On my most pessimistic days I view growing up as something like learning how to come to terms with disappointment. Summer doesn’t last forever. Sugar can kill you. And often, when researchers claim to have done their research, they really haven’t.
Speaking with a start-up working in nutrition a few weeks ago, I was surprised when one of the founders dismissed iodine fortification and supplementation as interventions that lack any clear significance. Upon further questioning, he directed me to the work of Charity Entrepreneurship, an offshoot of the Charity Science Foundation that attempts to advise new companies hoping to make an impact. As it turns out, iodine fortification and supplementation have been placed on their list of causes no longer worth consideration. According to their website:
“Salt iodization was one of the hardest interventions for us to evaluate. It scored middling to high in almost every area and this would normally put it in our top interventions. However the benefits of salt iodization are almost entirely IQ increases and we are unsure how to value IQ increases as we cannot find strong connections between increased IQ and other things, like happiness, income, or improved health in the third world. We are broadly utilitarian so we ultimately care about what leads to greater well-being in the world.”
Like Scott Alexander, I believe that “epistemological kindness” is valuable. Like Kolmogorov, I think that saviness is better than direct confrontation. But like any good effective altruist, I have no patience with inefficiency. Charity Entrepreneurship’s goals are incredibly admirable. And at the end of the day, we are both “broadly utilitarian” and want the same thing: for everyone to do good better. Yet as the staff at Charity Entrepreneurship currently do not have time for the consideration of counterevidence or the re-evaluation of their position, this post is an attempt to make sure that their advice related to nutritional interventions is taken with a pretty large grain of (iodized) salt.
The WHO estimates that iodine deficiencies affect roughly two billion people worldwide. The Global Health Data Exchange (2017) estimates that roughly 3,240,585.56 DALYs, or years lost to premature death and disability, over the past 15 years of alone. This insufficiency is typically measured via samples of urinary iodine concentration, with any amount less than 50 ug/l officially deemed deficient. (Morse, 2012) Accordingly, it may be argued that the WHO estimation should be taken as a baseline estimate for rates of iodine deficiency rather than an absolute value. Pregnant and nursing mothers need roughly 50% more iodine than the general population, due to both the dependence of fetal and infant brain development upon maternal thyroid hormones and the increased maternal glomerular filtration rate. (Morse, 2012; Glinoer, 2004) As iodine is passively secreted, this increase in renal glomerular filtration results in increased losses of ingested iodine. (Yarrington & Pearce, 2011) During pregnancy there is thus often more iodine present in the urine, though the mothers are not necessarily iodine sufficient. It is thus highly probable that the WHO estimate does not account for these mothers.
Facts like this are extremely alarming in light of the crucial role iodine plays cognitive development and function. (Zimmerman, 2009) Having been established as a mineral that is essential to the production of the thyroid hormones that “ensure the normal development of the brain and nervous system during gestation and early life”, it is required for normal neuronal migration, myelination, and synaptic transmission and plasticity during early fetal and postnatal periods. (Morse, 2012; Zimmerman, 2008) Severe iodine deficiencies during pregnancy are the root cause of maternal and fetal hypothyroxinenima, a condition which results in brain damage with mental retardation and neurologic abnormalities through various combinations of neurologic and myxedematous cretinism. (Halpern, 1994) This hypothryoxinenima has been deemed the world’s most frequent cause of preventable mental retardation in later life, because of both the irreversible nature of the negative effects it engenders and, as will be shown, the potential ease of iodine deficiency remediation. (Thyroid Public Health Association, 2006) Moderate, mild, and even asymptomatic iodine deficiencies during pregnancy have been linked with increased rates of spontaneous abortion, low birth weights, infant mortality, and serious cognitive impairment. (Bath, 2012)
