Does a Stronger Amphetamine Response Predict Heavy Drinking? What 192 Adults Actually Showed

Answer: In 192 selected young adults who received one controlled 10-mg oral dose of d-amphetamine, the relationship between a negative-emotionality personality score and participants’ previously reported alcohol consumption depended statistically on how stimulated they said they felt after the drug. In the 160 current drinkers included in three quantitative regression models, the interaction contributed modest additional explanatory variance for average drinking, recent binge episodes and recalled intoxication frequency. Neither negative emotionality nor subjective stimulation alone showed the same independent statistical contribution in those models. The investigation did not administer alcohol, measure an individual’s actual postdose amphetamine concentration, predict future alcohol-use disorder, test driving, or establish impairment in a legal case.

The underlying complete empirical source is Allen and Gabbay’s study, The amphetamine response moderates the relationship between negative emotionality and alcohol use, published in Alcoholism: Clinical and Experimental Research 37:348–360 (2013), original research DOI 10.1111/j.1530-0277.2012.01935.x. The central issue is an interaction among reported feelings, a personality scale and retrospective drinking measures—not a chemical test for alcohol or amphetamine.

What question did the experiment actually answer?

Investigators asked whether sensitivity to a stimulant experience changes the association between a person’s tendency toward negative emotionality and their reported history of alcohol use. Negative emotionality is a multidimensional personality construct. It is not a diagnosis of intoxication, an observed aggressive act or a laboratory concentration. Similarly, feeling energized after a pill is a subjective effect, not proof that a particular molecule reached a specific concentration in legally collected blood.

The word moderates has a specific statistical meaning: the estimated relationship between one variable and an outcome changes across levels of another variable. It does not mean that the investigators experimentally manipulated personality, caused the participants to drink more alcohol, or discovered a validated cutoff identifying future alcohol problems. In this experiment, they measured the response to one fixed drug dose and then related that response to reported earlier behavior. The temporal sequence matters before applying the result to any individual.

Who was studied, and which sample size belongs to each result?

The full study included 192 different volunteers, 99 women and 93 men, between 18 and 25 years old. Their average age was 20.5 years, with a standard deviation of 1.9 years. They were recruited around Washington, DC, for a wider research project that also involved event-related potentials; the present analysis used data from a distinct laboratory session occurring at least two days before the separate electrophysiology sessions. There were 192 independently enrolled people, not hundreds of subjects generated by questionnaire questions or repeated postdose ratings.

Screening selected relatively healthy young adults. Medical contraindications, several psychiatric diagnoses and potentially interacting prescriptions led to exclusion. The investigators screened for substance exposure and explicitly excluded positive urine drug screens at the experimental appointment. These exclusions constrain generalization to older adults, chronically prescribed stimulant patients, individuals with alcohol dependence, people using multiple drugs, or defendants undergoing forensic assessment.

Analytic denominators become smaller for particular outcomes. The binary analysis of whether someone had ever consumed alcohol used the full 192. The corresponding ever-used-illicit-substance analysis used 191 because one respondent lacked that information. Three quantitative drinking outcomes used only 160 current drinkers, meaning participants reporting drinking during the preceding month. An onset-age model included 144 people who reported regular alcohol use. A statistically significant result in 160 current drinkers must not be described as an estimate calculated on all 192 without qualification.

What drug, route and actual dose were used?

Each participant swallowed a capsule containing 10 milligrams of d-amphetamine with water during a supervised session of approximately four and one-half hours. It was a single oral administration, not an intranasal methamphetamine challenge, an injectable dextroamphetamine protocol, a mixed-amphetamine-salts dosing schedule, or conversion from the lisdexamfetamine prodrug. The report did not randomize these volunteers between several doses or include a matched placebo session in this particular experiment.

Before dosing, staff obtained an alcohol breath test and a urine drug screen. The breath result was required to be 0.00%, and no participant in the analyzed cohort had a positive breath alcohol screening result. Among 206 individuals screened, 14 were excluded for positive drug screens. These entry tests helped characterize immediate eligibility; they were not quantitative post-amphetamine drug concentration studies. Historical self-reported alcohol consumption was recorded separately through a questionnaire, not inferred from the breath screening result.

