Can a THC Driving Simulator Miss Road-Test Lane Weaving at 10 mg? A 24-Driver Crossover Study

Short answer: In a controlled study of 24 experienced drivers, 10 mg of oral dronabinol increased measured lane weaving significantly during an actual highway-driving test, but the 10-mg comparison did not achieve statistical significance in a fixed-base driving simulator. At 20 mg, both settings detected a significant increase. That difference between two separate significance tests is not itself proof that their drug responses were statistically different, nor that a particular simulator falsely cleared a particular driver. The same investigators reported no significant overall treatment-by-environment interaction for lane weaving. The study is useful precisely because it demonstrates both the promise and the limits of simulator validation for impairment litigation.

The primary evidence is Veldstra JL, Bosker WM, de Waard D, Ramaekers JG, and Brookhuis KA (2015), Comparing treatment effects of oral THC on simulated and on-the-road driving performance: testing the validity of driving simulator drug research, Psychopharmacology 232:2911–2919, doi:10.1007/s00213-015-3927-9. The comparison was conducted in the same volunteers, not in two independent cohorts.

What did investigators actually test?

The 24 participants were licensed, experienced drivers: 14 men and 10 women, mean age 23.6 years (standard deviation 3.0). Their license tenure exceeded three years and their reported driving exposure met the study’s minimum annual mileage requirement. In a randomized-treatment, double-blind, placebo-controlled crossover experiment, each person received an oral capsule containing placebo, 10 mg dronabinol, or 20 mg dronabinol on separate study days. At least four days separated treatment days. Dronabinol is administered oral Δ9-tetrahydrocannabinol (THC). This is not a study of smoked marijuana, vaporized cannabis, a roadside recreational-use population, or an arbitrary mixture of cannabis products.

Drivers first performed controlled tests in a specially instrumented car on a motorway, approximately two to four hours after administration. They then performed corresponding tasks in a fixed-base simulator, approximately four to five hours after administration. The road-versus-simulator order was not counterbalanced. Although the drug conditions were crossed over within participants, the two driving environments were also separated by time since dose, possible fatigue, adaptation, and differences in physical and visual cues. That design feature matters whenever an expert attributes a difference exclusively to simulator sensitivity.

What does SDLP measure—and what does it not measure?

The main road-tracking endpoint was standard deviation of lateral position (SDLP), usually described as lane weaving. Larger SDLP reflects more variability in the car’s sideways position over the measured segment. It is a recorded performance variable, not an officer’s impression and not a directly assayed drug concentration. In the actual-road test, an electro-optical system monitored vehicle position relative to a lane marking. In the simulator, the software recorded analogous virtual lateral position while the driver viewed a projected roadway.

Mean SDLP is informative for comparing the conditions in this experiment. It is not the probability that one named driver will collide, cross a lane line, commit an offense, or demonstrate every clinical feature associated with intoxication. SDLP can also depend on roadway geometry, traffic, steering dynamics, vehicle characteristics, and test protocol. Even a carefully standardized simulator should be evaluated against the specific behavior and drug effect being inferred.

What did the directly measured lane-weaving results show?

Table 2 of the original paper gives the mean and standard error (SE) for SDLP in centimeters. The placebo-to-treatment increments below are simple arithmetic differences between published group means; they are not individual observations or newly calculated confidence intervals.

The road test thus detected a significant 10-mg increment that the simulator’s separate test did not, despite mean weaving increasing in both. With 20 mg, mean weaving increased in both environments and both comparisons reached the paper’s nominal significance threshold. The authors also reported the overall treatment effect on road SDLP (p=.002), and a simulator overall effect at p=.05. The simulator’s estimated mean at the higher dose was not uniformly smaller than the road estimate; its observed 20-mg increment was larger. The central distinction concerns precision and dose-specific detection, not an all-purpose claim that simulators always underestimate THC effects.

Why is a significant road result plus nonsignificant simulator result not proof of a significant difference?

A p-value tests a specified contrast relative to its sampling variation. Two separate p-values can fall on opposite sides of .05 even when their underlying effects have not been shown to differ. Veldstra and colleagues directly evaluated the environment comparison. For the 10-mg treatment increment, their road-versus-simulator comparison yielded p=.55; for 20 mg, p=.83. Their treatment-by-driving-condition interaction was also not significant. Consequently, the defensible interpretation is dose-specific differences in detection within this sample, without a demonstrated treatment-by-environment difference. It would be incorrect to testify that a statistically significant simulator failure was conclusively proven merely because one arm crossed the .05 threshold and another did not.