Over the past two decades a number of studies have attempted to determine the potential affects of iodine supplementation during pregnancy upon the cognitive development of offspring. Quian et al. (2005) analyzed 37 studies (N = 12,291) in a recent meta analysis to conclude that children living in iodine sufficient communities possessed an IQ 12.45 points higher than those living in areas with no supplementation, 12.3 points higher than those living in areas with inadequate supplementation, and 4.8 points higher than those whose mother’s received supplementation before and during pregnancy. The total effect size was an increase of about 8.7 IQ points in the group that received iodine supplementation during pregnancy. Moreover, for those born more 3.5 years after the introduction of iodine supplementation, 12 to 17.5 point increases in IQ were demonstrated on the Binet and Raven Scale respectively. (Quian, 2005, p.33) These results are consistent with earlier meta-analyses, including Bleichrodt and Born (1994) and Scrimshaw (1998). Intervention trials in areas as remote as Papua New Guinea, Peru, Zaire, and rural China have demonstrated the efficacy of iodine treatments in severely deficient populations, with Pharoah and Connolly (1987) showing sharp reductions in endemic cretinism and Cao, Jiang, Dao, et al. (1994) reporting scores up to 10-20% higher in young children born to mothers that underwent treatment. (Zimmerman, 2008) Populations with only mild to moderate iodine deficiency have been shown to be similarly affected by iodine treatments. A study by Bath et al. (2013) on the Effect of Inadequate Iodine Supplementation revealed that the children of U.K. mothers with even mild iodine deficiencies were roughly 54-69% more likely to be in the lowest quartile for verbal IQ, reading comprehension, and reading accuracy. (Bath, 2013) It was also shown that these scores worsened depending upon the severity of the mother’s discerned deficiency, as seen in the graph below. Evinced by these studies, the effects of iodine deficiencies are clear and grave. Yet there is general consensus that iodine supplementation during pregnancy and infancy is effective in raising IQs by up to thirteen points, or nearly a full standard deviation, in deficient populations. It should be noted that there is some evidence that an oversufficiency of iodine can negatively affect the individual, but the level of iodine that qualifies as overly sufficient is extremely high and thus rarely attained. (Leung & Braverman, 2013)
Givewell’s analysis of current organizations working on salt iodization gives a thorough account of the flaws they possess and the pitfalls they have experienced. While there are a few charities in the space, none seem efficient or organized. This fact supports Charity Entrepreneurship’s claim that salt iodization efforts, like most charities, are “difficult to evaluate”. Yet the absence of good charities working on iodine deficiency only increases the need for further efforts in the space. Moreover, the difficulty of evaluating the significance of iodine and IQ demands that more research be done. Even a little knowledge of the field of intelligence renders their subsequent claim that increases in IQ do not possess any strong connections with positive life outcomes both false and disappointing.
Significance of IQ to Society:
What would happen if the IQ of the general population shifted by only three points? In 1994, Herrnstein and Murray utilized this thought experiment as the basis for a nationwide study on the significance of intelligence. For those who deemed IQ nothing more than a meaningless number, their results were shocking:
“For starters, [in America] the poverty rate falls by 25%. So does the proportion of males ever interviewed in jail. High school dropouts fall by 28%. Children living without their parents fall by 20%. Welfare recipiency, both temporary and chronic, falls by 18%. Children born out of wedlock drop by 15%. The incidence of low-weight births drops by 12%. Children in the bottom decile of home environments drop by 13%. Children who live in poverty for the first three years of their lives drop by 20%.” (Herrnstein & Murray, 1994, p. 365)
Though seemingly incredible, these results have been replicated by countless studies: the practical importance of general intelligence is too often underestimated. The most important findings regarding IQs contribution to society stem from the book IQ and the Wealth of Nations, which estimated the IQs of 81 different countries by amassing testing results and explored the correlation between IQ and GDP per capita (Lynn & Vanhanen, 2002). Since its publication in 2002, the IQ estimates have been revised and expanded twice. A study running an exponential regression model found that the correlation between IQ and GDP per capita to be r = 0.83 or 70% variance explained, which implies that a 10 IQ mean increase in a country predicts the doubling of its GDP per capita (Dickerson, 2006).
Similar analysis done by Rindermann et al. (2009) has drawn from these studies in an attempt to explain global developmental patterns, positing the Smart Fraction Theory, which claims that there exists an IQ threshold below which countries are not able to generate the civic foundations, inclusive of complex social networks and democratic engagement, upon which a thriving economy is built. The real GDP per capita of 185 nations is plotted against their mean IQs in the graph below.