The authors discussed an external estimate of what blood amphetamine concentrations may follow a 10-mg dose, but that contextual estimate was not a measurement in these 192 participants. It cannot be inserted into a results table as an observed individual or cohort amphetamine concentration. No blood amphetamine concentration-time series, measured peak concentration or validated ratio between self-rated stimulation and chemical exposure was reported for this experiment.

How was the stimulant response measured?

Investigators used the seven-item stimulant portion of the Biphasic Alcohol Effects Scale, or BAES, asking participants about feelings such as energy, excitement, stimulation and vigor. Although the questionnaire had been developed for alcohol-related subjective effects, the study expressly applied its stimulant scale to the response to d-amphetamine. The BAES was administered before the drug and at four postdose intervals, approximately 0.5, 1.5, 2.5 and 3.5 hours. No alcoholic beverage was administered during that sequence.

The central variable, termed Stim, was the change in the self-rated BAES stimulant score from before drug administration to 1.5 hours afterward. This was the group’s mean peak rating time, not an analytically measured pharmacokinetic time to maximum plasma drug concentration. The mean change across volunteers was +2.1 points, with a substantial 14.0-point standard deviation. Individual changes extended from −47 to +58. Some participants reported less stimulation than at baseline, emphasizing that a prescribed dose and a feeling are not interchangeable.

For an illustrative graph, the investigators separated participants into three response tertiles, approximately 63 high, 64 intermediate and 65 lower responders. Their main regression analyses, however, used continuous stimulation values. Turning those broad tertiles into fixed diagnostic categories would misrepresent the method: no classification accuracy, sensitivity, specificity, or forensic threshold was validated for a named person.

What was meant by negative emotionality?

The study used the Multidimensional Personality Questionnaire to measure negative emotionality, a construct informed by scales addressing stress reactivity, alienation and aggression-related tendencies. Such questionnaire scores describe reported patterns of disposition, not observed criminal intent, current alcohol impairment, aggressive behavior at an accident scene, or a unique biological genotype. Personality was not experimentally assigned or altered by the study protocol.

The investigators considered a model in which people with greater negative emotionality might differ in drinking behavior depending on sensitivity to stimulant experiences. This is an interaction hypothesis, not a laboratory demonstration that negative feelings caused an alcohol exposure. The authors acknowledged alternative explanations, including traits not measured in the regressions—such as novelty seeking, reward sensitivity or other correlated behaviors—and the limitations inherent in recalling prior alcohol use.

How were the actual alcohol variables obtained?

Participants completed shortened questions from a Department of Defense health-behaviors survey. The researchers computed average alcohol consumption in ounces of ethanol per day, a recalled frequency of days with five or more drinks in the preceding 30 days, and the number of occasions in the previous year on which a person recalled drinking enough to feel drunk. They also asked when drinking at least once a month began and whether the participant had ever consumed alcohol or an illicit substance.

These are not equivalent endpoints. Average daily intake summarizes reported beverage histories and stated volumes and can be affected by estimation of beer, wine and spirits. A binge-day count is a behavioral frequency over 30 days, not a test of peak blood alcohol. Feeling drunk over the prior year is the individual’s recalled subjective experience, not a medically observed impairment episode. Age at first monthly drinking is a distant retrospective milestone, not a documented pharmacological effect of the subsequent research dose.

The full-sample descriptive means were 0.5 ounces of ethanol per day (SD 0.7), 1.5 binge episodes in the preceding month (SD 2.6), and 18.4 recalled intoxication occasions in the preceding year (SD 32.2). Among the full sample, 83.3% reported ever consuming alcohol. These descriptive observations can frame the analyzed population; they cannot be translated into measured contemporaneous breath or venous whole-blood ethanol for any participant.

What were the main interaction results in current drinkers?

The researchers fitted hierarchical regression models that first entered sex, then the self-rated stimulant response, then negative emotionality, and finally their interaction. The quantitative outcomes were positively skewed, so the authors used transformed versions and a 2,000-iteration bootstrap for specified significance estimates. The relevant models each included 160 current drinkers. Changes in R² indicate the additional proportion of sample outcome variation associated with entering another term; they are not percentages of drinking or causal percentages of drug effects.

