Custodes Futurifor the advancement of humanity

All key figures

273 cited figures from across the nine factors, in one table. Figures that are model-based, debated, or correlational are labeled as such.

FactorMeasureValueSource
BiologicalDeaths from noncommunicable diseases, 2021At least 43 million, 75% of non-pandemic deathsWHO fact sheet on noncommunicable diseases
BiologicalNoncommunicable disease deaths before age 7018 million, 82% of them in low- and middle-income countriesWHO fact sheet on noncommunicable diseases
BiologicalPeople living with obesity, 20221 in 8 worldwide, including 890 million adultsWHO obesity and overweight fact sheet
BiologicalAdults who were overweight, 20222.5 billion (43%)WHO obesity and overweight fact sheet
BiologicalGrowth in obesity since 1990Adult obesity more than doubled, and adolescent obesity quadrupledWHO obesity and overweight fact sheet
BiologicalAdults not meeting recommended activity levels, 202231%, about 1.8 billion people, up about 5 percentage points since 2010WHO (2024)
BiologicalPhysical inactivity by sex, 202234% of women, 29% of menWHO (2024)
BiologicalOriginal ACE studyOf 9,508 adults, more than half reported at least one adverse childhood experience. Four or more categories carried 4 to 12 times the risk of alcoholism, drug abuse, depression, and suicide attemptFelitti et al. (1998)
BiologicalACE meta-analysis37 studies, 253,719 participants. Four or more ACEs raised risk for all 23 outcomes. Odds ratios were under 2 for inactivity, obesity, and diabetes. They were over 7 for problematic drug use and violenceHughes et al. (2017)
BiologicalChildren under 5 stunted, 2024150.2 million (23.2%), down from 40% in 1990WHO, UNICEF, and World Bank Joint Child Malnutrition Estimates, 2025 edition
BiologicalProgress toward the 2030 stunting targetAt current progress the target will be missed by 46 million childrenWHO, UNICEF, and World Bank Joint Child Malnutrition Estimates, 2025 edition
BiologicalChildren under 5 wasted, 202442.8 millionWHO, UNICEF, and World Bank Joint Child Malnutrition Estimates, 2025 edition
BiologicalLong-term effects of childhood undernutritionShorter adult height, less schooling, and lower economic productivity. Height-for-age at 2 years was the best predictor of human capitalVictora et al. (2008), cohorts from five countries
BiologicalChild deaths linked to undernutritionAround 45% of deaths among children under 5. Wasted children face a higher risk of death if not treated. The estimate comes from a 2013 Lancet analysis (Black et al., 2013), and WHO still cites it. These deaths remove potential future contributors, including possible pioneers, before any ability can developWHO malnutrition page
BiologicalLow birth weight, 202019.8 million newborns, 14.7% of all live births (under 2,500 grams), down from 16.6% in 2000. South Asia accounted for 44%. Birth weight data were missing for 28.7% of newborns, so the global figure is partly modeledUNICEF and WHO (Okwaraji et al., 2024)
BiologicalNorwegian twin study of birth weightThe effects of birth weight on long-run outcomes such as height, IQ, earnings, and education were significant. They were similar in size to simple comparisons across all children, even though short-run effects within twin pairs were much smallerBlack, Devereux, and Salvanes (2007)
BiologicalNewborn deaths from preterm or small-for-gestational-age births, 202055.3% of 2.4 million newborn deaths worldwide. About three quarters of these were preterm. The estimate covers these two groups. They overlap heavily with low birth weight but are not identical to itLawn et al. (2023)
BiologicalFirst-year mortality in Danish children born 1981 to 19950.6% overall and 6.6% among babies under 2,500 gramsMaruyama and Heinesen (2020)
BiologicalTwin studies of birth weight, United States and SwedenFound similar effects on education and income, though the U.S. effects were generally small. The Swedish study also linked lower birth weight to higher adult mortalityRoyer (2009), Bharadwaj et al. (2018)
BiologicalDanish study of birth weightAbout two million births from 1981 to 2013. The study used a pregnancy complication that causes early birth. It confirmed causal effects on infant death, health, and disability but found no strong evidence for effects on test scores, income, or IQ. The effect on infant deaths has weakened as newborn care has improvedMaruyama and Heinesen (2020)
BiologicalSickle cell disease worldwide, 2021An estimated 7.74 million people were living with it and 515,000 babies were born with it, mostly in the Caribbean and western and central sub-Saharan Africa. Deaths recorded as caused by it numbered 34,400. But deaths it contributed to totaled about 376,000, including 81,100 children under 5GBD 2021 Sickle Cell Disease Collaborators (2023)
BiologicalSickle cell disease and child mortality in AfricaAvailable data are consistent with 50% to 90% of children with the most common form dying in early childhood. The reviewers state that the data are inadequate for definitive statementsGrosse et al. (2011)
BiologicalSchool performance with sickle cell disease, United StatesStudents scored 0.70, 0.87, and 0.80 standard deviations below norms in reading, math, and spelling. They repeated a grade about 10 times as often as Black students nationally. Students who had a stroke or silent stroke scored lower. Environmental factors such as socioeconomic status are also associated with cognitive outcomes in this groupHeitzer et al. (2021), Prussien et al. (2020)
BiologicalNewborn deaths and child deaths from congenital disordersAn estimated 240,000 newborns in their first 28 days each year and 170,000 more children aged 1 month to 5 years. Nine of ten children born with a serious congenital disorder live in low- and middle-income countries. Congenital disorders have genetic and non-genetic causesWHO (2023)
BiologicalPhenylketonuria screening and treatmentNewborn screening and early dietary treatment greatly improve brain development. They prevent the severe intellectual disability the disorder otherwise causes. Even with early treatment, children often score below the IQ predicted from their parents and siblingsU.S. Preventive Services Task Force (2008), NICHD
BiologicalOrigin of the sickle cell gene in tropical regionsThe gene became common because carriers, who have one copy, are more likely to survive malaria. This links the barrier to the malaria barrierUyoga et al. (2019)
BiologicalSleep restriction experiment48 adults were randomized to 4, 6, or 8 hours in bed for 14 days. 6 hours or less produced cognitive deficits equal to up to 2 nights of total sleep deprivation. Participants were largely unaware of their declineVan Dongen et al. (2003)
BiologicalDeaths caused by alcohol and psychoactive drugs, 20192.6 million from alcohol (4.7% of all deaths) and 0.6 million from drugs. Men accounted for 2 million alcohol deaths and 0.4 million drug deathsWHO (2024)
BiologicalPeople with alcohol use disorders, 2019An estimated 400 million, including 209 million with alcohol dependenceWHO (2024)
BiologicalAdolescent cannabis use and schoolingThe study combined three Australian and New Zealand cohorts (3,765 people). After adjusting for many background factors, those who used cannabis daily before age 17 had about 63% lower odds of finishing high school than those who never used it. They also had 62% lower odds of earning a degree. Whether cannabis causes this or marks earlier disadvantage is debated. A published letter noted that daily users did no worse than weekly users on schoolingSilins et al. (2014)
BiologicalBody mass index and mortality, 239 studiesThe study pooled individual data from 239 prospective studies of 10.6 million adults in Asia, Australia and New Zealand, Europe, and North America. The main analysis was limited to 3.95 million never-smokers without chronic disease at the start. In that group, the lowest death rates were at a body mass index (BMI) of 20 to 25. The risk of death was about 45% higher with BMI 30 to 35 and 2.76 times higher with BMI 40 or more. Each 5 units of BMI above 25 carried more excess risk in adults aged 35 to 49 than in those aged 70 to 89. The study is observational, and critics questioned its restriction method and its selection of studiesGlobal BMI Mortality Collaboration (2016)
BiologicalMeasured physical activity and mortalityThis was a harmonized meta-analysis of eight cohorts that measured activity with accelerometers (36,383 adults, mean age 63, median follow-up 5.8 years, 2,149 deaths). The most active quarter had a 73% lower risk of death than the least active quarter. Even light activity was linked to lower risk. The most sedentary quarter had 2.63 times the risk of the least sedentary. The studies are observational, with mostly older adults and short follow-up. So part of the link may reflect illness lowering activityEkelund et al. (2019)
BiologicalGenetically higher BMI and income, UK BiobankThe study used genetic variants as a natural experiment (Mendelian randomization) among 119,669 UK Biobank participants aged 37 to 73. In women, a genetically higher BMI of 4.6 units was linked to £2,940 lower annual household income and greater deprivation. There was no such effect in men. A later study using siblings found that similar genetic estimates for BMI and education shrank sharply once family background was accounted for. So the causal size remains uncertainTyrrell et al. (2016), Brumpton et al. (2020)
BiologicalBucharest Early Intervention ProjectThe Bucharest Early Intervention Project was the first randomized trial of foster care as an alternative to institutional care. In it, 136 young children living in Romanian institutions (aged 6 to 31 months at the start) were randomly assigned to foster care or to continued care as usual. Children placed in families had better IQ, physical growth, and mental health. Gains in early childhood were largest for those placed youngest. The benefits stayed stable through ages 16 to 18. The sample is small, and the follow-ups were run by the original team. The trial also drew debate over research ethicsNelson et al. (2007), King et al. (2023)
BiologicalChild maltreatment and adult economic outcomesThe study took children with court-documented abuse or neglect from 1967 to 1971 in a Midwestern US area. It matched them with similar children and followed them to about age 41 (807 adults). After controlling for background, those maltreated had less education, lower earnings, and fewer assets as adults. They also had a 14% gap in the probability of being employed. The effects were larger for women. Because maltreatment was recorded at the time, the study avoids the recall problems of most ACE research. But it covers only cases that reached court in one areaCurrie and Widom (2010)
BiologicalDepression, anxiety, and costs linked to ACEsA meta-analysis of 23 studies estimated that adverse childhood experiences account for about 40% of depression and 30% of anxiety in North America. The same analysis estimated they account for more than a quarter of both in Europe. Illicit drug use was the risk factor most linked to them. The authors put the annual cost at US$581 billion in Europe and US$748 billion in North America. More than 75% of it was among people with two or more ACEs. These estimates treat the associations as causal. They also rely on adults recalling their childhoodsBellis et al. (2019)
BiologicalGuatemala nutrition trial and adult wagesIn Guatemala from 1969 to 1977, two villages were randomly assigned a nutritious supplement (atole) for all children. Two other villages got a less nutritious drink. Researchers traced 1,424 of the original children at ages 25 to 42. Men who had received atole from birth to age 2 earned hourly wages about 46% higher. No wage gain was found for exposure after age 3 or for women. Only four villages were randomized, and 60% of the children were tracedHoddinott et al. (2008)
BiologicalPreschool growth and schooling in rural ZimbabweIn rural Zimbabwe, drought and civil war hit siblings at different preschool ages. This let researchers compare brothers and sisters to estimate the effect of early growth. They estimated what would happen if the median preschool child had the height-for-age of a median child in a developed country. She would be about 3.4 cm taller in adolescence, complete about 0.85 more grades of schooling, and start school about six months earlier. The result comes from one panel study. It also depends on the shocks affecting schooling only through nutritionAlderman, Hoddinott, and Kinsey (2006)
BiologicalPeople facing hunger, 2025About 645 million people, 7.8% of the world's population, faced hunger in 2025. That was down from 8.1% in 2024 and 8.6% in 2022. About 2.1 billion people (25.8%) faced moderate or severe food insecurity. And 2.69 billion (32.7%) could not afford a healthy diet. Projections suggest 510 to 520 million people may still be hungry in 2030, missing the goal of ending hunger. These are modelled estimates by UN agenciesFAO et al. (2026)
BiologicalSleep duration and mortalityA meta-analysis pooled 16 prospective studies (27 cohorts, 1,382,999 participants, 112,566 deaths, 4 to 25 years of follow-up). It found that short habitual sleep was associated with a 12% higher risk of death. Long sleep was associated with a 30% higher risk. Sleep was self-reported, and long sleep may partly mark existing illness. A later dose-response meta-analysis found the lowest risk at about 7 hours a nightCappuccio et al. (2010), Yin et al. (2017)
BiologicalSleep and earnings, sunset timingThis study used US time-use diaries. It relied on the fact that a later local sunset shortens sleep while work hours stay fixed. An extra hour of average weekly sleep was estimated to raise earnings by about 1.1% in the short run and 5% in the long run. The method assumes sunset time affects wages only through sleep. A German panel study used the same approach. It found that an extra hour of weekly sleep raised weekly earnings by 3.4%Gibson and Shrader (2018), Costa-Font, Fleche, and Pagan (2022)
