EVIDENCE LIBRARY
Does 20 mg Dextroamphetamine Reduce Post-Error Slowing Without Improving Accuracy? What 55 Volunteers Showed
An original 55-person N-back analysis found 20 mg d-amphetamine reduced post-error reaction time without significantly improving accuracy or measuring blood drug levels.
Short answer: In a counterbalanced, double-blind experiment, 20 mg of oral d-amphetamine reduced the customary delay in responding immediately after a mistake on a difficult working-memory task. The change arose from faster correct responses following errors, not a statistically established improvement in overall three-back accuracy or responses following correct trials. Although the broader accuracy analysis included 110 compliant participants, the post-error analysis concerned only 55 volunteers who had enough qualifying responses, including a small number of imputed intermediate-dose sessions. The experiment measured computer-task behavior, not a forensic drug concentration in venous whole blood or an individual’s ability to drive, recognize hazards, or recount an actual event.
This article examines Margaret C. Wardle, Amy Yang, and Harriet de Wit, Effect of d-amphetamine on post-error slowing in healthy volunteers, Psychopharmacology 220(1) (2012), beginning at page 109, DOI: 10.1007/s00213-011-2462-6. The analysis separates the drug assigned, the task observed, the genuinely statistically supported effects, and the forensic questions that the primary study did not test.
What is post-error slowing, and why does it matter?
Post-error slowing describes a familiar laboratory pattern: after giving an incorrect answer, a participant often responds more slowly on the next occasion than after a preceding correct answer. Researchers study this as one observable part of performance adjustment. Slowing can be interpreted as greater caution, temporary reorientation, disruption caused by the error, or some combination. It is not a direct measurement of whether an individual has consciously detected the mistake or will make a better choice next time. The meaning depends on instructions, task difficulty, feedback, and the surrounding pattern of accuracy.
In litigation or professional fitness discussions, the tempting shortcut is to treat faster responding as necessarily superior function. That shortcut is unsound. A person may respond quickly but inaccurately, deliberately slow to avoid future errors, or change response strategy without obtaining a better outcome. The Wardle experiment was well suited to isolating a narrow behavioral signal under controlled conditions. It was not constructed to certify reliable human judgment across different situations.
How many independent humans were actually analyzed?
The recruitment and data-processing sequence is essential. A total of 198 volunteers participated; 177 had complete N-back data across sessions. Investigators then identified 67 whose response patterns showed that they were not continuously applying the task instruction. Excluding those records left 110 people in the reported three-back accuracy analysis. The more specialized post-error analysis required a valid next response after both an incorrect and a correct preceding response. Fifty-eight participants met the core placebo and 20-mg requirements. Missing one intermediate-dose session could be estimated under a specified within-person linear interpolation rule, producing a final 55 independent people for the primary post-error slowing analysis.
These figures do not describe 198 independent successful post-error observations, 110 people in every final statistical model, or 220 different drug recipients. The four dosing visits involved repeated observations on the same volunteers. Likewise, hundreds of individual number presentations and repeated reaction times are not independent human subjects. The exclusions and imputation narrow generalizability and deserve attention whenever the result is offered as broad evidence about healthy adults.
Which people were eligible for the experiment?
Participants were healthy English-speaking men and women aged 18–35, with at least high-school education and a specified body-mass-index range. The researchers recruited Caucasian volunteers because their broader project included genetic analyses. They screened against recent Axis I psychiatric diagnoses, substance dependence, significant problems related to drug use, medical contraindications to stimulant administration, and relatively high baseline nicotine or caffeine exposure. Women not using oral contraceptives were scheduled during the follicular phase. These restrictions yielded a relatively selected population, not a representative cross-section of adults with medical illnesses, sleep loss, stimulant tolerance, other medications, or substance-use disorders.
Source Table 1 describes the 110 task-compliant people as 55 men and 55 women, mean age 22.97 years with standard deviation 2.99 years. The publication stated that the smaller post-error group did not significantly differ from excluded task-compliant participants on measured demographics. A null demographic comparison does not establish equivalence across unmeasured traits such as impulsivity, prior stimulant experience, or error-response strategies.
