Does Methamphetamine Use Disorder Predict Less Trust and More Punishment? What 102 Adults’ Tasks Showed

Short answer: In a 2025 Australian case–control study, adults recruited with methamphetamine use disorder returned a smaller proportion of simulated investment proceeds and selected more intense noise punishments than methamphetamine-naïve comparison participants. They also needed stronger happy facial cues to identify happiness. Yet the study found no statistically supported overall group difference in its cognitive- or emotional-empathy outcomes and no demonstrated extra punishment immediately after being punished. The comparisons concern laboratory social decisions associated with a clinical history—not experimentally administered methamphetamine, measured blood drug concentrations, violence in daily life, or proof of any person’s intentions.

The original primary study by Mahlberg and colleagues is Social cognition and decision-making in people with methamphetamine use disorder, published in Addiction (2025). The researchers tested 52 treatment-seeking participants with methamphetamine use disorder (MUD) and 50 eligible methamphetamine-naïve controls after exclusion of one initially recruited control. That is a nominal combined analytical cohort of 102 distinct people; different tasks had missing observations and therefore smaller usable samples. Read the original primary publication, DOI 10.1111/add.70108, for the complete protocol and statistical models.

What exactly did this research ask?

The project examined whether a history of MUD was associated with a distinctive profile across several social-cognitive operations. Recognizing an expression is not the same function as experiencing empathy. Empathy is not identical to allocating resources in a bargaining game. Selecting a punishment level in a computer task is also not the same as causing physical injury. Treating all four as a single measure of a purported “methamphetamine personality” would discard the experiment’s central scientific distinctions.

The investigators recruited the clinical group through Melbourne-area treatment services and the controls from community sources. Their design was observational and cross-sectional. It compares groups observed at an assessment, rather than assigning otherwise comparable people to take methamphetamine versus placebo. Associations may reflect substance-use history, treatment selection, current circumstances, differences in other substance use, psychological symptoms, demographic imbalance, or combinations of these factors. They cannot alone identify a chemical causal pathway from methamphetamine to interpersonal behavior.

Who were the independent participants?

Fifty-two people with moderate or severe MUD completed study procedures. Fifty-one controls were recruited, but one was excluded after recent methamphetamine use; the resulting comparison cohort was 50. The source’s abstract describes recruitment with 51 controls, whereas the Results and sample tables describe 50 retained controls. The distinction matters when interpreting how many independent humans actually contributed to an outcome.

The groups were similar in mean age and estimated verbal intelligence, but they were not interchangeable in every characteristic. The MUD group had a larger proportion of male participants among those with gender data: 39 of 51 (76.47%) versus 26 of 48 (54.17%), with a reported group test p = 0.021. Mean education was 13.08 versus 14.32 years (p = 0.048). Depression scores, reported psychotic experiences, anger expression and quality-of-life measures also differed. Gender and education were included in task models, but that adjustment does not eliminate all possible confounding from treatment status or comorbid factors.

Only 43 MUD participants had complete multifaceted-empathy data, three lacked trust-game data and one lacked each of the face-morphing and punishment-task records. Individual clinical-screening denominators also varied. A repeated game with 25 rounds does not create 25 new patients; neither do 13 face trials per condition create 39 independent observers. Statistical models appropriately sought to account for within-person observations, subject to their stated modeling assumptions.

What physical specimens were collected—and what was actually assayed?

The publication reports screening urinalysis as part of the clinical characterization. Among 50 MUD participants with methamphetamine-screening data, 41 (82%) had a methamphetamine-positive urine result. Other urine-screening categories included amphetamines (35/50), THC (11/50) and benzodiazepines (18/48). These are reported screening classifications with differing denominators, not a complete validated forensic assay package for this study’s behavioral endpoints.

