Does Modafinil Slow Verbal Responses Without Improving Accuracy? What a Randomized Hayling Test Found

Short answer: In a randomized, double-blind study of healthy volunteers, a single oral 200 mg dose of modafinil was associated with slower verbal responses on the Hayling Sentence Completion Test, but not with fewer errors. Researchers enrolled 64 people, randomized them between modafinil and placebo, and excluded three non-native English speakers from the principal Hayling analysis. The resulting 61-person analysis showed an overall drug-group difference in response latency (p = .035) and no significant overall difference in error counts (p = .561). A slower answer, in this experiment, was not evidence of more accurate cognitive control.

This is a narrow experimental result about speaking words in response to sentence prompts. It is not a blood-drug analysis, a measurement of aviation vigilance, a driving study, a forensic impairment cutoff, or evidence that modafinil slows every kind of reaction. The original primary report is Mohamed and Lewis’s 2014 randomized trial, Modafinil Increases the Latency of Response in the Hayling Sentence Completion Test in Healthy Volunteers.

What question did the researchers actually test?

Modafinil is best known as a wakefulness-promoting medication. Claims that it might also make healthy people think more effectively raise a different question: does a change in performance reflect greater accuracy, greater alertness, a slower and more cautious response style, or a combination? The researchers focused on an executive-language task in which a participant had to supply a sentence-ending word promptly. Some prompts invited an obvious completion; other prompts required suppressing the obvious completion and producing an unrelated word instead.

That contrast is useful because the two instructions put different demands on verbal control. A respondent can take longer and still make the same number of mistakes. Conversely, a respondent could become faster without making better choices. The investigators therefore assessed elapsed response time and errors as separate endpoints, rather than treating a general impression of being “sharper” as proof of cognitive enhancement. Their prespecified interest was whether modafinil slowed verbal performance and whether such slowing brought any improvement in accuracy.

How many people were studied, and how were they assigned?

The report describes 64 screened healthy adults, comprising 31 men and 33 women, with an average age of approximately 25 years. Its abstract describes 32 people assigned to modafinil and 32 to placebo in a parallel-group experiment. The printed demographic table gives a different distribution when its sex counts are summed; that inconsistency should not be silently harmonized into a new invented allocation. The essential facts are that drug and placebo were between-person groups and that the Hayling comparison did not include all 64 randomized volunteers.

Three people whose first language was not English were excluded from the Hayling analysis because sentence comprehension and verbal retrieval were central to the measurement. The reported main task analyses have error degrees of freedom 59, consistent with 61 analyzed persons across two groups. This is not a 64-person confirmed task-completion sample, and the two Hayling sections are repeated observations from participants, not 122 independent people.

Investigators reported excluding people with relevant psychiatric or medical histories, contraindicating medication use, or substance-use concerns and requested avoidance of caffeine and alcohol for the preceding 12 hours. Such selection helps define the study population. It also prevents unrestricted generalization to sleep-deprived drivers, pilots, patients taking multiple medications, people with active psychiatric conditions, and individuals involved in a particular legal event.

What was administered, and when was the test performed?

Volunteers received a single swallowed 200 mg modafinil dose or a matching placebo. The experimenter and key testing clinicians were masked to assignment, while designated drug-administering staff knew allocations. The Hayling task was administered about two hours after dosing. The study also collected subjective mood ratings and cardiovascular observations during the session, but these are different endpoints with different measurement properties from the recorded verbal answers.

Two hours after an oral dose is a study clock time, not a directly measured peak concentration. The authors did not report drawing and analytically assaying a venous whole-blood or separated plasma sample for modafinil during Hayling performance. Consequently the nominal 200 mg tablet cannot be translated into a participant’s modafinil concentration, a therapeutic monitoring result, or an individual person’s impairment status. If the question concerns a measured drug specimen, see the site’s guide to biological matrices for why the material actually collected matters.

What does the Hayling Sentence Completion Test measure?

The Hayling task uses sentence stems with missing final words. During response initiation, the respondent must complete a sentence with a sensible, context-appropriate word. During response inhibition, the respondent must avoid the predictable ending and choose a word unrelated to the sentence. The researchers used 30 sentences distributed across the task’s two sections. A handheld stopwatch recorded the elapsed time for spoken answers, while scoring captured errors or failures to follow the relevant instruction.

These are operationalized measures of a specific verbal task. They are not chemical assays of dopamine or another neurotransmitter, direct images of prefrontal neurons, standardized roadside psychomotor examinations, or measurements of whether a witness’s account is true. “Inhibition” here describes complying with a laboratory language instruction. It does not automatically generalize to resisting a dangerous driving maneuver or deciding whether to commit an offense.

What was the principal response-time result?

Across the initiation and inhibition sections, the paper reported mean response latencies of 30.21 seconds in the modafinil group and 16.22 seconds in the placebo group. The between-group main effect was F(1,59) = 4.674, p = .035. The difference between those reported group summaries was 13.99 seconds, with the modafinil group responding more slowly. These are task-level, group-reported values; they should not be repackaged as a 13.99-second predicted delay for any given person.

