Short answer: In a randomized social-drinking experiment involving 156 young adults who regularly drank heavily, participants assigned alcohol were less accurate than placebo participants at identifying anger in a dynamic emotion-recognition task. Yet the same alcohol group reported feeling more able to relate to another participant’s emotions after a face-to-face interaction. Neither result establishes that an individual misunderstood a particular person’s anger, was aggressive, possessed a particular mental state, or gave unreliable testimony. The findings concern different measurements of social cognition, and their substantial measurement and design limitations matter.
The primary experiment is Kumar and colleagues’ 2025 Scientific Reports article, “Alcohol’s acute effects on emotion recognition and empathy in heavy-drinking young adults” (DOI: 10.1038/s41598-025-98153-2). It is especially useful where a forensic argument treats emotion recognition, empathy, anger, social conduct, and ethanol concentration as interchangeable. They are not.
What did the 156-person alcohol-and-placebo experiment actually test?
The researchers studied 156 adults averaging 22.51 years of age. Participants were recruited from a parent investigation of drinking behavior and were eligible only if they drank at least weekly and reported repeated recent binge-drinking episodes. The findings therefore reflect a selected group of relatively young, experienced drinkers; they do not define how all adults, nondrinkers, adolescents, or people with particular medical conditions react to ethanol.
Unacquainted participants were assigned to three-person groups, and the group was randomized to receive alcohol or a placebo beverage. The reported final allocation was 117 alcohol participants and 39 placebo participants. A placebo participant who did not believe the deception, along with that participant’s group, was excluded. Thus, the experimental condition was administered socially, not in an isolated individual laboratory booth. Members interacted during a 36-minute drinking period, creating an experimentally relevant social setting and also a possible pathway through which companions’ behavior affected subsequent responses.
The administered alcohol dose was 0.82 g/kg for men and 0.74 g/kg for women in a cranberry-vodka mixture. This was designed to produce a peak estimated blood-alcohol concentration around 0.08%; it was a dosing target, not an individual venous whole-blood assay result. Researchers then used a breathalyzer during the session to obtain study-reported alcohol values. The authors reported alcohol-condition mean BAC-labeled readings of 0.051% immediately after drinking, 0.058% at 30 minutes, and 0.060% at 60 minutes. Because the analytical measurement was breath based, those device-reported BAC-style figures must not be silently represented as independently measured forensic whole-blood results.
Assessments began at different times after drinking. Participants rated their experiences of others on the State Empathy Scale (SES) approximately six minutes after the drinking period; completed the Geneva Emotion Recognition Test (GERT) around 11 minutes afterward; and completed the photograph-based Multifaceted Empathy Test (MET) approximately 34 minutes afterward. The distinct instruments did not all sample the same psychological construct, stimulus, or time.
Which task detected an alcohol-associated change in anger recognition?
The GERT presents short recordings of actors conveying emotions through changing faces, voices, and bodily expressions. The full short-form instrument includes 42 clips spanning 14 emotions, but this analysis used an 18-clip subset assessing six selected emotions: joy, sadness, anger, fear, surprise, and disgust. Participants selected the emotion label they believed best matched each clip. This measures responses to staged audiovisual expressions, not the content of an actual altercation or anyone’s intent during a real encounter.
On the GERT, the alcohol group performed worse on identifying anger than the placebo group. The reported beverage-condition coefficient was B = −0.18, p < 0.01. In a sensitivity model that included demographic and drinking-related covariates, the association remained, with B = −0.23, p < 0.001. These are statistical model estimates for a particular task outcome. They are not individual error probabilities, forensic screening sensitivity, percentages of violence, or a universal change in recognition ability.
The researchers described incorrect answers to anger clips in 35% of relevant alcohol-group responses and 21% of placebo-group responses. Among the errors, irritation was the most common substitute label in both groups, followed by disgust. The investigators cautioned that their scoring method could not distinguish a diminished perceptual sensitivity to anger from a greater tendency to choose another negative-emotion label. That unresolved mechanism is important: calling an expression “irritation” rather than “anger” during a task does not establish how a person would react to real provocation.
