Did 5 mg/kg of Caffeine Slow Soccer Players’ Thinking? A 12-Player Crossover Trial

Quick answer: In a randomized, double-blind crossover experiment involving 12 professional male soccer players, a single oral dose of caffeine at 5 mg/kg was associated with slower responses on the congruent and neutral computerized Stroop conditions relative to placebo. The exercise protocol itself was associated with faster responses across all three Stroop conditions. Some accuracy contrasts also favored placebo. But the caffeine effect on incongruent response speed was not statistically significant, and perceived exertion did not differ significantly between caffeine and placebo. The study measured neither a caffeine concentration in blood nor real-world driving or soccer match decisions. It therefore cannot support a concentration-based or individualized forensic impairment opinion.

The original study was published by de Almeida and colleagues in 2023 in Nutrients: Effects of Acute Caffeine Ingestion on Cognitive Performance before and after Repeated Small-Sided Games in Professional Soccer Players. Its main contribution is not a universal conclusion that caffeine makes thinking worse. It is a closely specified comparison of a single administered dose, a short sports-exercise protocol, and one laboratory attention task in a small, specialized sample.

What was the study’s actual answer?

For these 12 soccer players under the study’s particular conditions, caffeine did not enhance the cognitive performance endpoints chosen by the investigators. On the contrary, response times were longer with caffeine in two of three Stroop conditions. Accuracy in the congruent condition was also lower with caffeine, while the more demanding incongruent condition displayed a significant treatment-by-time interaction and reported within-timepoint placebo–caffeine differences. That combination is more nuanced than saying that every measure deteriorated. The authors also found faster response times after repeated small-sided games regardless of caffeine assignment.

The interpretation must preserve the difference between a drug-assignment effect, an exercise-period effect, and a statistical interaction. A small p-value for exercise does not establish a caffeine benefit. A significant interaction does not necessarily mean that the overall treatment main effect was significant. And an absence of statistical significance in a 12-person study does not prove that an effect is exactly zero. Each endpoint answers a particular comparison, not an unrestricted question about cognition or safety.

Who participated, and how many independent observations were there?

The biological sample consisted of 12 professional soccer players recruited from a Brazilian club. Their reported mean age was 29 ± 4.1 years, mean body mass 78.1 ± 7.7 kg, and habitual daily caffeine intake 139 ± 85 mg. The investigators excluded goalkeepers from analysis and screened for characteristics such as injury, certain psychoactive medication use, and color-vision problems. This is a small sample of practicing athletes, not a population sample of licensed motorists, trauma patients, children, or individuals taking prescribed stimulants for ADHD.

Each player completed two experimental conditions. Thus, the study generated repeated treatment observations from the same 12 people; there were not 24 independently recruited humans. The five exercise bouts, repeated Stroop trials, and successive ratings of effort are further measurements nested within those people. Treating each response, trial, or exercise bout as if it represented another unrelated participant would artificially increase the apparent sample size. The authors analyzed the repeated design with repeated-measures methods, and forensic readers should use the participant-level n when explaining its strength.

How were caffeine and placebo administered?

On one visit, the intervention was caffeine at 5 mg per kilogram of body mass dissolved in 250 mL of artificially sweetened water. On the other, the drink contained the same volume of sweetened water without caffeine. Conditions were randomized and described as double-blind, with approximately seven days between the two test sessions. The 5 mg/kg value is an administered oral dose per unit of body mass; it is not a measured blood concentration, a plasma caffeine value, a limit of detection, or a pharmacokinetic parameter.

The athletes were asked to abstain from caffeine-containing foods and drinks for 48 hours before testing and to replicate breakfast between test days. The study reported that five participants correctly identified receipt of caffeine in the blinding assessment. Such imperfect blinding is relevant when interpreting subjective expectations but does not transform an observed reaction-time contrast into a measured drug concentration. The same participants provided their own within-person comparisons, which is useful for reducing some differences among players, but a crossover does not by itself guarantee control of every environmental or practice effect.