Thus, the interaction added approximately 3.4 percentage points to the sample’s explained variance for average daily consumption and recent binge episodes, and 2.4 points for recalled intoxication occasions. The entire final models explained about 10.1%, 10.6%, and 5.9%, respectively. Most individual variation remained outside those particular models, and the estimations do not isolate unique effects of the experimental capsule from preexisting behavior or unmeasured background factors.

The interaction evidence also coexisted with smaller or null main effects. The stimulation score alone did not significantly improve these models when entered at its stage; neither did negative emotionality alone. Reporting only the interaction p-values could create the incorrect impression that high negative-emotionality scores independently predicted heavy drinking for everybody, or that strong feelings after the drug invariably accompanied more drinking. Both claims exceed the actual conditional model.

Which results were genuinely nonsignificant?

Three key alcohol-consumption models produced an interaction, but the individual predictor results were not uniformly positive. For average daily consumption, entry of Stim alone had p = 0.522, and negative emotionality had p = 0.070. For recent binge episodes, those p-values were 0.756 and 0.332. For annual self-reported intoxication frequency, they were 0.541 and 0.284. These tests evaluate specified incremental model terms; a nonsignificant result is neither proof of absolutely no effect nor support for a hidden significant one.

At the full-sample level, whether a participant had ever consumed alcohol was not significantly predicted by the complete logistic model (p = 0.370). The stimulant-response term initially approached, but did not reach, the conventional threshold (p = 0.051). The Stim × negative-emotionality addition to this binary ever-used model was nonsignificant (p = 0.733). The historical distinction between drinking at all and how frequently current drinkers reported drinking is essential.

The age of onset of regular monthly drinking was analyzed among 144 people. An intermediate predictor step included a reported ΔR² = 0.049 for stimulation, but the model at that stage had overall p = 0.062 and the bootstrapped coefficient p = 0.158. The final four-predictor model was not significant (p = 0.163), and the interaction itself added virtually no explained variance (p = 0.890). This is not a prospective finding that amphetamine response causes drinking to start earlier.

These results are compatible with an association conditioned on another measured trait, not with a universal statement that all people with stronger stimulant feelings consume more alcohol. They also show why differing regression denominators and tests cannot be pooled into one overconfident verdict about an identifiable individual.

What did the Johnson–Neyman calculations add?

The investigators used a Johnson–Neyman method to locate regions along the observed stimulation-response scale in which the regression slope relating negative emotionality to a drinking outcome crossed the specified statistical-significance criterion. For daily ethanol intake, the calculated stimulus-response boundary was 2.5 score points; for binge frequency it was 10.6; for reported annual intoxication frequency it was 12.8. Each is expressed on the laboratory questionnaire-change scale, not in mg/L, ng/mL, blood alcohol percentage or elapsed time since drug ingestion.

Those three numbers should not be taken as clinical diagnostic thresholds. Their sizes depend on the sampled population, score distributions, covariates, scale construction, bootstrapping and model specification. A fitted significance boundary also does not measure sensitivity or specificity for detecting alcohol misuse. Applying one as a personal drinking-risk or legal-impairment cutoff would require prospective validation and an explicitly defined target outcome, neither of which was performed in this study.

Why is the absence of a placebo session important?

Every participant in this analysis received the active oral capsule during the subjective-response session. The authors expressly chose not to include a placebo session based on exploratory work from a different earlier sample. Therefore, the baseline-to-postdose difference includes not just any pharmacological response but also possible changes attributable to time, setting, anticipation, habituation, food intake, fatigue and other factors that a concurrently matched placebo might have helped disentangle.

The participant was told the capsule might contain one of several kinds of medication or placebo, which could reduce simple certainty about its contents; that procedural blinding of expectation does not itself create an experimental placebo group. The result should be described as an association involving post-amphetamine ratings minus baseline ratings, not as a placebo-corrected causal effect of amphetamine on mood or drinking. It certainly is not evidence that administering amphetamine changed a person’s past drinking, which by definition occurred earlier.

