Short answer: In a within-person study of 41 prescribed medicinal-cannabis users, a dose of their own THC-containing oil was not associated with a statistically significant change in response time on a video hazard-perception test. The same participants, however, selected slower hypothetical speeds and greater comfortable following distances after dosing. Their decisions about when to accept a gap in traffic did not change significantly. These are video-based responses and choices, not observed vehicle movements, a blood-THC impairment threshold, or proof of safe driving.
The distinction matters whenever a forensic opinion uses one passing cognitive test, a cautious answer on a driving questionnaire, or the presence of THC to draw a conclusion about a particular driver. Different measures ask different questions. In this experiment, no statistically detectable difference in hazard recognition coexisted with measurable differences in two driving-related choices. Neither finding established that the participants’ overall driving skill was normal or impaired.
This appraisal is grounded in the complete original investigation by Taren Mieran, Andrew Hill, Mark S. Horswill, Mathew J. Summers, and Kayla B. Stefanidis: The effects of orally ingested Delta-9-Tetrahydrocannabinol on drivers’ hazard perception and risk-taking behaviours: A within-subjects study of medicinal cannabis users, published online in 2025 and in Psychopharmacology 243:651–670 (2026).
What question did the researchers actually test?
The investigators asked whether established medicinal-cannabis consumers would respond differently to traffic videos after using their prescribed THC oil than at a separate baseline appointment. They looked at four behavioral outcomes: recognizing developing traffic conflicts, selecting a preferred speed relative to the camera vehicle, selecting a minimum comfortable following gap, and indicating when they would turn into a gap in oncoming traffic. They also asked participants how well they believed they performed and how safely they believed they would drive.
This design can identify within-person differences between two laboratory sessions. It is not a randomized, blinded THC-versus-placebo drug challenge. Participants knew whether they had consumed their medication, product THC and CBD compositions differed, and symptoms or other conditions might have changed between appointments. A change between sessions therefore should not automatically be treated as a pure pharmacological effect of THC.
Who participated, and how were the sessions organized?
Forty-three adult patients were recruited in Queensland, Australia; two withdrew before the second session, leaving 41 participants for the main analyses. The final group included 23 men and 18 women, with a mean age of 46 years and a range of 21–67. Each held an unrestricted driver’s license and drove at least once a week. Participants had been prescribed THC oil for a mean of 10 months and reported an average of 15 years of cannabis-use history. Twenty-eight reported chronic pain among the conditions for which cannabis was prescribed; sleep and mental-health conditions were also common. The treating conditions were self-reported, not verified from complete medical records.
One appointment was designated baseline, with no cannabis taken during that session. At the other appointment, participants took a single observed dose of their prescribed oil orally or sublingually. Sessions were approximately one week apart, order was counterbalanced, and testing occurred at approximately the same clock time for each participant. Everyone was asked to abstain from cannabis for at least 11.5 hours before either appointment. These controls are useful, but they do not make the baseline biologically THC-free.
What THC dose was given, and was blood THC measured?
Unlike a fixed-dose trial, this study allowed individual prescribed products and doses. The administered THC amount averaged 10.80 mg (standard deviation 11.95 mg; observed range 0.06–50 mg). The authors also reported a mean of 0.12 mg/kg THC, but this was calculated from reported administered amounts and body weight; it was not a laboratory determination of the THC entering circulation. Mean coadministered CBD was 16.05 mg, with wide variation and an observed range of 0–350 mg. Thirty-eight of the 41 users placed oil under the tongue, while three used the oral route.
No venous whole-blood THC concentration, serum THC, plasma THC, quantitative oral-fluid THC, or serial absorbed-dose profile was measured. Dräger DrugCheck 3000 oral-fluid screening was performed at the start of each session, but it is a qualitative screen for recent drug evidence, not a blood concentration or validated numerical impairment metric. THC screens were positive in 19/41 baseline appointments and 20/41 post-consumption appointments before the monitored dose. Fourteen participants screened positive at both sessions. The pre-dose screen cannot establish the concentration or functional effect at the later video assessment. Nor does a negative screening result guarantee no THC molecule was present.
What did each traffic-video task measure?