What did the five small-sided games involve?

At each test visit, players participated in five five-minute small-sided soccer games, separated by one-minute recovery intervals, after a brief warm-up. The stated format was three outfield players plus a goalkeeper on each side, played on a 36 by 27 meter area. Goalkeepers contributed to the exercise format but were excluded from the analyzed participant sample. Testing occurred at approximately the same morning time on the experimental days. The study reported relatively warm field conditions, with start temperature 31.9 ± 1.98°C, increasing to 32.45 ± 3.21°C at the end.

The protocol combined sport-specific activity with a computerized cognitive test administered before and immediately after the exercise bouts. Therefore, it can speak to pre–post changes around a standardized training session; it was not an instrumented assessment of actual game judgment in professional competition. The researchers did not measure every environmental or physiological feature that might influence a player’s performance. In particular, hydration status was not objectively measured, a limitation the authors acknowledged even though water was available during the games.

What did the computerized Stroop test measure?

Participants pressed colored keyboard keys to identify the displayed text’s ink color, not to read the word. The source separated stimuli into congruent conditions (word and color matched), incongruent conditions (word and ink color conflicted), and neutral conditions. The software recorded the mean response time in milliseconds and response accuracy as a percentage of correct responses. A faster response is a lower time value; an accurate response is a correctly selected ink color. Those dimensions should not be conflated.

Each test condition used a block of repeated one-second visual stimuli, with the source specifying 48 stimuli per Stroop condition and an overall test duration of approximately five minutes. The incongruent condition was intended to impose interference and response inhibition demands. But Stroop performance remains a computer-based behavioral proxy. It does not chemically measure caffeine, establish driving fitness, measure reaction to a real pedestrian, prove accurate tactical choices in a match, or diagnose brain injury. The forensic question must remain tied to this apparatus and task.

Reaction-time findings: caffeine and exercise did not point in the same direction

The first source-faithful evidence table separates the reported caffeine main effect from the pre/post small-sided-game effect and their interaction. Values are the source’s reported p-values from Figure 3; they are not fabricated participant-level responses or estimates read off plotted curves. The source reports caffeine-associated slowing for the congruent and neutral conditions but not for the incongruent condition. All three exercise main effects met the study’s stated significance threshold and pointed toward faster responses after play.

The table does not imply that caffeine slowed every response type or that soccer exertion chemically counteracted caffeine. It describes the fitted differences in this small repeated-measures experiment. None of the reaction-time treatment-by-exercise interactions shown in the original figure was statistically significant. Thus, the observed faster responses after play are not evidence of a demonstrated extra caffeine-specific exercise advantage. The paper does not provide individual paired time trajectories sufficient to compute a reliable person-specific caffeine effect.

Why was incongruent reaction time a null finding?

It may be tempting to emphasize the interference condition because it resembles situations in which a person must ignore distracting information. Yet this is precisely the response-time endpoint where the source reported a nonsignificant caffeine main effect, p = 0.071. The numerical trend in a plot does not convert that hypothesis test into a significant result. The pre/post exercise effect for incongruent reaction time was significant, but this was common to the study’s repeated exercise protocol rather than a demonstrated difference in the response to caffeine.

A conventional threshold of 0.05 is an analysis convention, not a biological boundary. A p-value of 0.071 does not demonstrate perfect equivalence between treatment conditions, just as a value of 0.045 does not establish a large, clinically meaningful effect on every participant. In a legal setting, one should avoid changing the scientific conclusion by rounding a near-threshold number downward, pooling unrelated task outcomes, or treating the most favorable contrast as though it were the study’s only planned analysis.

Accuracy findings were also selective

The second source-grounded evidence table reports accuracy tests and perceived effort separately from response times. It preserves the distinction between a significant caffeine main effect in the congruent condition and a significant treatment-by-exercise interaction in the incongruent condition. Figure 3 marks within-timepoint caffeine–placebo accuracy contrasts in both the congruent and incongruent conditions, but the incongruent overall caffeine main effect did not reach significance. The neutral accuracy tests were also nonsignificant.

