Can Feeling Sober Prove You Were Below the Alcohol Driving Limit? A 90-Person SAFE Trial

Short answer: No. Feeling sober, believing that you have reached a legal alcohol limit, or estimating that you remain below it does not reliably establish your measured alcohol concentration or your ability to drive. In a controlled study of 90 social drinkers, 33 of 85 study-day-one statements that the drinker had just reached the study’s 0.05% breath-alcohol threshold were already above that threshold; on day two, 49 of 93 such statements were above it. These were breath-test comparisons of self-estimates, not measurements of driving performance or proof of impairment in a particular legal case. The original primary research is Köchling and colleagues’ 2021 SAFE randomized crossover trial (DOI: 10.1186/s12954-021-00567-4).

What did the SAFE trial ask?

The investigators addressed a practical question faced by drivers and forensic experts: can someone drinking beer or wine accurately tell, from internal sensations, whether they have crossed an operational alcohol limit? Instead of asking participants to complete a road-driving task, the research team compared two contemporaneous quantities: a person’s self-estimated breath alcohol concentration and a measured breath alcohol concentration (BrAC). The study’s legal-limit reference was the authors’ German threshold, expressed in the paper as 0.05% BrAC. That notation must not be silently converted into a different jurisdiction’s statutory blood-alcohol definition or instrument unit.

This distinction matters in DUI testimony. A person’s confidence that they are “under the limit” is a subjective judgment. A breathalyzer reports an instrument-based breath result. Neither the judgment nor this particular study independently measures the driver’s steering control, divided attention, response inhibition, actual driving conduct, or responsibility for a collision.

How was the experiment conducted?

The authors recruited healthy adult social drinkers and ultimately analyzed 90 participants (mean age approximately 23.9 years; 50% women). The larger recruitment process screened 272 applicants; 105 were randomized and 90 completed the relevant trial. The 90 completers comprised an information/intervention group of 62 and a control group of 28. The design used matched randomization, two drinking occasions, a crossover of beverage order, and a washout interval of at least one week.

On separate visits, participants consumed beer and/or wine under supervision. Intervention participants were told when they had reached the study’s driving-limit threshold; controls were not told their BrAC. Drinking proceeded under safety limits, with a planned maximum reported BrAC of 0.11%. The study used an AlcoQuant 6020+ breath instrument with repeat measurements at approximately 45–60-minute intervals and an observation period of 15 minutes without oral intake before measurements. Each analyzed instrument reading was paired with the volunteer’s estimate of their current alcohol level.

Participants were not exposed to a blinded ethanol-versus-placebo challenge: everyone included in this analysis drank alcohol. The intervention also differed in information disclosure and beverage sequence, so the design does not isolate every potentially relevant learning mechanism. More importantly, this was not a direct test of legal driving competence.

What did the paired self-estimates and breath measurements show?

The investigators’ original Table 2 classified paired judgments according to whether the measured BrAC was on the safe side of the study threshold. The following table restates the four analytically distinct conditions. Denominators represent judgment/measurement pairs in the indicated category, not four independent populations or a count of actual drivers on the road.

The complementary category of people who said they had reached the threshold but whose measured BrAC was not above it comprised 52 of 85 pairs on day one and 44 of 93 on day two. The original paper reported 202 such selected comparisons across the study days; it also analyzed a larger set of serial paired estimates at other alcohol concentrations. A repeated comparison from the same participant cannot be treated as statistically independent without accounting for that design.

The reported 39% and 53% should therefore be described precisely: they refer to statements by participants who believed they had reached the study threshold, not to every volunteer, every breath reading, or the prevalence of intoxicated motorists. Similarly, the 75% and 41.6% figures for statements of being below the threshold were based on just 12 judgments per study day. They are not field-validated probabilities for an individual defendant.

Why can perceived sobriety differ from a measured BrAC?

Subjective sensations integrate many influences beyond the ethanol amount in a breath sample: expectancy, rate of drinking, learned tolerance, fatigue, attention, individual variability and the reference point a person uses when deciding how “drunk” they feel. A drinker may accurately report feeling relatively normal yet inaccurately estimate a numerical breath result. Conversely, someone feeling impaired need not have a particular measured BrAC. The SAFE experiment measured the gap between two observations; it did not separately quantify the biological contribution of each mechanism.

In this study, the authors defined the difference as measured BrAC minus self-estimated BrAC. A positive value meant the participant underestimated their measured result; a negative value meant they overestimated it. The paper found that underestimation tended to become more pronounced at higher measured BrACs across the serial observations. This is a group-level pattern, not a validated way to infer a missing breath reading from how a witness says they felt.

Did the second drinking session make people more accurate?

Across the analyzed within-person comparisons, the reported mean difference between measured and estimated BrAC moved from approximately −0.005 ± 0.015 on the first study day to 0.000 ± 0.013 on the second, with an exploratory paired comparison of p = 0.009. The authors interpreted that overall difference as suggesting a possible practice or awareness effect.

However, they also reported that a clear between-session difference was not identified when the intervention and control groups were examined separately. Importantly, receiving direct information about reaching the limit did not show a demonstrated advantage over the control condition. It would therefore overstate this study to claim that one warning, a simple drinking exercise, or an app necessarily teaches a particular drinker to predict their legal alcohol status.

Any educational interpretation is exploratory. Repeated exposure, beverage order, a controlled setting and study participation itself may influence self-appraisal. There was no third-party confirmation that improved estimates persisted outside the study, under different drinking conditions, or at another jurisdiction’s alcohol threshold.

Did beer versus wine explain the error?

The original paper examined beverage type and several participant characteristics. It described no meaningful beverage-type difference in self-estimation on the second study day and only a slight first-day difference favoring more accurate wine-related estimates. The investigators’ repeated-measure models associated higher peak BrAC with larger self-estimation errors. Certain exploratory analyses also implicated the time taken to reach peak BrAC on day one and vomiting on day two.

