Can Inconsistent Reaction Times Predict Simulated Driving Problems in Methamphetamine Dependence? A 90-Man Study

Answer first: In an observational study of 90 men, 35 with a history of methamphetamine dependence and 55 comparison participants, reaction-time consistency on a 14-minute vigilance task was worse in the methamphetamine group. Within those 35 men, greater response-time variability predicted more simulated driving crashes and traffic penalties and poorer performance on a laboratory everyday-skills assessment, even after accounting for mean response speed and recency of self-reported methamphetamine use. But the direct between-group comparison of the simulator crash-and-ticket score was not statistically significant (p = .07), and the study did not chemically measure methamphetamine in legal venous whole blood. The association is informative about a laboratory cognitive marker; it does not validate a drug-blood cutoff or determine why any individual road collision happened.

The primary evidence is Morgan, Doyle, Minassian, Henry, Perry, Marcotte, Woods, Grant and the Translational Methamphetamine AIDS Research Group, Elevated intraindividual variability in methamphetamine dependence is associated with poorer everyday functioning, Psychiatry Research 220 (2014):527–534, original DOI 10.1016/j.psychres.2014.07.008. The distinction between a group difference, a within-group statistical association, and an unmeasured personal impairment finding controls the interpretation below.

What, precisely, did this study establish?

The central measured phenomenon is intraindividual variability (IIV): fluctuations in a single participant’s speed of responding to repeated targets. The researchers measured those fluctuations while men performed a sustained-attention computer task, then examined whether unusually variable responding accompanied worse functional performance. The methamphetamine-history group had a higher IIV score than the comparison group, and higher variability within the methamphetamine group tracked multiple simulated or laboratory outcomes. Neither observation is the same as demonstrating a pharmacological effect of a known contemporaneous dose.

A particularly important limit is that the group differences were selective. The groups did not differ significantly on mean CPT reaction time (p = .12), reaction-time change across task blocks (p = .26), signal detectability (p = .26), omissions (p = .47), or commissions (p = .37). The increased variability is therefore not a general statement that every aspect of attention was impaired. Likewise, the observed functional associations were not uniformly positive across self-reported everyday difficulties. Statistical selectivity is part of the result rather than an inconvenient exception.

Who was included—and why are there only 90 independent people?

Investigators studied 35 methamphetamine-dependent men and 55 men without a history of methamphetamine dependence, recruited from San Diego community and treatment settings as a subset of a larger research program. The total biological sample was 90 men, not 90 independent driving events, not hundreds of people because the attention task repeated targets, and not an additional sample for every regression. Thirty-five men constituted the methamphetamine group in all principal within-group analyses, subject to any outcome-specific missingness not resolved in the manuscript.

Only men were included because the available female methamphetamine subgroup was too small to support the intended analyses. The groups were not randomly assigned to use drugs. The methamphetamine group had fewer years of education, more depressive symptoms, and more remote other-substance diagnoses. HIV prevalence was comparatively similar across the two cohorts and was considered in statistical adjustment. These facts matter because the research estimates associations among selected people, not the isolated effect of randomly administered methamphetamine.

What exposure did the investigators actually observe?

Methamphetamine dependence was assessed through structured psychiatric history and a timeline-follow-back interview. Participants reported lifetime patterns and recency of use; the median interval since last methamphetamine use was 61 days, with an interquartile range of 21 to 243 days. Recent-use grouping subsequently divided the 35 men into 10 whose reported last use was within 30 days, 16 more than one month but no more than six months earlier, and 9 at least six months earlier. Those bins describe the study sample, not biologically confirmed last-use dates.

The research team did not give the men a controlled methamphetamine dose. Breath alcohol screening and qualitative urine drug screening had to be negative on the assessment day. A negative screening result under a particular testing protocol cannot independently verify a month or six months of continuous abstinence, and a reported use interval cannot be converted into a verified dose clock. The paper contains no timed mass-spectrometric series of parent methamphetamine concentrations for these participants.

How was reaction-time variability measured?