What drug and dose were administered?
The study administered d-amphetamine, also called dextroamphetamine, in oral capsules. Each person received placebo, 5 mg, 10 mg, and 20 mg conditions in a double-blind, counterbalanced order, at sessions separated by at least 72 hours. Placebo capsules contained dextrose. The protocol did not administer methamphetamine, a mixed amphetamine-salt product, lisdexamfetamine, or ethanol as experimental agents. Those distinctions matter scientifically because different drugs and formulations cannot be assumed to produce interchangeable concentration–time curves or identical subjective and behavioral effects.
Participants swallowed the assigned capsule at approximately 9:30 a.m. and performed the computerized N-back task at about 11 a.m., roughly 90 minutes later. The investigators chose the interval partly because earlier pharmacokinetic research suggested a near-peak exposure period. They did not chemically measure each participant’s actual plasma peak or report a matching whole-blood dextroamphetamine concentration. An assigned 20-mg oral dose is an administered amount, not a finding of 20 ng/mL in a biological specimen.
What did participants actually do during the N-back task?
The task displayed numbers one through four in changing screen positions. In a zero-back trial the participant responded to the currently displayed number. In a one-, two-, or three-back trial, the correct response depended on the number observed one, two, or three positions earlier. This placed progressively greater demands on working memory while maintaining a straightforward keyboard response. After an attempted response, the screen displayed immediate correct or incorrect feedback; time-outs were treated differently and provided no such feedback.
The protocol included 120 trials across the task, with shorter blocks of harder and easier conditions. The easiest portions produced too few errors for reliable post-error comparisons: the source reports only 16 people with any zero-back errors, two with any one-back errors, and 26 with any two-back errors. Therefore, the central analysis concerned the three-back condition. The result cannot be assumed to occur on every task, at every difficulty level, or when no immediate feedback follows a mistake.
How was a post-error response distinguished from an ordinary response?
Researchers compared the reaction time for a correct response after an error against the reaction time for a correct response after a correct response. The distinction isolates what happens on the immediate next trial, not whether a participant eventually learns a new rule or performs better for the remainder of a test. Trials with missing responses also affect how many qualifying pairs can be analyzed. That is why the sample for this outcome was substantially smaller than the sample retained for ordinary accuracy.
A measured millisecond difference here is a property of task performance. It is not a direct observation of a neurological transmitter, a clinical diagnosis, or a measure of legal culpability. Even when a drug reduces the measured delay, further evidence would be necessary to decide whether the participant became less cautious, more confident, less distracted by an error, or simply faster in a particular response class.
Evidence table 1: denominators, exposure, and task design
| Original study element | Reported quantity | What it does and does not count |
|---|---|---|
| Participants initially tested | 198 people | Not the final post-error model sample |
| Complete N-back sessions | 177 people | Before task-compliance exclusions |
| Excluded for incorrect task strategy | 67 people | Incorrect strategy, not a positive drug test |
| Task-compliant accuracy sample | 110 people; 55 male, 55 female; mean age 22.97 ± 2.99 years | Three-back overall accuracy analysis |
| Primary post-error slowing sample | 55 people | Repeated observations and limited intermediate-dose interpolation |
| Within-person oral conditions | 0, 5, 10, 20 mg d-amphetamine; ≥72 hours apart | Four visits are not four biologically different people |
| Time from dosing to task | Approximately 90 minutes | Scheduled assessment, not measured individual drug peak |
Was there genuine post-error slowing under placebo?
Yes. In the placebo session, responses after an error were significantly slower than responses after a correct answer: the source reports a paired comparison of t(54) = −2.26, p = .03. This baseline comparison matters because an intervention cannot be said to diminish post-error slowing unless slowing was actually present under the reference condition. Investigators also reported that greater three-back accuracy under placebo was associated with more post-error slowing, r = .26, p = .05.
