Why Did 3 mg/kg Caffeine Change Stroop Speed and Prefrontal fNIRS Signals When 9 mg/kg Did Not? A 10-Man Crossover Trial

Answer: In a small randomized crossover experiment involving ten healthy young men, an oral caffeine dose of 3 mg per kilogram of body weight was followed by faster performance on both congruent and incongruent Stroop trials, and by an increase in an optical measure of oxygenated hemoglobin during the congruent task in three prefrontal regions. The 6-mg/kg condition improved reaction time only on incongruent trials, while the investigators did not report a statistically significant reaction-time improvement at 9 mg/kg. Accuracy and the principal incongruent-task fNIRS regional comparisons did not show equivalent treatment effects. This is evidence of a dose- and task-specific group-level laboratory response, not proof that 3 mg/kg is universally optimal, that 9 mg/kg causes impairment, or that an individual person’s blood caffeine level or driving fitness can be inferred.

The original primary study is Zhang, Liu, Wang, Deng and Zheng, Cognition and Brain Activation in Response to Various Doses of Caffeine: A Near-Infrared Spectroscopy Study, Frontiers in Psychology 2020;11:1393, doi:10.3389/fpsyg.2020.01393. Its distinctive contribution is that the experiment connected a color-word reaction-time task with functional near-infrared spectroscopy (fNIRS) during four randomized caffeine conditions. fNIRS records light-based changes associated with cerebral oxygenated hemoglobin. It does not measure the concentration of caffeine in venous whole blood or directly identify the biochemical cause of every behavioral change.

What forensic question can the experiment actually answer?

It can help answer whether increasing an administered dose of an ordinary stimulant produces a monotonic improvement in these particular tasks, or whether different levels of cognitive conflict show different responses. It also indicates whether a simultaneously measured cortical hemodynamic proxy follows the same pattern. This question is narrower than asking whether caffeine makes people generally safer, more perceptive, better witnesses, or less impaired.

That distinction is especially useful when a scientific claim passes through several unsupported steps: a dose is treated as though it were a measured blood level; a group average is described as an individual’s response; and a response on a fixed computer task is promoted to real-world driving or incident causation. The original investigators did not test those additional links. Their strongest measured result is a contrast between nominal caffeine exposures, Stroop reaction times, and condition-specific optical hemoglobin signals.

Who participated, and why is the true sample size ten?

Ten healthy nonsmoking men participated. Their reported mean age was 20 ± 1 years and average body weight 70.5 ± 4.8 kg. They were described as low habitual caffeine users, consuming less than 50 mg a day. This selection limits the ability to generalize findings to women, older adults, people with high caffeine tolerance, workers with sleep loss, or patients taking interacting medications. There was no clinical intoxication cohort.

Each man completed four treatment sessions, allowing within-person comparisons of the placebo and three caffeine doses. Forty participant-visits do not mean forty independent biological participants. Likewise, the many color-word presentations, forty optical detector channels, repeated 3.91-Hz instrument samples, and separately calculated regional responses were observations nested within ten people. The experiment is informative about within-person treatment contrasts but remains vulnerable to sampling imprecision and outlying individual responses because the underlying cohort was small.

What exactly was administered?

The investigators used oral capsules containing caffeine hydrate, with caffeine mass scaled to each participant’s body weight: 3, 6, or 9 mg/kg. In the control condition the capsules contained calcium carbonate. For blinding, all conditions contained the same total 9 mg/kg weight of caffeine plus calcium carbonate, distributed among three red capsules: the 3-mg/kg caffeine condition had 6 mg/kg calcium carbonate, the 6-mg/kg condition had 3 mg/kg calcium carbonate, and the 9-mg/kg caffeine condition had caffeine alone. Participants swallowed the capsules with 200 mL water.

A 60-minute interval preceded post-administration Stroop testing. Each man returned for all four conditions, separated by approximately a week, and the sequence was double-blind and crossover. The study used standardized conditions and a prior task familiarization, but no design can eliminate every practice, expectancy, circadian or diet-related difference. A 3-mg/kg dose is a mass administered per unit body weight. It is not a concentration of 3 mg/L, 3 µg/mL or 3 ng/mL in any biological specimen.

How was cognitive conflict created with the Stroop task?

The computer displayed Chinese words denoting green, blue, or red in colored type. Sometimes the color of the letters agreed with the meaning of the word; this was the congruent condition. Other times the ink color and the word meaning disagreed; this was the incongruent condition. Participants responded to the visual color using one of three assigned number-keypad buttons. The incongruent condition requires resisting the more automatic tendency to read the word rather than name its color.