Participants took the Conners Continuous Performance Test II (CPT-II), a roughly 14-minute computerized vigilance task. A letter appeared for 250 milliseconds per trial. The instruction was to respond rapidly to every letter except X; withholding response to X required inhibition. Interstimulus intervals varied among one, two, and four seconds, and six larger blocks each included smaller groups of trials. Accordingly, a participant could exhibit a normal average response speed while producing an uneven series of fast and slow responses across targets.

The study analyzed “Hit SE per block” as an index of within-block response stability and the standard CPT-II “Hit SE Variability” T-score as its summary of fluctuations across sub-blocks. A larger score signifies less consistent timing. This is a psychometric score derived from repeated button presses, not an actual speedometer measurement, a blood drug concentration, or a continuous monitor worn while driving. Mean hit reaction time, by contrast, summarizes central response latency and can hide alternating very fast and slow trials.

Why variability and mean reaction time are different scientific questions

Imagine two people who average the same latency over a succession of target presentations. One consistently responds close to the average; another alternates between unusually rapid and delayed responses. They can have similar mean response times but different variability. A sustained task may reveal the second pattern even if a small set of discrete responses does not. That is why the researchers separately controlled for mean reaction speed when relating variability to functional outcomes.

In this sample the mean hit-reaction-time T-scores were 48.5 in the comparison group and 51.9 in the methamphetamine group, with p = .12. Hit SE Variability T-scores were 48.1 versus 53.0, with the source table reporting p = .04. The results support a narrower distinction between average speed and timing stability. They cannot establish that the variability in a particular participant was caused by recent methamphetamine, nor can a T-score be converted into a millisecond delay at a specific traffic signal without the original task-level records.

What did the unadjusted group comparisons actually show?

The source’s table of observed outcomes reported several modest or nonsignificant contrasts. The methamphetamine-history group showed more subjective cognitive complaints: mean Patient Assessment of Own Functioning Inventory score 5.9 versus 2.8 (p = .01). A daily-living complaint severity comparison was also significant (p = .02), although both groups had a printed median of zero, illustrating why distributional tests cannot be reduced to comparison of two medians.

On the UCSD Performance-Based Skills Assessment, which approximates tasks such as communications, household activities, finances and transportation, the group means were 91.6 versus 96.5 and the direct group test had p = .08. In the driving simulator, median total crashes and tickets were 4 versus 3, but the direct group comparison had p = .07. Neither should be described as a statistically significant between-group result at a conventional .05 threshold. The within-methamphetamine predictive analysis asks a different question and must not be substituted for either nonsignificant group comparison.

Evidence table 1: the source’s different group-level outcomes

The T-scores are standardized test scores, not reaction times in milliseconds. The source printed some means with standard errors and some measures as medians with interquartile ranges; only the central values appear here to avoid mixing distributional formats. The p-values are the original published results, not newly calculated tests. The same participants contributed several measures and therefore cannot be counted repeatedly as independent human samples.

What did the driving simulator ask participants to do?

The simulator supplied a steering wheel and pedals, a route to complete within approximately ten minutes, and an incentive to collect a modest reward. Drivers could lose portions of it through simulated crashes with hazards, speeding, or running traffic signals. The source combined simulated crash events and ticket-like penalties into a “Total Crashes and Tickets” (TCT) count. It was a constrained, incentivized computer simulation—not naturalistic monitoring of travel on public streets, a validated clinical drug impairment examination, or an adjudication of an actual traffic event.

This endpoint combines different event types into one unweighted count. Two virtual speeding penalties are not physically equivalent to two real vehicle collisions, and the published model does not demonstrate a rate of injury among these men. The game-like incentive and time limit may also influence strategies. The results can provide evidence about within-task behavior under those particular conditions, but they cannot identify a universal exposure-to-crash risk function or establish how one named driver would behave in an unobserved road scenario.

How strongly was variability associated with simulated crashes and tickets?