The hazard-perception task used 30 driving-view clips per appointment. Participants clicked the other road user associated with a developing traffic conflict as soon as they recognized it. Different, counterbalanced sets of clips were used across sessions. The primary score was an average standardized response-time measure in seconds, not brake force, steering control, distance traveled, or a vehicle collision.
The Video Speed Test displayed 16 road clips and asked how much faster or slower, in km/h, a person would choose to drive than the camera vehicle. It did not give a speedometer reading, require actual speed regulation, or record speed on a road. The Following Distance Test asked for a comfortable minimum following distance across 20 traffic clips; responses were converted into time gaps in seconds. The Gap Acceptance Test asked when a waiting driver would turn across oncoming traffic in 23 clips and scored time until the selected gap. These tasks are meaningful controlled proxies but not full driving examinations.
| Study component | Actual observation or measurement | Boundary on inference |
|---|---|---|
| Prescribed THC/CBD oil | Observed oral or sublingual use; reported product dose and composition; average THC 10.80 mg | Administered label-derived dose, not absorbed THC or venous blood concentration |
| Pre-session oral-fluid testing | DrugCheck 3000 screening: THC positive 19/41 baseline and 20/41 post-consumption, before dosing | Qualitative evidence, not quantitative impairment or session-at-testing blood THC |
| Hazard recognition | Computer mouse response time to developing conflicts in traffic videos, n=41 | Not actual braking, collision avoidance, or complete on-road skill |
| Chosen speed and following gap | Responses to video scenarios in km/h relative to a camera car and preferred following-time gaps, n=40 | Expressed choices, not instrumented vehicle speed or distance |
| Gap acceptance | Time to select a gap for a proposed turn in traffic videos, n=41 | Not an observed physical turn or intersection outcome |
| Self-assessment | Visual analogue self-ratings and a separate drug-effects questionnaire | Subjective appraisal, not chemical analysis or proof of individual driving safety |
Did oral THC oil slow the hazard-perception test?
No statistically significant group change was detected. Mean hazard-response time was 6.233 seconds at baseline (SD 1.717) and 6.144 seconds after oil consumption (SD 1.735), with a paired-test p=.644 in 41 participants. These are the study’s standardized scores. The slight numerical decrease is descriptive and not reliable evidence that THC improved hazard recognition. The authors’ bias-corrected bootstrap interval for the within-person contrast encompassed no difference.
The investigators also divided participants by a median split of administered THC per kilogram. That exploratory model found no significant condition effect (F[1,39]=0.188, p=.667), dose-group main effect (F[1,39]=1.784, p=.189), or condition-by-dose-group interaction (F[1,39]=0.027, p=.871). It does not follow that every person at every THC dose has unchanged hazard recognition. The sample was modest, doses were heterogeneous, and there was only one post-consumption measurement for this outcome.
Did people select slower speeds after taking their medication?
Yes, on the video-choice measure. Average preferred-speed response shifted from −1.058 km/h relative to the depicted camera vehicle at baseline (SD 4.447) to −3.302 km/h post-consumption (SD 5.106), a descriptive difference of −2.244 km/h. The paired result was p<.001 in 40 complete datasets and remained significant after the study’s familywise multiple-comparison correction. The reported Cohen’s d was 0.73.
The participants therefore indicated that they would choose to drive more slowly than the camera vehicle under the post-consumption condition. The study did not show that their actual road speed fell by precisely 2.244 km/h, that everyone made the same choice, or that taking THC reduced crash probability. A choice to slow down could reflect caution, expectations, symptom changes, familiarity with the assessment, or some combination; the design cannot isolate the motive.
Did participants choose more following distance?
Yes, for the video-based preferred gap. Mean minimum comfortable following distance rose from 1.537 seconds (SD 0.188) at baseline to 1.645 seconds (SD 0.325) post-consumption, a mean difference of 0.108 seconds. The paired test gave p=.019 in 40 complete participants, with reported Cohen’s d=0.50; the result survived the authors’ Holm adjustment for the related video-choice comparisons.
These were selected gaps on a screen, not time headways calculated from a participant’s instrumented vehicle. The observed increase also does not show that other performance dimensions were compensated to a level of adequate driving safety. In actual traffic, appropriate distance depends on roadway, speed, visibility, reaction capability, vehicle and circumstances; this experiment was not designed to prescribe an individualized safe time gap.