Crucially, the researchers did not assign a known oral or injected methamphetamine dose, report timed quantitative methamphetamine concentrations in legal venous whole blood, or construct individual serum/plasma pharmacokinetic curves. Nor did they report quantified active methamphetamine in contemporaneous oral fluid or hair. A urine-screening positive may support exposure history within a method-dependent detection window; it does not yield a numeric concentration in another physical matrix or by itself establish the momentary neural or behavioral effect during a game.

For related methodological context, see the separately published biological-matrices guide and confirmatory drug-testing guide. Those pages explain why specimen identity, chemical specificity, collection time and analytical purpose must travel with any toxicology claim.

How did researchers measure a difference in recognizing happy faces?

The face-morphing task presented a digital face with a continuum from happy to angry and a neutral midpoint. Participants moved through a sequence of 100 expression frames and identified a neutral appearance, the first appearance of anger, or the first appearance of happiness. Each instruction appeared in 13 trials. The 0–100 values were constructed morph positions, not blood concentrations, probabilities of dishonesty or classifications of a particular witness.

After adjusting for gender and education, the MUD group selected faces containing more happiness as still neutral, with a model contrast of −5.05 morph-scale units (standard error 2.05; 95% confidence interval −9.12 to −0.98; p = 0.016). For identifying the initial appearance of happiness, the group contrast was −6.34 (SE 3.09; 95% CI −12.47 to −0.20; p = 0.043). These findings support lower sensitivity to subtle positive expressions within that task. They do not establish an inability to recognize all facial expressions.

The group comparison for anger sensitivity was not statistically significant: estimate 5.23, 95% CI −0.90 to 11.36, p = 0.094. There was also no statistically supported additional MUD-specific happy-by-group interaction in the broader mixed model (p = 0.67). Reporting the supported happy-face contrasts without the nonsignificant comparisons would make the evidence appear more sweeping than it was.

Were people with MUD less empathetic?

Not on the overall group contrasts reported for the study’s multifaceted empathy test. That distinction is essential. Participants encountered emotionally evocative images and completed two different operations: selecting a label corresponding to the depicted person’s emotional state, and rating how strongly they empathized with that person. The former is a cognitive-empathy accuracy outcome; the latter is a self-reported emotional-empathy intensity outcome.

For cognitive empathy, the model’s MUD group term was not statistically significant (p = 0.588). A significant positive-valence effect across participants (p < 0.001) did not amount to a between-group methamphetamine effect, and the positive-valence-by-MUD interaction was nonsignificant (p = 0.379). For emotional empathy, the group term was also nonsignificant (p = 0.27). The positive-valence-by-group term reached only p = 0.07, while a within-MUD positive-versus-negative contrast was p = 0.06.

Neither p = 0.07 nor p = 0.06 meets this paper’s conventional 0.05 threshold. Those values are not proof that the groups are equivalent, but they do not justify describing an established global empathy deficit. Missing empathy-task data and a relatively modest clinical sample further constrain confidence. A forensic assertion that methamphetamine use automatically renders a person incapable of feeling empathy would be inconsistent with this source.

What did the investment-return game actually measure?

Participants played the role of trustee in a simulated economic interaction. An ostensible investor received $20 in digital currency, decided how much to invest, and had that investment multiplied threefold before the trustee chose how much to return. Participants were told they were working with another player, but investor behavior came from an adaptive algorithm using historical human-response data. The task comprised 10 rounds per participant, not 10 independent people.

The MUD group returned a smaller proportion of the investment proceeds than controls, with an adjusted group estimate of −0.11 (SE 0.04; 95% CI −0.18 to −0.04; p = 0.002). This finding is a laboratory measure of returned share, often interpreted in the article as reduced interpersonal trust or prosocial reciprocity. It is worth preserving the behavioral mechanics: the person being measured was deciding how much to give back after an algorithmic investor had already invested. They were not simply deciding how much of their own money to entrust to a stranger.

Calling the output “less trust” can serve as the source’s interpretive shorthand; claiming it directly proves that a named person distrusts everyone or would misappropriate property does not follow. A computer-game return fraction does not adjudicate honesty, contractual intent, fiduciary conduct or the reasons a real person made a financial decision. Prior experiences, strategic game behavior and comprehension of payoffs can all influence such measures.