Did alcohol impair all emotion recognition?
No statistically significant overall difference was detected between the alcohol and placebo conditions on the analyzed GERT total score. Nor did the study detect significant beverage-condition effects on recognition of joy, sadness, fear, surprise, or disgust as individual GERT outcomes. The emotion-recognition component of the photograph-based MET also showed no statistically significant alcohol-versus-placebo difference.
The appropriate reading is selective, not global. A significant anger result alongside nonsignificant other outcomes shows that the reported task measures did not all change detectably under the researchers’ analyses. It does not prove that the five nonsignificant abilities were unaffected, exactly equivalent, or clinically normal. Nor does it mean that the anger finding alone describes every aspect of social perception.
These distinctions also prevent an invalid forensic syllogism: if someone’s general emotion-recognition score looked ordinary, a case expert cannot simply conclude that their recognition of a specific expression was necessarily intact. Conversely, even a group-level difference for anger offers no validated method for inferring one witness’s perception of a particular person.
How were empathy during real interaction and empathy toward photographs different?
The study used two notably different approaches. The SES asked participants to retrospectively evaluate how they understood, shared, or related to the emotions of someone they had just interacted with. Each person rated the participant seated to their left. The MET, by contrast, displayed photographs of strangers in emotionally suggestive situations, asked participants to identify emotions, and separately asked how strongly they empathized with the depicted person.
Alcohol was associated with a higher SES associative empathy score—the perceived ability to relate to an interaction partner’s emotional experience—at B = 0.31, p < 0.05. That association remained after specified covariate adjustment at B = 0.27, p < 0.05. A higher SES total score in the primary model (B = 0.25, p < 0.05) did not retain the same significance after adjustment. There was no significant beverage-condition change in the distinct SES cognitive and affective-empathy subscales.
The photograph-based MET perceived-empathy outcome likewise showed no statistically significant alcohol effect. Therefore, the experiment does not justify the unqualified claim “alcohol increases empathy.” More precisely, the alcohol group reported greater ability to relate to a recent real interaction partner on one SES subscale, while separate photographs and other empathy measures did not show the same effect. A subjective retrospective statement about feeling connected is not a direct observation of helpful action, genuine emotional understanding, or an objectively verified empathic response.
What are the exact results that may matter in a forensic appraisal?
| Measured outcome | Reported alcohol versus placebo finding | Boundary of the finding |
|---|---|---|
| GERT identification of anger | B = −0.18, p < 0.01; covariate-adjusted B = −0.23, p < 0.001 | Group model for staged expressions; not an individual witness error rate |
| GERT overall recognition | No statistically significant beverage-condition difference | Not evidence of equivalent performance across every emotion |
| GERT joy, sadness, fear, surprise, disgust | No significant separate beverage-condition effects | Not proof that alcohol can never affect these judgments |
| MET photograph emotion recognition and perceived empathy | No significant alcohol-versus-placebo difference reported | Different third-person task, different time and stimuli |
| SES associative empathy after interaction | B = 0.31, p < 0.05; adjusted B = 0.27, p < 0.05 | Retrospective subjective relatedness, not demonstrated compassion or conduct |
| SES total empathy | Primary B = 0.25, p < 0.05; adjusted model did not preserve significance | Does not support a robust unqualified global empathy effect |
These are the authors’ reported statistics. They are not derived thresholds for diagnosing a particular witness, defendant, complainant, or driver. The models accounted for participants clustered in three-person groups; reported hypothesis tests were uncorrected for multiplicity. When several different emotions and empathy dimensions are tested, selective attention to a single small p value risks exaggerating the strength of the overall evidence.
Does a breathalyzer value in this experiment equal a forensic blood sample?