Neither result shows that people who slow the most are always more accurate in daily activities. A correlation within a selected experimental sample is not a universal predictive standard. The baseline findings primarily establish the setting in which the later drug contrasts are interpretable, including the possibility that participants differ materially in how they react to an error.
Did dextroamphetamine improve three-back accuracy?
No statistically significant improvement in three-back accuracy was detected. The repeated-measures analysis reported F(3,327) = 0.48, p = .70, with reported effect-size η² = .004. This genuine null result should accompany any claim about reduced post-error slowing. It means that the source did not establish a group-level gain in the percentage of correct responses across the four administered conditions, even though one component of response timing changed.
It is equally important not to overstate the null as proof that amphetamine can never improve working memory under other conditions. A nonsignificant result in this particular selected task, dose scheme, sample and statistical analysis bounds what the paper demonstrated. It is not a universal demonstration of no effect. The careful description is that reduced error-associated response delay did not coincide with a demonstrated increase in the accuracy endpoint measured here.
Which reaction times changed with treatment?
The central analysis found an interaction between drug condition and the kind of preceding response, F(3,162) = 3.06, p = .03, η² = .05. Investigators then examined the two response categories separately. Post-error reaction times changed across treatment conditions, F(3,162) = 6.41, p < .001, η² = .11; post-correct reaction times did not demonstrate a significant drug effect. The reported figure identified the 20-mg condition as different from placebo in the post-error comparison under the figure’s conventional threshold.
This selective pattern is the paper’s defining result. Calling it generalized motor speeding would overlook the absence of a supported post-correct effect. Calling it restored cognition would overlook the accuracy null. Conversely, calling it evidence of impaired error awareness would also outrun the measurements: conscious awareness, electroencephalographic error signals, and practical recovery after real mistakes were not measured as validated outcomes of this experiment.
Evidence table 2: significant findings alongside genuine null outcomes
| Endpoint or comparison | Original reported statistic | Narrow interpretation |
|---|---|---|
| Post-error slowing during placebo | t(54) = −2.26; p = .03 | Baseline slowing present in the 55-person group |
| Three-back accuracy, drug across conditions | F(3,327) = 0.48; p = .70 | Genuine null on correct-response accuracy |
| Drug × preceding-response category | F(3,162) = 3.06; p = .03 | Drug relationship differs by previous correct/error status |
| Post-error reaction times across doses | F(3,162) = 6.41; p < .001 | Significant response-timing effect |
| Post-correct reaction times across doses | Reported nonsignificant; exact p not given | Do not invent an exact null-test value |
| 20-mg-induced elation over placebo | t(53) = 3.60; p = .001 | Subjective positive mood changed |
| Drug × category after elation adjustment | F(3,156) = 2.68; p = .05 | Effect remained at a boundary after modeled adjustment |
| Placebo constraint vs post-error slowing | r = −.45; p = .001 | Exploratory baseline association, not drug-response prediction |
Why is an increase in drug-induced elation not a complete explanation?
As expected for an active stimulant dose, 20 mg dextroamphetamine increased self-reported elation relative to placebo. The original paired comparison was t(53) = 3.60, p = .001. Investigators therefore asked whether the behavioral effect could be accounted for simply by an uplift in reported mood. They included the difference in peak elation changes as a covariate. The drug-by-response-category term remained at approximately the conventional decision boundary, F(3,156) = 2.68, p = .05.
This does not prove an independently isolated neural mechanism. Covariate adjustment in a relatively small selected sample cannot rule out every behavioral, psychological, or physiological contributor. Nor does it show that elation is irrelevant to every stimulant response. The appropriately limited conclusion is that the measured mood change did not fully account for the particular response-timing pattern under the authors’ specified model.
Did personality identify who would respond to the drug?
Researchers examined three Multidimensional Personality Questionnaire factors: positive emotionality, negative emotionality, and constraint. Under placebo, higher constraint was associated with less post-error slowing, r = −.45, p = .001, after excluding two participants with incomplete personality-related data. That is an association between a questionnaire trait and a baseline laboratory response. Constraint was not an assayed biomarker and was not presented as an operational medical test.