What counted as aggression in the study?

The competitive reaction-time task consisted of 25 interactions against a simulated opponent. Before each trial, the participant selected one of eight noise-punishment intensity levels to be delivered to the opponent if the opponent lost. Outcomes and opponent settings followed a programmed sequence. The dependent measure was selected sound-punishment intensity, not assault, threats, violent offending or injuries observed outside the laboratory.

The MUD group selected higher punishment intensities overall: adjusted difference 1.48 noise-intensity scale units (SE 0.38, 95% CI 0.73 to 2.23, p < 0.001). This is a relatively clear group separation within the simulation. However, its meaning must remain restricted to the actual experimental condition. The original data do not turn a quantitative game setting into the probability that any given participant will commit an act of interpersonal violence.

There was no statistically supported overall increment in punishment settings after the participant lost the preceding round (p = 0.144), and no group-by-prior-loss interaction (p = 0.64). Thus, while the MUD group set higher punishments on average, the experiment did not establish a distinctive immediate retaliatory escalation. Conflating higher overall selected intensity with confirmed trial-by-trial revenge is a specific and avoidable analytical error.

What is the evidence ledger for the principal social-task comparisons?

These estimates are not all expressed on a common scale. A facial morph unit, returned-investment fraction, and eight-level punishment setting are different constructed measures. The table therefore reports source-specific effect estimates rather than combining them into a composite “dangerousness” score or converting them into physical drug concentration.

Did the study find correlations among traits and outcomes?

Exploratory correlations related task results to other reported psychological measures. Emotional empathy for positive stimuli correlated negatively with anger expression (Spearman ρ = −0.37, p = 0.005). The proportion returned in the trust game correlated negatively with selected punishment intensity (ρ = −0.32, p = 0.02). More punishment was associated with more self-reported psychotic experiences (ρ = 0.29, p = 0.03) and lower quality of life (ρ = −0.29, p = 0.03).

The paper reports a false-discovery-rate approach to this correlation matrix. A correlation in a mixed clinical and control sample does not demonstrate a mechanistic chain in which an acute drug concentration causes a specific hostile act. The outcomes could move together because of shared clinical status or psychosocial differences. Their direction is also not proof that improving facial recognition necessarily improves trust, reduces punishment or prevents subsequent adverse events.

Which characteristics differed before considering the laboratory games?

The clinical group had substantially worse reported quality of life on multiple components and higher scores on several symptoms. For example, the total quality-of-life mean was 73.71 in MUD versus 104.58 in controls (p < 0.001). Mean anger-expression index was 36.90 versus 26.82 (p < 0.001). Persecutory-ideation scores were 9.56 versus 6.82 (p < 0.001), while the perceptual-abnormalities component did not show a statistically significant group difference (p = 0.40).

The sample was not a random draw from all people who have ever used methamphetamine. The MUD participants were treatment-seeking, and 47 of 52 (90.39%) met severe symptom-count criteria. Thirty-five of 50 tested positive for the broad amphetamine screen and 41 of 50 for methamphetamines, with possible co-exposures represented in the clinical table. These details reduce the defensibility of generalizing findings to one-time users, people in stable remission, workers using prescribed stimulants, or all defendants with a positive toxicology screen.

What denominators and comparison conditions must stay together?

Model degrees of freedom need not equal 102 minus a fixed number because task availability, repeated measures, mixed-model estimation and covariates affect the tests. A reader should look to the source’s Methods, participant flow and each task’s missingness rather than treating every reported coefficient as one identical full-sample comparison.

Can urine methamphetamine screening explain the social task results?

Not at the level needed for a causal toxicological attribution. The screening table does show substantial contemporaneous methamphetamine detection within the MUD cohort. But it does not present a validated quantitative concentration-to-task model, synchronized collection intervals to every trial, active enantiomer distinction, or an experiment showing that an administered dose changes the trust or punishment indices. A screening category can coexist with long-term use history, differential tolerance, concurrent medications, illness severity and other influences.