The accuracy results illustrate why response speed and correct choices must be read separately. The same soccer-exercise period can be followed by faster button pressing without a demonstrable improvement in the percentage of correct answers. A claim that participants were “thinking better” merely because they responded sooner would overstate the observed outcome. Likewise, saying “caffeine worsened all accuracy measures” would contradict the neutral task results and the nonsignificant treatment main effect for incongruent accuracy.

How should the incongruent accuracy interaction be interpreted?

The investigators reported an incongruent accuracy treatment-by-time interaction of p = 0.032 and indicated significant placebo–caffeine contrasts at the measured pre- and postgame time points in Figure 3. The same paper states that Tukey post hoc tests did not show significant pre-versus-post changes within either the caffeine or placebo condition. Those are not mutually exclusive comparisons. One comparison asks about differences between treatments at a time point; another asks whether a treatment group changed between time points.

Neither comparison establishes a numerical individual accuracy decrement absent the individual observations. More importantly, an interaction does not automatically show a significant overall drug main effect. The source’s incongruent accuracy drug main effect was p = 0.082, which must remain described as nonsignificant. In forensic communication, explicitly naming the contrast is preferable to an undifferentiated claim that the effect was “significant.” This prevents a statistical sentence from being interpreted as a direct impairment finding.

What happened to ratings of perceived exertion?

After each of the five games, players rated their perceived effort using a CR-10 scale. Those ratings rose significantly over the exercise series (reported time effect p < 0.001), an expected pattern with repeated activity. However, the caffeine-versus-placebo main effect was reported as p = 0.23, and the interaction was also p = 0.23. Those two findings were not statistically significant. The source therefore does not show that this caffeine dose reliably lowered perceived effort in these athletes during the particular small-sided-game sequence.

Perceived exertion is a self-reported intensity measure. It is not a blood lactate value, an objective caffeine assay, an indicator of drug absorption, or a personal clinical diagnosis. It is one physiological-experience endpoint relevant to this sport protocol. Different studies under other conditions may find different results; this source by itself cannot settle a population-wide controversy about caffeine and physical fatigue. A forensic report should not convert the absence of a caffeine RPE difference into proof of an absence of all cardiovascular, central nervous system, or metabolic actions.

Did exercise improve attention, or simply shorten test responses?

Observed reaction-time improvement following five small-sided games is a performance change in a computerized color-word task. Whether that improvement reflects practice, arousal, motor speed, attention, habituation, or a combination of mechanisms is not determined by the reported contrast alone. The crossover design involved repeated test exposure, and there was no separate rest-only condition matched to the elapsed time and repeated testing without the exercise regimen. The repeated-measures analysis can identify differences observed around the activity; it cannot isolate every proposed physiological explanation.

The authors’ discussion describes cognitive performance after exercise in favorable terms, but the accuracy results did not show uniform improvement. For a scientifically faithful summary, the narrow conclusion is that participants responded faster after the standardized exercise sequence across the three Stroop response-time conditions. It is not that they necessarily evaluated hazards more reliably, committed fewer game errors, became less intoxicated, or acquired a validated protection against stimulant-related cognitive effects. No such events or individualized tests were measured.

Was caffeine actually measured in whole blood, plasma, saliva, or urine?

No quantitative caffeine concentration was reported in a collected biological specimen in this experiment. The caffeine amount was calculated as a prescribed oral dose by body mass. The observed study variables were response speed, accuracy, self-reported effort, and experimental conditions. There was no reported validated GC–MS or LC–MS/MS assay for caffeine or its metabolite paraxanthine in the trial participants; there were no assay calibration results or reporting limits from which to reconstruct participant concentrations.

Specimen distinctions matter. Legally collected venous whole blood contains cells and plasma; plasma and serum are separated fluid matrices and are not automatically numerically interchangeable with whole blood. Urine may answer an exposure or excretion question, oral fluid may have its own collection and matrix effects, and sweat, hair, vitreous humor, milk, tissue, or breath involve yet other sampling considerations. The soccer study collected no reportable quantitative caffeine result in any of these forensic matrices. Its oral 5 mg/kg administration should never be relabeled as a blood concentration.

