Does a Year Without Cannabis Improve Football Reaction Speed? An 18-Athlete Study

Short answer: Eighteen male NCAA Division I football players who had reported frequent cannabis use scored better on several computerized, sport-relevant cognitive measures after approximately one year of program-verified abstinence. Their overall standardized S2 Cognition score increased from −0.773 to 0.493, and the authors reported faster responses and improvements in decision complexity, distraction control, impulse control, tracking, and target detection. Those are within-person associations, not proof that cannabis caused the earlier scores or that stopping cannabis alone caused their improvement. The study did not measure contemporaneous venous whole-blood THC, directly test game performance, or validate a forensic impairment threshold.

The primary source is Ally, Mentele, and Wylie (2026), Cognitive performance recovery following marijuana cessation in NCAA division I athletes, a seven-page original research report in Frontiers in Sports and Active Living (doi:10.3389/fspor.2026.1821656). Its distinctive question is whether a change over a year in football-specific computerized performance can be translated into claims about individual cannabinoid exposure, impairment, sports injury, or legal causation. Those are separate questions and require separate evidence.

What did the researchers actually test?

The investigators evaluated football-specific speeded cognition, meaning performance on time-constrained perception, choice, visual search, response inhibition, and tracking exercises. The tool was the S2 Cognition Football Battery, administered on calibrated computers using a Cedrus button-box response interface. The researchers examined nine primary cognitive processes and an overall composite, along with secondary speed and accuracy scores. Most results were z-scores, not concentrations of cannabinoids or raw milliseconds.

Twenty-two male Division I football players initially met the study’s THC-screening and self-reported-use conditions. Eighteen were included in the paired analysis after maintaining program-verified abstinence between assessments, approximately 363 days apart. Thus the principal biological denominator is 18 repeat-tested athletes, not 22 completed participants, not 4,600 participants, and not hundreds of test responses. The reference dataset of 4,600 collegiate athletes supplied the scoring norms; its members were not randomized concurrent controls in this comparison.

At baseline, participants acknowledged daily or near-daily smoking and had screened positive during preseason onboarding at a reported threshold of at least 150 ng/mL. The paper does not provide an accompanying forensic venous whole-blood THC concentration, a same-day confirmatory toxicology series, or a detailed laboratory chain of custody for both visits. A screening threshold must not be silently relabeled as a blood concentration. Likewise, program verification of abstinence is described, but the report does not furnish a full independent specimen-by-specimen validation record for that year.

What changed after the year of abstinence?

Compared with the baseline assessment, the average overall S2 composite moved from −0.773 (SD 0.669) to 0.493 (SD 0.356). All 18 athletes improved on this composite, as presented in the paper’s Table 1. Standardized SPEED moved from −0.722 (SD 0.961) to 0.202 (SD 0.461); 15 improved and three worsened on that measure. These are changes in norm-referenced scores, not the number of seconds needed to recognize an actual on-field play.

Decision Complexity rose from −0.539 to 0.509, with 16 of 18 athletes improving. Distraction Control rose from −0.627 to 0.439, with 17 improving. Impulse Control rose from −0.281 to 0.495, also with 17 improving. Multiple-Object Tracking increased from −0.329 to 0.241, and Speeded Target Detection from −0.382 to 0.445. The researchers reported paired-test statistical support for those domains, including p < 0.001 for decision complexity, distraction control, impulse control and target detection, and p = 0.001 for tracking.

These are encouraging observations about measured laboratory performance. The strength of a paired design is that an athlete can be compared with his earlier self. Its limitation is that a year is also enough time for conditioning, coaching, experience, maturation, altered sleep, new habits, and familiarity with the instrument to change. Without randomized assignment or a comparable nonabstaining group, the improvement cannot be partitioned reliably among those possibilities.

Table 1. The actual tested athletes, variables, and measurements

Source: original Table 1. These are norm-referenced task scores from the same 18 athletes at two visits. Counts of improved athletes are descriptive, not independent treatment trials. The table does not report serum, plasma, whole-blood, urine, breath, or oral-fluid THC.

Was reaction speed actually 20% slower with cannabis use?