Were sex, personality and other confounders considered?

Sex was entered first into each hierarchical regression because prior drinking studies show meaningful differences. The main results retained sex as a covariate before evaluating stimulation and negative emotionality. Covariate adjustment can reduce association with the included variable; it cannot eliminate confounding from variables absent from the model, inaccurate historical reports, selection into the study, or differences between eligible and excluded participants.

The original paper discusses other plausible traits, including reward sensitivity, novelty seeking and impulsivity, that might covary with both the stimulant experience and drinking behavior. They were not comprehensively measured and separately adjudicated in the reported interaction models. In addition, negative emotionality came from a questionnaire rather than direct observation, while alcohol consumption came from retrospective answers. Treating two self-report constructs as independently validated causal biological mechanisms would exceed their measurement properties.

The single-dose young, relatively healthy sample also lacks representation of many patients encountered in clinical forensic medicine: individuals with substantial alcohol dependence, acute polysubstance toxicity, advanced age, altered metabolism or significant psychiatric illness. The study cannot supply a general medical dosing recommendation and does not establish safety or effective alcohol-use treatment.

What did the illicit-drug-use comparison establish?

For 191 participants with complete historic illicit-drug-use information, including stimulation improved the binary logistic model of whether they reported ever having tried an illicit drug: the incremental χ² was 5.38, with p = 0.020. The combined sex-plus-stimulation model had χ² = 7.72 and p = 0.028. The authors stated that adding the other proposed predictors did not improve the model meaningfully.

This endpoint encompassed various substances rather than chemically verified current stimulant exposure. The authors noted that more than four-fifths of respondents reporting an illicit drug had reported marijuana only. A past affirmative survey answer does not establish when a specific substance was last taken, whether it remained detectable in blood, or whether a toxicological result was positive during an incident. The association was not a prospective diagnosis of addiction and should not be turned into an estimate of what another person will use.

What specimens were collected and what was chemically tested?

The screening visit included a venous blood sample for general medical screening; the original paper did not report a quantitative drug-assay result from that blood. At the active dosing visit, breath was checked for alcohol at zero as an eligibility condition, and urine received a qualitative screen for selected drugs of abuse. A qualifying breath result and preliminary urine screen do not supply the postdose pharmacokinetics of amphetamine or document historical blood ethanol concentrations.

Critically, the published experiment contains no analytically assayed human venous whole-blood d-amphetamine concentration. Nor does it present quantitative parent amphetamine or metabolite measurements from centrifuged plasma or serum, urine, oral fluid, hair, sweat, vitreous humor, milk or exhaled breath. No experiment-specific amphetamine LC–MS/MS or GC–MS calibration, reporting limit, enantiomer-specific quantitation or validated forensic drug cutoff appears in its results.

Venous whole blood, separated plasma and serum must not be relabeled as equivalent matrices. The numerical questionnaire value +2.1 ± 14.0 is measured in subjective scale points—not ng/mL, milligrams administered, or percent BAC. The 0.00% prestudy breath screening result is not a calculated daily alcohol dose or direct chemical confirmation of what the person drank during the prior month. Proper application of a forensic laboratory result requires knowing the actual collected specimen, tested analyte, calibration, timing and acceptance criteria.

Why can a relationship in a group not classify an individual?

Regression estimates describe patterns across the observed sample under a specified statistical model. People with identical personality scores can have different drinking histories, and a similar BAES change can accompany very different reported behavior. Even the largest of the three final quantitative model R² values was 0.106, leaving most observed outcome variation unexplained by the included variables. A nominally significant group interaction is not an individualized confidence statement and does not authorize assigning a particular person’s behavior to an alleged acute drug effect.

The original design neither prospectively followed participants for incident alcohol-use disorder nor validated a score-to-risk probability model outside its recruited cohort. It did not supply an external test sample, case-specific toxicology, sensitivity/specificity for intoxication, a last-use time estimator, or a measure of on-road vehicle control. Personal forensic conclusions would require relevant individual records, validated tests and a coherent factual timeline, not an ecological translation of statistical coefficients.