Why can’t we reverse-calculate a blood concentration or last-use time?

An administered dose and a reported task change do not identify the concentration of drug circulating in one particular person at an unknown time. That inference would require actual sampling and additional validated pharmacokinetic information, including timing, distribution, clearance, possible competing exposures, formulation, and interindividual variability. None of those data can be reverse-engineered from a Stroop p-value or from the mean body weight of the 12-person group. Body mass is useful for computing the nominal administered dose, not for proving a person’s analyzed sample result.

The trial was not designed to estimate time since caffeine ingestion in an unrelated event or to discriminate habitual exposure from a particular morning drink. Although the test schedule included a controlled administration window, a controlled research schedule does not serve as a validated forensic last-use clock. The absence of measured concentrations is a substantive limitation, not a missing number that can safely be replaced with a generic textbook peak or assumed half-life. An expert should explicitly distinguish the actual study’s dose assignment from any outside model they might separately choose to analyze.

What are the principal statistical limitations?

The trial analyzed numerous outcomes across three Stroop congruency types, response time and accuracy, different exercise stages, and post hoc contrasts. The repeated-measures framework is appropriate to the research question, but examining many contrasts increases the opportunity for nominally significant results. The paper did not present a single validated individual-prediction algorithm, comprehensive multiplicity-controlled clinical decision threshold, or external replication cohort. Reported p-values can support descriptions of prespecified or reported comparisons; they are not likelihood ratios for determining a driver’s actual impairment.

The sample included only 12 participating athletes, so precision and generalizability are constrained. Several reported statistics in the original text and figure warrant source-level checking before any formal secondary inference. This article preserves the printed figure p-values rather than fabricating corrected statistics. In particular, a reader should not infer an effect size solely from a p-value or estimate exact mean response times by eye from the plotted source image. For litigation, obtain the underlying data, analysis code, and prespecified analysis plan when a more exact calculation would materially matter.

What did the study leave unmeasured?

The investigators did not directly measure caffeine or paraxanthine in forensic blood, separated plasma or serum, urine, oral fluid, sweat, hair, vitreous, or another analytical matrix. They did not establish a concentration-effect curve, chemical detection window, validated toxicological cutoff, or exposure timing model. The protocol did not conduct an actual road-driving assessment, record crash risk, evaluate witness recall, quantify clinical toxicity, diagnose a cognitive disorder, or assign responsibility for an injury. Even the sport-related endpoint was a laboratory Stroop test before and after simulated practice games, not instrumented tactical decision-making in a competitive match.

Similarly, environmental heat was described but participant hydration was not objectively evaluated. The design was not built to disentangle dehydration, heat exposure, practice effects, exercise arousal, habitual stimulant tolerance, and placebo expectations in an individual participant. A reader can reasonably discuss these possibilities as limitations without presenting them as established alternative causal pathways. Absence of a measured endpoint is not a nonsignificant hypothesis test; no p-value should be manufactured for an experiment that the investigators did not conduct.

How applicable are these results beyond professional soccer?

Professional soccer players represent a relatively fit and specialized population with shared training demands. Their habitual caffeine exposure, exercise intensity, thermal environment, reaction to sport-specific stress, and familiarity with computerized testing may differ from a fatigued worker, commercial driver, patient with a sleep disorder, occasional caffeine user, or someone who consumed caffeine with alcohol or prescription medications. The single tested dose was 5 mg/kg, and the experiment did not randomly compare several different active caffeine doses. There is no basis to extrapolate this dose’s Stroop pattern to all smaller or larger doses as an experimentally demonstrated dose-response function.