The authors describe reaction speed as approximately 20% slower at the first assessment, corresponding in their interpretation to roughly 80–100 milliseconds of decision-execution delay. They report a paired test statistic of t(17) = −5.84, p < 0.001. However, the principal source table presents SPEED as a standardized z-score, not a row of raw pre/post millisecond latencies from which every reader can recalculate that percentage or absolute time difference.

It would therefore overstate the disclosed measurement record to say that the study directly demonstrated an extra 100 milliseconds in every player, every play, every reaction or every driving decision. The millisecond figure is the authors’ approximate translation of a laboratory speed effect, not a validated forensic adjustment to an individual’s stopping distance or reaction-time reconstruction. The measured outcome was performance on the specific computerized assessment.

Even small delays can matter in time-limited sport. The sound scientific statement is that the S2 speed measure improved within this selected cohort; neither an independently sampled THC concentration nor actual NFL or NCAA play error rates were collected alongside the cognitive difference. The direction of a group-level change cannot furnish a precise numerical deficit for a defendant, employee, eyewitness or injured player.

Did all kinds of cognition recover equally?

No. The largest and most consistent gains were concentrated in the overall composite and selected speeded-control measures. Accuracy improved in ten athletes and worsened in eight, with its mean z-score moving from 0.160 to 0.332. Visual Learning moved from −0.482 to −0.246, and Perception Speed from −0.334 to −0.016. The original paper did not establish statistically significant improvement for those latter endpoints after the reported multiple-testing treatment.

Instinctive Learning also showed an increase in mean standardized performance but no statistically reliable adjusted signal. It is incorrect to turn an upward mean into a claim that every cognitive domain normalized, every player benefited, or overall intelligence recovered. A battery average can improve even if some constituent outcomes remain variable or statistically unresolved.

Improvisation requires particular care. The original Table 1 places a significance marker beside that measure, whereas one paragraph of the Results describes Improvisation among gains not surviving correction. A speed-adjusted model lists an Improvisation coefficient of 0.680 with p = 0.003. These are not interchangeable analyses; a careful interpretation does not silently reconcile inconsistent descriptions of the paired-test correction. The unambiguous message is that several distinct speeded-control measures improved, while some other measures did not show comparably supported change.

Does adjustment for general speed resolve the explanation?

The investigators also fitted mixed-effects models with athlete-specific intercepts and the overall SPEED variable as a covariate. Several domains retained associations with testing occasion after this adjustment. For example, Decision Complexity had an adjusted coefficient of 1.049 (95% CI 0.586–1.511; p < 0.001). Distraction Control had a coefficient of 1.066 (95% CI 0.548–1.583; p < 0.001). These values refer to changes in the battery’s modeled standardized-score scale, not milligrams of THC or a percentage increase in true field performance.

That analysis addresses an important alternative: perhaps everyone simply pressed buttons faster on the second occasion. It suggests the selected domain changes were not wholly described by the measured global SPEED score. But adjusting for one statistical covariate does not control for all changing conditions. Repeated exposure to the S2 tests, player development, coaching practice, sleep patterns, stimulant or caffeine use, and varying participation could still contribute to a measured difference. A model coefficient does not upgrade an observational pre/post comparison into a randomized cannabis-cessation experiment.

Table 2. Speed-adjusted findings and important null comparisons

Source: original Table 2, mixed-effects models controlling for SPEED. Confidence intervals and coefficients represent statistical outcomes for this sample. They are not individual impairment thresholds, measured cannabinoid levels, or a test of a legal standard.

Was there a comparison group, and what does it tell us?

The paper mentions an independent football reference sample of 146 athletes measured about a year apart. Its overall S2 performance was fairly stable across that interval, with an average shift of about +0.14 z-score and reported longitudinal correlation of r = 0.897. This reference is useful context: a large, stable comparison distribution makes the observed +1.266-point mean composite shift in the 18-person selected cohort noteworthy.