In the 35-man methamphetamine group, the investigators regressed simulator TCT on the variability score while accounting for mean hit-reaction time and recency of self-reported use. The overall adjusted model explained a reported adjusted R² of .40 (model F = 7.78, p = .001), and Hit SE Variability had a standardized model coefficient B = +.72, p < .0001. Larger timing variability thus tracked more penalties within the methamphetamine group, after those statistical adjustments. This was an association, not a randomized manipulation of variability.

Critically, the table comparing all 35 methamphetamine participants with the 55 controls listed median simulated events 4 versus 3, with p = .07. It is scientifically incorrect to change that nonsignificant group contrast into a positive result merely because a within-group model was strongly associated. Conversely, a nonsignificant between-group TCT contrast does not erase the observed within-group regression. They condition on different hypotheses and reference populations. Both must appear in a complete account.

Did variability also track practical living skills?

Yes, within the methamphetamine group, higher variability was associated with worse laboratory functional skills. The adjusted UPSA regression had adjusted R² = .19, model F = 3.41, p = .03, and a variability coefficient B = −.58 with p = .005. The negative coefficient reflects lower scores on a scale where higher performance means better functional capacity. The study adjusted for mean reaction time and methamphetamine-use recency in this performance-based model. Those covariates do not eliminate every unmeasured difference in education, motivation, mental health or daily experience.

The direct group comparison of UPSA was still nonsignificant at p = .08. This is the same inference problem as with simulated driving: a significant within-clinical-group predictor does not show that a drug-dependent group’s average performance differs significantly from a control group’s average. Nor does an association between two tests prove that variability caused poor financial management or transportation skills in everyday life; the measured UPSA outcome was a laboratory capacity assessment.

Which self-reported functioning findings were significant?

Higher variability within the methamphetamine group predicted a larger number of cognitive complaints on the PAOFI questionnaire. The multivariable model reported adjusted R² = .23, F = 4.1, p = .02, with variability coefficient B = +.56, p = .002. The authors included mean RT and depressive symptom burden in those self-report analyses. Greater instability on the task aligned with more complaints about cognitive performance in everyday life, but complaints are self-reports, not corroborated records of particular failures.

The parallel model for instrumental activities of daily living (IADL) complaint severity was genuinely nonsignificant: adjusted R² = .01, model p = .36 and the variability term p = .12. This null is important because the article’s broad “everyday functioning” language can otherwise sound as if each domain showed a relationship. Self-report difficulties with memory and attention, perceived ability to shop or manage medication, and directly observed functional tasks are related but nonidentical constructs. Positive findings for one cannot be silently assigned to another.

Evidence table 2: adjusted models, recency, and real null findings

The regression coefficients and p-values are taken from the original manuscript. Clinical subgroup sizes are people, not simulated crash-event denominators or independent signal measurements. The abstinence-binned analysis used the same 35 methamphetamine-history men already included above, not an additional experimental cohort. Comparisons were numerous and not presented as a validated prospective prediction algorithm.

Does a relation to abstinence establish when methamphetamine was taken?

No. Among the 35 men, the interval since the last reported use was inversely related to IIV (Spearman rho = −.37, p = .03). The ten men reporting use within 30 days showed larger variability than the 16 reporting an intermediate duration and the nine reporting at least six months; omnibus F(2,32) = 4.81, p = .01. The two more distant abstinence groups were not significantly different (p = .69). That describes a cross-sectional association across self-reported histories, not an experimentally observed recovery trajectory in the same person.

No one was repeatedly assayed for months to document continuously decreasing drug exposure and improving vigilance. The negative testing requirement on the assessment day does not transform self-reported abstinence into chemically verified last-use time. Differences in baseline cognitive functioning, living conditions, treatment engagement, depression, HIV care, and other exposures may correlate with both reported recency and task performance. A finding that a group with more recent reported use fluctuated more cannot date an unknown specimen or identify whether an individual was using methamphetamine during a specific road event.

Were the results explained by HIV or global cognitive scores?