BiologicalEarly class start times, US Air Force AcademyAt the US Air Force Academy, 6,165 first-year students who entered from 2004 to 2008 were randomly placed into course sections. A first-period class lowered grades across that day's courses by 0.14 standard deviations when the day began at 7:00 a.m. When the day began at 7:50 a.m., the drop was a statistically insignificant 0.014. The authors compare the benefit of the later start to raising teacher quality by one standard deviation. Sleep was not measured, and cadets differ from typical teenagersCarrell, Maghakian, and West (2011)
BiologicalAlcohol as a leading risk factor for deathIn 2016, alcohol use was the leading risk factor for death worldwide among people aged 15 to 49. In that age group it accounted for 12.2% of male deaths and 3.8% of female deaths. The same study concluded that the level of drinking with the least health risk is zero. Its 2020 update found instead that this level varies by age and region. It ranged from 0 to about 0.6 standard drinks a day for ages 15 to 39. For people aged 40 and older it ranged from about 0.1 to 1.9GBD 2016 Alcohol Collaborators (2018), GBD 2020 Alcohol Collaborators (2022)
BiologicalLegal drinking age and academic performanceThe US Air Force Academy strictly enforces the minimum drinking age. There, gaining legal access to alcohol at 21 lowered course performance by about 0.1 standard deviations among 3,884 students. The drops were larger among the strongest students. The study compares grades just before and just after the 21st birthday, so it identifies a causal effect. But it does not measure drinking directly. A later study compared each student's grades before and after turning 21. It also found a drop. The drop was smaller than this estimate, but it was lasting and present for womenCarrell, Hoekstra, and West (2011), Lindo, Swensen, and Waddell (2013)
BiologicalSmoking, quitting, and life expectancyThe study interviewed about 202,000 US adults from 1997 to 2004. Current smokers died at about three times the rate of people who had never smoked. They lost more than 10 years of life expectancy. Quitting before age 40 removed about 90% of the excess risk. The study is observational, but it adjusted for age, education, body fat, and alcohol. It also matches the 50-year study of British doctors. In that study, continuing smokers died about 10 years earlier than non-smokersJha et al. (2013), Doll et al. (2004)
BiologicalFetal alcohol spectrum disorder worldwideA meta-analysis of 24 studies estimated that about 7.7 per 1,000 children and youth worldwide have fetal alcohol spectrum disorder (FASD). FASD is a range of lifelong physical, learning, and behavioral problems caused by alcohol exposure before birth. The highest regional rate was in Europe (19.8 per 1,000). For many countries the figure was estimated from drinking in pregnancy rather than measured. In four US communities, direct assessment of first graders found conservative rates of 1.1% to 5.0%Lange et al. (2017), May et al. (2018)
BiologicalOpioid prescriptions and men's labor force participationKrueger compared US counties. He estimated that the rise in opioid prescriptions since 1999 could account for about 43% of the fall in men's labor force participation. This figure was corrected in 2019 from about 20% after a calculation error. A Federal Reserve Bank of Cleveland study looked at 2006 to 2016. It found that a 10% higher local prescription rate was associated with about 0.5 percentage points lower participation among prime-age men and 0.1 among women. Another study looked at county employment rates from 2006 to 2014. It found no relationship for men and a small positive one for women. So whether prescriptions cause lower work remains unsettledKrueger (2017), Aliprantis, Fee, and Schweitzer (2022), Currie, Jin, and Schnell (2019)
BiologicalCosts of excessive drinking and opioids, United StatesIn the United States, excessive drinking was estimated to cost $249.0 billion in 2010, about $2.05 per drink. Binge drinking accounted for 76.7% of this, and government paid 40.4%. A CDC study estimated the cost of opioid use disorder and fatal opioid overdose at $1.02 trillion in 2017. Most of that cost came from reduced quality of life and the value of lives lost. These depend on how they are valued in moneySacks et al. (2015), Florence, Luo, and Rice (2021)
PsychologicalPeople living with a mental health conditionIn 2023, nearly 1 in 7 people (1.2 billion) lived with a mental disorder. Anxiety and depressive disorders were the most common. WHO's September 2025 release had described more than 1 billion people, roughly 1 in 8, living with a mental health conditionWHO (2026), WHO (2025)
PsychologicalWorking days lost to depression and anxiety each yearAn estimated 12 billion, costing about US$1 trillion in lost productivityWHO fact sheet on mental health at work
PsychologicalSize of loss aversionPeople often quote the figure that losses weigh 2.25 times as much as gains. It comes from Tversky and Kahneman (1992), a study of 25 graduate students, not from the 1979 prospect theory paper. A 2024 meta-analysis of 607 estimates from 150 articles found an average of 1.96. A separate meta-analysis of risky choices found a smaller average of 1.31. So the size of loss aversion depends on the settingBrown et al. (2024), Walasek et al. (2024)
PsychologicalLearned helplessness, revised viewPassivity is the default, control must be learnedMaier and Seligman (2016)
PsychologicalReplication of 100 published psychology studies97% of the originals had significant results, but only 36% of the replications did. Replication effects averaged half the size of the originalsOpen Science Collaboration (2015)
PsychologicalMindset and achievement, two meta-analyses273 studies (365,915 people) on mindset and achievement and 43 studies (57,155 people) on mindset interventions. Overall effects were weak in both. But students with low socioeconomic status or at academic risk may benefitSisk et al. (2018)
PsychologicalOnline mindset intervention of under one hourBetter grades among lower-achieving students and more enrollment in advanced math, mainly in schools where peer norms supported the messageYeager et al. (2019), nationally representative U.S. sample
PsychologicalDark Triad traits and work behaviorA meta-analysis of 245 samples (43,907 people) found that narcissism, Machiavellianism, and psychopathy were all associated with counterproductive work behavior. It found that Machiavellianism and psychopathy were also associated with lower job performance. The effects depended on context, including authority and cultureO'Boyle et al. (2012)
PsychologicalNarcissism and leadershipNarcissism predicts who emerges as a leader, largely through extraversion. But it does not predict how effective leaders are. Narcissistic leaders rate themselves as more effective, while supervisors, subordinates, and peers do not. Moderate narcissism appeared bestGrijalva et al. (2015)
PsychologicalNarcissistic CEOsAmong 111 computer-industry CEOs, narcissism was linked to bold strategies and to more and larger acquisitions. It was also linked to extreme, fluctuating results. But it was not linked to better or worse performance overallChatterjee and Hambrick (2007)
PsychologicalDispositional greedPeople higher in dispositional greed kept more money for themselves in economic games. They also judged transgressions more acceptable. In a laboratory study, they were more likely to accept a bribe. Greed was linked to lower self-control and life satisfactionSeuntjens et al. (2015, 2019)
PsychologicalSocial class and unethical behaviorA widely cited claim says that higher social class causes unethical behavior (Piff et al., 2012). Its field studies failed two well-powered, preregistered replications. Greed as a trait and wealth are different thingsJung et al. (2023)
PsychologicalChildhood conduct problems and later costs, Dunedin cohortThe 9% whose childhood conduct problems lasted into adulthood accounted for 53% of all criminal convictions and 25% of welfare benefit monthsRivenbark et al. (2018)
PsychologicalConcentration of costs, Dunedin cohortOne-fifth of members accounted for 81% of criminal convictions, 66% of welfare benefits, and 57% of hospital nights. Poor brain health at age 3 predicted this groupCaspi et al. (2016)
PsychologicalConcentration of costs in 4 million peopleRecords of 1.7 million New Zealanders and 2.3 million Danes showed the same pattern. Costs were concentrated in a small part of the population. That group had early-life disadvantages, including low education and poor mental healthRichmond-Rakerd et al. (2020)
PsychologicalEconomic impact of violence, 2025$21.8 trillion, or 10.5% of global GDP. Most of this is military and internal security spending, not individual behavior. Homicide was about 6% of the total in the previous year's estimateInstitute for Economics and Peace (ThisDay, 2026, and Vision of Humanity)
PsychologicalSelective exposure to informationA meta-analysis of 91 studies with nearly 8,000 participants found a moderate preference for information that supports existing views over information that challenges them (d = 0.36). The bias was stronger in more closed-minded people and when beliefs were tied to values. It reversed when challenging information was useful for a current goalHart et al. (2009)
PsychologicalGym memberships and present biasThe study covered 7,752 members of three U.S. health clubs. Those who chose a flat monthly fee of over $70 went on average 4.3 times a month. They paid more than $17 per visit, when a 10-visit pass cost $10 per visit. On average they gave up about $600 over their membership. The authors attribute this to overconfidence about future self-control. Later research debates the explanation. A membership may serve as a commitment device. People may simply overpredict how often they will go (one experiment found predictions about three times actual attendance). Or they may prefer a single paymentDellaVigna and Malmendier (2006), Acland and Levy
PsychologicalHow common learned helplessness isNo large study of how common learned helplessness is in people was found. Maier and Seligman (2016) revised the original theory using animal neuroscience. They concluded that passivity is the default response to prolonged adversity and that a sense of control must be learned. The Ethiopian aspirations trial under cultural factors is the closest human evidence that expectations of what is possible can be changed. Laboratory studies in people show a pattern consistent with the theory. In one, a single failure improved later performance while four failures worsened it (Mikulincer, Kedem, and Zilkha-Segal, 1989). Critics argue such deficits can arise from causes other than expecting future lack of control. They also argue that artificial lab tasks may not generalize (Silver, Wortman, and Klos, 1982). No study linking learned helplessness to real-world achievement was found, so this barrier is rated CMaier and Seligman (2016), Mikulincer et al. (1989)
PsychologicalCognitive dissonance replicationA preregistered replication ran in 39 labs with nearly 4,900 people. It found that writing an essay against one's own view shifted attitudes. But having a choice about writing it made no difference, contradicting the classic prediction. The selective exposure meta-analysis above supports a related prediction: people avoid information that conflicts with their beliefsVaidis et al. (2024), Hart et al. (2009)
PsychologicalCognitive biases in medical diagnosisA systematic review covered 20 studies with 6,810 physicians. It found that overconfidence, anchoring, availability bias, and tolerance to risk were associated with diagnostic inaccuracies in 36.5% to 77% of case scenarios. In the review, 5 of 7 studies linked biases to treatment or management errors. Most studies used case vignettes, and 60% were rated low quality. Only one of two studies on patient outcomes found an effectSaposnik et al. (2016)
PsychologicalOverconfidence and household stock tradingThe study covered 66,465 U.S. households with accounts at a large discount broker from 1991 to 1996. The households that traded most earned 11.4% a year, against 17.9% for the market. The average household earned 16.4% and turned over 75% of its portfolio each year. Before costs, returns were close to the market, so the losses came from trading costs. The authors attribute the heavy trading to overconfidence. This is an interpretation rather than a direct measureBarber and Odean (2000)
PsychologicalOverconfident CEOs and acquisitionsThe study covered 394 large U.S. firms from 1980 to 1994. Some chief executives were classified as overconfident because they held company stock options into their final year while the options were well in the money. They had 65% higher odds of making an acquisition. Investors reacted to their bids with an average loss of 90 basis points (0.9%) in firm value over three days. The loss was smaller for other chief executives (12 basis points in the abstract, 27 in the body text of the author version). The measure is indirect. Still, the authors found that inside information, signaling, and risk tolerance did not explain the resultsMalmendier and Tate (2008)
PsychologicalLoss aversion among professional golfersResearchers studied 2,525,161 putts by 421 professional golfers on the PGA Tour from 2004 to 2009. Players made putts for par (avoiding a loss relative to the hole's reference score) about 2 percentage points more often than equally difficult putts for birdie (a gain). This happened although only total strokes count. The gap was not related to world ranking. The authors report it among the best golfers, including Tiger Woods. The gap narrowed in later rounds but did not disappearPope and Schweitzer (2011)