A finding with this degree of societal importance and with a correlation value higher than almost any seen in the social sciences provides ample justification for skepticism. However, questioning the validity of the IQ estimations, three follow-up analyses breaking down and cross-referencing their estimations have found them to be legitimate (Whetzel & McDaniel, 2006) (Jones & Schneider, 2006) (Hunt & Wittmann, 2008) (Sandberg & Savulescu, 2011). A similar relationship can also be found between the economic performance of US states and their estimated IQ score averages (Kanazawa S., 2006).
Even with the existence of accurate data, the most prominent rebuttal to studies concerning IQ and the wealth of nations is that correlation does not equal causation. In other words, it is often asserted that reverse causation may be to blame whereby having a lower GDP per capita led to a lower IQ. Facts relating to the heritability of IQ weaken these claims to a degree, but not entirely. Thus Christainsen (2013) decided to intensify its analysis and employ GDP per capita, under 5 mortality rate, malnutrition, average years of schooling, parasitic load, and ancestral origin to try and predict the IQs of the currently adult population in across countries when they were born and developing. Using non-linear regression this study found that when malnutrition was measured, all of the other environmental conditions no longer statistically significant while all of the dummy variables regarding ancestral origin remained not only significant, but also possessed large coefficients. For example, when looking at the difference in IQ between Switzerland and Nigeria, the model found that just 4 points of the 22 point difference could be accounted for by malnutrition.10 The regression model was also overall very accurate in its ability to explain 86% of variance (r=0.93) (Christainsen, 2013).
It seems clear from these findings that the genetic and cultural implications of a country’s region of ancestry are the chief influence in regression analyses of IQ and the national wealth. Taking this fact and the scientific consensus surrounding heritability of IQ into account, the pattern of causality claimed by Smart Fraction Theory is evidentially sound. It is most likely that the mechanization relates to the influence of intelligence on the interrelated phenomena of innovation, accomplishment, and national development. The positive relationship between intelligence (g) and educational attainment and occupational success has been demonstrated repeatedly. The correlation values (r) are roughly .56 and .53 respectively, indicating that differences in intelligence account for an average of 30% of the variance in outcome typically observed in these areas (Strenze, 2007; Hunter & Hunter, 1984). Examining representative samples (N = 2254) of CEOs, billionaires, senators, and federal judges, intensive studies of the American elite vouchsafe that one-third to one-half of those in positions of power fall in the top 1% of the nation’s cognitive ability distribution (Wai, 2013). Those in both elected and unelected positions draw from this bracket. Moreover, the results of longitudinal studies of young individuals who demonstrate extreme intellectual talent further support the notion that these types of individuals consistently possess a serious capacity to contribute to society through innovation and excellence in fields ranging from the natural sciences to the creative arts: “constituting the far edge of a population whose continued success will be further emphasized – globally – for the foreseeable future” (Kell, Lubinski & Benbow, 2013).
Beyond tangible metrics of societal well-being, it has been shown that higher IQ leads to more prosocial action and societal cohesion, both of which are crucial to the formation of stable, productive societies (Fehr, Fischbacher, & Gatchter, 2002). In 2014, Proto, Rustichini, et al. conducted a study in which a group of participants were divided into two groups based upon their score on an administered Raven’s Progressive Matrices assessment and then asked to participate in an iterative series of prisoner’s dilemma games with legitimate monetary payoff. Even after controlling for personality type, risk aversion, and other possible confounders, their results showed significant differences between the performance of the high scoring Raven’s group and the low scoring Raven’s group. The graph below displays the cooperation trends for both groups, charting the high-scoring Raven’s group in blue and the low scoring Raven’s group in red with 95% confidence intervals (Proto, Rustichini, et al., 2014).
It is clear that the average cooperation for the high-scoring Raven’s group was significantly higher than that of the low-scoring Raven’s group. Much more likely to engage in reciprocal cooperation, the higher-scoring Raven’s group consistently utilized more effectively strategies: “Low Raven subjects play Always Defect with probability above 50 per cent, in stark contrast with high Raven subjects who play this strategy with probability statistically equal to 0. Instead, the probability for the high Raven to play more cooperative strategies (Grim and Tit for Tat) is about 67 per cent, while for the low Raven this is lower (around 45 per cent).” (Proto, Rustichini, et al., 2014, p.14) This finding has extended and replicated earlier findings (Al-Ubaydli, Jones, & Weel, 2016; Jones, 2008).