HIV was intentionally represented in both groups. The reported methamphetamine-by-HIV interaction on variability was nonsignificant (p = .55); the source also reported no trial-block-by-group interaction. Those null terms narrow interpretation but do not prove HIV has no relationship with vigilance under other conditions. The study was not designed or powered to estimate all possible HIV-related effects, particularly across immune disease stages. It described the HIV comparison as one confound-sensitive component of the analysis rather than a separate trial of an HIV treatment.

Global cognitive impairment score and mean CPT response speed were included as covariates in the principal analysis, and both related to variability. The published cognitive domain comparisons in Table 1 were mostly nonsignificant between clinical groups. Within the methamphetamine group, variability correlated with executive-function deficit score (rho = .34, p = .04) and processing-speed deficit score (rho = .35, p = .04), but not every other domain. Multiple marginal comparisons and small subgroup sizes temper mechanistic and clinical extrapolation.

How should an original statistical-reporting inconsistency affect interpretation?

The manuscript reports a main methamphetamine-group effect in its adjusted repeated-block analysis with F(1,83) = 0.08 and p = .03. Taken literally, those two values do not mathematically correspond under the usual F distribution; the published text does not provide the correct replacement value. The independent group-summary table separately reports higher Hit SE Variability at p = .04, and the figure illustrates separation across blocks. Those additional pieces support discussing the observed direction, but they do not license silently repairing the internally inconsistent adjusted F test or treating it as independently verified.

That inconsistency is a reason to keep original effect estimates, model definitions and primary tables separate in any expert explanation. Exact adjusted inference would require the original analysis output or a published correction. The article still supplies useful measured results and tested nulls, but the discrepancies prevent the strongest possible claim about an adjusted biological group effect. Recalculation from an assumed F-value or invented raw observations would be less reliable than naming the uncertainty.

Why the biological specimen question is separate from these findings

The investigators screened for recent substances using a breathalyzer and urine toxicology, but did not publish confirmatory mass-spectrometric quantitative methamphetamine or amphetamine concentrations measured in legally collected venous whole blood. Nor did they publish a separate human plasma or serum concentration series, quantified oral-fluid or hair drugs, sweat or vitreous chemistry, validated enantiomer-specific testing, or a synchronized toxicological drug-versus-performance trajectory. “Methamphetamine dependence” in this paper is a clinical-history classification, not a measured parent-drug result.

Urine screens are fundamentally different from separated plasma, serum, or unseparated whole blood, both in analyte behavior and interpretation. A blood concentration is not a CPT-II response-variability T-score, and a simulator crash-event count is not a metabolite. The site’s verified biological-matrices explanation and confirmatory-testing guide address the distinction between a tested material, analytical method and ultimate forensic opinion.

What can this study say about a particular collision or driver?

It can say that within this selected male clinical group, more variable computerized vigilance responding correlated with worse performance on a ten-minute simulated driving task. It cannot identify how a specific driver behaved during a crash. No participants were observed in an actual road collision as the outcome, and no event-specific exposure timing, vehicle dynamics, distraction, sleep history, medical status, or witness observations were modeled as real-case inputs. A simulator challenge can be hypothesis-generating for safety research, not a legal adjudication instrument.

Even if a person has a known methamphetamine history and a blood result from another case, one cannot insert that person into the Morgan model without an independently validated individual prediction equation and test results collected under comparable conditions. This paper reports relationships among laboratory measures, not a validated impairment cutoff, accident probability for named individuals, criterion validity in forensic DUI cases, or testimony about a particular witness’s honesty. Such conclusions demand distinct evidence of the event and analytical record.

What confounding and selection limits remain?

This is an observational subset of a larger program, with 35 methamphetamine-dependent men and 55 comparison men. Years of education, depression, and other substance-use diagnoses differed. Model adjustment can reduce some measured confounding, but it cannot guarantee that unmeasured sleep quality, motivation, nicotine use, other treatments, housing instability or psychiatric symptoms explain none of the observed relationships. The study excluded some neurological and psychiatric conditions and people with positive same-day screens, so the recruited cohort is not representative of all acutely intoxicated people or emergency forensic cases.