PsychologicalInertia in health insurance choiceA large U.S. employer changed its insurance options. About 8,000 employees were tracked over six years, and they kept plans that had become worse for them. Because of inertia, they gave up an estimated $2,032 a year on average. In the author's model, nudging people to choose better would have made adverse selection worse (healthier people leaving generous plans). That would have doubled the existing welfare loss. So reducing inertia does not always helpHandel (2013)
PsychologicalDisposition effect among individual investorsResearchers studied trading records for 10,000 accounts at a large discount broker. Investors were more likely to sell stocks that had gained than stocks that had lost. This pattern is called the disposition effect. Rebalancing and trading costs did not explain it, and later portfolio performance did not justify it. In taxable accounts the habit lowered after-tax returns. A theoretical analysis found that prospect theory sometimes predicts this pattern and sometimes the opposite. So its link to loss aversion is debatedOdean (1998), Barberis and Xiong (2009)
PsychologicalPresent bias across convex time budget studiesA meta-analysis pooled 220 estimates from 28 articles that used the same experimental method (convex time budgets). It found that people are present-biased on average. But estimates varied widely, and selective reporting modestly overstated the bias. The main difference was the reward. Present bias was weaker for money than for non-monetary rewards such as effort. This difference shrank after correcting for selective reporting. In the working paper version, corrected estimates were close to no bias for money and about 0.90 to 0.93 for effort (1 means no bias)Imai, Rutter, and Camerer (2021)
PsychologicalDelay discounting and addictionA meta-analysis covered 64 studies (138 effect sizes, 11,395 participants). Some people more steeply preferred smaller immediate money over larger delayed money. The analysis found that these people had more severe addictive behavior across alcohol, tobacco, cannabis, stimulants, opiates, and gambling (r = 0.14, a small effect, and r = 0.17 for severity). The studies are correlational, so the authors state that cause and effect cannot be determined. Most samples were adultsAmlung et al. (2017)
PsychologicalSelf-control at work, Indian data entry workersA year-long field experiment studied 102 data entry workers in Mysore, India. Output rose about 8% from the start to the end of each pay week as payday approached. Workers often chose contracts that could only penalize low output, never reward high output. These contracts raised production. Workers with larger payday effects were more likely to choose them. Summaries give different figures for how often such contracts were chosen (28% and 36%). The study covers one firmKaur, Kremer, and Mullainathan (2015)
PsychologicalHelpless explanatory style and depressionA meta-analysis covered 104 studies with nearly 15,000 people. Some people explain bad events as caused by themselves, as lasting, and as affecting everything (the "helpless" explanatory style in the revised theory). The analysis found this style was reliably associated with depression, in both psychiatric patients and students. Links for good events were weaker. A review of statistical power found most studies in this literature too small to detect modest effects. This explains some inconsistent resultsSweeney, Anderson, and Bailey (1986), Robins (1988)
PsychologicalLocus of control and academic achievementA review looked at studies on locus of control. This is the belief that outcomes depend on one's own actions (internal) or on outside forces (external). The review found that more internal beliefs were associated with higher academic achievement, with a small to medium effect. The link was stronger in adolescents than in children or adults. It was also stronger with standardized tests than with teacher grades. The evidence is correlational and concerns a related construct, not learned helplessness itselfFindley and Cooper (1983)
PsychologicalHopelessness and later suicidal behaviorA meta-analysis covered 166 longitudinal studies published from 1971 to 2014 (357 prediction cases). It looked at hopelessness, the expectation that bad outcomes will happen and cannot be changed. It found that hopelessness roughly doubled the odds of later suicidal thoughts (odds ratio 2.19), attempts (1.95), and death (1.98). The authors call these effects weak in absolute terms. Adjusting for publication bias reduced them further. Most studies treated hopelessness as a fixed trait rather than tracking how it changed. Follow-up across all included studies averaged about 9 yearsRibeiro et al. (2018)
PsychologicalAdolescent depression, schooling, and unemploymentA meta-analysis covered 31 prospective cohort articles (24 cohorts). It found that depression in adolescence was associated with higher odds of not finishing secondary school (odds ratio 1.76). It was also associated with higher odds of unemployment in adulthood (odds ratio 1.66). The studies are observational, so earlier disadvantage may explain part of the linkClayborne et al. (2019)
PsychologicalEarly depression and lifetime earnings, DenmarkA register study covered all 2,390,127 people born in Denmark from 1955 to 1990. Some were diagnosed in hospital with depression before age 25. Over ages 25 to 61, they earned about 51% of the income of people without mood disorders. At age 30, 53% of them were outside the workforce, compared with 19% of the general population. The study covers only hospital-diagnosed cases and shows association, not causeHakulinen et al. (2019)
PsychologicalMath anxiety and achievementResearchers looked at 747 effect sizes from studies run from 1992 to 2018. Math anxiety was associated with lower math achievement (correlation r = -0.28), from childhood through adulthood. The link likely runs both ways, since poor performance can also cause anxietyBarroso et al. (2021)
PsychologicalBeliefs about intelligence and reading, PISA 2018PISA 2018 surveyed about 600,000 15-year-olds in 78 countries and economies. Some students disagreed that their intelligence "can't change very much". On average across OECD countries, they scored 31.5 points higher in reading, after accounting for socioeconomic background. Disadvantaged students were more likely to hold a fixed view. The OECD notes these are cross-sectional correlations that cannot show the direction of cause. The link was weaker in East Asian systems and absent in fourOECD (2021)
PsychologicalGrowth mindset trial in ArgentinaA randomized trial ran across 202 public secondary schools in Salta, Argentina. Final-year students got a growth mindset lesson, which was delivered as designed. But it had no detectable effect on attitudes toward hard tasks, grades, test scores, or plans after school. Small effects could be ruled outGanimian (2020)
PsychologicalTest of the core premises of mindset theoryA preregistered study of 438 participants tested six key premises of mindset theory. One example is that mindset shapes motivation and responses to failure. The study found no support for most of them. All associations were weaker than r = 0.20. The strongest ran opposite to the predicted directionBurgoyne, Hambrick, and Macnamara (2020)
PsychologicalChoice and preference change, corrected designsA classic dissonance finding is about choosing between two similar options. Afterward, people rate the chosen one higher and the rejected one lower. Chen and Risen (2010) showed that the standard test can produce this pattern even when preferences do not change. A meta-analysis of 43 studies (2,191 participants) used corrected designs. It still found a moderate effect of choice on preference (Cohen's d = 0.40) and no evidence of publication bias. All the studies were laboratory experimentsEnisman, Shpitzer, and Kleiman (2021)
PsychologicalSunk-cost effect meta-analysisA meta-analysis of 45 studies (100 effect sizes) found a moderate sunk-cost effect. People were more likely to continue a project or use a purchase because of money or effort already spent. For decisions about using things already bought, the effect was smaller in older adults. It also faded with time. Most studies used hypothetical scenarios, so real-world costs are less certainRoth, Robbert, and Straus (2015)
PsychologicalEvidence and attitude change without backlashResearchers ran three survey experiments on gun control, the minimum wage, and the death penalty. The studies used different samples (one nationally representative sample of 2,122 people and two online samples). Supporters and opponents mostly shifted their views in the direction of the evidence they were shown (23 of 24 estimates). No subgroup showed backlash. Effects were small in the national sampleGuess and Coppock (2020)
EconomicPeople in extreme poverty (below US$3.00 a day)About 831 million in 2025, roughly 1 in 10, down from about 2.3 billion in 1990. A March 2026 update revised the 2024 count to 847 million. It projected the rate to fall from 10.4% in 2024 to 10.0% in 2026World Bank (Fall 2025, March 2026)
EconomicPeople in poverty at the US$8.30 a day line, 2025An estimated 3.7 billionWorld Bank (Fall 2025)
EconomicChildren in households below US$3.00 a day, 2024An estimated 412 millionWorld Bank (Fall 2025)
EconomicChange to the extreme poverty line, June 2025Raised from US$2.15 to US$3.00 a day, adding about 125 million people. Older figures near 700 million used the previous lineWorld Bank (June 2025)
EconomicShare of global wealthTop 10% own about 75%, bottom 50% hold about 2%, top 1% hold 37%World Inequality Report 2026
EconomicShare of global incomeTop 10% receive 53%, bottom 50% receive 8%World Inequality Report 2026
EconomicUpward mobility, bottom to top income fifth (U.S. cities)4.4% (Charlotte) to 12.9% (San Jose)Chetty et al. (2014)
EconomicInventor rates by parents' incomeChildren whose parents are in the top 1% of income become inventors at a rate of 8.3 per 1,000. For children of below-median-income parents, the rate is 0.85 per 1,000. Early ability explains little of this gap. Top-scoring third graders are much more likely to become inventors only if their families have high incomes. By 8th grade, test scores explain half the gap, because low-income children steadily fall behindBell et al. (2019), authors' summary
EconomicThird-grade math scores and the innovation gapIn the conference version of the study, third-grade math scores explain less than one-third of the income gap in innovationBell et al. (2018), conference version
EconomicPatenting at the most innovative collegesAt colleges such as MIT, students from low- and high-income families patent at similar rates. The authors conclude that most of the gap comes from childhood environment and exposure, not from lack of funding or aversion to riskBell et al. (2019)
EconomicGreat Gatsby CurveAcross rich countries, higher income inequality goes with lower mobility between generations (Corak, 2013). Corak has said it is not a causal relationship. He has also argued that dismissing it as mere correlation is too glib. Critics argue that with a different mobility measure, the United States has upward mobility similar to Canada and Sweden, despite very different inequality (Winship, 2015). Durlauf and Seshadri (2018) propose that inequality reduces mobility by increasing residential segregation. This is a model-based argument. Within the United States, Chetty et al. (2014) found lower mobility in areas with greater inequality. This is also a correlationCorak (2013), Winship (2015), Durlauf and Seshadri (2018), Chetty et al. (2014)
EconomicIncome versus wealth in the mobility and inventor studiesThese studies measure parents' income, not wealth. The global wealth figures above are rated C, because many countries' wealth shares are estimated rather than measuredStrength ratings
EconomicPerry Preschool Program, annual social rate of returnAbout 7% to 10%. That is lower than earlier claims but above the historical return on U.S. stocks. The program was small (58 treated children), so its results should be applied more widely only with cautionHeckman et al. (2010)
EconomicResearchers needed to double computer chip density vs. early 1970sMore than 18 timesBloom et al. (2020)
EconomicLong working hoursIn 2016, an estimated 488 million people (8.9% of the world's population) worked 55 or more hours a week. WHO and ILO attributed 745,000 deaths from stroke and heart disease that year to these long hours. That was a 29% increase since 2000. Working 55 or more hours was associated with a 35% higher risk of stroke than working 35 to 40 hours. It was also associated with a 17% higher risk of dying from heart disease. Independent epidemiologists argued that the heart disease conclusion does not meet the reviewers' own evidence standards. They said the link was weaker and not statistically significant in the higher-quality studies. The review authors replied that their conclusion stands. So the death estimates are contestedPega et al. (2021), Kivimäki et al. (2020)
EconomicShort-termism, survey of executivesA survey asked 401 financial executives and held 20 interviews. In it, 78% said they would give up economic value to smooth reported earnings. Many said they would delay valuable projects to meet short-term earnings targets. This measures what they said they would do, not decisions anyone observedGraham, Harvey, and Rajgopal (2005)
EconomicShort-termism, public versus private firmsA study compared matched U.S. firms. Private firms invested about 6.8% of total assets a year. Public firms of similar size in the same industry invested about 3.7%. Private firms were also about four times more responsive to new investment opportunities. In the study, firms invested more in the five years before going public than after. The gap was largest where stock prices react most to earnings news. The study assumes private firms face less short-term pressure. Later studies using other data found public firms invest as much or more (Gilje and Taillard, 2016, Feldman et al., 2021). So the finding is debatedAsker, Farre-Mensa, and Ljungqvist (2015)