Was turning into a traffic gap also affected?
The study found no statistically significant change on its gap-acceptance task. Mean response or waiting time was 14.935 seconds (SD 6.837) at baseline and 14.580 seconds (SD 5.780) after oil administration, with p=.534 across 41 observations. This result is distinct from the speed and following-distance choices. It also provides no proof that a particular left-turn decision was safe, unsafe, or unchanged after cannabis use. A non-significant result is not an equivalence test.
| Outcome | Baseline mean (SD) | Post-oil mean (SD) | n | Paired p and interpretation |
|---|---|---|---|---|
| Hazard-perception response time, study seconds | 6.233 (1.717) | 6.144 (1.735) | 41 | .644; no significant difference |
| Preferred speed relative to camera car, km/h | −1.058 (4.447) | −3.302 (5.106) | 40 | <.001; slower selected speed |
| Preferred minimum following gap, seconds | 1.537 (0.188) | 1.645 (0.325) | 40 | .019; longer selected gap |
| Gap-acceptance response time, seconds | 14.935 (6.837) | 14.580 (5.780) | 41 | .534; no significant difference |
| Self-rated performance on hazard-perception task, 0–100 | 76.93 (14.01) | 71.45 (15.18) | 40 | .031; lower self-rating |
The self-rating outcome belongs to a separate appraisal question; it should not be treated as another objectively measured driving outcome. The significant speed and following-distance results survived the specified correction for the related risk-choice tests. The parenthesized spreads are standard deviations, not standard errors.
How did objective hazard detection compare with perceived performance?
After completing the hazard-perception videos, participants judged their own performance against other drivers on a 0–100 scale. Their mean rating fell from 76.93 (SD 14.01) at baseline to 71.45 (SD 15.18) after oil, a significant paired difference at p=.031 in 40 participants. Yet the group hazard-detection response time itself did not significantly worsen. This contrast demonstrates that subjective confidence and measured performance can move differently within one experiment.
Neither session demonstrated a statistically significant Pearson correlation between the self-rating and the measured hazard-perception response time: baseline r=.21, p=.190 and post-consumption r=.17, p=.296, as reported in the paper. A nonsignificant correlation is not proof that everyone misjudged their ability or that the true population correlation is exactly zero. It does limit attempts to substitute one person’s confidence for an independently validated objective measurement.
Did participants believe they were safer or less safe on public roads?
Four additional ratings asked how participants imagined their current ability to predict traffic conflicts, drive skillfully, drive safely, and manage crash risk relative to other local drivers. Some unadjusted comparisons produced nominal p values below .05, including perceived traffic-conflict prediction (.022) and perceived driving safety (.042). None of these four comparisons remained significant after the authors applied the Holm correction, so they should not be promoted to four independent positive effects.
The authors also collected a separate 17-item subjective drug-effects questionnaire. At approximately 85 minutes after dosing, mean rated drug effect was 27.66/100 (SD 25.53) and self-rated high was 20.17/100 (SD 25.68). Those numbers describe reported experiences. They are not a blood concentration, a known impairment percentage, or an observed hazard-response deficit. A person may acknowledge feeling drug effects without any one test settling overall functional status.
Why do the positive and null findings need to be interpreted together?
The study did not produce a single all-purpose ‘driving impairment’ score. It observed a null difference in hazard recognition, two positive changes in expressed speed and gap preferences, a null gap-acceptance result, and lower self-appraisal of hazard recognition. These outcomes are not logically contradictory: detecting a conflict and deciding how fast to go are distinguishable processes, with separate operational definitions and response methods.
One positive result cannot automatically be carried across other dimensions, and a null result cannot automatically erase a different positive finding. Nor do slower stated speeds and wider stated gaps prove that every underlying THC-related deficit, if present, has been fully corrected. The word compensation describes one possible interpretation of the choices; it is not the experimentally identified psychological cause or a validated measure of risk offset.
What are the major confounders and generalizability limits?
These 41 individuals were established, licensed medicinal-cannabis patients, not randomized occasional users or a representative roadside population. The repeated-session design controls some stable between-person characteristics, but the sessions differed in medication exposure, symptoms, expectations, potentially sleep and other day-specific factors. There was no blinded placebo oil and no fixed pure-THC dose across patients. A participant’s medication could relieve pain or anxiety while introducing other effects, and this study did not isolate those pathways.