Two blocks containing 120 presentations each mixed 60 congruent and 60 incongruent events per block. Response time was reported in milliseconds, and accuracy as the fraction of correct answers. These separate outcomes matter: a faster response that sacrifices correct answers has a different interpretation from a faster response that retains accuracy. The response apparatus did not simulate a motor vehicle, quantify real hazard detection, measure long-duration vigilance, or establish the validity of testimony under cross-examination.

What numerical reaction-time changes were actually printed?

The paper’s original Table 2 reports means and standard deviations before and about an hour after administration for each caffeine condition. In the incongruent task, the 3-mg/kg session changed from 685.43 to 649.70 milliseconds, and the 6-mg/kg session changed from 658.83 to 624.10 milliseconds. Both within-condition comparisons were marked significant at p<0.05. The 9-mg/kg session changed numerically from 654.54 to 649.65 milliseconds, without a statistically significant treatment difference reported.

In the congruent task, only 3 mg/kg was accompanied by a significant reported shortening, from 580.66 to 553.60 milliseconds, with the narrative identifying p<0.01. The corresponding 6-mg/kg values were 576.28 and 580.37 milliseconds; 9-mg/kg values were 585.74 and 574.05 milliseconds. Neither higher dose had a significant within-condition improvement in the source’s stated post-hoc analysis. The placebo session actually showed a numerical increase in mean time in both task conditions. These before-after numbers are descriptive means within each assigned dose visit, not estimates of a causal effect for a separately recruited group.

Evidence table 1: dose-specific reaction-time findings

The overall dose-by-time interaction was reported as F(3,27)=5.61, p<0.01 for incongruent response time and F(3,27)=3.14, p<0.05 for congruent response time. Those model tests are not equivalent to proving that the 3-mg/kg dose is statistically superior to every other dose in every person. In particular, the source reports no significant direct difference between the 3- and 6-mg/kg conditions for incongruent response time. The correct interpretation keeps omnibus effects, before-after comparisons and between-dose superiority claims separate.

Did participants become more accurate, or only faster?

The primary dose-related accuracy interactions were not significant in either the congruent or incongruent condition. The paper’s original Table 2 shows accuracy means generally near 0.92–0.99 across conditions. For example, the placebo incongruent accuracy mean changed from 0.92 to 0.95, whereas the 6-mg/kg incongruent mean moved from 0.95 to 0.93. Neither numerical direction supports a new claim of a statistically reliable dose effect without its associated model test.

This negative finding places an important limit on the phrase “cognitive enhancement.” The source demonstrates a narrow reaction-time result, not evidence that all components of cognitive control improved. Absence of a significant accuracy difference is not a formal proof that the doses are equivalent, especially in a cohort of ten. The experiment also was not designed to test retention of a complex episode, truthfulness, divided-attention driving, visual search in a moving scene, or emergency judgment under real consequences.

What did fNIRS physically measure?

Investigators placed a near-infrared optical sensor arrangement over frontal and parietal scalp regions. Sixteen light sources and fifteen detectors yielded forty channels, using two optical wavelengths, 780 and 830 nanometers. Signals were sampled at 3.91 Hz and converted using a modified Beer–Lambert approach into a change signal associated with oxygenated hemoglobin (HbO). The authors selected channels for regional analysis and related them to dorsolateral prefrontal cortex (DLPFC), frontal pole area (FPA), and ventrolateral prefrontal cortex (VLPFC).

This is a physiological optical hemodynamic signal, not a tube of blood assayed for caffeine. It reflects changes in light absorption associated with oxygenated hemoglobin in sampled head tissue and is subject to vascular and optical influences. It does not directly measure neuronal firing, dopamine receptor occupancy, synaptic caffeine concentration, or a person’s blood caffeine amount. Although the authors discussed cortical activation, they did not demonstrate that caffeine solely altered the signal through a particular neurotransmitter mechanism. The analytical system was a NIRScout instrument, not a mass-spectrometric drug-testing platform.

Why did the result depend on congruent versus incongruent trials?

In the congruent condition, the paper reports significant dose-by-time interactions in each of its three selected prefrontal regions. The 3-mg/kg session was associated with increased mean HbO, while placebo showed a decrease; the authors did not identify corresponding significant mean HbO increases after 6 or 9 mg/kg. DLPFC had F(3,27)=4.21, p<0.01; frontal pole had F(3,27)=3.21, p<0.05; and ventrolateral prefrontal cortex had F(3,27)=4.02, p<0.05.

