Answer: A small human experiment found that participants were slightly less accurate when reporting how many briefly displayed objects they had seen after drinking alcohol than when performing the same task sober. In Harvey and Seedhouse’s 2021 study, average exact-count accuracy fell from approximately 67% before drinking to 64% after drinking. The decrement was statistically detectable, but the study did not test actual eyewitnesses, establish an individual alcohol-impairment threshold, or show that a particular witness miscounted people, vehicles, or events. Its design also prevents alcohol from being isolated completely from testing order and practice.
The useful forensic question is narrow: can a brief numerical observation become less accurate under alcohol exposure, even when simple small-number perception looks relatively intact? This experiment provides task-specific evidence consistent with that possibility. It is not a method for translating a breath alcohol reading into a witness-reliability score.
What did the original visual-counting study test?
Alistair J. Harvey and Molly Seedhouse published “Influence of Alcohol and Cognitive Capacity on Visual Number Judgements” in Perception in 2021 (DOI: 10.1177/0301006620984105). The researchers recruited 40 university students aged 18–21 years, including 30 women and 10 men. All had normal or corrected vision and met their alcohol-use eligibility conditions.
The protocol began while participants were sober. They completed a working-memory assessment and then a visual enumeration test. After a vodka-and-tonic beverage, they completed the enumeration test again. Three participants’ enumeration records were excluded because their patterns of responses were considered outlying or inconsistent with the task. The authors’ statistical results for the principal accuracy comparison are consistent with an analyzed group of 37, not a new sample of 47.
In the visual task, a fixation cross preceded a display of one to nine randomly positioned squares for just 200 milliseconds. A visual-noise mask followed for 400 milliseconds. Participants then entered the number of squares they thought had appeared. They were not asked to remember a crime, identify a person, drive, judge distance, or estimate a moving crowd.
This separation matters. A screen-based count of briefly presented squares can reveal something about controlled attention and rapid number judgments. It does not reproduce the viewing conditions, distractions, stress, lighting, divided attention, or later questioning involved in a real eyewitness account.
How was alcohol exposure measured—and was it a blood alcohol concentration?
The investigators administered vodka at 1.5 mL/kg to men and 1.35 mL/kg to women, mixed with tonic water. The beverage was consumed over approximately 15–20 minutes, followed by a short rest, a water mouth-rinse, and a breath test. Alcohol was measured with a Dräger Alcotest 3000 breath instrument, which recorded milligrams of ethanol per liter of breath.
The paper reported alcohol levels as BAC-style percentages derived from breath measurements, applying a fixed 2,300:1 blood-to-breath partition ratio. The reported mean was approximately 0.05%, with participants ranging from roughly 0.03% to 0.08%. These are the paper’s transformed breath-derived values, not independent ethanol measurements from venous whole blood. The study did not validate an individual’s true blood-to-breath ratio or perform a forensic blood gas-chromatography comparison.
Neither the planned vodka dose nor an instrument-derived BAC equivalent predicts identical neurocognitive effects in each subject. The study’s measurements support a comparison between two group testing occasions; they do not estimate the amount of alcohol a particular witness drank at another place and time.
Did alcohol make people worse at counting?
For exact-number answers across the array sizes, the authors reported average accuracy of about 67% sober and 64% after alcohol. The main alcohol-condition result was F(1,35) = 8.31, p = .007. The numerical difference was only about three percentage points in overall accuracy—not a threefold increase in errors, not a 3% BAC shift, and not a measure of how often actual witnesses make inaccurate statements.
The investigators also assessed enumeration precision using a Weber-fraction metric, which reflects the variability of numerical responses relative to perceived quantity. Lower values on that measure indicate greater precision. Mean Weber fraction was 0.115 sober versus 0.133 after alcohol, with a reported alcohol-condition comparison of p = .033. This outcome supported reduced task precision, but it did not provide an individual error probability for a real incident.
The experiment helps distinguish two claims often conflated in forensic discussion: the ability to see a briefly displayed item and the ability to accurately enumerate a set of visible items. An observation may be visually registered without the exact quantity being reported correctly.
| Study measure | Original finding | What it does not establish |
|---|---|---|
| Participants | 40 recruited; 3 enumeration records excluded | Representativeness of all witnesses, ages, or drinking patterns |
| Exposure | Mean reported BAC-equivalent about 0.05% (range 0.03%–0.08%) from a breath instrument | Direct laboratory whole-blood ethanol concentration for any person |
| Exact-count accuracy | Approximately 67% sober versus 64% after alcohol; F(1,35) = 8.31, p = .007 | A 3% reduction for every person, or a case-specific probability of error |
| Enumeration precision | Mean Weber fraction 0.115 sober versus 0.133 after alcohol; p = .033 | Validity of a particular numerical statement or estimate |
| Small versus larger arrays | Counts of 1–4 remained relatively accurate; errors were more apparent with larger sets | That all basic vision was unaffected, or that only large crowds can be miscounted |
Why are four objects different from eight or nine?
