What the Largest U.S. Governtment’s Drugged-Driving Study Actually Found
When prosecutors, the press, or a juror’s intuition tell you that “drugs cause crashes,” there is one study they should be made to confront: the 2015 NHTSA case-control study by Richard Compton and Amy Berning, Drug and Alcohol Crash Risk (DOT HS 812 117). It remains the largest, most carefully designed crash-risk study ever conducted on U.S. roads, and it tested for stimulants, narcotic analgesics, sedatives, antidepressants, THC, and “illegal drugs” as a category. Its bottom line is one that surprises people who have not read it:
Once you account for the things that actually do cause crashes — age, sex, race, and especially alcohol — drivers who tested positive for any drug class were no more likely to crash than drivers who tested negative.
Not stimulants. Not opioids. Not sedatives. Not antidepressants. Not cannabis. Not illegal drugs as a group. Not legal/medicinal drugs as a group. None of them.
That is not a defense talking point. That is the finding of the federal agency tasked with reducing highway deaths.
What the Study Did
Over twenty months, NHTSA-funded researchers in Virginia Beach, Virginia, collected samples from roughly 3,000 crash-involved drivers and 6,000 control drivers who were not in a crash. The control drivers were not randomly grabbed from the general population. Each control was matched to a crash driver by the same location, day of the week, time of day, and direction of travel — exactly one week later. That matching is what makes the study trustworthy. It controls for the obvious confounders (weekend nights are more dangerous than Tuesday mornings; certain intersections are deadlier than others) before any statistics are even run.
Researchers obtained breath samples from 10,221 drivers, oral fluid from 9,285, and blood from 1,764. Participation topped 96%.
The Headline Numbers
Every Drug Class, After Proper Adjustment
This is the table that does the work. These are the odds ratios after adjusting for age, sex, race/ethnicity, and alcohol use:
| Drug Class | Adjusted Odds Ratio | Statistically Significant? |
|---|---|---|
| THC (cannabis) | 1.00 | No |
| Stimulants | 0.92 | No |
| Sedatives | 1.19 | No |
| Narcotic Analgesics (opioids) | 1.17 | No |
| Antidepressants | 0.86 | No |
| Illegal drugs (all) | 0.99 | No |
| Legal / medicinal drugs (all) | 1.02 | No |
Read those numbers carefully. After accounting for who the drivers actually were, and whether they had been drinking, every single drug class hovers around 1.0 — the value that means “no elevated risk.” Stimulants and antidepressants actually came in below 1.0. None of the differences from 1.0 reached statistical significance.
How the Raw Numbers Collapsed
The raw, unadjusted numbers looked a little more dramatic at first. THC came in at 1.25; “illegal drugs” as a category came in at 1.21; both were statistically significant before adjustment. But the moment the researchers controlled for demographics, those numbers fell to 1.05 and 1.04 respectively. When alcohol was added to the adjustment, they fell again, to 1.00 and 0.99. The apparent drug effects were not drug effects at all. They were the fingerprints of other variables that travel with drug use — variables that independently predict crashes.
Alcohol Behaved Completely Differently
Alcohol was the one substance the study found to be genuinely associated with crash risk, and its risk curve climbs exponentially with BrAC:
- 0.05 BrAC: about 2× the risk of a sober driver
- 0.08 BrAC: about 4× the risk
- 0.10 BrAC: about 5.5× the risk
- 0.15 BrAC: about 12× the risk
- 0.20+ BrAC: over 23× the risk
When the researchers combined alcohol with drugs, the alcohol did essentially all of the work. Drivers at ≥0.05 BrAC with no drugs were 6.75× more likely to crash; drivers at ≥0.05 BrAC with drugs were 5.34× more likely. The difference was not statistically significant, and there was no significant interaction between any drug class and any alcohol level. Drugs were not amplifying alcohol’s effect.
In other words: in the largest U.S. study ever done on this question, alcohol is the impairing substance of consequence on the road, and drugs — across every category tested — are not.
What Is an “Adjusted Odds Ratio,” and Why Does It Matter So Much Here?
This is the concept that does the heavy lifting in the study, and the concept that most people — including a lot of expert witnesses — get wrong.
An odds ratio (OR) compares the odds of something happening in one group to the odds in another group.