Across both treatment groups, the inhibition section was slower than the initiation section: the reported section means were 32.92 and 13.30 seconds, respectively, with F(1,59) = 15.47 and p < .001. That section effect confirms the language task was more time-demanding when the obvious word had to be suppressed. The researchers did not find a significant treatment-by-section interaction in latency, F(1,59) = 0.022 and p = .88. Therefore the trial did not establish a separate, special slowing effect confined to the inhibition condition.

What happened to accuracy and error counts?

The error analysis did not show a statistically significant benefit from modafinil. Its overall drug main effect was F(1,59) = 0.34, p = .561. The study also reported no significant treatment-by-section interaction for errors. Within both groups, the inhibition task was harder: mean error counts were 3.8 for response inhibition versus 0.34 for response initiation, with a section main effect F(1,59) = 167.541, p < .001. This comparison of task sections must not be confused with a between-drug benefit.

A nonsignificant treatment effect does not prove that the two drugs produce mathematically identical accuracy under every condition or in every population. The trial supports a narrower reading: under this task, dose, timing and sample, the investigators found slower responding with modafinil but failed to demonstrate an accuracy improvement. There was no active-driving test, no objective measure of workplace decisions, and no repeated-dose assessment establishing long-term performance outcomes.

Two evidence tables: the measured signal and its boundaries

The table reports the authors’ original aggregates, not participant-level raw data. Response seconds and error counts are not interchangeable measurement scales. The fact that the section effect and the medication effect were both detected in latency does not establish a medication-induced improvement in the inhibition task. An absence of a statistically significant interaction also does not establish a formal equivalence margin.

Did the study measure physiologic stimulation or perceived mental improvement?

The investigators collected a 16-dimension visual analogue mood questionnaire at multiple times and monitored pulse and blood pressure. The source did not find statistically significant treatment differences across the reported mood comparisons. Its cardiovascular table likewise lists no significant modafinil-versus-placebo differences for pulse (p = .17), systolic blood pressure (p = .14), or diastolic blood pressure (p = .581). These null findings are real findings within the procedures described; they are not proof that modafinil never changes cardiovascular measurements in a different group, dose or clinical setting.

Importantly, a subjective mood score is not the same endpoint as a timed verbal response. A person might report feeling attentive while completing a verbal task slowly, or might report no marked mood change despite changes in reaction timing. The paper does not provide direct molecular evidence about why the latency difference occurred. Neither pulse rate nor the Hayling response score can serve as an alternative identifier for drug concentration.

Does a longer response mean better inhibitory control?

No. Waiting longer before saying a word could reflect deliberation, difficulty retrieving language, caution, unfamiliarity, altered attention or some combination. The Hayling study was structured to ask whether additional time yielded a measurable accuracy advantage. The principal result was more time without fewer errors. To claim “improved inhibition” solely because the group hesitated longer would replace the actual error endpoint with an untested explanation for response time.

Likewise, the direction of the finding should not be inverted into a claim that modafinil broadly degrades intelligence or verbal competence. Accuracy on this single task was not significantly worse under modafinil, either. The observation is bounded by the intervention, healthy-volunteer selection, laboratory verbal task and measurement window. Understanding other cognitive domains would require experiments expressly designed to assess them.

Was drug exposure verified by a forensic biological specimen?

No quantitative modafinil concentration was reported from directly analyzed human venous whole blood, separated plasma, serum, urine, oral fluid, hair or another biological matrix during the Hayling experiment. Investigators controlled administration of a nominal oral dose in a study setting. That is a valid way to assign treatment, but it is not the same as a calibrated laboratory result establishing a person’s concentration at a particular moment.

A forensic laboratory might identify a substance using a validated confirmatory technique and report a concentration in a defined specimen; such evidence has its own collection timing, calibration and interpretive limits. None of those analytical details can be invented for the Hayling trial. The site’s confirmatory drug testing overview explains why identifying an analyte and interpreting its effects are separate questions. The therapeutic drug monitoring overview similarly distinguishes measured concentrations from assumptions based on dose alone.

How was response latency timed, and what uncertainty follows?

Verbal response latencies were recorded by an experimenter using a handheld stopwatch rather than by an automated microphone-triggered latency system. The authors described its hundredth-of-a-second display and cited work suggesting handheld chronometry can be reliable under some conditions. Nevertheless, human triggering and the decision about precisely when an answer starts can introduce observer-related measurement variability. This is especially relevant when translating findings into a different test setting or an individual case.

The group contrast was sufficiently large to motivate the authors’ interpretation, but their source aggregates are not a validated individual prediction interval. The paper’s descriptive labels and certain reported effect-size calculations also warrant careful review against underlying data before quantitative secondary modeling. The defensible public takeaway relies primarily on the explicitly reported means, test design, model F statistics, nominal p values and the unchanged error endpoint.

How should the 61-person analysis be distinguished from the 64-person study?

Recruitment, assignment and analysis populations answer different questions. The randomized study involved 64 volunteers; language eligibility removed three from the Hayling analysis. A headline describing “64 people randomized” is consistent with the source, while a claim that all 64 contributed to the main Hayling effect is not. Further, the study presented a participant under both task sections, so those two observations must not be counted as two separately randomized people.