By contrast, the paper’s separate incongruent-condition analyses reported no significant dose-by-time interaction for mean HbO in DLPFC, FPA or VLPFC. This is scientifically interesting precisely because some incongruent reaction-time differences were significant. An optical signal and a behavioral response can diverge. It would be a category error to declare that faster incongruent responding proved increased prefrontal HbO in that same condition, or that the lack of an HbO result negated the recorded reaction-time changes.

Evidence table 2: significant and genuinely null prefrontal results

These numbers should not be treated as three independent biological replication experiments. The same ten people wore one optical array that generated spatially related channels, and the regions were selected from those channels. The numerical changes illustrated in the source plots carry dispersion and are not accompanied by a validated cutoff for diagnosing intoxication or individual cognitive enhancement. Both the small biological denominator and selection of regions matter when interpreting isolated significant results across multiple outcomes.

Did higher caffeine doses cause a measurable deterioration?

The results do not establish that conclusion. A null or nonsignificant finding at 9 mg/kg does not show that the dose caused a deficit, and the small trial was not powered to establish equivalence or comparative safety among these doses. The authors discussed possible adverse effects at high doses as a mechanistic explanation, but this protocol did not generate a clinical adverse-event rate table proving that nervousness or gastrointestinal distress caused the nonsignificant performance findings. That distinction avoids turning background discussion into new experimental evidence.

The design also did not track repeated occupational intake, delayed sleep disruption, combinations with prescribed stimulants, cardiovascular adverse events in vulnerable patients, or toxicity in caffeine-naive or older populations. It studied a single short postdose interval in healthy young men with little habitual caffeine consumption. There is no basis here for recommending one administered dose to a real driver or for characterizing another dose as legally impairing.

Could the measured HbO response establish a particular neural mechanism?

Not by itself. Caffeine affects adenosine signaling and can alter both neuronal processing and vascular responses. fNIRS records changes consistent with altered local tissue oxygenation and hemodynamics, so both pathways may influence the signal. The authors proposed neurovascular explanations but did not administer an isolated receptor antagonist comparison, chemically quantify caffeine exposure, directly measure extracellular neurotransmitters, or calibrate fNIRS against an individual’s cortical dopamine release.

Because optical signals are influenced by superficial circulation, motion, sensor geometry and preprocessing, a change in HbO cannot be reduced to a universal “more brain activity equals better cognition” formula. The positive result was conditional: the congruent Stroop trials and three selected regions. No corresponding treatment interaction was demonstrated for the three regions during the incongruent trials. That selective pattern is valuable evidence about limitations, not a flaw to conceal.

Why does body-weight dosing not establish a forensic concentration?

An oral mass-per-kilogram prescription is an exposure administered under research conditions, but it does not indicate what a later sample would contain. To reconstruct a human drug result, one would need a collected matrix, sampling time, validated assay, identity and quantitation of parent caffeine or metabolites, individual pharmacokinetic context, and information about other intake. Biological matrices in forensic toxicology explains why results should remain attached to the material actually tested.

In this experiment, no participant’s caffeine or paraxanthine was quantified in legally collected venous whole blood. The paper likewise reports no direct plasma, serum, urine or oral-fluid caffeine drug assay and no caffeine concentration-time curve. A 9-mg/kg administration must not be reinterpreted as 9 mg/L in plasma, and the fNIRS HbO signal must not be relabeled a quantitative blood-caffeine result. Confirmatory drug testing describes the separate analytical questions necessary to identify and measure a chemical in a real specimen.

Does the result say anything about actual driving or eyewitness perception?

Only indirectly, as a reminder that attention and speed have multiple separable components. The test required pressing keys to identify the printed color of a familiar word. Driving also requires maintaining lane position, anticipating hazards, dividing attention, making decisions and coordinating responses while moving through an unpredictable environment. None of those operational endpoints was tested here. No participant drove on a road, encountered a surprise collision hazard or produced an objectively validated crash-risk outcome.

Similarly, an experimental difference in reaction time does not determine whether a specific witness correctly perceived a face, remembered a sequence or accurately recounted an event. The researchers did not test testimony, identification, memory confidence, an actual incident, or the relationship between fNIRS changes and credibility. Any case-specific inference would require independent facts; a study of ten healthy male volunteers cannot supply a numerical conversion from one task or optical variable to courtroom reliability.

Which limitations most affect scientific weight?

First is sample size: ten low-caffeine-consuming men, with no female or older-adult replication. Second is experimental duration: the investigators measured responses at baseline and roughly one hour after one caffeine administration, not an entire day or repeated dosing. Third is outcome multiplicity: reaction time, accuracy, congruence conditions, three regions and interactions produce multiple possible comparisons. The isolated positive signal cannot establish a robust individual-level prediction without replication and appropriately controlled error rates.