In the same vein, cognitive ability has been shown to anticipate a vast array of antisocial and criminal behaviors across populations. In fact, apart from age and sex, intelligence has proven to be one of the best predictors of this behavior. It has been found and replicated that those who display delinquent behaviors have a mean IQ of 92, independent of race, social class and gender (Lynam, Moffitt, & Stouthamer-Loeber, 1993; Short & Strodtbeck, 1965; Wolfgang, Figlio, & Sellin, 1987; Stattin & Magnusson, 1990). This correlation exists on a more macro level with the mean IQ of different US states (Bartels et al., 2010). Recent studies have also examined the relationship between county-level IQ and county-level crime analysis, as well as state-level IQ and state-level crime analysis (Beaver & Wright, 2011). Significant negative associations were found to exist between these intelligence averages and rates of larceny, aggravated assault, burglary, motor-vehicle theft, violent crime, property crime, and robbery. This propensity for criminal and antisocial behavior correlated with IQ also manifests itself in a heightened susceptibility to drug and alcohol abuse. The hazard ratio for a one standard deviation decrease in intelligence is 1.29, meaning even a moderate dip in cognitive ability results in a 29% increase in risk for substance misuse (Latvala et al., 2016).
As rigorous twin and intergenerational studies suggest, a significant portion of the effect is genetic (Latvala et al., 2016; Latvala et al., 2014). Controlling for poverty, race, or “concentrated disadvantage” does not attenuate the strength of the correlation between criminality and IQ (Koenen et al., 2015). The relationship is even evident during early childhood. Lower cognitive ability predicts the persistency of antisocial behavior. And life-course persistent antisocial individuals present the highest risk of succumbing to a range of undesirable and even criminal behaviors, including “substance abuse, drug-related violent crime, and violent crime against women and children (Moffitt et al., 2002). Though a number of researchers have recently attempted to prove that there is a significant “variation in the IQ-offending association across subscales of intelligence”, these studies often focus not upon intelligence, but upon the “executive functioning”, which they divide into an unorthodox set of abilities including: crystallized intelligence, fluid intelligence, “shifting”, and “inhibition” (Herrero, Escorial, & Colom, 2010; Herrero, Escorial , & Colom, 2010; Isen, 2010 ) While IQ and executive functioning as a whole do not exhibit extremely significant correlations, their results still clearly indicate an inverse relationship between both subsets of intelligence and criminality: “inmates score lower than controls on crystallized intelligence tests”, there is “worse [inmate] performance on updating processes” that are most closely “related to intelligence”, and there exists a “significant difference between offenders and controls on the administered fluid intelligence measure” (Herrero, Escorial, & Colom, 2010). More efforts should always be made to control for confounders, but it is noteworthy that studies like these produce the results that they do, though they take pains to account for the influence of socioeconomic status and educational attainment.
Replicated by numerous other studies, findings of this nature call into question the genius fallacy’s clichéd image of the highly intelligent individual – unadjusted and awkward, alone with their books in the library. Not only are the intelligent socially competent, they are able to perform considerably better in situations, like the prisoner’s dilemma above, that require an extreme degree of intrapersonal skill. It is worth mentioning that intelligence correlates significantly with other politically desirable effects as well. Examining the connection between cognitive ability, prejudice, and political orientation, a number of interesting studies have found that lower measured levels of general intelligence (g) or verbal intelligence in childhood predict racism in adulthood (Hodson & Busseri, 2012). Authoritarian leanings and aggressive tendencies toward social dominance have likewise been linked with low general intelligence (Heaven, Ciarrochi & Leeson, 2011). In light of the extensive and well-documented relationship between cognitive ability, antisocial disorders, these facts should not be extremely surprising. What is perhaps more surprising, is the significant correlations that have been found to exist between intelligence and height, as well as intelligence and attractiveness, further undermining the genius fallacy (Sundet et al., 2005) IQ also correlates with creativity, another crucial component of innovative and flourishing societies. Peter Thiel in Zero to One outlines the power of creative innovation as a way to use things around us more efficiently and create multiplier effects (Thiel & Masters, 2014).