Almost half screened oral-fluid THC-positive before each session despite the requested abstinence interval. Some participants had concurrent medical conditions or prescribed medications, and one showed positive screens for methamphetamine and cocaine; excluding that participant in a sensitivity analysis did not change the study’s t-test conclusions. Self-reported THC use, oil content and medication history are not equivalent to verified blood concentration and measured pharmacokinetics. The chosen 90-minute wait, together with approximately 20 minutes of preliminary neuropsychological testing, yielded only one principal post-dose video-test occasion. A time course, peak-effect verification, dose-response curve and individualized impairment threshold were not established.
How should lawyers and toxicologists use this evidence?
Begin by identifying the proposed evidentiary proposition. If the proposition concerns chemical exposure, this paper reports oral-fluid screening and observed prescribed-oil administration, not numerical blood THC. If it concerns how quickly someone recognizes a hazard in videos, use the hazard-perception result and its uncertainty. If it concerns stated speed and following choices, use the separate video proxy measurements. If it concerns actual lane keeping, a collision or a particular driver’s fitness, acknowledge that those endpoints were not measured in this trial.
A case-specific analysis would require independent records appropriate to the actual question: the product and dosing history, relevant specimen and collection time, screening versus confirmation methods, concurrent substances, observed driving or task performance, and plausible non-drug explanations. No inference about a defendant’s driving intent, a witness’s legal capacity, or an individual’s accident causation should be attributed to this group study alone.
For related but separate study designs, the published THC potency and driving research brief discusses a randomized smoked-cannabis trial with simulator and blood measurements; its design and measured outcomes must not be conflated with this oral-oil/video-choice study. The published research library organizes specimen, analytical and inference questions, while the source and editorial policy explains the limits of literature-based briefs.
Frequently asked questions
1. Did the study prove that prescribed THC oil does not impair driving?
No. It found no statistically significant difference on one video hazard-perception response-time task. It did not conduct an actual on-road safety trial, measure collisions, or establish equivalence of overall driving performance before and after medication.
2. Were the measured speed changes actual reductions in vehicle speed?
No. The −1.058 to −3.302 km/h shift was a change in participants’ hypothetical preferred speed relative to a vehicle shown in 16 videos. No participant had to operate a real vehicle in this measure.
3. Was 10.80 mg the THC concentration found in participants’ blood?
No. It was the mean administered amount of THC in prescribed oil, in milligrams. The study did not measure blood THC concentration, which would be reported in a concentration unit such as ng/mL, or quantify absorbed THC.
4. Why were oral-fluid THC screens positive during the no-dose session?
Participants were regular users who reported at least 11.5 hours of abstinence before testing. A qualitative pre-session THC-positive screen does not identify an exact last-use time or the quantitative drug concentration during later task performance.
5. Does a longer preferred following gap mean THC made these drivers safer?
Not by itself. A preferred time gap increased by 0.108 seconds on average in the study’s video-choice task. Whether this choice translated to actual safe headway or counteracted any other impairment was not evaluated.
6. Does the nonsignificant hazard response result show that every participant was unimpaired?
No. Statistical nonsignificance at the group level does not validate individual normal performance, clinical equivalence, or impairment cutoffs. It also does not rule out effects at other doses, times or traffic demands.
7. Can self-rated sobriety, confidence, or perceived skill replace objective testing?
No. Participants lowered their average self-assessment despite no significant group change in hazard-response time, and self-assessment was not significantly correlated with the task in either session. These are population findings about particular scales, not a validated individual credibility or impairment test.
Primary research and interpretive boundary
Original study: Mieran T, Hill A, Horswill MS, Summers MJ, Stefanidis KB. The effects of orally ingested Delta-9-Tetrahydrocannabinol on drivers’ hazard perception and risk-taking behaviours: A within-subjects study of medicinal cannabis users. Psychopharmacology. 2026;243:651–670. DOI: 10.1007/s00213-025-06869-w. The interpretation above is limited to the original Methods, Results, tables and Discussion and should not be mistaken for an individualized legal, toxicological or medical finding.