It is not the same as a contemporaneous randomized nonabstaining control group. The 146 comparison athletes were not described as being randomized to cannabis exposure versus abstinence, or matched to all baseline characteristics and changing training conditions of the 18. Their reference data cannot by themselves remove selection, retest, coaching, maturation, or confounding. The 4,600-athlete normative set serves a third, different function: standardizing scores. Neither reference sample should be counted as another 146 or 4,600 individuals receiving a studied abstinence intervention.

What did the screening and abstinence records establish?

At enrollment, the authors report a positive THC screen at ≥150 ng/mL and participants’ statements of frequent smoking. The methods do not supply a detailed paired analytical record specifying a validated confirmation method, exact specimen matrix, parent THC versus metabolite, blood collection times or individual pharmacokinetic curves. Nothing in this report supplies a directly measured venous whole-blood Δ9-THC concentration at either cognitive-testing visit.

That distinction matters because a positive drug screen may reflect a detection condition without measuring contemporaneous neurobehavioral function. The study’s score changes came from button-box performance, not an equation converting screening concentration to impairment. A drug-screening result cannot be repurposed as a quantified whole-blood level without evidence of the tested matrix, analyte and analytical method. See the site’s discussion of why biological matrices are not interchangeable.

Likewise, the investigators characterized participants as maintaining program-verified abstinence for twelve months, but did not publish the kind of sequential, individually authenticated toxicological documentation needed to reconstruct every date of last exposure. The source does not directly measure residual blood THC, oral-fluid detection, urine THC-COOH, hair incorporation, or brain THC over the follow-up. A claim about the entire route and time course of cannabinoid elimination would demand a different experimental design.

Does a better laboratory score prove better football or safer driving?

No. The battery attempts to sample skills relevant to reading defensive motion, selecting among options, tracking multiple objects and suppressing inappropriate responses. That construct validity gives the results practical relevance, but the investigators did not record real pass-rush recognition time, tackle mistakes, penalties, field injuries, pass completions or in-game collisions. They also did not conduct a road-driving protocol or assess an individual’s position against a legal driving-impairment threshold.

Improvement in a computerized decision task is not the same as a measured reduction in injury frequency or a causal explanation for a sports accident. In litigation, counsel may properly ask whether a specific person’s attention, decisions or reaction timing were affected. This study supplies neither an individualized baseline capacity nor exposure-measurement link capable of deciding that question. The difference between a measured task result and an asserted real-world outcome should appear explicitly in any expert assessment.

A driver who performs normally on one reaction-time task has not necessarily demonstrated intact divided attention, hazard perception or lane control. Conversely, a slower task performance is not proof that THC, rather than a host of other factors, produced a particular driving error. The study did not test either legal proposition, and transferring its group-level athlete results to a roadside driver would exceed the observed sample and endpoints.

Why do sample selection and testing conditions matter?

The 18 analyzed players were drawn from 22 initially eligible individuals. Repeated-test completion after a verified-abstinence program can select for adherence, motivation and other unmeasured traits. The reported mean age of the 18 was 19.6 years (SD 1.5). At that life stage, both ordinary maturation and sport-specific training may alter speed and executive skills over nearly a year. A pre/post comparison cannot separately measure each contribution without additional controls.

All athletes were men in one Division I program. Results cannot automatically be generalized to women, adolescents outside collegiate sport, older adults, people using prescribed cannabinoids, or drivers with medical conditions. The paper did not systematically establish exact THC dose, potency, smoking frequency beyond self-report, other route histories, or quantitative co-exposure to nicotine, caffeine and additional substances. Indeed, players were encouraged to test in the state in which they usually played, including permitted caffeine or prescribed ADHD medication, making a tightly standardized pharmacological exposure comparison unavailable.

These limitations do not erase the observed results. They define what those observations mean. The strongest defensible description is an association between an earlier frequent-use period and later higher scores during program-documented abstinence, observed in 18 selected male football athletes. The weaker and unwarranted description is a known percentage loss caused by THC in any athlete or a validated recovery period for every person.

How would a forensic examiner evaluate a comparable case?

The first step is to identify the actual proposition. Is the question past cannabis consumption, recent intoxication, impairment during a game or collision, reliability of a screening test, or recovery during a period of monitored abstinence? Each calls for different records. A complete chain of custody and documented specimen collection are material to an analytical claim; authenticated cognitive-test instructions and repeated scores are material to a performance claim.