The researchers studied a distinction between subitising and more effortful counting or estimation. Subitising refers to the rapid recognition of very small quantities—often up to three or four items—without sequential counting. Above that range, identifying the quantity generally places greater demands on attention and cognitive processing.
In this experiment, exact counts of the small arrays stayed relatively close to ceiling. The clearer errors involved larger arrays. That pattern is consistent with difficulties in controlled numerical judgments rather than a simple inability to register a small number of items. It does not prove where in the brain alcohol acted, or show that vision itself was universally preserved.
For litigation purposes, the distinction suggests asking what numerical operation the witness actually performed. Someone who briefly saw several objects and offered an approximate number is engaged in a different task from someone who slowly and deliberately counted a stable, clearly illuminated line of people.
Did working-memory capacity change the result?
Before drinking, participants completed an operation-span assessment that combined mental arithmetic with remembering letter sequences. The researchers used these scores to examine differences in working-memory capacity (WMC). They initially divided the recruited group into lower- and higher-WMC halves, and then examined numerical-task performance.
Among alcohol-condition responses, higher operation-span scores correlated positively with exact-count accuracy for displays containing six, seven, eight, or nine squares. The reported correlations were r = .398, .426, .432, and .484, respectively, with nominal p-values from .015 to .002. The same paper did not report a significant overall alcohol-by-WMC group interaction for accuracy (p = .619). Those two facts must be interpreted together.
It is therefore too strong to claim that a high working-memory score “protects” an individual witness from intoxication, or that a low score diagnoses vulnerability. These exploratory associations were made in a small convenience sample, and a median split can obscure continuous variation. Working-memory capacity was not itself a measurement of alcohol concentration or a validated forensic risk classifier.
Were responses always slower after alcohol?
No. The authors expected intoxicated participants to take longer to enumerate larger arrays. Instead, their reaction-time pattern showed faster responses for some larger arrays after drinking, especially among participants classified as having lower working-memory capacity. The authors discussed the possibility of reduced deliberation or more impulsive choices, but treated that as a tentative interpretation rather than a demonstrated causal mechanism.
Speed and correctness must be kept separate. A faster entry on a number keypad is not proof of more efficient perception, and a slower entry is not necessarily an error. A person may make a faster but less accurate judgment, or a correct but slower one. The study recorded both outcomes precisely because no single latency value captures the accuracy of a numerical report.
The response-time pattern does not justify calculating a new performance threshold or claiming that alcohol universally speeds reactions; the appropriate interpretation remains task-specific and qualified.
Does this establish that alcohol causes eyewitnesses to miscount?
No. The experiment supplies evidence about numerical judgments made under its particular laboratory sequence. It does not measure eyewitness identification, event reconstruction, line-of-sight limitations, post-event memory, or courtroom testimony. It measured immediate keypad responses to square arrays displayed for 200 milliseconds.
There is also a major design limitation: the sober task always came before the alcohol task. This was not a placebo-controlled, counterbalanced trial in which beverage order varied across participants. Practice, fatigue, expectancy, timing and other session effects could contribute to differences. The authors themselves proposed future counterbalanced beverage designs and more demanding dual-task conditions. Even a statistically significant within-person group comparison does not eliminate these alternative explanations.
Nor did the study supply sensitivity, specificity, a receiver-operating-characteristic curve, a likelihood ratio, or a validated cutoff for classifying an individual as accurate or unreliable. It should not be used to multiply a witness’s account by 0.64, discount a statement by three percentage points, or assert that one identified BAC determines whether a witness was truthful.
What is the difference between numerical accuracy, memory, and credibility?
Numerical accuracy concerns whether a reported quantity matches the objects presented. Memory concerns retention and retrieval after time passes. Credibility is a broader evaluation that may involve perception, circumstances, consistency, bias, sincerity and corroboration. The experimental response was entered almost immediately; the paper did not measure delayed recollection or deception.