Because many outcomes, predictor steps, associations and exploratory regions of significance were investigated, the possibility of false-positive findings across multiple analyses must also be considered. Bootstrapping evaluates the stability of selected coefficients under resampling assumptions but is not automatically equivalent to a familywise correction across all evaluated outcomes and analyses. Source interpretations should preserve both significant and nonsignificant endpoints without pretending that small p-values alone demonstrate a broadly replicable causal mechanism.

What would a defensible follow-up study require?

A prospective design could measure baseline negative emotionality, establish standardized placebo-controlled and active dose conditions, assay d-amphetamine in specifically identified matrices at timed intervals, and record drinking behaviors over later weeks or months through diaries and, where appropriate, independently observed outcomes. Such a program could test whether results generalize beyond young healthy adults and whether any score adds meaningful predictive value over existing risk factors. This is a research proposal, not an additional experiment completed by Allen and Gabbay.

For forensic work, a separate study would have to address an independently defined legal or functional endpoint—such as validated driving behavior under known conditions—rather than assuming that questionnaire responsiveness predicts impairment. The original work offers a hypothesis about correlated psychological traits and retrospective drinking. Its significance lies in motivating careful follow-up, not in replacing case-specific analytical and behavioral evidence.

Seven frequently asked questions

1. Did the researchers give alcohol and amphetamine together?

No. Participants received one oral 10-mg d-amphetamine capsule at the experimental session. Alcohol use was reported retrospectively. Breath alcohol at the session was screened as zero. The study did not experimentally dose ethanol with amphetamine or measure a combined acute drug interaction.

2. Were 192 people included in every alcohol model?

No. There were 192 independent adults in the overall cohort, 191 in the ever-illicit-use analysis, 160 current drinkers in the three quantitative drinking models, and 144 people in the regular-drinking-onset analysis. The subset denominators must remain attached to their own results.

3. Does a strong subjective response alone mean more binge drinking?

No. In the reported n160 binge model, stimulation alone added ΔR² = 0.001 with p = 0.756, while its interaction with negative emotionality added ΔR² = 0.034 with a step-test p = 0.017. Conditional association does not establish causation or reliably classify a particular person’s drinking.

4. Was the result based on an amphetamine blood concentration?

No. A questionnaire supplied the stimulant-response change score. The paper reported neither postdose amphetamine concentration in legally collected venous whole blood nor a quantitative plasma, serum or oral-fluid drug assay. Oral capsule milligrams and subjective scale points are not blood-drug values.

5. Did the researchers find a significant interaction for whether someone had ever drunk alcohol?

No. The final full-sample logistic model had p = 0.370, and the stimulation-by-negative-emotionality interaction for ever drinking had p = 0.733. More detailed patterns appeared only for specific historical drinking-frequency measures in current drinkers.

6. Are 2.5, 10.6 and 12.8 legal cutoffs for alcohol impairment?

No. They are model-dependent Johnson–Neyman boundaries on a subjective stimulation-change scale used to describe statistical regions in this sample. They are not breath or blood alcohol percentages, legal standards, validated clinical risk classifications or instructions about dosing.

7. Can the study prove why a defendant drank, drove, acted aggressively or caused a crash?

No. It measured retrospective drinking reports, a personality inventory and acute self-rated stimulation in young volunteers. It did not examine a specific defendant, validate driving or intent, measure postdose whole-blood drug concentrations, or establish causation in any disputed legal incident.

Primary source and live same-site laboratory context

The scientific basis for this interpretation is Allen KJD and Gabbay FH, The amphetamine response moderates the relationship between negative emotionality and alcohol use, Alcoholism: Clinical and Experimental Research 2013;37(2):348–360, full original empirical source DOI 10.1111/j.1530-0277.2012.01935.x. The separate original research article should be consulted for exact regression footnotes, transformed outcomes and methods when presenting any scientific conclusion.

For the distinction between a drug-exposure history and analytical evidence, consult the site’s published background on biological specimen matrices and specific confirmatory drug testing. Neither guide supplies an original missing amphetamine concentration for this experiment. Their relevance is methodological: a psychometric response, a breath alcohol eligibility screen and an assay result must not be substituted for one another.