- OR = 1.0 → the two groups are equally likely. No association.
- OR > 1.0 → the first group has higher odds.
- OR < 1.0 → the first group has lower odds.
So an unadjusted odds ratio of 1.25 for THC means: in the raw data, drivers who tested positive for THC had 25% higher odds of being in the crash group than drivers who tested negative.
That sounds damning — until you ask the next question: why?
The Confounding Problem
Suppose I tell you that people who carry lighters are more likely to develop lung cancer. The raw odds ratio looks alarming. But lighter-carriers and non-lighter-carriers differ in another important way: whether they smoke cigarettes. Once you statistically control for smoking, the apparent link between lighters and lung cancer disappears. The lighter never caused anything. Smoking did. The lighter was just correlated with the real cause.
That is exactly what an adjusted odds ratio does. It uses regression to ask: holding the other variables constant, does this variable still predict the outcome? If the answer is no — as it was for every drug class in this study — then the unadjusted association was being driven by something else.
What Was Driving the Raw Numbers?
Demographics, and to a lesser extent alcohol co-use. Young men crash more than older drivers and more than women. Young men also use drugs — cannabis, stimulants, opioids — at higher rates than older drivers and women. So a raw comparison of “drug-positive” versus “drug-negative” drivers is partly a comparison of young men versus everyone else. Once you adjust for age, sex, and race/ethnicity, the apparent drug effects collapse. Once you add alcohol — because some drug-positive drivers had also been drinking, and alcohol is what actually elevates crash risk — they collapse further, in some cases below 1.0.
This is not a statistical trick. It is the entire purpose of the methodology. The whole reason to run a careful case-control study, instead of just looking at percentages, is to separate the variable you care about from the variables that travel with it.
Why “Drug Presence” Is Not the Same as “Drug Impairment”
The authors flag this explicitly:
“Caution should be exercised in assuming that drug presence implies driver impairment. Drug tests do not necessarily indicate current impairment. Also, in some cases, drug presence can be detected for a period of days or weeks after ingestion.”
This caution applies to every drug class the study tested. A positive test for THC, an opioid, a benzodiazepine, a stimulant, or an antidepressant tells you that the person consumed that substance at some point in a wide window. It does not tell you they were impaired when they got behind the wheel. Detection windows vary by drug and matrix — blood is shorter than urine, urine can be days to weeks — but in every case the lab result is a record of past consumption, not a measurement of present effect.
This is why per-se drug DUI statutes, which treat a number on a lab report as proof of impairment, are scientifically indefensible. Concentration is not impairment. Presence is not impairment.
How This Study Fits the Broader Evidence
Compton and Berning’s findings are not an outlier. The authors note that across the existing literature, studies that actually measure drugs in blood or oral fluid — rather than relying on self-report, a urine screen showing past use, or other weak proxies — consistently find lower or no elevated crash risk. The studies producing scary numbers (3×, 7×, even higher) are almost always the ones using the weakest methods: self-report, urine-as-proxy, no proper control group, or no adjustment for confounders.
The 2015 NHTSA study is the methodological gold standard, and across every drug class it tested, it found no elevated crash risk after appropriate adjustment.
The Practical Takeaway
If you are a juror, a judge, a defense attorney, or a policymaker, the science says three things very clearly:
- Alcohol is the impairing substance of consequence on the road. Its risk curve is steep, exponential, and well-established for sixty years.
- The presence of drugs in a driver’s body — stimulants, opioids, sedatives, antidepressants, cannabis, illegal drugs as a category, legal drugs as a category — does not predict crash involvement. The largest and best-designed U.S. study found relative risks at or near 1.00 across the board after appropriate adjustment.
- Drug presence is not drug impairment. A positive lab result establishes consumption within a window, not impairment at the wheel.
None of this means that someone acutely intoxicated on a drug is safe to drive. It means that the cartoon version of “drugged driving” — test positive, ergo dangerous — does not survive contact with the actual evidence, according to the U.S. Government.
Reference: Compton, R. P., & Berning, A. (2015, February). Drug and Alcohol Crash Risk. Traffic Safety Facts Research Note, DOT HS 812 117. Washington, DC: National Highway Traffic Safety Administration.