No. A breathalyzer measures alcohol-related content in an exhaled breath sample, typically presenting a converted or calibrated value under that instrument’s operating conventions. The original study labels its repeated results “BAC” but describes breathalyzer recordings; it does not report paired independently headspace-gas-chromatography-confirmed venous whole-blood ethanol specimens in this experiment. A dosing formula used to target approximately 0.08% is a prediction, not an assay.
Nor did the experiment measure ethanol concentrations inside the brain, quantify cortisol or another physiological empathy biomarker, or derive an individual-specific rate of elimination. The observed GERT classifications and SES answers are behavioral responses and self-reports. The three categories—ethanol-exposure estimate, task behavior, and subjective social impression—must remain separate.
This distinction also applies when someone tries to reason backwards to the drinking level at an incident. A later alcohol reading and a model do not substitute for a complete timed sampling history. For background on why blood results at different times need careful interpretation, see the separately focused two-blood-sample elimination research brief.
Why is the reliability of the emotion-recognition tasks consequential?
The authors explicitly identified low internal consistency for both tested recognition measures. The six-emotion analyzed GERT subset had Cronbach’s α = 0.25; including all 14 emotions gave α = 0.54. The MET emotion-recognition component had α = 0.35. The SES associative-empathy measure, by comparison, had α = 0.85. These are study-specific reliability summaries, not sensitivity or specificity against a real-world event.
Low internal consistency does not automatically erase the group-level anger comparison, but it significantly limits confidence in treating one brief task score as a stable characteristic of a particular person. It also limits arguments that a nonsignificant overall score proves normal real-world perception. The GERT uses actors and selected clips; the MET uses still photographs; neither reconstructs a disputed encounter, its context, or the emotions of the people involved.
For an expert, identifying the exact item set, number of observations, reliability evidence, scoring conventions, adjustment choices, and tested error types is more informative than merely repeating the word “significant.” Most importantly, the experiment did not establish an individual clinical or forensic cutoff.
Could drinking companions’ behavior explain part of the result?
Possibly, and the authors said their design cannot cleanly separate these pathways. All three members of an experimental group received the same beverage condition and were free to interact while drinking. Alcohol might have affected the person completing the later questionnaire, the behavior of the person being rated, the interaction between them, or some combination. The research team modeled group clustering statistically, but a statistical random intercept does not by itself isolate an individual pharmacological mechanism.
Other limitations include an unbalanced 117-to-39 allocation, differences in group sex composition between beverage conditions, a largely male and White heavy-drinking sample, different administration times for the GERT and empathy tasks, and reliance on self-reported recollections for the SES. The researchers controlled for several covariates in selected sensitivity models; that helps test some measured explanations without eliminating residual confounding or transforming the findings into individualized causation.
The placebo deception was unusually convincing, which is a strength when evaluating expectancy effects. It does not remove the effect of group interaction or mean that every laboratory measure would respond identically in a different population.
What would the study mean for testimony about perceived anger or aggression?
It supports a question, not a verdict. In a disputed confrontation, a witness may claim to have recognized anger, hostility, or fear in someone else’s tone or facial expression. This experiment illustrates why emotion perception under ethanol exposure merits careful contextual examination: a group difference was observed specifically for anger labeling. But it does not provide a validated percentage by which that witness’s account should be discounted, and it does not establish what emotion the other person actually felt.
Importantly, the authors did not measure subsequent assaults, threatening conduct, accurate eyewitness recollection of a crime, or truthful versus deceptive testimony. They did not test whether any individual participant became aggressive. Their anger clips were performed by actors, not extracted from real altercations. It would therefore be scientifically unsound to use this paper alone to claim that alcohol caused a defendant’s aggression, proved a complainant misread a gesture, or negated criminal intent.
A careful case review would examine the witness’s actual contemporaneous observations, drinking history relevant to the event, independently verified ethanol evidence and timing, lighting and auditory conditions, alternative sources of context, prior familiarity among the participants, and inconsistencies or corroboration in the whole record. No one can substitute an average experimental coefficient for those case-specific facts.
What does the social-empathy finding not establish?