EconomicAdult learning barriersIn the 2023 Survey of Adult Skills, about one in four adults across OECD countries reported a barrier to taking part in adult learning. Among them, the most common reason was lack of time due to work or family (48%). It came ahead of a lack of suitable training (14%) and cost (13%). These are self-reported reasonsOECD, Trends in Adult Learning (2025)
EconomicExtreme poverty estimates, September 2026 updateThe World Bank's Poverty and Inequality Platform put out a release in September 2026. It revised the 2024 count again, to about 824 million (10.12%). It estimates about 817 million (9.95%) for 2025 and 811 million (9.79%) for 2026World Bank Poverty and Inequality Platform (September 2026)
EconomicIncome and life expectancy, United StatesResearchers used 1.4 billion US tax records from 1999 to 2014, linked to Social Security death records. The gap in life expectancy at age 40 between the richest 1% and poorest 1% was 14.6 years for men and 10.1 years for women. From 2001 to 2014, the top 5% gained about 2.3 to 2.9 years. The bottom 5% gained 0.04 to 0.32 years. The study shows association. Life expectancy for low-income people varied by about 4.5 years across areasChetty et al. (2016)
EconomicChildren at risk of not reaching their developmental potentialAn estimated 249.4 million children under 5 in low- and middle-income countries, 43%, were at risk of not reaching their developmental potential in 2010. This was because they were stunted or lived in extreme poverty. The number was down from 279.1 million (51%) in 2004. Sub-Saharan Africa had the highest share. The estimate uses poverty and stunting as stand-ins, not direct measures of developmentLu, Black, and Richter (2016)
EconomicMothers' Pensions and sons' later outcomesMothers' Pensions (1911 to 1935) was the first US welfare program. Under it, sons of accepted applicants lived about one year longer than sons of rejected applicants. They also had about one-third of a year more schooling, were less likely to be underweight, and had higher adult incomes. The study compares accepted with rejected families. It could not follow daughtersAizer et al. (2016)
EconomicEssential health services and financial hardship, 2025The 2025 WHO and World Bank monitoring report estimated that 4.6 billion people still lack access to essential health services. It also estimated that 2.1 billion face financial hardship paying for health care. The report defines this as a household spending more than 40% of its discretionary budget on out-of-pocket health costs. In 2022, about 3 in 4 people in the poorest part of the population faced such hardship. Among the richest, fewer than 1 in 25 did. The figures are modeled from national surveys (service coverage data to 2023, hardship data to 2022). They undercount hardship, because people who cannot afford care and go without it record no spendingWHO and World Bank (2025)
EconomicReturns on 133 U.S. policiesHendren and Sprung-Keyser (2020) compared 133 U.S. policy changes from the past half century. For each policy, they estimated the benefit to recipients per dollar of net government cost, including later tax revenue and savings. In their estimates, direct investments in the health and education of low-income children had the highest values, averaging above 5. Many of these paid for themselves. Most policies aimed at adults fell between 0.5 and 2. The estimates depend on projecting long-run earnings and tax effects from the underlying studiesHendren and Sprung-Keyser (2020)
EconomicDeaths from poor-quality health careKruk et al. (2018) compared case fatality (the share of people with a condition who die) for 61 conditions. They compared 137 low and middle income countries with 23 high income countries with strong health systems. They estimated 8.6 million excess deaths in 2016 that health care could have prevented. Of these, 3.6 million were attributed to people not using health care and 5.0 million to poor-quality care. The figures are modeled from Global Burden of Disease data and household surveys. So they depend on the comparison countries and on how deaths were split between the two causesKruk et al. (2018)
EconomicUnstable work schedules and wellbeingA survey covered 27,792 hourly workers at 80 large U.S. retail and food service companies (2016 to 2017). It found that unstable and unpredictable schedules were associated with psychological distress, poor sleep, and unhappiness. These links were much stronger than those with low wages. In the authors' estimates, 64% of workers who had a shift canceled reported distress, against 43% of those who did not. In the same survey, work-life conflict explained more of the link than household money problems. The data come from a single-time survey recruited through Facebook ads. So cause and effect cannot be separatedSchneider and Harknett (2019)
EconomicJob insecurity and depressionA meta-analysis pooled 20 prospective cohort studies. It found that feeling one's job was insecure was associated with higher odds of later depressive symptoms (odds ratio 1.29). This was slightly more than for unemployment (1.19). Effects were stronger in studies of workers under 40. Job insecurity was self-reported, and the studies were observationalKim and von dem Knesebeck (2016)
EconomicWorking 55 hours or more and depressive symptomsResearchers pooled data on 189,729 workers from 35 countries. Working 55 or more hours a week was associated with a 14% higher risk of new depressive symptoms over one to five years. The link was moderate in Asia, small in Europe, and absent in North America and Australia. A later review for WHO and ILO judged the evidence inadequate to show harm. That review found a smaller association that was not statistically significantRugulies et al. (2021), Virtanen et al. (2018)
EconomicVoters and disaster preparedness spendingHealy and Malhotra (2009) used U.S. data on natural disasters, government spending, and presidential election returns. They found that voters rewarded the incumbent party for disaster relief spending. But voters did not reward it for disaster preparedness spending. The authors estimated that $1 spent on preparedness was worth about $15 in future damage avoided. This implies large losses from underinvestment. The study is observational, and the $15 figure is the authors' estimateHealy and Malhotra (2009)
EconomicShare buybacks to meet earnings forecastsFirms that would have just missed analysts' earnings-per-share forecasts were more likely to buy back shares to meet them. They did so more often than firms that would have just beaten the forecasts. According to the authors, comparing the two groups, these buybacks were followed by cuts. The cuts equaled about 5% of average employment, 10% of average capital spending, and 3% of average R&D spendingAlmeida, Fos, and Kronlund (2016)
EconomicCEO equity vesting and investmentIn U.S. firms, some quarters had large amounts of a CEO's stock and options scheduled to vest. Those quarters saw slower growth in R&D and capital spending. They also saw more positive earnings guidance and analyst forecast revisions. Vesting dates were set by grants made years earlier. So the dates are unlikely to reflect current investment opportunities. A separate study used a 2004 accounting rule that led 723 firms to speed up CEO option vesting. It found those firms cut investment by about $14 million at the medianLadika and Sautner (2020), Edmans, Fang, and Lewellen (2017)
EconomicDecline of disruptive papers and patentsResearchers studied 45 million papers and 3.9 million patents over six decades. Over that time, papers and patents became steadily less likely to break with past work and push fields in new directions. This was measured by how later work cites them (the "CD index"). The decline appeared in all fields. A reanalysis argued the decline is largely a side effect of growing reference lists and changing citation habits. So whether science has truly become less disruptive is disputedPetersen, Arroyave, and Pammolli (2025), Park, Leahey, and Funk (2023)
EconomicEroom's law in drug developmentThe number of new drugs approved per billion U.S. dollars of research and development spending halved roughly every 9 years from 1950. After adjusting for inflation, it fell about 80-fold. The authors called this pattern "Eroom's law" (Moore's law spelled backwards). Scannell and colleagues later wrote that the trend was already being broken around the time the 2012 paper was published. So the decline may not have continuedRingel et al. (2020), Scannell et al. (2012)
EconomicCitation entrenchment in large fieldsChu and Evans (2021) analyzed 1.8 billion citations among 90 million papers in 241 subjects. They found that when a field publishes very many papers a year, citations flow disproportionately to papers that are already highly cited. They also found that the list of most cited papers stops changing, and new papers rarely become highly cited or displace existing work. The study is correlational. The authors' explanation (overloaded attention) is a proposed mechanismChu and Evans (2021)
SocietalPeople forcibly displaced, end of 2024123.2 million, nearly double the figure a decade earlierUNHCR Global Trends 2024
SocietalPeople forcibly displaced, end of 2025117.8 million. That is 5.4 million fewer than a year earlier and the first decrease in a decade. But it is still about 1 in 70 people worldwideUNHCR Global Trends 2025
SocietalCorruption Perceptions Index, 2025The global average dropped to 42 out of 100, the first drop in more than a decade. 122 of 182 countries (67%) scored below 50. Only 31 countries have improved their scores since 2012Transparency International
SocietalRule of Law Index, 2025 (143 countries, 95% of the world's population)Rule of law declined in 68% of countries, up from 57% the previous year. It was the eighth straight year in which more countries declined than improvedWorld Justice Project
SocietalRule of Law Index, specific measures, 2025Freedom of opinion and expression fell in 73% of countries. Judicial limits on government power declined in 61%World Justice Project
SocietalPremature chronic disease deaths in low- and middle-income countries, 202182%WHO (see the biological factor)
SocietalHealth gradient by social positionDeclines step by step at each lower levelMarmot (2004)
SocietalU.S. hiring field experimentsA meta-analysis of every available U.S. hiring field experiment (28 studies, 55,842 applications for 26,326 positions) found that since 1989, white applicants received on average 36% more callbacks than African Americans and 24% more than Latinos. The meta-analysis found no change in discrimination against African Americans over 25 years. It found modest evidence of a decline for Latinos. The authors treat this cautiously because few studies included Latino applicantsQuillian et al. (2017)
SocietalHiring discrimination in nine countriesResearchers pooled 97 field experiments with more than 200,000 applications in nine countries in Europe and North America. Discrimination against nonwhite natives appeared in every country. In the countries with the most discrimination, white natives received nearly twice the callbacks of nonwhites. In the countries with the least, white natives received about 25% more. France had the highest rates, followed by SwedenQuillian et al. (2019)
SocietalDiscrimination at large U.S. employersA field experiment sent more than 83,000 fictitious applications to 108 of the largest U.S. employers. It found that distinctively Black names reduced the chance of employer contact by 2.1 percentage points. Discrimination was concentrated. The fifth of firms that discriminated most accounted for nearly half of lost contacts to Black applicants. There was no average gender gap, but some firms favored men and others favored womenKline, Rose, and Walters (2022)
SocietalAllocation of talent, 1960 to 2010In 1960, 94% of U.S. doctors and lawyers were white men. By 2010 the share was 62%. Over that time, women and Black men entered jobs that had been closed to them. A model-based study estimated that this better allocation of talent explains 20% to 40% of U.S. growth in output per person from 1960 to 2010Hsieh, Hurst, Jones, and Klenow (2019)
SocietalWomen's legal rights worldwideWomen have about two-thirds (67%) of the legal rights of men worldwide. Only 4% of women live in economies close to full legal equality. No economy grants women equal economic opportunities. Fewer than half (47%) of the policies and institutions needed to make these laws effective are in place. Between 2023 and 2025, 68 economies passed 113 reformsWorld Bank, Women, Business and the Law 2026
SocietalDiscrimination in science fundingBlack applicants' chance of winning a major NIH research grant was about 10 percentage points lower than white applicants' (Ginther et al., 2011). A 2019 NIH analysis looked at applications from 2011 to 2015. It found award rates of 10.7% for African American or Black applicants and 17.7% for white applicants. In that analysis, topic choice explained over 20% of the gap after controlling for prior achievements. Most of the gap was left unexplained. An unexplained gap is not by itself proof of reviewer biasGinther et al. (2011), Hoppe et al. (2019), Taffe and Gilpin (2021)
SocietalGender bias in academic hiring, a contested findingMoss-Racusin et al. (2012) found that 127 science faculty rated an identical lab-manager application as more competent and more worth hiring when it carried a male name. Williams and Ceci (2015) found that 873 faculty preferred equally qualified women about 2:1 for hypothetical tenure-track positions. But critics argue that their design did not reflect real hiring. Jussim and colleagues reported a larger replication of Moss-Racusin et al. that found the reverse. The authors list it as accepted by Meta-Psychology, but it had not appeared there as of October 9, 2026. So the direction of gender bias in academic hiring depends on the career stage and study designMoss-Racusin et al. (2012), Williams and Ceci (2015)