The same 35 methamphetamine-history men contributed to multiple outcome regressions, creating related tests rather than independent replications. The model’s adjusted R² describes sample fit conditional on included variables; no independent external sample was presented to establish reproducibility or individual calibration. The investigators analyzed self-reported abstinence intervals retrospectively and did not assign participants randomly to different use histories. These limits keep causal claims and long-term recovery forecasts provisional.

Which inferences remain unsupported or genuinely unmeasured?

The reported IIV association does not establish that methamphetamine caused persistent structural brain damage, that an individual has permanent executive dysfunction, or that the next act of driving will be unsafe. The study did not chemically quantify drug dose in a legal blood specimen, independently time last exposure, assess real roadway crashes as outcomes, test field sobriety signs, determine personal witness credibility, or replicate an individual model across independent forensic datasets. These are unmeasured endpoints, not nonsignificant test comparisons.

The actually tested nonsignificant findings include several CPT dimensions, the direct group UPSA and simulated TCT contrasts, the within-MA IADL complaint regression, the HIV interaction, and the absence of a significant difference between the two longer abstinence groups. Conflating these tested nulls with entirely unobserved drug concentrations or incident outcomes obscures the original evidentiary boundaries. A scientifically responsible account preserves both kinds of limitation and avoids claiming proof of exact equivalence from a nonsignificant p-value.

Frequently asked forensic-science questions

1. Does the study show that all methamphetamine users have impaired reaction time?

No. It reports higher within-person variability as a group tendency, while the direct mean reaction-time contrast had p = .12. A selected group result is not a diagnosis for every person.

2. How can a nonsignificant simulator group difference coexist with a significant predictor?

The median TCT direct group contrast p = .07 compares the two groups. The B = +.72, p < .0001 adjusted model instead relates variation within the 35 methamphetamine-history men to their simulator penalties. They test different hypotheses.

3. Did 90 people experience real vehicle crashes?

No. Ninety men contributed to clinical and cognitive analyses; the crash and ticket score concerns virtual events in an incentivized ten-minute simulator. Trials and simulated penalties are not 90 independent road crashes.

4. Were methamphetamine concentrations measured in legal human venous whole blood?

No. Participants underwent same-day urine and breath screening, but no published quantitative confirmatory methamphetamine mass-spectrometric whole-blood, plasma, or serum concentrations were paired with task outcomes.

5. Does the association with 30-day reported use determine last-use time?

No. Self-reported timing, an eligibility screen and three retrospective intervals do not establish an individually validated toxicological use clock. The longer two abstinence bins were not significantly different, p = .69.

6. Did every everyday-functioning outcome significantly track response-time variability?

No. Within-group UPSA, simulator TCT, and PAOFI associations were significant, but the IADL complaint model had p = .36 overall and p = .12 for variability. Direct group UPSA p = .08 and TCT p = .07 were nonsignificant.

7. Can an expert infer why a named driver crashed or whether a witness is truthful?

No. The source contains no event reconstruction, individual blood-concentration–effect validation, actual crash outcome, or eyewitness-truthfulness test. Any case-specific opinion needs independent specimen and event evidence.

Bottom line

Morgan and colleagues’ original 90-man observational study distinguishes consistency of attention from average reaction speed. Methamphetamine-history men had higher CPT-II variability, and variability within that group predicted more simulated driving penalties, lower laboratory skills performance and more perceived cognitive problems. Equally central are the nonsignificant direct simulator and UPSA group comparisons, the null IADL model, and the manuscript’s inconsistent adjusted F/p pair. These facts cannot be collapsed into a forensic blood threshold or causal explanation of a particular human act.

The proper conclusion is narrower: a repeated-response cognitive marker was associated with selected laboratory outcomes among men with a history of dependence, amid clinical confounds and without contemporaneous quantitative blood-drug testing. For exact participant definitions, published comparisons and original limitations, consult Morgan and colleagues’ complete primary research record, DOI 10.1016/j.psychres.2014.07.008.