Figure 1

The atlas

Click a mark to open the variable: its edges, the sign of each arrow, and the papers. Drag a mark to move it. Click a curve to open the edge. Hover for a short card. The "Strata" layout groups nodes by layer. The URL is the view.

32 edges in this cut — click a mark, a curve, or a row

1. SMOKE+ INCREASESBreastfeeding → CognitionIn three European birth cohorts, longer breastfeeding tracked with higher child IQ on crude models. After maternal IQ, education, and class, Generation R's 0.32 IQ points per month fell to 0.09 and crossed zero. Sabadell's full breastfeeding past six months still sat 7.4 McCarthy points higher. Nobody randomized the feed.
2. SMOKE~ NON-LINEARCaffeine → MortalityLi 2020 pooled 21 cohort articles (10,103,115 people, 240,303 deaths) and found a nonlinear inverse between coffee and all-cause death: three cups a day against none or rare sat at relative risk 0.87 (0.84 to 0.89). Tsujimoto 2017 saw the same U in caffeine milligrams. Nobody assigned a dose.
3. SMOKE0 NULLDHEA-S → StatusDHEA-S on an optimization panel is not a status intervention.
4. SMOKE? CONDITIONALE2 → DominanceCycle-phase 'preference for dominance' is a contested meta-finding, not a hormone dial.
5. SMOKE− DECREASESAerobic → MortalityMartinez-Gomez 2024 pooled 2,011,186 adults in four megacohorts and found that meeting recommended leisure-time activity tracked a 22 percent lower all-cause death rate (hazard ratio 0.78, 0.77 to 0.79). Nobody randomized walking against death.
6. SMOKE+ INCREASESPrenatal famine → HOMA-IRPeople exposed to the Dutch Hunger Winter in utero had higher 120-minute glucose at age 58, and a smaller IVGTT subsample showed worse glucose tolerance after mid-gestation famine. Nobody randomized a siege.
7. SMOKE0 NULLIncome → HypergamyAs the income gap closes, strict economic hypergamy yields to educational and cultural homogamy. Incentives, not a frozen instinct.
8. SMOKE? CONDITIONALIncome → RiskHolt–Laury measures the preference. Income changes the budget. Do not call either 'testosterone'.
9. SMOKE0 NULLIncome → Youth pref.Richer men can act on a preference that poorer men also often hold. Money changes the feasible set, not necessarily the utility function.
10. SMOKE? CONDITIONALJob strain → CortisolHair cortisol ran higher with effort-reward imbalance among 81 high-workload workers, and morning saliva ran 21.7 percent higher in 48 high-strain teachers, but Karasek job strain tracked a lower diurnal cortisol area in 254 matched MESA adults. Nobody assigned the job.
11. SMOKE+ INCREASESJob strain → DepressionMadsen 2018's IPD-Work pool of 120,221 employees found a 27 percent higher rate of first hospital-treated depression after job strain (HR 1.27, 1.04 to 1.55). In Smith 2012, 3,735 Canadians free of depression who reported a rise in psychological demands had later depression odds of 2.36. Nobody assigned the job.
12. SMOKE− DECREASESLead → CognitionHigher childhood blood lead tracked lower IQ in Rochester, the Lanphear pool, and Dunedin at age 38 after maternal IQ was in the model. Nobody randomized lead.
13. SMOKE+ INCREASESLoneliness → DepressionHysing 2026 followed 10,460 Norwegian university students: those in the highest fifth of a 2022 Three-Item Loneliness Scale score, against the lowest fifth, had more CIDI major depressive episodes a year later (relative risk 2.02 in women and 2.64 in men after age and baseline Hopkins symptoms).
14. SMOKE+ INCREASESLoneliness → MortalityWang 2023 pooled 90 prospective cohorts in 2,205,199 adults and found loneliness associated with a 14 percent higher all-cause death risk (effect size 1.14, 1.08 to 1.20). Nobody randomized loneliness.
15. SMOKE+ INCREASESTraffic noise → CV eventsIn a 2023 dose-response meta-analysis of 23 cohort papers and 18.5 million adults, mostly in Europe, each extra 10 dB of modelled road-traffic noise (as Lden) sat with a 2 percent higher risk of incident cardiovascular disease (RR 1.020, 1.006 to 1.035). Nobody randomized the street.