SocietalLimits of field experimentsField experiments measure discrimination only at the first step, the callback. So they do not capture what happens in interviews, pay, or promotion. They also face a methodological critique (Heckman and Siegelman, 1993). If employers see more variation in unobserved qualities for one group, callback gaps can misstate discrimination in either direction. Reanalyses using a correction method developed by Neumark (2012) found that most estimates held up. But a Swedish reanalysis found the critique mattered for some studiesNeumark (2012), Neumark and Rich (2016)
SocietalRacial gaps in mobility across generationsU.S. tax and census records cover nearly the whole population. In these records, Black and American Indian children have much lower rates of upward mobility than white children, and higher rates of downward mobility. Among children from families with the same income, the Black-white gap is driven by lower wages and employment among Black men. There is no such gap between Black and white women. Parental marital status, education, and wealth explain little of the gap. Black boys have lower adult incomes than white boys from comparable-income families in 99% of Census tractsChetty, Hendren, Jones, and Porter (2020)
Societal1930s redlining mapsA federal agency graded neighborhoods for lending risk. The lowest-graded areas often had mostly African American residents. A study compared areas on either side of the map boundaries in 149 cities. It found that the maps led to lower homeownership, house values, and rents, and to more racial segregation in later decades. Later research disputes the mechanism. It says the agency that drew the maps lent widely in low-graded neighborhoods and drew them after most of its lending was done. It also says the Federal Housing Administration had already been excluding these neighborhoods by its own methods. The long-run gaps are well documented, but whether the maps themselves caused them is contestedFishback, Rose, Snowden, and Storrs (2021), Aaronson, Hartley, and Mazumder (2021)
SocietalRacial gaps in mortgage lendingLatinx and Black borrowers with the same credit risk paid significantly higher interest rates on government-backed mortgages. This cost them over $450 million a year. Algorithmic lenders showed similar gaps for some loans and smaller gaps for others. The widely quoted figures of 7.9 basis points and $765 million come from the 2019 working paper, not the published versionBartlett, Morse, Stanton, and Wallace (2022)
SocietalRacism and healthA meta-analysis of 293 studies, mostly from the United States, found that reported racism was associated with poorer mental health (r = -0.23), including depression and anxiety. Associations were stronger in cross-sectional and non-representative studies. So how much of the link is causal is uncertainParadies et al. (2015)
SocietalSame-race teachers, what helpsIn the Tennessee STAR experiment, some Black students were randomly assigned at least one Black teacher in kindergarten through third grade. These students were 9 percentage points (13%) more likely to graduate from high school. They were also 6 percentage points (19%) more likely to enroll in college. There was no effect on white students, and North Carolina data gave similar results. The authors suggest role-model effects as one possible channel. A review of 37 studies found benefits when students and teachers share the same race, especially for Black and Latino studentsGershenson et al. (2022), Redding (2019)
SocietalRacial discrimination in the European UnionA survey of 6,752 people of African descent in 13 EU countries found that 45% experienced racial discrimination in the previous five years. That was up from 39% in 2016. In the same survey, 34% experienced it when looking for work. Only 9% reported it. Germany had the highest rateEU Agency for Fundamental Rights (2023)
SocietalA caution on implicit biasMeasured discrimination, as in the field experiments above, is stronger evidence than scores on implicit bias tests. A meta-analysis of 492 studies (87,418 participants) found that procedures can change implicit bias scores. But the same meta-analysis found those changes produced trivial changes in behavior and did not explain behavior change. An earlier meta-analysis described the Implicit Association Test as a poor predictor of discriminatory behavior. The test's developers replied that small effects can still matter across large populations. But they acknowledged the test is problematic for identifying which individuals will discriminateForscher et al. (2019), Oswald et al. (2013), Greenwald, Banaji, and Nosek (2015)
SocietalScope of the mobility, redlining, and lending studiesThe mobility, redlining, and lending studies show large, persistent racial gaps. For redlining, whether the maps themselves caused these gaps is contested. These studies do not each identify present-day discrimination as the sole cause of the gaps they measureStrength ratings
SocietalPublic goods experimentsContributions fell to very low levels when free riders could not be punished. By the final round, 53% to 75% of participants contributed nothing. When participants could punish free riders, contributions rose toward full cooperationFehr and Gächter (2000)
SocietalManaging shared resourcesA review looked at 91 studies of real communities that manage shared resources. It supported most of Ostrom's design principles for successful local governanceCox et al. (2010)
SocietalCOVID-19 excess deaths, 2020 and 2021WHO estimated 14.83 million excess deaths worldwide. That is 2.74 times the 5.42 million deaths reported as caused by COVID-19. Only about half of countries had complete death data, so many estimates are modeledMsemburi et al. (2023)
SocietalExistential risk estimatesOrd's probability estimates for existential catastrophe are his own judgments, not measurements. So they are not given here as figuresOrd (2020)
SocietalPeru's colonial mining mitaIn Peru, the mita was a colonial forced mining labor system in force from 1573 to 1812. Districts just inside its boundary have about 25% lower household consumption today than districts just outside it. They also have about 6 percentage points more child stunting. The effect persisted through land tenure and public goods such as roads. The boundary was drawn centuries ago. So this comparison isolates the effect of an extractive institution better than cross-country studies canDell (2010)
SocietalUnderinvestment in pandemic preparednessA model combined the frequency and severity of pandemics. It estimated expected global losses from future pandemics at over $700 billion a year. It found that spending $60 billion up front and $5 billion a year to expand vaccine production capacity would yield an expected net gain of more than $500 billion. A University of Leeds policy brief argues that claims about returns on preparedness rest on unreliable assumptions. So the size of the underinvestment is uncertainGlennerster, Snyder, and Tan (2022), University of Leeds
SocietalDemocracy and economic growthA study used a country panel with country fixed effects and controls for past GDP. It found that countries that democratized had GDP per capita about 20% higher in the long run. Matching and instrumental variable methods gave similar results. The authors trace the gain to more investment, schooling, economic reforms, public goods, and less social unrest. A reanalysis found the long-run effect robust but about half as large. The reanalysis also found the effect sensitive to which countries are includedEberhardt (2019), Acemoglu et al. (2019)
SocietalCorruption in education funds, BrazilResearchers used federal audit reports on how Brazilian municipalities spent education grants. Students in municipalities where auditors found corruption in education funds scored about 0.35 standard deviations lower on standardized tests. They also had higher dropout and failure rates. Their schools had fewer computer labs, teaching supplies, and teacher training. Audits were assigned by lottery, but which municipalities were corrupt was not. So the comparison is observational with controlsFerraz, Finan, and Moreira (2012)
SocietalColonial rubber concessions in the CongoIn the north of the Congo Free State, private companies ran colonial rubber concessions through violence and indirect rule. The concessions had arbitrarily drawn borders. People living just inside former concessions today have about 1.4 fewer years of education than people just outside. They are about 26% less wealthy. Children and women inside have about 9 percentage points lower height-for-age percentiles. These findings are based on Demographic and Health Survey data from 2007 and 2014. Village chiefs inside the former concessions provide fewer public goods. They are also 35 percentage points less likely to be elected. Even so, residents are more trusting and more supportive of sharing incomeLowes and Montero (2021a)
SocietalOverfished marine fish stocks, 2023The UN Food and Agriculture Organization assessed marine fish stocks. Of these stocks, 62.4% were within biologically sustainable levels in 2023, down from 64.5% in 2021. So roughly 38% were overfished. FAO attributes part of the drop to changes in method and to the addition of new stocks. Larger stocks are better managed: 72.6% of landings came from sustainable stocks. A study of 230 fisheries found that stock assessment models tend to overstate the health of overfished stocks. So the true share may be worseEdgar et al. (2024), FAO (2026)
SocietalAntisocial punishment across 16 societiesIn public goods experiments run in 16 comparable participant pools around the world, many participants punished people who gave generously, not only free riders. In some pools this "antisocial punishment" was strong enough to wipe out the gains in cooperation that punishment normally brings. Antisocial punishment was more common in societies with weak norms of civic cooperation and weak rule of law. This is a correlation across 16 sitesHerrmann, Thöni, and Gächter (2008)
SocietalCommunity forests in NepalNepal has more than 18,000 community forests. The study compared subdistricts that were similar at baseline, from 2000 to 2012. Community forest management was associated with net reductions in both deforestation and poverty. Gains for forests were smaller where poverty was high. They were larger where community forests were bigger and older. The study gives the effects as a 37% relative reduction in deforestation and a 4.3% relative reduction in poverty, compared with similar areas without community forest management. The study also cites estimates that about a quarter of Nepal's forests are managed directly by more than a third of its populationOldekop et al. (2019)
SocietalDeaths in organized violence, 2025The Uppsala Conflict Data Program recorded about 244,600 deaths in organized violence in 2025, up from about 187,000 in 2024. Killings of civilians rose fivefold, driven largely by violence in Sudan. They reached the highest level since the 1994 Rwandan genocide. The number of state-based armed conflicts reached a record 65. These counts rely on reported deaths and are revised as information arrives (the 2024 total was first reported as almost 160,000). A survey of the program's coders suggests the counts tend to undercountVesco et al. (2026), Davies, Pettersson, and Öberg (2026)
SocietalEconomic cost of terrorism, Basque CountryTerrorism broke out in the Basque Country in the late 1960s. After that, its GDP per capita fell about 10 percentage points relative to a "synthetic" Basque Country. This synthetic version was built from a weighted mix of other Spanish regions without terrorism. During the 1998 to 1999 truce, shares of firms with much of their business in the region rose relative to others when the truce became credible. These shares fell when the truce endedAbadie and Gardeazabal (2003)
SocietalArmed conflict and infant mortality in AfricaResearchers linked 1.99 million births in 35 African countries from 1995 to 2015 with the locations of armed conflicts. Infants born within 50 km of a conflict had a 7.7% higher risk of dying before age 1 than infants born in the same region in years without conflict. For conflicts with more than 1,000 deaths, this higher risk rose to 26.7%. The raised risk reached up to 100 km away and lasted for 8 years after conflicts. The estimated infant deaths related to conflict were 3.2 to 3.6 times the number of direct combat deaths. The design is observationalWagner et al. (2018)
SocietalOccupational position and mortality, 48 cohortsIn 48 prospective cohort studies covering 1,751,479 adults in seven high-income countries, people in low occupational positions had higher death rates than those in high positions (hazard ratio 1.42 for men and 1.34 for women). The link remained after adjusting for smoking, alcohol, inactivity, obesity, hypertension, and diabetes. Low socioeconomic status was associated with 2.1 fewer years of life between ages 40 and 85. That is more than high alcohol intake (0.5), obesity (0.7), or hypertension (1.6). But it is less than physical inactivity (2.4), diabetes (3.9), or smoking (4.8). The cohorts are observational and from high-income countries onlyStringhini et al. (2017)
SocietalMoving to Opportunity and adult healthIn the Moving to Opportunity experiment, 4,498 mothers living in high-poverty U.S. public housing were randomly offered vouchers usable only in low-poverty neighborhoods, unrestricted vouchers, or neither. Follow-up measurements took place from 2008 to 2010. In them, the low-poverty voucher group had lower rates of severe obesity (body mass index 40 or more) than the control group, by 3.4 percentage points. This group also had lower rates of diabetes-range blood sugar (HbA1c 6.5% or more), by 4.3 percentage points. Unrestricted vouchers had no significant effect. The authors say the mechanisms are unclearLudwig et al. (2011)