The second step is to identify which physical specimen was analyzed, which cannabinoid or metabolite was measured, what analytical technique was used, and when samples were collected in relation to alleged events. A screening threshold without matrix, confirmatory method and timing is not a case-specific brain-exposure measurement. A useful overview of confirmatory analytical testing explains why a preliminary screen should not be described as a definitive whole-blood result.

The third step is to examine performance evidence independently: exact test scores, baseline capacity, fatigue, concussion history, medication, sleep, environmental demands, learning effects and task validity. A forensic opinion should not infer a legal blood concentration from a norm-referenced reaction score, or infer on-field causation from a positive toxicological screen alone. Where a person-specific impairment opinion is requested, the expert should distinguish observed facts, statistical association, assumptions and unmeasured information rather than treat all four as equivalent.

What is the appropriate evidentiary weight?

Ally and colleagues provide a useful preliminary observation: several S2 speeded-cognition domains improved substantially across roughly twelve months in a small selected cohort, with additional analyses suggesting changes not entirely reducible to the single SPEED score. The work is therefore relevant to designing prospective studies and identifying which laboratory constructs deserve attention.

It does not establish a causal effect of cannabis cessation isolated from time and training; it does not quantify venous whole-blood THC at testing; it does not validate an athlete-specific injury prediction; and it does not prove who was impaired at an incident. The original authors themselves note that the applied evaluation lacked a matched randomized control and detailed exposure history and did not capture actual field outcomes. No claim of universal impairment, permanent harm, absolute cognitive restoration or case-specific accident causation follows from the reported means.

Bottom line: Better computerized game-speed scores following a year of reported verified abstinence are a measured association in 18 athletes. They are not a THC blood test, not an individual forensic impairment diagnosis and not evidence that a particular sports event or collision was caused by cannabis. The complete original methods, study tables and discussion should be consulted before using any headline change as expert evidence: Ally et al., original primary study, DOI 10.3389/fspor.2026.1821656.

Frequently asked questions

1. How many athletes actually completed the study?

Eighteen male NCAA Division I football athletes completed the paired assessments. Twenty-two initially met study-entry conditions, but four did not enter the twelve-month paired analysis. The 146-player reference cohort and 4,600-player norming database are not additional abstinence-treatment participants.

2. Was THC directly measured in venous whole blood during the cognitive tests?

No such whole-blood measurement is reported. The baseline onboarding THC screen had a reported ≥150 ng/mL criterion, without the detailed specimen/analyte confirmation needed to equate it to a forensic blood result. S2 scores are cognitive-performance measures, not biological THC concentrations.

3. Did the researchers prove that quitting cannabis caused faster decisions?

No. They documented improvements across two assessments during a period described as program-verified abstinence. In the absence of randomized exposure conditions or a matched concurrent nonabstaining group, practice, training, maturation, motivation and other changes remain possible contributors.

4. Did reaction time improve by exactly 100 milliseconds for every athlete?

No. The authors described approximately 20% slower baseline reaction speed and an estimated 80–100 ms group-level decision delay. Their primary table gives standardized speed scores, not separate raw millisecond values for each participant or every scenario. The estimate is not an individualized time correction.

5. Were accuracy and visual learning significantly better after abstinence?

The source reports upward average changes, but Accuracy, Visual Learning, Perception Speed and Instinctive Learning did not have the same supported corrected improvements as the principal speeded-control measures. Results vary by domain; a better composite does not imply every individual ability changed significantly.

6. Can the S2 battery establish whether a driver was impaired by THC?

Not from this report. The S2 battery measured sport-relevant computer tasks. The study included no on-road driving, no person-specific validated DUI classification, and no paired quantified forensic whole-blood THC. A person’s alleged driving impairment requires independent incident, clinical and analytical evidence.

7. Does this study establish how long cannabis impairment lasts after stopping?

No. It compared two assessments about 363 days apart and did not map daily withdrawal, drug elimination, serial whole-blood THC, or performance recovery over intervening days and weeks. It cannot set a universal abstinence duration or reliable individualized return-to-fitness date.