Even within the report, not all secondary measures necessarily have the same analyzed denominator. A mood or cardiovascular table is not automatically based on exactly the same complete cases as the verbal-task model. For a systematic review or legal expert report requiring exact allocation and follow-up accounting, it would be prudent to inspect the trial protocol, CONSORT flow and original data rather than infer any missing number from an abstract.

Could this result establish individual driving impairment or occupational fitness?

No. The research did not put subjects in a vehicle, simulator, cockpit, industrial control station, or real-world safety-critical decision system. Its measured action was producing a sentence completion. Slowing the production of a word does not translate to an identified number of extra feet of stopping distance or to a probability of making an aviation error. There is no validated conversion between the Hayling seconds in this trial and the threshold for an individual driver’s legal impairment.

For case analysis, a contemporaneous toxicology specimen, confirmed identification and quantitation where available, testing method, chain of custody, clinical observations, driving behavior and alternative causes could be relevant. The Hayling study cannot replace those facts. Equally, the absence of an accuracy benefit on one task does not disprove a potential beneficial effect of modafinil on sustained vigilance in a separately tested sleep-deprivation paradigm.

How does this differ from modafinil studies of prolonged wakefulness?

The healthy volunteers in this experiment were not characterized as military aviators facing an extended wake period. This was a single-dose, non-sleep-deprivation executive-language experiment. Studies examining repeated modafinil doses under sleep deprivation may use psychomotor vigilance tasks, self-reported sleepiness, or performance consistency rather than sentence completion and word inhibition. Those endpoints do not yield interchangeable estimates simply because the administered molecule has the same name.

Scientific interpretation should be anchored to the task and cohort. A PVT might capture lapses of visual attention; a Hayling test captures timed verbal completions and errors under rule changes. Neither alone amounts to an all-purpose legal cognitive impairment assay. Comparing studies responsibly therefore requires identifying their populations, exposure schedules, assessment timing, measured outcomes and any biological assays before claiming agreement or contradiction.

What are the main limitations of the randomized experiment?

Its strengths include random assignment, masking of the testing staff, an active-placebo comparison, explicit accuracy and timing outcomes, and a fixed postdose task window. Important limits include a relatively small healthy cohort, one 200 mg dose, no dose-response curve, three exclusions from the key task, reliance on handheld verbal-response timing, and no biological determination of drug concentrations. Participants’ language proficiency was sufficiently important that three records were excluded, emphasizing that performance on this test depends on language as well as inhibitory control.

Secondary descriptive tables and reported effect-size statistics contain details requiring verification before precise quantitative synthesis. The primary treatment finding remains a group-level association in a randomized design, not an individual diagnostic rule. There is no independent replication in this report and no direct study of long-term dosing, therapeutic patients, different occupations, crashes, or impairment statutes. An honest account preserves those boundaries rather than replacing a specific experiment with a sweeping description of modafinil as “good” or “bad” for cognition.

Seven focused questions about the original evidence

1. Did everyone randomized contribute to the Hayling result?

No. Sixty-four volunteers participated in the randomized study, but three whose first language was not English were excluded from the Hayling analysis. The principal analysis’s F(1,59) denominator supports an analyzed task sample of 61 across the two between-person groups.

2. Did 200 mg modafinil make answers faster?

No. The source reported longer average Hayling latency with modafinil: 30.21 versus 16.22 seconds with placebo, and a significant overall treatment effect (p = .035). That describes the tested verbal task, not every form of reaction time.

3. Did taking longer reduce Hayling errors?

Not detectably. The main between-drug error comparison was not significant (p = .561). The stronger error contrast was between task sections, with more mistakes when suppressing the expected sentence ending.

4. Was the inhibition section specifically changed by modafinil?

The treatment-by-section interaction for response latency was not significant (p = .88). The study therefore did not establish an additional drug effect uniquely limited to the inhibition instruction, despite greater overall difficulty in that section.

5. Was a modafinil blood level measured during performance?

No. Dose administration was recorded, but the paper did not report quantified modafinil in venous whole blood, plasma, serum or any other specimen collected during the verbal task. A swallowed dose is not a measured concentration.

6. Did participants report markedly better mood or show significant cardiovascular changes?

Not on the reported between-group comparisons. The study found no significant treatment difference for the listed mood dimensions or its pulse and blood-pressure outcomes. This does not prove an absence of all effects under other circumstances.

7. Can the result determine whether a driver or pilot was impaired?

No. Neither driving nor flying was tested, there was no case-specific biological specimen and no validated personal impairment threshold. An individual’s performance and legal interpretation require separate evidence tied to that person and event.

Bottom line for forensic interpretation

The 2014 randomized Hayling experiment provides a clear caution against treating slower responses as automatic evidence of improved inhibitory control. Modafinil changed the timing of verbal answers at the tested 200 mg dose without a significant advantage in error counts. It neither measures the drug in a forensic specimen nor validates conclusions about intoxication, individual driving ability, real-world aviation safety or an incident’s cause. The correct citation for this result is the original primary study by Mohamed and Lewis, PLoS ONE 9(11):e110639 (2014); its scope should accompany every practical use of the finding.