The distinction between a respondent’s felt sense of relatedness and demonstrably accurate understanding of another person is central. In this study, the SES asked about remembered interactions. A higher score could reflect genuine social connection, changes in attention, self-presentation, memory, or interaction dynamics. The design was not built to adjudicate which explanation applies to an individual.
The separate photographic empathy task did not show the same increase. Nor did the SES results establish uniformly increased cognitive understanding or shared emotional experience. Repeating “alcohol increases empathy” would erase the particular measure on which the association arose and risk an unsupported inference about a person’s kindness, motive, trustworthiness, or capacity to identify another’s emotions accurately.
Readers interested in other forensic interpretation questions can explore the research library’s evidence-based reading paths. Related research may study different constructs, such as neurological responses to visual threat cues or impulse control, but findings from those separate methods should not be merged into this experiment’s estimates.
How should opposing experts test an inference drawn from this study?
| Question to ask | Why it changes the inference |
|---|---|
| What physical specimen was measured? | The study reports breathalyzer alcohol values, not a paired forensic venous whole-blood chromatogram. |
| Was the number a modeled target, measured value, or task coefficient? | 0.08% target dosing, serial breathalyzer readouts, and GERT model B coefficients represent different quantities. |
| Which emotion and instrument? | The significant effect involved GERT anger labeling, not global empathy or all six emotions. |
| How reliable was the measure? | Low GERT and MET recognition α values weaken person-level interpretation. |
| Were multiple statistical outcomes tested? | The reported tests were uncorrected for multiplicity, making selective citation hazardous. |
| Did real behavior, a crime, or road driving occur? | The experiment did not validate any individual act, guilt, witness credibility, driving outcome, or causal threshold. |
An appropriate expert opinion states whether the study actually investigated the proposed proposition, identifies the design’s level of inference, addresses null findings as carefully as positive ones, and does not impute a particular participant’s mechanism to someone never enrolled in the trial. The scientific source and editorial policy describes the importance of separating primary evidence from interpretation.
Frequently asked questions about alcohol and anger recognition
Does drinking necessarily make someone misinterpret anger?
No. The investigators found an average group difference in one task. That is not a universal effect or an individually validated diagnostic rule.
Did participants become more aggressive after drinking?
The study did not measure aggression as its relevant outcome. Anger labeling during staged clips does not establish actual hostile behavior.
Does the study show alcohol increases empathy?
Only one specific retrospective interpersonal relatedness subscale retained a significant association after covariate adjustment. Photograph-based perceived empathy and other SES components did not show matching effects.
Was the approximately 0.08% alcohol figure a measured blood concentration?
No. It was the dose-targeted estimate. The researchers collected breathalyzer readings during the session and reported BAC-style values; they did not report paired forensic whole-blood GC results here.
Can a normal total emotion-recognition score rule out alcohol effects?
No. The overall recognition contrast was nonsignificant while the anger-specific comparison was significant. A nonsignificant aggregate does not establish exact equivalence or a person’s normal performance.
How did the group drinking design affect the interpretation?
Because all three members of a group received the same condition and interacted, partners’ behavior as well as participants’ own alcohol exposure could have affected their reported social impressions.
Can the paper determine whether a particular witness was accurate or credible?
No. It reports research-task responses in a selected sample, not case-specific emotion truth, memory accuracy, or witness credibility. Those questions require the actual record and properly matched evidence.
Primary scientific source and interpretation boundary
Kumar L, Creswell KG, Brown KW, Lyons G, Feeney BC. Alcohol’s acute effects on emotion recognition and empathy in heavy-drinking young adults. Scientific Reports. 2025;15:13807. https://doi.org/10.1038/s41598-025-98153-2.
The study offers group-level evidence about a specific anger-recognition task and a distinct self-reported interaction-related empathy measure. It is not a blood-alcohol assay validation, a personal intoxication classifier, a demonstration of aggression, or a method for deciding the truth of a contested encounter. The useful forensic conclusion is limited to what the investigators measured, how they analyzed it, and where those measurements end.