Fourth is dose interpretation: a missing significant result at the highest tested dose does not demonstrate harm or a formally established optimum. Fifth is measurement: the optical hemodynamic proxy does not quantify stimulant concentrations, and the experiment collected no matrix-specific blood drug levels. Sixth is ecological validity: a computer-based Stroop task is not a tested substitute for driving, emergency response, forensic cognitive capacity or personal incident causation. These limitations do not negate the genuine results; they define the domain within which those results remain accurate.

What would a stronger concentration-and-performance experiment require?

For pharmacokinetic questions, investigators could pair rigorously timestamped whole-blood caffeine collection with separated plasma or serum, collected oral fluid, and validated caffeine/paraxanthine assays, while reporting analytical ranges, quality controls and uncertainty for each matrix. Repeated sampling at prespecified times could then distinguish nominal dose from observed individual concentrations. The current paper does not contain those data. Therapeutic drug monitoring outlines why specimen time and intended interpretation belong alongside any concentration.

For a stronger behavioral conclusion, a larger sample including women, older people and a range of habitual caffeine users would be necessary. Preregistered primary Stroop contrasts, region definitions before inspecting responses, multiple-comparison control and independent replication could help evaluate whether a low-dose advantage persists. Separately validated driving or safety-critical outcomes would be needed before discussing operational fitness. Adding these measurements might answer new questions; their absence cannot be repaired by treating the published fNIRS plot as a forensic blood assay.

Bottom line for interpreting dose, behavior, and physical evidence

The most defensible interpretation is selective rather than universal. A short crossover experiment in ten healthy young male low-caffeine consumers found significant within-session speed improvements with 3 mg/kg during both Stroop congruence conditions, with 6 mg/kg during incongruent trials only, and with neither task at 9 mg/kg. It also found congruent-task optical prefrontal HbO changes in three regions after 3 mg/kg, while incongruent regional HbO and answer-accuracy interactions were not significant.

That combination is scientifically useful because the behaviors and physiological proxies did not move in lockstep across doses and conditions. It does not identify a caffeine blood concentration, individual state of intoxication, last use, driving competence, witness credibility or causal contribution to any crash or injury. A sound forensic review should retain the original specimen/method boundaries and ask which independently measured facts bridge a research outcome to the actual issue in dispute.

Frequently asked questions

1. How many different people participated?

Ten independent healthy young men. Each completed four oral dose conditions and many repeated Stroop and fNIRS observations. Forty visits and hundreds of trials do not create forty or hundreds of independent biological participants.

2. Did 9 mg/kg caffeine make cognitive performance worse?

The researchers did not show a statistically significant within-session Stroop reaction-time benefit at 9 mg/kg, but the study did not establish that this dose impaired performance or caused a clinical adverse event. A nonsignificant result is not proof of harm or equivalence.

3. Did 6 mg/kg improve all cognitive outcomes?

No. It showed a significant shortening of reaction time during incongruent color-word trials but not a significant improvement during congruent trials, nor an established improvement in accuracy. The effect was task-dependent.

4. Was oxygenated hemoglobin obtained from a venous blood sample?

No. HbO was an optically derived fNIRS signal recorded through scalp sensors over selected cortical regions. It was not collected blood analyzed in a laboratory, and it cannot be interpreted as a chemical caffeine concentration.

5. Did the investigators chemically measure caffeine in whole blood, plasma, or oral fluid?

No quantitative caffeine or paraxanthine testing of those human specimens was reported. The paper assigned oral doses and measured computer-task performance plus optical hemodynamics. It contains no validated caffeine whole-blood drug assay or person-specific pharmacokinetic curve.

6. Why could a reaction-time result be significant while the fNIRS result was null?

The measures reflect different aspects of a task. Incongruent response times changed at certain doses, but the source did not find significant regional dose-by-time fNIRS HbO interactions in that condition. One measurement does not automatically predict another, and both are affected by sampling and statistical uncertainty.

7. Can this study establish impairment or accident causation in one person?

No. It did not measure driving, crashes, an individual’s blood caffeine, dose timing in a forensic case, or incident mechanisms. It supports only the reported average task and fNIRS comparisons in the studied volunteer cohort.

Primary empirical source: Zhang B, Liu Y, Wang X, Deng Y, Zheng X. Cognition and Brain Activation in Response to Various Doses of Caffeine: A Near-Infrared Spectroscopy Study. Frontiers in Psychology. 2020;11:1393. https://doi.org/10.3389/fpsyg.2020.01393.