report flows between zones and by time band instead of individual paths when the question allows it.
- Separate the keys. Keep the participant roster and the linking key on a different system from the trip and video data, with different access.
- Blur before sharing. Blur faces and plates in any frame that leaves the secure enclave, and log who saw unblurred footage and why.
- Limit derived fields. A field like "most frequent overnight location" is an address by another name. Do not create it unless the analysis needs it.
- Set a destruction date. State in the protocol when raw video and raw traces are deleted, and keep a record that it happened.
Consider the release question at the start. If the funder expects a public dataset, the consent form has to say so, and the de-identification method has to be strong enough for public release, which is a higher bar than internal analysis. Retrofitting a public release onto a study whose consent form promised confidentiality usually means asking the board for an amendment, and sometimes it is not possible at all.
Two other reviews that are not the IRB
IRB approval answers one question: are the people in the study protected? It does not clear the study for every other rule. Two others come up often in transportation work.
The Paperwork Reduction Act. When a federal agency conducts or sponsors a collection of information from ten or more members of the public, the collection generally needs clearance from the Office of Management and Budget. A survey a contractor runs on an agency's behalf can fall under it. That clearance runs on its own track and its own calendar, entirely separate from IRB review. Ask the agency program office early whether it applies.
Agency privacy review. If the study collects personal information into an agency system, the agency may require its own privacy assessment. The IRB's data plan and the privacy office's questions overlap, so write the data section once and reuse it.
Building review into the schedule
Review time varies widely by board, by review level and by how complete the submission is. Rather than guessing a number, plan the sequence and get the board's own calendar into the schedule as soon as you know which board it is.
From protocol to approval
Three habits keep this off the critical path. First, put the determination in the proposal stage, not the kickoff. Second, write the protocol in parallel with the instruments, because a reviewer will ask about a question on the survey and you want the survey final. Third, treat changes as amendments: a new recruitment channel, an added survey question or a new data field goes back to the board before it goes into the field. An unapproved change is the most common way a well-run study ends up with data it cannot use.
Plan for continuing obligations too. Approved studies have reporting duties: unanticipated problems, protocol deviations, and in some cases periodic check-ins with the board. A crash or a data exposure during a naturalistic study is a reportable event, and the protocol should say who reports it and how fast.
Where transportation studies go wrong
The same handful of mistakes account for most delays and most unusable data.
- Recruiting before approval. Even a "just gauging interest" email is a recruitment contact if it is part of the study.
- Treating a de-identified label as a fact. Trip data with endpoints intact is not de-identified, whatever the file name says.
- Forgetting the subcontracted field team. Anyone who interacts with participants is engaged in the research and needs training and coverage.
- Changing the survey after approval. A reworded question is an amendment. Submit it.
- Assuming a determination carries over. A new task or phase with a new design needs its own look.
- Leaving destruction to memory. If the protocol says raw video is deleted at study close, put the date on the schedule and log the deletion.
Frequently asked questions
Often not, if the footage is of public traffic, nobody is recruited, and no one is identified or tracked. Once you read plates, follow individuals across cameras or link footage to other records, you are obtaining information about identifiable people and should ask the IRB office for a determination. Get the answer in writing either way.
At most institutions, the IRB office, not the investigator. Exemption categories carry conditions, and an independent determination protects the study if anyone later questions it. Keep the determination letter in the project file with the protocol it applies to.
Yes. It can rely on a partner institution's board through a reliance agreement, or contract with an independent IRB. If the research is federally supported, the organization also needs a Federalwide Assurance on file with the Office for Human Research Protections that designates the board it uses.
It depends on what the records contain. An extract with names, plates, addresses or other identifiers is identifiable private information, and using it for research usually brings the study under the rule, though it may qualify for an exemption. A properly de-identified extract with no way to link back to individuals usually does not.