Evidence.

Stimulants reduce ADHD symptoms acutely vs placebo with effect sizes of 0.55–0.84 on teacher and clinician ratings, but the 14-month MTA advantage disappeared by 36 months, long-term functional outcomes do not differ by initial treatment assignment, and parent-teacher rating disagreement is not measurement error—it reflects situation-specific behavior across settings.

Guanfacine and clonidine show child clinician SMDs −0.67 and −0.71 (Cortese 2018), but atomoxetine and viloxazine carry boxed warnings for suicidal ideation. Adult lisdexamfetamine Castells 2018 SMD −1.06; bupropion Verbeeck 2017 pooled −0.50 but clinician-only ns. Adult ER-MPH Boesen 2022 investigator SMD −0.42; IR-MPH Cândido 2021 investigator MD −20.70 but participant ns. Adult GXR Iwanami 2020 Japan diff −4.28; child GXR Sallee 2009 LS-mean −5.41 to −7.88.

Prospective cohort studies find that childhood ADHD persistence to adulthood is low across named cohorts (Dunedin 5%, Pelotas 17.2%, E-Risk 21.9%), most adult-ADHD cases lack prospectively diagnosed childhood ADHD (Dunedin 90%, Pelotas 84.6%, E-Risk 67.5%), and after excluding substance use and other disorders, fewer than 1% of originally undiagnosed children meet late-onset criteria.

Antidepressants work modestly better than placebo, publication bias inflated their apparent efficacy by 32%, and the serotonin-deficiency model lacks consistent evidence—but neither the chemical-imbalance story nor the 'antidepressants don't work' narrative survives contact with the data.

Forty years of looking for a pathophysiology of anxiety — and what the well-powered nulls, demonstrated confounds and outright reversals actually leave standing. Companion to the stress critical review.

What the evidence actually supports about how chronic stress degrades energy and functioning — and why the folk model of cortisol exhaustion and depletable willpower is wrong. Companion to the anxiety critical review.

Benzodiazepines reduce anxiety symptoms versus placebo in short-term trials (SMD −0.58, low certainty), but physical dependence can develop within days to weeks (FDA FAERS: onset 'days to weeks,' withdrawal 'weeks to years'). Hip fracture risk is substantial (RR 1.52 overall, RR 2.40 new users). Dementia association (HR 1.60, observational) cannot establish causation. Among incident users, 39.4% → long-term use; 50% of patients dispensed oral benzodiazepines received them ≥2 months in 2018.

All antipsychotics beat placebo in acute trials (SMD 0.33–0.88), but 74% of CATIE participants discontinued within 18 months, dose reduction in first-episode patients doubled 7-year recovery rates versus maintenance treatment, and long-term observational data show better outcomes off medication—confounded by selection, but challenging the 'indefinite treatment for all' assumption.

Lithium and valproate work comparably for acute mania (no head-to-head difference), lithium shows superior maintenance versus valproate, suicide prevention evidence is contested (early meta-analyses significant, Riblet 2022 not significant), lamotrigine prevents depressive but not manic episodes, and the folk models—lithium as miraculous or lithium as toxic relic—both overclaim what opened trials support.

ADHD prevalence is stable at 5.3–7.6% when standardized procedures are used, but diagnostic thresholds—symptom count, impairment requirement, informant rules—determine who gets counted. The youngest children in a classroom are diagnosed 60% more often than the oldest, not because of neurological differences, but because teachers compare them to older peers. The evidence refutes both 'diagnoses are exploding' and 'diagnoses are fake.'

Recognizable side effects unblind psychiatric trials: appetite suppression, sleep disturbance on stimulants; SSRI sexual dysfunction. Active-placebo trials mimicking side effects show SMD 0.17 versus 0.39 with inert placebo. Kirsch's severity interaction is real—placebo response drops in severe depression, drug response stays flat. Publication bias inflated efficacy by 32%. Neither 'it's all placebo' nor 'double-blind means the estimate is clean' survives.

Waitlist controls inflate psychotherapy effect sizes by 50%: depression versus waitlist g=0.95, versus care-as-usual g=0.63. Response rates are modest: 24-42% across disorders. Head-to-head trials show medication 46.4%, psychotherapy 46.3%, mixed controls 24.4% remission. Common factors account for large variance. Treatment differences exist for some disorders, not others. Neither 'huge effects' nor 'just talking to a friend' survives.