SocietalLife expectancy gaps between countries, 2025WHO's 2025 world report on social determinants of health equity found that people born in the country with the lowest life expectancy live 33 years less than those born in the country with the highest. The report also found that children born in poorer countries are 13 times more likely to die before age 5. WHO estimates, from models, that closing health gaps in low- and middle-income countries could save the lives of 1.8 million children a year. WHO also notes that data on gaps within countries are scarceWHO (2025)
SocietalDeaths in the 1918 influenza pandemicA historical review of the 1918 to 1920 influenza pandemic concluded that it killed on the order of 50 million people. That was up from about 21.5 million in a 1920s estimate. The review noted that records were incomplete because of missing registration and misdiagnosis. A later study used mortality data from 13 countries and British India. It estimated about 17.4 million deaths and judged totals above 25 million unrealistic. So the size of the toll is disputed, though every estimate is in the tens of millionsJohnson and Mueller (2002), Spreeuwenberg et al. (2018)
SocietalFamine after nuclear warA peer-reviewed study linked climate, crop, and fishery models. It estimated that soot injections above 5 million tonnes from burning cities would cause mass food shortages in almost all countries. The study projected that more than 2 billion people could die from a nuclear war between India and Pakistan. It projected that more than 5 billion could die from one between the United States and Russia. These are model projections, not observations. The amount of soot such fires would produce is disputed. A Los Alamos team modeled much less smoke reaching the upper atmosphere. The original modelers replied that this team's assumptions were unrepresentativeXia et al. (2022), Reisner et al. (2018)
SocietalHeat deaths attributable to climate changeResearchers used daily death and temperature records from 732 locations in 43 countries from 1991 to 2018. They attributed 37% (range 20.5% to 76.3%) of warm-season heat-related deaths to human-caused warming. They found increases on every continent. The estimate covers only the locations studied. It also covers only heat, not other climate effects such as floods or droughtVicedo-Cabrera et al. (2021)
SocietalLand titles in a Buenos Aires suburbSquatters occupied a Buenos Aires suburb in 1981. Some families there received land titles while their neighbors did not. This happened because some former owners accepted state compensation and others contested it in slow courts. Titled families invested more in their houses and had smaller households. Their children received more education. But these gains did not come through better access to credit, where effects were modest. The study covers one community of about 1,000 parcels. An independent reproduction confirmed the main resultsGaliani and Schargrodsky (2010)
SocietalDebt recovery tribunals in IndiaIndia introduced debt recovery tribunals state by state from 1993 to speed up large debt claims. One study compared loans above and below the size threshold, before and after a state got a tribunal. It found the tribunals reduced delinquency on the average loan by 28% and lowered interest rates on larger loans. The data come from one lender. A later study found the same reform reduced credit access for small borrowers while expanding it for wealthy onesVisaria (2009), von Lilienfeld-Toal, Mookherjee, and Visaria (2012)
SocietalUnmet justice needs worldwideThe World Justice Project estimated in 2019 that about 5.1 billion people, two thirds of the world's population, have at least one unmet justice need. This includes 1.5 billion who cannot resolve civil, administrative, or criminal justice problems. It includes 4.5 billion who are shut out of opportunities the law provides, such as identity documents, land or housing tenure, and formal work. It also includes 253 million living in extreme conditions of injustice. The total combines 11 data sources, with extrapolation where data were missing and an adjustment for people counted more than once. So it is an approximate estimateWorld Justice Project (2019)
CulturalGender Social Norms Index, 2023 (80 countries, 85% of the world's population)Nearly 9 in 10 men and women hold at least one fundamental bias against womenUNDP
CulturalBeliefs about political and business leadershipAbout half of people believe men make better political leaders. More than 40% believe men make better business executives. There has been no improvement over the previous decadeUNDP
CulturalChild marriageAn estimated 640 million girls and women alive today were married in childhood. About 1 in 5 young women aged 20 to 24 were married as children, down from nearly 1 in 4 a decade earlier. At the current rate, ending the practice would take about 300 yearsUNICEF (2023)
CulturalTrust in scientists across 68 countries71,922 people were surveyed. In most countries, most people trust scientists and want them to engage more in society and policymaking. Trust varied within and between countries, partly by political orientationCologna et al. (2025)
CulturalScience literacy and climate polarizationHighest science literacy and numeracy were the most culturally polarized on climate risk, not the leastKahan et al. (2012)
CulturalAspirations trial in rural EthiopiaHouseholds were randomly shown a one-hour documentary about people from similar communities who escaped poverty. Six months later, these households had higher aspirations. They also had more saving, borrowing, and spending on children's schooling. Five years later, they had invested more in farming, children's education, and assetsBernard et al. (2026)
CulturalMaterialism and wellbeingA meta-analysis of 259 samples found that materialist values were associated with lower wellbeing (r = -0.19 for the most widely used measures). The link was strongest for risky health and spending behaviors and negative self-views. It was weakest for life satisfaction. The studies are correlational. Three longitudinal studies and a small intervention trial found wellbeing improved as people became less materialistic (Kasser et al., 2014). But a three-year study of 6,551 people using a stricter method found no within-person link between changes in materialism and changes in life satisfaction (reported by PsyPost, 2023)Dittmar et al. (2014)
CulturalTight and loose culturesA 33-nation study measured cultural "tightness," meaning strong norms and low tolerance of deviant behavior. Tighter nations (such as Pakistan, Singapore, and Japan) tended to have histories of threats such as dense populations, resource scarcity, conflict, and disease. Looser nations included New Zealand, the Netherlands, and the United States. Tightness can support coordination but leaves less room for nonconformity. The measure is disputed. A six-country replication found different country rankings and did not support averaging individual answers into country scores (Treviño et al., 2021). Another study argued the scoring method can distort results (Venaik et al., 2021). The country examples should be treated as uncertainGelfand et al. (2011)
CulturalTrust in scientists in 12 countries, 2020Panel surveys ran in 12 countries from March to December 2020. In them, trust in scientists was the key driver of support for and compliance with non-drug public health measures. It was also the key driver of favorable attitudes toward vaccinationAlgan et al. (2021)
CulturalTrust in experts and vaccination intentionsIn 25 national samples across 12 countries (25,334 people), trust in medical and scientific experts was among the strongest and most consistent factors linked to willingness to get a COVID-19 vaccine (odds ratio 1.28). These are surveys of stated intentionsKerr et al. (2021)
CulturalMaterialism and children's learningIn 9- to 11-year-olds in Britain and Hong Kong, materialism was linked to lower exam performance through weaker interest in mastering material. In Hong Kong, materialism predicted worse grades a year later. In an experiment, British children shown a materialism prime chose easier tasks focused on performance. These children also gave up on a learning task sooner. Samples were modest. An independent longitudinal study of 404 Hong Kong secondary students found that earlier materialism predicted lower engagement and lower later achievement. That study found this worked partly through reduced motivation (King and Datu, 2017)Ku, Dittmar, and Banerjee (2014)
CulturalIndividualism and innovationA cross-country study used the frequency of certain genes and historic disease prevalence as instruments for culture. It found that individualism raised innovation, productivity, and income per worker. The link held after controlling for institutions. The instruments rest on strong assumptions that they affect growth only through culture, so this is rated CGorodnichenko and Roland (2017)
CulturalAge at marriage and schooling, BangladeshIn rural Bangladesh, the study used the age at which girls reach puberty (menarche) as a natural source of variation in when they marry. Each additional year that marriage was delayed was associated with about 0.22 more years of schooling and higher literacy. Later marriage was also linked to more use of preventive health care. The study covers one country. A later working paper using surveys from 12 countries found broadly similar links between later marriage, schooling, and employmentField and Ambrus (2008)
CulturalMissing female births from sex-selective abortionSon preference, combined with prenatal sex determination and falling fertility, has led to sex-selective abortion. A Bayesian model used 10,835 observations from vital registration, censuses, and surveys in 202 countries. It found strong evidence of imbalanced sex ratios at birth in 12 countries during 1970 to 2017. This amounted to an estimated 23.1 million missing female births worldwide (uncertainty range 19.0 to 28.3 million). Of these, 11.9 million were in China and 10.6 million in India. These are modeled estimatesChao et al. (2019)
CulturalNorm against wives out-earning husbandsUsing U.S. data, Bertrand, Kamenica, and Pan reported that the share of household income earned by wives drops sharply just above one half. They also reported that wives likely to out-earn their husbands were less likely to work and earned less than predicted. In their data, couples where the wife out-earns the husband reported less marital satisfaction and more divorce. They read this as evidence of a norm against wives out-earning husbands. Two later studies disputed the central evidence. One found that ordinary assortative matching closely reproduces the same income pattern, so the pattern says little about whether such a norm exists. Another, with Finnish data, traced the drop to spouses who work for the same employer (rated D)Bertrand, Kamenica, and Pan (2015)
CulturalLocus of control and job search, GermanyIn a panel of newly unemployed people in Germany, those with an internal locus of control searched harder for jobs. Having an internal locus of control means believing your own actions shape your future. Those who believed outcomes are determined by outside factors set lower reservation wages (the lowest pay they would accept). The design is observational. Locus of control was measured during unemployment. A later study found that unemployment temporarily shifts how people describe their own sense of controlCaliendo, Cobb-Clark, and Uhlendorff (2015)
CulturalSelf-efficacy program for women in Uttar PradeshIn rural Uttar Pradesh, India, 1,022 women were randomly assigned to one of two groups. One group got a nine-session program to raise generalized self-efficacy, the belief that one can handle challenges and reach goals. The other group got control meetings. The program raised self-efficacy, and the gain lasted 13 months. Work off the family farm rose by 6.1 percentage points at six weeks and 8.7 points at five months. The control levels were 13.1% and 18.7%. But there was no effect on employment at 13 months. A six-minute job promotion video shown to family members raised employment by itself. But women who received both the program and the video worked no more than the control group. This is a single unpublished working paperMcKelway (2021)
CulturalFatalism and lung cancer screeningA systematic review of 72 studies looked at why people take up, or plan to take up, lung cancer screening. It found that fear was the most often reported barrier. This was either fear of a cancer diagnosis rooted in fatalism, the belief that cancer cannot be prevented or survived (22 studies), or fear of the procedure (13 studies). Cost (28 studies) and inconvenience (24 studies) came next. Most included studies measured intention to screen rather than actual screening. The review counts how often a barrier was reported rather than measuring its sizeSalman et al. (2025)
CulturalPartisan bias in evaluating informationA meta-analysis of 51 experiments with more than 18,000 participants found that people consistently rated otherwise identical information more favorably when it supported their political side (average correlation r = .245). It found the bias was about equal among liberals (r = .235) and conservatives (r = .255). Critics argue the included studies were not designed to detect differences between the sides. So the equal-bias conclusion is disputed, though the existence of the bias is notDitto et al. (2019)
CulturalMotivated numeracy, failed replicationsThe idea that more numerate people use their skill to defend their side has not held up in replication. A preregistered replication of the motivated numeracy experiment found no good evidence for it. Another test covered five contested issues in a large nationally representative probability sample of Americans. It found that people were more accurate when a result fit their politics. It also found that more numerate people were more accurate whether or not the result fit. Partisan bias remained, but skill did not generally amplify itPersson et al. (2021), Stagnaro, Tappin, and Rand (2023)