The original Figure 1 showed wider 95% confidence intervals around simulator treatment differences. This is consistent with less precision in the tested sample and raises a legitimate test-sensitivity concern, particularly when the goal is a low-dose inference. Yet interval width is not a validated individual false-negative probability. The 24-person crossover was not designed to estimate diagnostic sensitivity and specificity for one person’s impairment in a roadside population.

What can a simulator validate: absolute agreement, relative change, or an individual diagnosis?

These are three different questions. Absolute agreement asks whether the observed numerical outcome is similar in road and simulated settings. Relative validity asks whether a treatment changes performance in similar directions or magnitudes across both environments. Individual diagnostic validity would require separate evidence about correctly classifying a specified driver’s relevant condition. The paper examined the first two questions through performance contrasts; it did not create a legal impairment classifier.

The reported average SDLP levels and treatment differences were compatible with a degree of absolute and relative correspondence for road tracking. The larger simulator variability, however, is relevant when study results are used to claim that a weak or absent individual simulator signal rules out a drug effect. The inference should remain at the level of the actual protocol and population unless additional external validation supports broader application.

Did car-following measures tell the same story as lane weaving?

No. The experiment also asked drivers to follow changes in a lead vehicle’s speed. Its component measures included coherence (how closely speed changes matched), gain (over- or under-reaction), and delay (time to speed adaptation). Absolute lead-car-following characteristics differed between a real motorway and the simulated following task. The authors did not find a significant treatment main effect for coherence in either environment (road p=.34; simulator p=.12), nor for gain (road p=.12; simulator p=.38). Average delay was greater on the road, and the dose-specific delay patterns differed qualitatively.

For interpretation of the delay data, the original manuscript contains a printed p-value anomaly in one road comparison. It would not be scientifically sound to correct that entry silently or convert it into an invented significance level. The more secure conclusion is that car-following dynamics did not transfer as cleanly between environments as the lane-position endpoint. A simulator validated for one outcome under one protocol is not therefore validated for every possible driving task, dose, or driver.

What did the study measure chemically?

The investigators drew venous blood at 1.5, 4.25, and 6 hours after dosing and centrifuged it to obtain serum. A gas-chromatography/mass-spectrometry method quantified serum THC, 11-hydroxy-THC, and THC-COOH. Those are serum laboratory measurements, not simultaneously determined forensic whole-blood THC values. They cannot be renamed as whole-blood results, and the article did not establish a universal serum-to-whole-blood conversion for forensic litigation.

Serum THC values tended to be highest at the first sampling time and lower later. The source reports no statistically significant difference in a combined molar THC-plus-11-OH-THC comparison between the broad road and simulator periods (p=.38). That null test does not show that every participant had identical concentrations across testing windows, that pharmacodynamics were identical, or that time and order confounding were removed. Serum findings belong to a different measurement domain from the behavioral driving tasks.

Why are the paper’s alcohol-equivalence lines not measured BAC evidence?

The researchers included reference thresholds based on earlier ethanol-driving experiments when considering the magnitude of THC-related SDLP changes. Those benchmarks can help frame how much weaving was observed under a particular laboratory or road protocol. They do not mean that volunteers in this study drank alcohol, that their venous blood ethanol was assayed, that 10 mg or 20 mg dronabinol has a fixed equivalent BAC, or that a legal alcohol per se threshold applies to THC. Converting a group-level performance contrast into a person’s measured BAC or a case-specific collision conclusion would add unsupported assumptions.

Could testing order, time since oral dose, or driver selection matter?

Yes. Oral THC undergoes absorption and first-pass metabolism that may vary with dose and person. In the source experiment, the same volunteers experienced drug conditions on separate days, but the driving settings always occurred in a fixed sequence: road first and simulator second. Road testing happened about 2–4 hours after the capsule, simulator testing about 4–5 hours afterward. Time since ingestion and drive order are therefore partly aligned with setting. The paper sought to address this through serum measurements and within-person comparisons, but did not randomly reverse the setting order.