16. SMOKE− DECREASESTraffic noise → SleepIn a 2022 update of the WHO sleep review, each extra 10 dB of night-time road, rail, or aircraft noise raised the odds of being highly sleep disturbed about two- to three-fold when the question named the noise, across 36 observational studies and 173,160 responses. Nobody randomized the street.
17. SMOKE0 NULLOmega-3 → EmploymentFish-oil buyers are not a random sample of the labor force.
18. SMOKE0 NULLOT → TrustIntranasal oxytocin does not raise trusting behavior: a 2026 registered report found no effect in 211 people, and a pooled equivalence test with the 2020 registered replication (532 people) put the leftover effect inside a minimal range.
19. SMOKE− DECREASESAir pollution → CognitionPregnancy-average outdoor PM2.5 tracked 1.79 fewer full-scale IQ points per 3 µg/m³ at age 10.5 in 568 Salinas Valley children, while six European birth cohorts (9,482 children) found no association between pregnancy air pollution and cognition at ages 1 to 6. Nobody randomized the air.
20. SMOKE+ INCREASESAir pollution → MortalityHao 2023's Medicare cohort of about 73.4 million adults 65 and older found a 4.0 percent higher all-cause death rate per 3.68 µg/m³ of annual PM2.5 (HR 1.040, 1.039 to 1.041). Nobody randomized the air.
21. SMOKE0 NULLPower pose → RiskExpansive posture is not a Holt–Laury intervention. The later RCTs are nulls on the gambling choice.
22. SMOKE0 NULLPower pose → TTwo minutes in a stance does not identify an androgen change. The larger RCTs are nulls. The TED talk is not a protocol.
23. SMOKE0 NULLPower → Youth pref.Command over people is not a separately identified cause of youth preference. It usually arrives with money.
24. SMOKE0 NULLT/C → StatusT/C is a derived index. Stamp it smoke unless the paper pre-registered the ratio and showed the interaction.
25. SMOKE? CONDITIONALScreens → CognitionIn 3,625 Pelotas cohort members, more television and games at 11 and more computer time at 11 and 15 tracked with better Digit Span backward scores at 22 in men, and with nothing in women. In 356 Illinois toddlers, more daily screen minutes tracked with worse parent-rated executive function. Nobody assigned the screens.
26. SMOKE~ NON-LINEARSleep → MortalityYin 2018's dose-response meta of prospective cohorts found a U-shaped link between hours of sleep and all-cause death: each hour under 7 tracked a 6 percent higher risk, and each hour over 7 tracked 13 percent. Nobody randomized time in bed.
27. SMOKE+ INCREASESSmoking → HOMA-IRIn Cho 2022, 4,043 Korean adults without diabetes who met both urine NNAL and cotinine smoker cutoffs had higher triglyceride-glucose odds than people who met neither (men 1.74, women 2.01 after age, BMI, drink, walk, income, and pack-years); Eliasson 1997's 17 quitters, against 23 who kept smoking, improved clamp insulin sensitivity in eight weeks despite modest weight gain.
28. SMOKE0 NULLSOI-R → HypergamyUnrestricted sociosexuality is not a stealth measure of 'women want richer men'.
29. SMOKE0 NULLVitamins → EmploymentPeople who buy vitamins look more employed because they already differ in ways that keep jobs, not because the pill caused the job. Shrank (2011) named that healthy-user bias. Simpson (2006) measured the same selection as death, not employment: in 21 studies and 46,847 people, good adherence, including to placebo, had an odds ratio of 0.56 for mortality.
30. SMOKE0 NULL25(OH)D → IncomePeople with money and outdoor lives have higher 25(OH)D. That is not D causing the money.
31. SMOKE0 NULLWealth → HypergamyAssets change the feasible set in a mating market. That is not a new instinct.
32. SMOKE+ INCREASESWidowhood → MortalityIn 373,189 elderly US couples followed from 1993 to 2002, a wife's death sat with 18 percent higher all-cause death hazard in the husband (HR 1.18, 1.16 to 1.19) and a husband's death sat with 16 percent in the wife (HR 1.16, 1.14 to 1.17). Nobody assigned widowhood.