Critically, neither these personality factors nor initial accuracy significantly moderated the drug-induced reduction in post-error slowing in the exploratory analyses reported. It would therefore be misleading to claim that high or low constraint gives a validated individual dosing forecast. Similarly, the absence of demonstrated moderation in a subset does not prove uniform treatment response. Predicting a particular individual’s functional change would require a prospective, independently validated model with precision and calibration assessments.
How did the investigators handle missing intermediate-dose information?
Post-error analysis required enough correct responses immediately after both preceding event categories on the placebo and 20-mg days. Even among otherwise eligible volunteers, a few lacked one intermediate 5- or 10-mg measurement. For these selected instances, the authors interpolated a missing dose response using the two adjacent within-person dose values, according to a preliminary linear dose–response rule. Participants missing more than one session or either endpoint condition were excluded from that analysis.
Imputation does not manufacture independent participants, but it can influence the estimated shape of a dose-response relationship and the resulting inferential model. The study did not supply an independent complete-case replication of every reported contrast. Thus, the cleanest statement remains the observed selective 20-mg-versus-placebo difference and the documented full analysis method, rather than a claim that every individual followed an exact linear response to 5, 10, and 20 mg.
Why do the statistical denominators need careful explanation?
Statistical notation such as F(3,327) or F(3,162) might be mistaken for 327 or 162 independent people. Those numbers are residual or denominator degrees of freedom arising from repeated-measures analyses. They cannot replace the biological participant counts of 110 for accuracy and 55 for post-error slowing. Likewise, p < .001 is evidence against a model-specific null, not an assignment of near-certainty that a clinical or legal proposition about an individual is true.
Two task features further matter: the three-back block was chosen partly because easier conditions generated insufficient errors, and the result was analyzed after excluding many people whose task behavior did not meet the intended instructions. This selection is methodologically understandable but imposes real scope limits. The source was not a prospective population-based performance validation in all adults who use amphetamine.
What biological matrix did the researchers chemically test?
Participants provided urine and breath specimens at session entry for abstinence screening, and blood pressure and pulse were repeatedly measured. These procedures do not amount to validated quantitative measurement of d-amphetamine in human venous whole blood, centrifuged plasma, separated serum, or any other matrix during the task. The publication contains no original participant-by-participant LC–MS/MS or GC–MS parent amphetamine concentration series, no LOD or LOQ for such an assay, and no calibration curve tying keyboard performance to ng/mL drug concentrations.
That absence matters for forensic interpretation. Human whole blood includes red cells and other cellular components, whereas plasma and serum are separated compartments. Even a separately published therapeutic plasma concentration cannot automatically be substituted for a particular person’s venous whole-blood level. This principle is treated more broadly in the site’s independently published biological-matrices guide and confirmatory-drug-testing guide. Neither guide converts this experiment into an analytical toxicology study.
Can these results establish an individual driver’s impairment?
No. The investigators did not put participants behind the wheel, conduct a validated driving simulator, measure lane keeping, record near-collisions, examine divided attention to road hazards, or develop sensitivity and specificity against an independently adjudicated driving-impairment criterion. The stimulus was a numbered screen and the action a keyboard keypress. A statistically supported shortening of one response category in a controlled task can coexist with unmeasured adverse effects in other performance domains.
Nothing in the article sets a legal per se concentration, allows an exact last-use calculation, or demonstrates that someone who took a dose was safe or unsafe at a later incident. A case-specific conclusion would need actual specimen findings and their matrix validation, dosing chronology and corroboration, clinical and functional observations, co-exposures, and the circumstances under investigation. The relevant question is not whether the pharmacological intervention changed a group mean but whether independently verified evidence supports the particular inference being offered.
What does this study mean for assessing witnesses and reported mistakes?