The participants were relatively young experienced licensed drivers, not an unselected DUI-arrest group. Training and screening procedures, supervision, and the specific motorway and virtual road tasks make the setting controlled rather than equivalent to spontaneous driving during an actual alleged event. A result may inform a scientifically justified hypothesis about lane keeping; it cannot substitute for the traffic, roadway, vehicle, person, pharmacology, and evidence specific to a legal case.

Does nonsignificant performance at 10 mg in the simulator mean the person was unimpaired?

No. A nonsignificant group contrast is not affirmative proof that every participant performed normally or that the test has sufficient power to exclude a clinically important effect. In fact, the same study found a significant mean road SDLP effect at 10 mg, while the simulator mean also moved in the direction of more weaving. Conversely, one should not label a particular driver impaired solely because the experiment’s average road SDLP moved after oral THC. Both categorical leaps ignore dispersion, individual baselines, precision, and missing diagnostic validation.

How should a toxicologist or trial lawyer use this evidence?

The strongest supported use is methodological. When a party offers simulator data as proof that an exposure did or did not affect real driving, ask which endpoint, drug formulation, dose, observation window, simulator, comparison environment, and type of validity were actually tested. Then inspect the study’s participant denominator, counterbalancing, effect estimates, confidence intervals, and interaction tests rather than only comparing individual p-values. A study may support external relevance for mean SDLP while simultaneously disclosing limits for low-dose precision and other tasks such as lead-car following.

Laboratory confirmation requires the same discipline. Specify whether a figure is a directly measured concentration in serum, plasma, whole blood, breath, urine, oral fluid, or another matrix. Separate the administered oral THC dose, measured serum cannabinoid concentration, observed lane weaving, reference alcohol-equivalence concept, and legally relevant individual behavior. These are related investigative facts but not interchangeable measurements. Relevant background resources include the site’s discussions of specimen collection and forensic matrices, toxicologic pharmacokinetics, and its scientific research library.

Frequently asked forensic questions

1. Did this trial test smoked or vaporized cannabis?

No. It administered oral dronabinol capsules at 10 mg or 20 mg, plus placebo, in a double-blind crossover. Absorption and metabolite patterns need not match smoked or inhaled THC.

2. Were the road and simulator tests performed simultaneously?

No. Road testing came first about 2–4 hours after oral administration; simulator testing followed about 4–5 hours afterward. The order of environments was not counterbalanced.

3. Did the simulator fail to detect every THC effect?

No. The 20-mg simulator SDLP comparison was significant (p=.04). The 10-mg simulator comparison was not (p=.36), despite a significant 10-mg road effect (p=.02). The formal treatment-by-environment comparison was not significant.

4. Does a nonsignificant simulator result demonstrate a false-negative impairment test?

No. This experiment did not measure a reference-standard impairment diagnosis with a matched individual confusion matrix. It therefore cannot supply validated false-negative rates, sensitivity, specificity, or negative predictive value for one driver.

5. Is serum THC equivalent to a measured venous whole-blood THC value?

No. The researchers analyzed separated serum using GC-MS after venous collection. A serum concentration cannot simply be reported as a directly measured forensic whole-blood concentration.

6. Did investigators measure a participant’s blood alcohol concentration?

No. The alcohol comparison was an external performance benchmark drawn from earlier experiments. It was not a fresh venous ethanol assay in these orally THC-dosed volunteers.

7. Can the group mean SDLP increase establish a specific crash’s cause?

No. The study provides controlled task-performance evidence; it recorded no actual crashes and did not reconstruct any named driver, event, or collision. Event-specific inference needs independent case evidence and appropriate expert analysis.

Primary source and the evidence boundary

Veldstra JL, Bosker WM, de Waard D, Ramaekers JG, Brookhuis KA. Comparing treatment effects of oral THC on simulated and on-the-road driving performance: testing the validity of driving simulator drug research. Psychopharmacology. 2015;232:2911–2919. https://doi.org/10.1007/s00213-015-3927-9. This article draws empirical claims from that complete original primary report, not from abstract-only descriptions or secondary generalizations. Its observed means, standard errors, and study-specific p-values do not constitute a validated test of individual impairment or incident causation.