A sex-related difference appeared on day two (reported p = 0.009), but not consistently across both study days. The authors cautioned against generalizing that isolated result. Age, preference for beer or wine, BMI and the other measured characteristics were not universal individual predictors. These are exploratory associations, not validated decision rules for a specific person.

Can the trial establish an actual blood alcohol concentration?

No. The comparison in the published results used a breath instrument and the study’s BrAC notation. It did not report contemporaneous confirmatory forensic whole-blood ethanol testing by gas chromatography for every self-estimation pair. The manuscript’s percentage formatting for BrAC should be quoted as its source-reported unit, not silently relabeled as a U.S. blood-alcohol percentage, given that breath and blood are different specimens with different instrument and statutory conventions.

Whole-blood ethanol testing, evidential breath testing, and estimated alcohol levels can differ in sampling time, specimen matrix, measurement principle, uncertainty and legal interpretation. A scientific opinion about a particular accused person’s BAC at a disputed earlier time needs the actual measurement records, instrument basis, collection timestamps, relevant absorption or elimination evidence and jurisdiction-specific definitions. A participant’s confidence cannot substitute for those measurements.

Does saying “I felt fine” prove—or disprove—impairment?

Neither. The SAFE trial addresses estimation accuracy relative to a study threshold. It provides evidence against treating self-appraisal as a reliable quantitative alcohol assay. It did not measure standardized field sobriety tasks, lane deviation, real-world driving, collisions, reaction-time test batteries or a threshold for impairment in an identified motorist. There was no trial-based estimation of a specific witness’s driving fitness, intent, memory accuracy or criminal responsibility.

Subjective self-confidence, a measured BrAC, performance on a defined task, statutory alcohol limits and causation are distinct questions. Forensic testimony should not convert the trial’s percentage of erroneous estimates into the probability that any defendant was impaired, that a given crash was alcohol-caused, or that particular testimony about feeling sober is false.

What does the study allow a forensic expert to say?

It supports the limited conclusion that people can underappreciate their measured alcohol level while drinking and may cross a defined breath-testing threshold before recognizing that they have done so. It also documents that some drinkers estimated their threshold more safely than others. The appropriate inference is variability and the limitations of subjective estimation—not automatic dishonesty, incapacity or a numeric reconstruction of an unmeasured concentration.

For deposition or trial examination, asking which variable was actually observed is more useful than treating the study as a referendum on the general reliability of a witness’s feelings. One can accurately testify that a person reported feeling safe to drive without inferring what their blood test would have shown.

What limitations matter most when applying this study?

The 90 completers were predominantly young adults who regularly drank beer or wine, selected under multiple eligibility exclusions and tested in a medically supervised, semi-naturalistic environment. Participants knew they were being studied and were asked about a numerical driving threshold. Repeated measurements and two intervention days create within-person dependencies. Differences in disclosure and beverage sequences limit clean attribution of improvement to a single educational mechanism; there was no alcohol-free masked placebo in this phase.

The reported outcome is relative to a particular German threshold in the original research, not a universal standard for legal driving. The study did not sample older, medically complex or broadly representative DUI defendants, and it did not validate a clinical diagnostic test of impairment. Its supplementary ecological comparison of national legal limits and traffic deaths was not a randomized country-level experiment and cannot establish that changing a statute would produce a specific causal change in deaths. These are scope restrictions, not reasons to discard the careful paired-measurement findings.

Frequently asked forensic questions

Can someone feel sober while already exceeding a measured alcohol threshold?

Yes. Some participants in this controlled trial estimated they were at or below the study’s breath threshold even though contemporaneous instrument readings were higher. That does not tell us the unmeasured BAC of a different individual.

Does the reported 53% mean that 53% of drivers are illegally intoxicated without knowing it?

No. It describes 49 of 93 selected day-two paired statements in this experimental protocol. The denominator is not a representative roadside sample, and the study threshold and breath units must not be transplanted into another jurisdiction.

Does a positive self-estimation error prove that the person was impaired?

No. It identifies underestimation of a breath reading, not an independently established deficit in driving performance, divided attention or cognition.

Are the study’s 0.05% BrAC numbers direct blood-alcohol laboratory results?

No. Those are the article’s breath-instrument results expressed using its notation. A forensic whole-blood concentration requires its own specimen, assay and unit interpretation.

Did being warned about the threshold reliably prevent underestimation?

The study found overall improvement in estimates on the second visit, but did not demonstrate a clear advantage from threshold feedback over the control condition. That result should not be restated as validated prevention.

Can beer-versus-wine choice, sex or drinking speed predict one person’s estimate?

The paper explored such associations, but none established an externally validated individualized forecasting tool. In particular, the day-two sex association was not consistent across visits and was cautioned against by the authors.

How should this study be used in litigation?

As primary evidence that self-estimated intoxication can diverge materially from paired breath measurements under controlled drinking conditions. It cannot independently supply a missing BAC, prove an individual motorist’s impairment or assign collision causation.

Primary source and related forensic science

Primary peer-reviewed research: Köchling J, Geis B, Chao CM, Dieks JK, Wirth S, Hensel KO. The hazardous (mis)perception of Self-estimated Alcohol intoxication and Fitness to drivE—an avoidable health risk: the SAFE randomised trial. Harm Reduction Journal. 2021;18:122. https://doi.org/10.1186/s12954-021-00567-4. This article discusses only findings attributable to that primary investigation.

For scientific context, see our published resources on breath alcohol measurement, forensic pharmacokinetics, and the forensic evidence library. These links provide background, not independent validation of any individualized extrapolation made above.