CulturalParty registration and excess deaths after COVID-19 vaccinesResearchers linked 2017 voter registration to 538,159 deaths of adults aged 25 and over in Florida and Ohio. COVID-19 vaccines became available to all adults in May 2021. The researchers found that after that, excess deaths among registered Republicans were 43% higher than among registered Democrats. Before, there was no clear gap. The gap was larger in counties with lower vaccination rates. The study is cross-sectional and covers two states. It had no individual vaccination records or causes of deathWallace, Goldsmith-Pinkham, and Schwartz (2023)
CulturalColonial medical campaigns and vaccine trustFrom 1921 to 1956, French colonial governments in central Africa ran campaigns against sleeping sickness. The campaigns forcibly examined villagers and injected them with drugs that had severe and sometimes fatal side effects. The authors used 30 years of digitized archival records covering five countries. Areas with greater past exposure to the campaigns have lower child vaccination rates today. People there are also less willing to consent to a free blood test today, and World Bank health projects there were less successful. Exposure was not randomly assigned, so the result rests on the authors' controls and comparisonsLowes and Montero (2021b)
CulturalFake CIA vaccination campaign in PakistanIn July 2011 the Pakistani public learned that the CIA had used a vaccination campaign as cover in the search for Osama Bin Laden. The Taliban then ran anti-vaccine propaganda. The study compared birth cohorts and districts. It found that vaccination rates fell by 23% to 39% more in districts with high support for Islamist parties (90th percentile) than in districts with low support (10th percentile). The authors conclude that information that discredits vaccination campaigns can reduce trust in health services and demand for immunizationMartinez-Bravo and Stegmann (2022)
CulturalTuskegee disclosure and Black men's healthResearchers compared older Black men with other groups before and after the 1972 disclosure of the Tuskegee syphilis study. They also compared men by distance from the study's victims. The disclosure was associated with more medical mistrust, fewer doctor visits, and higher mortality. The authors estimated that life expectancy at age 45 for Black men fell by up to 1.5 years. That is about 35% of the 1980 gap with white men. A 2024 reanalysis argued the data do not credibly show that the study raised mortality. So the size of the effect is contestedKaestner (2024), Alsan and Wanamaker (2018)
CulturalExperiments exposing people to materialistic cuesA meta-analysis looked at 27 experiments (3,649 people) that exposed people to materialistic cues, such as images of luxury goods or framing people as consumers. It found lower individual wellbeing (δ = -0.39) and lower concern for others and the environment (δ = -0.41). The studies were small (50 to 487 people). The authors judged 37% of them to be underpowered. Publication bias was checked only with a funnel plotMoldes and Ku (2020)
CulturalConsumer framing and trust, Many Labs 2An experiment found that people asked to think of themselves as "consumers" trusted others less to conserve water in a shortage (77 people, d = 0.87). Many Labs 2 repeated it with 6,608 people across many labs and countries. The effect was in the same direction but much smaller (d = 0.12). The same project replicated only 15 of 28 psychology findings. The median effect size was 0.15, compared with 0.60 in the originalsKlein et al. (2018)
CulturalMaterialism and environmental attitudesA meta-analysis of 13 independent samples (about 2,500 people) found that people who held more materialistic values reported less favorable environmental attitudes (r = -0.22). These people also reported fewer pro-environmental behaviors (r = -0.24). Tests found no sign of publication bias. But the studies were cross-sectional, behavior was self-reported, and most samples were US studentsHurst et al. (2013)
CulturalUnpaid parking tickets of UN diplomatsBefore 2002, diplomatic immunity meant UN diplomats in New York faced no enforcement of parking tickets. Between November 1997 and the end of 2002, they built up more than 150,000 unpaid tickets. Diplomats from countries ranked as more corrupt in survey indices had significantly more unpaid violations. In 2002, authorities gained the power to confiscate diplomatic plates. After that, violations fell sharply, so enforcement mattered as well as normsFisman and Miguel (2007)
CulturalAncestral culture, women's work, and fertilityThe study looked at 6,774 married US-born women in the 1970 census whose fathers came from 25 countries. Women whose ancestral country had higher female labor force participation in 1950 worked more. Women whose ancestral country had higher fertility had more children (about 0.4 more children per standard deviation). The pattern held after controlling for education and husbands' characteristics. Later work argues this method can be biased by which families migrateBeblo et al. (2020), Fernández and Fogli (2009)
CulturalInherited trust and national incomeA cross-country study used the trust of US descendants of immigrants from 24 countries as a measure of trust inherited from each origin country. It tracked how this trust changed between 1935 and 2000. The study estimated large effects of inherited trust on income per head. Under its full model, income in 2000 would have been 69% higher in Russia and 59% higher in Mexico with Sweden's level of inherited trust. A published comment found much weaker links between inherited trust in 1935 and trust in the origin countries when using more reliable World Values Survey waves. The comment also noted very small samples behind the 1935 trust estimatesMüller, Torgler, and Uslaner (2012), Algan and Cahuc (2010)
SocialPeople who report feeling lonelyAbout 1 in 6 worldwide (15.8%)WHO Commission on Social Connection (2025)
SocialDeaths linked to lonelinessAn estimated 871,000 per year, about 100 every hourWHO Commission on Social Connection (2025)
SocialLoneliness by group21% among adolescents, 24.3% in low-income countries, and 11% in high-income countriesWHO Commission on Social Connection (2025)
SocialSurvival advantage of strong social relationshipsAbout 50% higher likelihoodHolt-Lunstad et al. (2010)
SocialAsch conformity experimentsWhen people judged line lengths alone, they erred under 1% of the time (0.7%). With a unanimous group giving wrong answers, 36.8% of responses conformed. 94 of 123 participants (about 76%) conformed at least once, and 29 of 123 (about 24%) never conformedAsch (1956)
SocialEffect of one dissenting voice in the Asch experimentsWhen one other person gave the correct answer, conformity dropped to about 5% of trialsAsch (1956)
SocialYoung people who never had an adult mentorAbout 1 in 3 (an estimated 16 million), including about 9 million considered at riskMENTOR, The Mentoring Effect (2014), nationally representative U.S. survey
SocialPostsecondary enrollment among at-risk young people45% of those with a mentor vs. 29% of those without. This is a correlation from a survey, not proof that mentoring caused the differenceMENTOR, The Mentoring Effect (2014)
SocialIn-group favoritismA meta-analysis of 212 studies found that people cooperate more with members of their own group than with outsiders (d = 0.32). It found that this reflects favoritism toward the in-group more than hostility toward the out-groupBalliet, Wu, and De Dreu (2014)
SocialPartner violence against womenAn estimated 27% of ever-partnered women aged 15 to 49 have experienced physical or sexual violence from a partner, and 13% in the past year. This estimate is based on 366 studies covering 2 million women in 161 countries. Rates are higher in low-income countries. Childhood abuse and household dysfunction are covered under early-life adversity in the biological key figuresSardinha et al. (2022)
SocialNeighbors' earnings and happinessU.S. panel data on 8,944 adults showed that people whose neighbors earned more reported lower happiness, holding their own income constant. The author found this was not explained by who chose to live where. German data found the opposite for neighbors. Once comparisons with coworkers and people of the same sex were accounted for, higher neighborhood income was linked to more happiness. Comparisons with those peers predicted lower wellbeing (Mayraz, Wagner, and Schupp, 2009). Who people compare themselves with appears to matter more than neighbors as suchLuttmer (2005)
SocialPartner violence and women's workAmong 1,311 Illinois welfare recipients, recent partner violence was linked to unstable employment over three years, largely through health problems (National Institute of Justice). In Tanzania, women who had been abused earned less than women who had not (World Bank). A Colombian study found the opposite. In that study, partner violence increased women's employment, possibly as a way to escape (EAFIT seminar abstract). A peer-reviewed study using Colombian survey data also reported this positive link (Fajardo-Gonzalez, 2021). All are observational, so the effect of adult partner violence on achievement is rated CNIJ, World Bank, EAFIT, Fajardo-Gonzalez (2021)
SocialTrickle-down consumptionThis study used variation across U.S. states and years. Non-rich households exposed to higher top incomes spent a larger share of income, especially on visible goods. They reported more financial strain and lived in states with more personal bankruptcies. Middle-income households might have saved an additional 2.6% to 3.2% of their income by the mid-2000s had top incomes grown at the same rate as median income. The authors judge status-driven spending the most consistent explanation. Other work is mixed. A related study supports such "expenditure cascades" (Frank, Levine, and Dijk, 2014). But a working paper found no effect when comparison groups were defined by census division (Wang, working paper)Bertrand and Morse (2016)
SocialSocial isolation, loneliness, and mortalityA meta-analysis of 90 prospective cohort studies (2,205,199 adults) found that social isolation was associated with a 32% higher risk of death from any cause (pooled effect 1.32). It found that loneliness was associated with a 14% higher risk (1.14). Isolation was also associated with higher cardiovascular (1.34) and cancer (1.24) mortality. The studies are observational, so the link may partly reflect poorer health causing isolation. An independent reproduction found the main results robust, but most publication bias adjustments gave somewhat smaller effectsWang et al. (2023), Molo et al. (2024)
SocialSocial relationships, heart disease, and strokeA meta-analysis of longitudinal studies in high-income countries (16 datasets, 4,628 coronary heart disease and 3,002 stroke events over 3 to 21 years) looked at social ties. It found that poor social relationships were associated with a 29% higher risk of coronary heart disease and a 32% higher risk of stroke. In 479,054 UK Biobank participants, isolation and loneliness were associated with about 36% to 49% higher risk of first heart attack and stroke after adjusting only for demographics. But conventional risk factors and existing chronic conditions explained most of this riskValtorta et al. (2016), Hakulinen et al. (2018)
SocialLoneliness and dementiaA meta-analysis of 21 longitudinal samples (608,561 people) found that loneliness was associated with a 31% higher risk of dementia (hazard ratio 1.31). In 16 samples (103,387 people), loneliness was associated with a 15% higher risk of cognitive impairment. The associations held after accounting for depression and social isolation. The studies are observational, and they measured loneliness in different ways. Early, undiagnosed dementia may itself cause lonelinessLuchetti et al. (2024)
SocialConformity across 133 Asch-type studiesA meta-analysis of 133 Asch-type line judgment studies from 17 countries found that conformity in U.S. studies had declined since the 1950s. It also found that countries scoring as more collectivist generally showed higher conformity. The culture comparison is at the country level. A later Asch replication with 140 Japanese students found Japanese participants conformed no more than AmericansBond and Smith (1996), Takano and Sogon (2008)
SocialSocial influence in an online music marketIn an online "music market," 14,341 participants were randomly assigned to see or not see other people's download counts. Stronger social influence made song success both more unequal and less predictable. Quality only partly determined success. The best songs rarely did poorly and the worst rarely did well, but any other result was possible. The setting was one website with unknown songs, so the size of the effect in real markets is uncertainSalganik et al. (2006)
SocialHerding in online ratingsOn a social news website, researchers randomly gave 101,281 comments an extra up-vote, a down-vote, or no change over five months. A single positive vote made the next rating 32% more likely to be positive. It raised final ratings by 25% on average. Other users tended to correct negative votes. The effect varied by topic, and the study covers one siteMuchnik et al. (2013)
SocialJudges' in-group bias in Israeli courtsIn Israeli small claims courts from 2000 to 2004, cases were effectively randomly assigned to Arab or Jewish judges. Claims were about 17 to 20 percentage points more likely to be accepted when the judge shared the plaintiff's ethnicity. The bias was larger after more terrorist violence near the court. The main sample was 1,748 cases. A study of juvenile courts in one U.S. state with random judge assignment found the opposite pattern, with judges harsher on defendants of their own raceShayo and Zussman (2011), Depew et al. (2017)