A witness could be sincere but mistaken about a rapidly glimpsed quantity. Another could give a precise count based on a long unobstructed view. The study cannot adjudicate either scenario without an event-specific factual record. It also offers no validated conversion from a breath reading or experimental accuracy percentage into the evidentiary weight of an individual’s testimony.
| Evidence level | What the study actually offers | Additional evidence needed for an individual claim |
|---|---|---|
| Alcohol exposure | Administered vodka and contemporaneous breath measurements reported as BAC equivalents | Actual case drinking timeline; properly documented specimen/instrument, if relevant |
| Visual display | One to nine squares presented on a computer screen for 200 ms | Scene geometry, duration, lighting, distance, visual obstructions and distractions |
| Immediate response | Keypad number, accuracy, precision and response time | Witness’s contemporaneous description and how the original quantity was established |
| Working-memory association | OSPAN score correlations in a small student group | Validated individual prediction model; none supplied by this paper |
| Witness credibility or impairment | Not directly tested | Case-specific corroboration, circumstances, differential explanations and appropriate legal standards |
How should lawyers and experts apply the paper carefully?
For a case involving disputed quantities—a claimed number of people in a crowd, objects handled, vehicles passing, or items briefly seen—the paper can support asking whether the observer had time to count or merely estimated, whether attention was divided, and whether the quantity exceeded a very small instantly recognizable set. But those questions are hypotheses for factual investigation, not conclusions supplied by the experiment.
Identify what was directly observed, how long it was in view, whether it moved, whether the witness had an unobstructed perspective, how soon the quantity was first recorded, and whether another independent record exists. If alcohol is at issue, separate an instrument’s measured breath ethanol from an inferred BAC and from an opinion about cognitive effects. Do not reverse the group result to calculate an earlier drinking level or to declare that an individual account is invalid.
For background on how to inspect source limits rather than treating a citation as an expert conclusion, see the Attorney Scientists research library. A separate brief on two blood samples and alcohol-elimination calculations illustrates a related analytical principle: a correctly reported observation need not justify every proposed extrapolation. The editorial and research source policy explains how these literature-based briefs are prepared and the limits of their use.
Frequently asked questions
Did everyone perform worse after drinking?
The paper reports a group-level reduction in exact-count accuracy, not that every participant became less accurate. Individual responses varied, and its group averages cannot diagnose a particular witness.
Was the mean alcohol exposure really a measured 0.05% whole-blood BAC?
No. The authors measured alcohol in breath with a Dräger instrument and converted the result to a BAC-style percentage using an assumed 2,300:1 partition ratio. They did not report laboratory-measured whole-blood ethanol as the study exposure.
Does a normal ability to identify two or three items mean someone was unimpaired?
No. Near-ceiling performance on small arrays describes that narrow task. It does not exclude deficits on larger number estimates, sustained attention, balance, driving or other functions.
Does faster counting after alcohol prove greater cognitive efficiency?
No. Faster responses coexisted with diminished overall accuracy and precision. The paper discusses speed–accuracy tradeoffs only tentatively, not as a validated individual diagnosis.
Can the study tell whether an eyewitness counted six people accurately?
No. The experiment did not test that witness, the incident setting, viewing duration, attention or later recall. Its task findings can identify relevant questions but cannot determine the truth of an individual count.
Does a 3-percentage-point group difference create a 3% forensic error allowance?
No. The 67%-versus-64% comparison is an average laboratory exact-answer difference. It is not a measurement-uncertainty interval, a probability that a specified testimony is incorrect, or an adjustment to any alcohol test.
Was this a double-blind randomized alcohol-placebo crossover?
No. Sober testing occurred first for participants, followed by vodka and a repeat task. The lack of a counterbalanced placebo condition limits causal interpretation and generalization.
Bottom line: a narrower, defensible forensic conclusion
Harvey and Seedhouse’s original experiment found modestly lower accuracy and precision in a rapid visual number-judgment task after drinking, with more pronounced difficulty for larger sets and a complicated relationship to working-memory scores. It illustrates why seeing items, counting them, remembering them and giving credible testimony are separate questions. It does not establish a universal intoxication threshold, an individual witness’s error probability, real-world event causation, or measured whole-blood BAC. Its fixed sober-then-alcohol design makes cautious interpretation essential.
Primary source: Harvey AJ, Seedhouse M. Influence of Alcohol and Cognitive Capacity on Visual Number Judgements. Perception. 2021;50(1):39–51. doi:10.1177/0301006620984105. AI-assisted literature appraisal of the complete original publication; not a case-specific legal or clinical opinion, and no personal review or adoption by Okorie Okorocha is represented.