Post-error slowing is not eyewitness memory accuracy, truthfulness, suggestibility, or the correctness of an incident narrative. A shorter delay after feedback on a number task does not establish that a person would remember a conversation accurately, acknowledge a mistake, resist social pressure, or provide reliable testimony. This experiment had no interviews about real events, delayed recall assessments, legal witness populations, or externally verified truth criteria.
The distinction also runs in the opposite direction. It would be unsound to use decreased slowing as proof of dishonest behavior or deficient remorse. Participants were not instructed to conceal errors, and the data cannot identify their motives. The result addresses the speed of a specific next response under immediate feedback—not how they would behave when accuracy, incentives, uncertainty, and consequences differed.
How should an expert present the positive and negative results together?
Start with the actual research question and unit of measurement: oral d-amphetamine dosing within people and post-error reaction times on a difficult three-back task. Then distinguish the full N-back accuracy sample from the restricted post-error sample, including the exclusions and interpolation procedure. Present the significant post-error treatment comparison adjacent to the genuine nonsignificant accuracy result and the reported nonsignificant post-correct response effect. Do not convert the statistical result into a claim of generally enhanced decision-making or clinically important impairment.
Next address concurrent mood and personality measurements without elevating exploratory associations into clinical diagnostics. Finally, identify important absent evidence: original quantitative whole-blood pharmacology, direct EEG error responses, real driving or occupational outcomes, witnessed historical events, and prospective individualized classification. This approach preserves the experiment’s genuine contribution while avoiding an inference that the investigators did not validate.
Frequently asked scientific questions
1. Did everyone in the experiment receive 20 mg?
No. The crossover design exposed the same participants to placebo, 5, 10, and 20 mg oral d-amphetamine on separate sessions. The 20-mg result was a comparison of within-person conditions, not a separate 55-person treatment arm.
2. Why were only 55 people included when the abstract reports 110?
The 110 people were retained for task-compliant accuracy analyses. Post-error slowing required qualifying correct responses after both preceding event types under relevant dose conditions. More stringent event-count requirements and limited permitted intermediate-dose interpolation yielded 55 independent people.
3. Did taking dextroamphetamine improve correct working-memory responses?
The original three-back accuracy comparison was nonsignificant, p = .70. The supported finding was a change in reaction time after errors, not a statistically established increase in the percentage of correct responses.
4. Does faster post-error performance prove a person noticed the mistake?
No. The task provided immediate feedback, and response latency changed, but awareness, physiological error monitoring, and deliberate strategy were not independently validated as individual mechanisms. More than one explanation may fit the observed behavior.
5. Was the 20-mg dose measured in blood?
No. It was an assigned oral dose. This original experiment did not publish chemically confirmed postdose concentrations of parent d-amphetamine in venous whole blood, plasma, or serum, and its eligibility urine screening did not quantify a task-time blood level.
6. Did stimulant-induced happiness explain the reduced slowing?
Not fully under the investigators’ covariate analysis. Elation increased significantly at 20 mg, but the drug-by-response-category term remained at approximately the conventional significance boundary when modeled with the mood change. That does not exclude every unmeasured mediator.
7. Can an expert use the result to decide whether a driver or witness was impaired?
No individualized classification was validated. The study did not measure actual driving, eyewitness veracity, an incident event, or a forensic blood-drug concentration. It can inform a bounded discussion of laboratory performance adjustment, not decide a person’s impairment or credibility.
Primary-study conclusion
The experiment supports a selective finding: 20 mg oral d-amphetamine reduced the usual post-error response delay in a 55-person subset of healthy volunteers under difficult computerized working-memory conditions, while the broader accuracy endpoint showed no significant improvement. The analysis of elation and personality did not supply a validated personal response predictor. Neither the milligram dose nor the laboratory reaction-time statistic substitutes for a quantitatively identified drug concentration or an actual event-specific performance assessment.
For primary documentation, consult the original Wardle, Yang, and de Wit article. Interpretation should stay with the specific sample, the 3-back error-feedback paradigm, the statistical distinctions between positive and null outcomes, and the unmeasured forensic endpoints.