The direction of a group-level test difference in this setting cannot determine whether an unrelated person performed better or worse after a given caffeine exposure. Cross-study comparisons require attention to cognitive tasks, baseline arousal, dosing schedule, sex and age distributions, and matrix measurement. This research is useful for challenging oversimplified statements that caffeine uniformly speeds every cognitive response. It is not a replacement for case-specific toxicology, validated psychomotor testing, or evidence about what occurred during a real event.

What questions should a forensic expert ask before citing this paper?

Start with the exact proposition being offered. Is someone arguing that caffeine generally improves vigilance, that a particular person reacted slowly, that a measured whole-blood value caused a crash, or that fatigue was relieved? Then identify which original outcome could actually address that proposition. For this paper, the most defensible answers concern a 5 mg/kg administered intervention and Stroop response-time or accuracy contrasts before and after standardized exercise in 12 professional male soccer players.

Ask whether the proposed inference changes the specimen, the population, the task, or the level of analysis. If the case concerns a venous whole-blood result, this source has no comparable quantitative caffeine measurement. If the issue is dangerous driving, the source has no driving task. If the dispute involves one individual’s concentration, it contains no individual-prediction calibration. Finally, distinguish reported statistical significance from reliability, direction, clinical relevance, and causal specificity; those are separate inferential requirements.

Frequently asked questions

1. Did 5 mg/kg of caffeine improve Stroop performance in these players?

No consistent improvement was demonstrated. The investigators reported slower congruent and neutral responses on caffeine, with no significant caffeine main effect on incongruent response speed. Some accuracy contrasts favored placebo. That is the result for this dose and protocol, not a universal conclusion about all uses of caffeine.

2. How many players were actually tested?

Twelve professional male soccer players participated. Each completed caffeine and placebo visits. Repeated game bouts and repeated Stroop stimuli were within-person observations, not independently recruited participants. The correct biological n for the crossover cohort remains 12.

3. Did the games themselves make reaction times faster?

The source reported statistically significant faster postexercise response times in congruent, incongruent, and neutral conditions. Accuracy did not show a corresponding significant exercise main effect across all conditions. That supports a narrow response-time observation, not proven improvement in real match decisions or accident avoidance.

4. Was the interference-task response speed significantly worsened by caffeine?

No. The incongruent response-time caffeine main-effect p-value was 0.071, which did not meet the paper’s stated 0.05 threshold. Do not confuse that nonsignificant result with the separate significant incongruent accuracy interaction, p = 0.032.

5. Did caffeine lower perceived exertion during the games?

Not significantly in this experiment. Perceived exertion increased over the five games, but the caffeine-versus-placebo main effect and interaction were each reported as p = 0.23. This does not test every form of exercise fatigue or physiologic action.

6. Can 5 mg/kg be converted into a forensic caffeine blood level?

Not from this trial. It reported an administered dose but did not quantify caffeine in venous whole blood, plasma, serum, oral fluid, urine, or other biological matrices. A dose, reaction-time measurement, and p-value cannot substitute for an analyzed specimen.

7. Does this study establish that caffeine caused or prevented a collision?

No. It assessed soccer-exercise conditions and computerized Stroop performance in a small group, not actual driving, crashes, or an individual’s legal intoxication. Forensic event causation requires separate validated measurements and event-specific evidence beyond this study.

Bottom line and original primary source

In a 12-player randomized crossover experiment, oral caffeine at 5 mg/kg was associated with some slower or less accurate computerized Stroop outcomes while repeated small-sided soccer games were followed by faster reaction times. Several other outcomes—including incongruent response-time caffeine main effect, neutral accuracy, and caffeine’s effect on perceived exertion—were nonsignificant. These competing results are valuable precisely because they defeat an overly simple narrative of uniformly improved or uniformly impaired cognition.

The complete primary paper is de Almeida RF, de Oliveira M, Furigo IC, Aquino R, Clarke ND, Tallis J, and Guimaraes-Ferreira L (2023), Nutrients 15(14):3094: original study and DOI 10.3390/nu15143094. It provides specific laboratory findings, not quantified legal blood drug concentrations or a scientifically validated opinion about the impairment, credibility, or causal conduct of any individual.