Whole blood, plasma, serum, urine and oral fluid do not supply interchangeable numerical concentrations. Even a confirmed quantitative drug concentration in an appropriate specimen would require timing, assay uncertainty and relevant effect-validation evidence before a specific behavioral inference. Here that chemically anchored dose–response analysis was not performed. The separate therapeutic drug-monitoring primer provides additional context on the relationship between timed concentrations, biological variation and the intended analytical question, but it does not supply the missing participant-level chemical measurements for this research.

Was this a trial of acute methamphetamine intoxication?

No. The contrast was between people with a diagnosed disorder and people reporting no lifetime methamphetamine use. The authors did not randomly assign healthy adults to methamphetamine, administer a fixed amount on the testing day or obtain blood concentrations over time to identify a peak response. Accordingly, the manuscript should be described as a case–control study of associations with MUD, not as a controlled dose experiment determining how an acute methamphetamine-positive person behaves.

This distinction is material in courtroom or investigative settings. A diagnosis spanning years, a urine-screening result indicating exposure, and the behavior at a specific time are separate propositions. To address an actual disputed event one would need records of that event, appropriate witness observations, specimen-specific analytical results when available, other health and medication context, and a validated method of interpreting those observations. This article does not supply that individualized bridge.

Could laboratory punishment intensity predict violence or dishonesty?

The noise-punishment game has experimental validity as a defined behavioral analogue, not as a calibrated actuarial measure of future assault. An intensity setting is not measured injurious force. A participant may respond to competitive framing, misunderstanding, incentives or the structure of a simulated opponent differently from a real interpersonal encounter. The confidence interval around an average MUD-versus-control difference cannot be read as the probability that a named individual will choose violence.

The trust game is similarly not a diagnostic test for fraud. The subject was the trustee allocating returns already generated through a programmatic investor. Whatever the normative interpretation of returning fewer resources, the experiment did not measure actual promises, fiduciary duties, intent to deceive, contractual facts, or real-world property transfers. The scientifically legitimate conclusion is narrower: compared groups made different task-specific allocation choices after statistical adjustment for two listed covariates.

How do null findings change the interpretation?

Null findings are not peripheral. They tell the reader which proposed generalizations the study failed to support. A group difference in noise-punishment intensity did not coincide with demonstrable increased punishment immediately after losing a trial. A happy-expression sensitivity contrast did not prove an equally strong anger-expression deficit. A group difference in a financial-return task did not imply broad cognitive-empathy impairment. The researchers therefore observed a patterned result, not a universal dysfunction across every aspect of social cognition.

At the same time, a nonsignificant p value is not proof that two populations are exactly equal. Confidence intervals, sample size, measurement range and model design still matter. For example, the anger-face CI extended from −0.90 to 11.36 and the group emotional-empathy CI from −0.41 to 1.45. Both include zero but span nonzero possibilities. The evidence should be presented as no statistically supported difference in this study, not as a demonstrated absence of all clinically meaningful effects.

What limitations are most consequential for causal and forensic interpretation?

First, treatment-seeking ascertainment creates a selected population; participants may differ from non-treatment-seeking users in illness severity and comorbidity. Second, controls and cases were not perfectly balanced on sex or education, and case participants had more depression, perceived psychotic experiences and anger symptoms. Covariate adjustment cannot be assumed to eliminate all residual confounding, especially when other relevant social and medical differences were not experimentally controlled.

Third, outcomes were measured using specialized computer paradigms with fixed stimuli and simulated actors. Generalization beyond the task conditions is an empirical question, not an assumption. Fourth, missing participant-level task observations varied among tests, particularly the empathy battery. Fifth, the analysis was explicitly exploratory and not preregistered; multiple related hypotheses and correlations create additional interpretive caution even where adjusted values are reported. Finally, the absence of timed quantitative drug specimens disconnects these observations from a pharmacokinetic or concentration-effect model.