SocialJury pool racial composition and convictionsIn felony trials in two Florida counties from 2000 to 2010, juries drawn from all-white jury pools convicted Black defendants 16 percentage points more often than white defendants. The gap disappeared when the pool included at least one Black member. The study used day-to-day random variation in who was called for jury duty. It covered about 700 trials, and about 40% of pools had no Black membersAnwar et al. (2012)
SocialPartisan favoritism in scholarship choicesIn a U.S. experiment, online panel members chose between two scholarship applicants (610 in the party version). About 80% of Democrats and Republicans picked the applicant from their own party. The authors found that "partisanship simply trumped academic excellence." Racial in-group preference was weaker, with European Americans slightly favoring the African American applicant. A follow-up in Belgium, Great Britain, Spain (the Basque Country), and the United States found that partisans discriminated against opposing partisans more than against religious, linguistic, ethnic, or regional out-groups. Both are hypothetical choices in surveys. Another study found that even strongly polarized partisans were no more likely to intentionally harm opponentsIyengar and Westwood (2015), Westwood et al. (2018), Lelkes and Westwood (2017)
SocialWomen and girls killed by partners or family, 2024An estimated 83,000 women and girls were intentionally killed worldwide in 2024. Of these, about 50,000 (60%) were killed by intimate partners or family members, compared with 11% of male homicides. The UN notes that country data are incomplete. The UN also notes that the drop from the 2023 estimate mostly reflects differences in data availability, not a real decline. An earlier systematic review of 66 countries found that intimate partners alone committed a median of 38.6% of female homicides, against 6.3% of male homicidesStöckl et al. (2013), UNODC and UN Women (2025)
SocialPartner violence and later depressionA meta-analysis of 16 longitudinal studies with 36,163 participants looked at partner violence. Women who experienced partner violence had about twice the odds of later developing depressive symptoms (pooled odds ratio 1.97 from six studies). Partner violence was also associated with later suicide attempts. The link ran both ways, since depressive symptoms also predicted later partner violence. Few studies controlled for confounders beyond demographicsDevries et al. (2013)
SocialDisruptive classmates and adult earningsEconomists used school records from one Florida county linked to later earnings. They compared school cohorts that by chance had more classmates from families linked to domestic violence. Earlier work had shown this group disrupts classrooms. Having one such disruptive peer in a class of 25 throughout elementary school was associated with about 3% lower earnings at ages 24 to 28 for classmates. It was also associated with a nearly one percentage point lower chance of a college degree. Unequal exposure to these peers explained about 5% of the gap in earnings between rich and poor children in the dataCarrell, Hoekstra, and Kuka (2018)
SocialMentoring for low-track students in GermanyA randomized trial tested a German program that paired students in low-track secondary schools with university student mentors. The relationships were meant to last about two years. After one year, mentoring raised an index of math grades, patience and social skills, and job market orientation by over half a standard deviation for students from low socioeconomic backgrounds. It had no effect for better-off students. After three years, gains in grades and job market orientation persisted, and early school-to-work transitions improved. Gains in patience and social skills faded. The estimation sample was 308 students (153 assigned to mentoring, 155 controls) in 19 schools in 10 citiesResnjanskij et al. (2024)
SocialMentoring workshop for junior women economistsThe American Economic Association randomly assigned 365 junior women economists who applied to a mentoring workshop between 2004 and 2014 to attend or not. Attendees later published about 1.6 more papers before tenure. They were 6.9 and 9.3 percentage points more likely to hold tenure at a top 30 or top 50 ranked institution. The program did not significantly change the overall chance of getting tenure anywhereGinther et al. (2020)
SocialFirst jobs at a parent's workplace, SwedenResearchers used Swedish population records on graduates from 1988 to 1995. They found that about 10% of young people found their first job at a workplace where a parent worked. This rose to 14% of those leaving after compulsory school and fell to 3.2% of university graduates. According to the discussion-paper version, these graduates found work about half a year sooner and were more likely to stay, though they started on lower wages. Parental ties mattered most for less-educated graduates and in years of high unemployment. Weaker ties, such as classmates' parents, showed no such patternKramarz and Skans (2014)
SocialNegative social comparison in depression and anxietyA systematic review of 54 studies, 14 of them pooled in a meta-analysis, looked at people with clinical depression or anxiety. Among them, more negative views of oneself relative to others were associated with more severe symptoms (weighted correlations of minus 0.53 for depression and minus 0.39 for anxiety). Experimental studies suggested that depression and anxiety are related to how people react to comparisons with those better off. The studies are mostly correlational and few used clinical samples, so the direction of cause is unclearMcCarthy and Morina (2020)
SocialIncome inequality and depressionA review of 26 studies found that nearly two-thirds reported higher depression risk where income inequality was higher. Five of six longitudinal studies reported this too. The pooled risk ratio was 1.19 across 12 studies. The review proposed social comparison as one of several pathways. These are mostly cross-sectional comparisons between places. A preregistered study of about 550,000 Norwegian adolescents from 2010 to 2019 found that changes in inequality within municipalities did not meaningfully predict depressive symptomsNilsen et al. (2026), Patel et al. (2018)
SocialIncome rank and mental distressA British cohort of 30,000 people completed health questionnaires yearly for up to 17 years. Both income and income rank within a region or comparison group predicted current and future mental distress. But when both were included, only rank remained significant. Distress was self-reported, and the design is correlational. A later study by members of the same team found a similar pattern for depressive symptoms in U.S. and English cohorts, with rank explaining a small share of variationHounkpatin et al. (2015), Wood et al. (2012)
EnvironmentalPremature deaths, ambient plus household air pollution6.7 million per yearWHO
EnvironmentalPremature deaths, ambient air pollutionAbout 4.2 millionWHO
EnvironmentalHousehold air pollution deaths, 20212.9 million (over 309,000 children under 5)WHO
EnvironmentalIQ loss, blood lead from 2.4 to 30 µg/dL6.9 points (1,333 children)Lanphear et al. (2005)
EnvironmentalAmericans with high early-childhood lead exposureOver 170 million, about 824 million IQ pointsMcFarland et al. (2022)
EnvironmentalGlobal lead burden, 2019765 million IQ points lost in children under 5, about 5.5 million adult cardiovascular deaths, over 90% of the harm in low- and middle-income countries, and a cost of about US$6.0 trillion (roughly 6.9% of global GDP)Larsen and Sánchez-Triana (2023), model estimate with wide uncertainty
EnvironmentalAll pollution combinedAbout 9 million deaths per year, one in sixFuller et al. (2022)
EnvironmentalGrowth in "modern" pollution deaths since 2000Over 66%Fuller et al. (2022)
EnvironmentalLearning lost per 1°F hotter school year, no air conditioning1 percent of that year's learning. Hot school days accounted for roughly 5 percent of the racial achievement gapPark et al. (2020), 10 million students who retook the PSAT
EnvironmentalAverage annual temperature at which productivity peaksAbout 13°CBurke et al. (2015)
EnvironmentalProjected cut in global incomes by 2100, unmitigated warmingRoughly 23%. This depends on adaptation assumptions and is debatedBurke et al. (2015)
EnvironmentalMalaria, 2024282 million cases and 610,000 deaths, up from a revised 273 million cases in 2023. About 95% of deaths occurred in the WHO African Region, mostly among children under 5WHO World Malaria Report 2025
EnvironmentalSchool-aged children in the American South with hookworm40%Rockefeller Sanitary Commission
EnvironmentalEffect of hookworm treatmentAreas with more infection saw larger gains in enrollment, attendance, literacy, and later income. An independent replication questioned whether these benefits are robustBleakley (2007), Roodman (2018)
EnvironmentalLack safely managed drinking water2.1 billion, including 106 million who drink untreated surface waterWHO/UNICEF JMP 2025
EnvironmentalLack safely managed sanitation3.4 billion, including 354 million who practice open defecationWHO/UNICEF JMP 2025
EnvironmentalPlanetary boundaries exceeded7 of 9: climate change, biosphere integrity, land system change, freshwater use, biogeochemical flows, novel entities, and ocean acidification. Only aerosol loading and stratospheric ozone remain in the safe zonePlanetary Health Check 2025
EnvironmentalPsychiatric disorder risk, least vs. most childhood green spaceUp to 55% higher (correlational)Engemann et al. (2019)
EducationalLearning poverty, low- and middle-income countriesAbout 70% (57% pre-pandemic)World Bank and partners (2022)
EducationalLearning poverty, Sub-Saharan AfricaAbout 89%World Bank and partners (2022)
EducationalLearning poverty, Latin America and the CaribbeanPredicted to reach about 80%, up from around 50% before the pandemicWorld Bank and partners (2022)
EducationalProjected lifetime earnings lost by this generationUS$21 trillion, about 17% of global GDPWorld Bank
EducationalChildren and youth out of school273 million in the 2026 estimate, about 1 in 6 of school age, and up for seven years in a row. UNESCO's June 2025 estimate was 272 million. Conflict-affected countries add at least 13 million more who are not in the main countUNESCO (2026, 2025)
EducationalOut-of-school rate by level, June 2025 estimate11% of primary-age children (78 million), 15% of lower-secondary-age (64 million), and 31% of upper-secondary-age youth (130 million)UNESCO (2025)
EducationalChange in out-of-school numbers since 2015Fallen by less than 1%UNESCO (2025)
EducationalLearning lost to COVID-19 disruptionAbout 35% of a normal school year. Losses were stable over time. They were larger for lower socioeconomic backgrounds and poorer countries, and larger in math than in readingBetthäuser et al. (2023), 42 studies, 15 countries
EducationalBottom-5% teacher replaced by average, per classroomAbout US$250,000 lifetime income. A published comment found these long-run estimates sensitive to modeling choices, so the figure is uncertainChetty, Friedman, and Rockoff (2014), Rothstein (2017)
EducationalTaRL language gains0.15 SD (Haryana), 0.70 SD (Uttar Pradesh)Banerjee et al. (2016), J-PAL
EducationalHigh-stakes exams and pollution, IsraelIsraeli students scored lower on matriculation exams taken on days with more fine-particle pollution. The same students had lower post-secondary attainment and earnings years later. Later studies found similar exam-day effects of pollution on China's College English Test and of heat on New York City exit exams. On the New York City exams, hot exam days lowered scores by up to 13% of a standard deviation and reduced high school graduationEbenstein, Lavy, and Roth (2016), Yao et al. (2023), Park (2022)
TechnologicalInternet users and people offline6 billion use the internet and 2.2 billion remain offline. 96% of those offline live in low- and middle-income countriesITU Facts and Figures 2025
TechnologicalShare online by group, 2025High-income countries 94%, low-income countries 23%, urban areas 85%, rural areas 58%, men 77%, women 71%ITU (2025)
Technological5G coverage, high-income vs. low-income countries84% vs. 4%ITU (2025)
TechnologicalPeople without basic electricity, 2024About 655 million, leaving about 92% of the world with basic access. 89 million people gained access that year, but population growth meant the total fell by only about 11 million. About 86% of those without power (563 million people) live in sub-Saharan AfricaTracking SDG7 (2026)
TechnologicalPeople lacking clean cookingAbout 2 billionTracking SDG7 (2026)
TechnologicalAverage lag from invention to adoption, 15 technologies in 166 countries over two centuries45 years, with wide variation. Newer technologies have been adopted faster than older onesComin and Hobijn (2010)
TechnologicalIncome differences accounted for by differences in technology adoptionAt least 25% of per capita income differencesComin and Hobijn (2010)
TechnologicalOne Laptop per Child, Peru (318 rural schools)Computers per student rose from 0.12 to 1.18, with no evidence of effects on math or language test scores after 15 monthsCristia et al. (2017)
TechnologicalOne Laptop per Child, long-term follow-upNo significant effects on academic performance, school completion, or university enrollment, and some evidence of slower grade progression. Computer skills improved, but teachers' digital skills did not, and classroom computer use was lowCueto et al. (2025)
TechnologicalMindspark gains after 4.5 months0.37 SD math, 0.23 SD HindiMuralidharan et al. (2019)
TechnologicalSpread of true and false news on Twitter, 2006 to 2017About 126,000 verified stories studied. False news spread significantly farther, faster, deeper, and more broadly than true news in every categoryVosoughi et al. (2018)
TechnologicalDigital technology use and adolescent wellbeing (debated)Three datasets totaling 355,358 adolescents. The association was negative but small, explaining at most 0.4% of the variation in wellbeingOrben and Przybylski (2019)
TechnologicalRobots: employment and wage effect per robot per 1,000 workers-0.2 percentage points, -0.42%Acemoglu and Restrepo (2020)