What would a stronger follow-up study need to measure?

One approach would recruit a broader and better-characterized sample across different stages of use and recovery, with prospectively specified outcomes, careful control of mood, psychosis, medications, education and socioeconomic variables, and replication in an independent cohort. Repeated measures across time could help determine whether a particular participant’s scores change with abstinence or treatment. Direct observation of real social outcomes would then be needed to study external validity, rather than treating noise intensity or game returns as stand-ins for crimes.

If the research question truly concerns acute drug effect, that demands a different design: ethically appropriate administration or tightly time-resolved naturalistic sampling with directly validated chemical measurement, matrix identification, timing and uncertainty, alongside control conditions. Even then, a population-average association would not automatically supply an individual legal threshold. A controlled intervention could improve causal identification for a particular task while still leaving real-world behavior and event causation untested.

What are the precise conclusions that can be carried into forensic discussion?

The defensible summary is that a group of treatment-seeking people with MUD showed, relative to retained methamphetamine-naïve controls, less sensitivity to subtle happy facial expressions, lower returns in a trustee resource-sharing game and greater selected simulated noise punishment. The paper did not establish global group deficits in the measured empathy domains or greater trial-by-trial revenge following a loss. The measured outcomes are laboratory decisions and self-report/psychophysical responses, not chemically assayed impairment or actual violent incidents.

Any use of this paper should state the comparative nature of the data, the task-specific numeric results, the nonsignificant findings, the participant and trial denominators, and the absence of quantified postdose blood drug exposure. Doing so allows the study to inform scientifically careful questions about social cognition without converting clinical diagnosis or drug-screening status into a presumption about a particular person’s honesty, emotional capacity, guilt, or dangerousness.

Frequently asked questions about the study

1. Did the study show that methamphetamine makes people violent?

No. The measured difference was higher noise-punishment settings in a computer competition among people with a clinical diagnosis, not actual violent behavior after experimentally taking methamphetamine. The study was cross-sectional and did not randomly assign a dose. The group estimate cannot classify a person’s likelihood of assault.

2. Did people with MUD have a proven empathy deficit?

Not in the overall measured cognitive- and emotional-empathy group contrasts. The source reported p = 0.588 for cognitive empathy and p = 0.27 for emotional empathy. Some emotional-valence patterns were suggestive but did not meet its p < 0.05 criterion. Non-significance does not itself prove equivalence.

3. Were users more likely to retaliate after losing the previous round?

The study found higher punishment intensity overall in the MUD group, but the previous-loss effect (p = 0.144) and previous-loss-by-group interaction (p = 0.64) were not statistically supported. Describing the finding as increased immediate revenge would misstate the particular statistical tests.

4. Does a positive urine result explain an individual’s trust score?

No. Screening urine was reported for clinical characterization, but the study did not establish a quantitatively confirmed methamphetamine concentration–trust relationship or the time of a participant’s precise exposure. A urine-positive classification cannot be translated into whole-blood concentration or a causal effect on a trustee game’s return decisions.

5. Did the participants distrust an actual human investor?

They were told another player invested, but the investor’s decisions were simulated by a computer algorithm informed by historical responses. The experimentally observed choice was the proportion of investment proceeds the participant returned. That is narrower than proving how they would trust another identifiable person in ordinary financial dealings.

6. Can these numbers establish that a defendant intended to deceive or assault someone?

No. The study reports average adjusted differences across groups on abstract experimental tasks. It does not assess a defendant, adjudicate motives, reconstruct a disputed incident, evaluate eyewitness accuracy or validate a diagnostic threshold for individual intent or violence. Case-specific evidence would be essential.

7. Where can the original primary research and its limitations be checked?

The complete report is Mahlberg and colleagues, Addiction (2025), DOI 10.1111/add.70108. Its Methods, Tables 1–3, figures, model notes, discussion and published supplementary-method references define what was tested. The open-access original publication should govern any quotation, exact variable definition or effort to reproduce the analysis.