This drives economic growth and improves quality of life. For example, just consider the way economic growth, driven by globalization and technology, has reduced global absolute poverty from 42% in 1981 to below 10% now (Economist, 2017).
Findings surrounding the correlation between intelligence and creativity are messy due to the difficulty of testing creativity and the range of different measurements for it, not to mention the different ways of measuring IQ. An important distinction should be made between creative potential – what one can achieve in a test environment – and creative achievement – what creative accomplishments one makes in the real world. Starting with accomplishments, it is largely agreed that creative achievements correlates linearly with intelligence (Kaufman & Plucker, 2011). Across three different studies, researchers were able to find that, using Math and Verbal SAT scores of 13 year-old children, who scored in the top 1% on these tests, they were able to predict creative accomplishments 20 years later. Even among these high performers, they found high odds ratios across domains. For example, between the top and bottom quintiles of these top performers in SAT math, those in the top quintile were 4.8 times more likely to have a patent (Park, Lubinski, & Benbow, 2007) (Park, Lubinski, & Benbow, 2008) (Wai, Lubinski, & Benbow, 2005). Park and others have used similar data to demonstrate that early “manifestations of [cognitive] abilities foreshadow the emergence of exceptional achievement and creativity in the world of work”. The graph below shows the bivariate means for age-13 SAT math (x-axis) and SAT verbal (y-axis) scores within “creative accomplishment” categories when participants were only 38 years old.
Creative potential is more highly debated due to the many ways in which it can be measured. An extensive meta-analysis of 21 studies around creative potential found the mean correlation to be r=0.17 and failed to confirm the threshold model whereby “intelligence is a necessary but not a sufficient condition of creativity” (Kaufman & Plucker, 2011, p. 772) (Kim, 2005). However, this study noted the heterogeneous results emergent from different forms of tests. The study is also questionable in that many of the studies used were over 30 years old and employed antiquated IQ tests (Kaufman & Plucker, 2011). A more recent study which cited this meta-analysis did find the threshold model for creative potential while confirming a linear model for achievement (Jauk et al., 2013). This paper discovered that having an IQ above 89 was necessary for the quantity of ideas one can generate, 104 for the quality of one’s best ideas, and 119 for the average quality of one’s ideas.11 Above these thresholds of IQ, personality factors were found to matter far more with the particular personality traits relevant depending upon the creative domain.
Significance of IQ to the Individual:
Turning to the effect that IQ has on the individual level, a higher IQ does not just reduce one’s chances of being unemployed, in jail, or homeless. A seminal paper in 2001 found that ” a person with IQ 115 (85th percentile) is 20% more likely to survive to age 76 than an average person with IQ 100.” (Whalley & Deary, 2001) .
This finding sparked a flurry of replication studies including a 2010 meta-analysis of 16 papers, which found that IQ in fact had an even higher predictive power at 24% and that a significant amount could not be controlled for by educational attainment or socio-economic status (Calvin et al., 2011). A 2015 paper looked at monozygotic (identical) and dizygotic (fraternal) twins intelligence versus life expectancy and found that the overall heritability of life expectancy, independent of IQ in particular, was 0.28, while the estimated correlation between intelligence and lifespan was r=0.32, or ~9% variance explained (similar to the finding from the meta-analysis when controlling for education and SES). Their model also found that 95% of this correlation was due to genetics (Arden et al., 2016).
It has also been found that the health and efficiency of one’s nervous system matters even more than IQ in affecting lifespan. Measuring reaction times to two different tasks, one study found that these tasks were more predictive of life span than IQ but that both IQ and reaction times were significant influencers of life expectancy even after controlling for smoking, education, and adult SES. Using forwards and backwards regression, the model picked only choice and simple reaction time which, when one standard deviation faster than the mean, gave 28% and 18% higher chances of dying at a given age, respectively. Taking into account these two variables made the IQ effect become no longer statistically significant. The reaction time variables were chosen instead of IQ because its effect size was slightly larger and because the tasks correlated with IQ at 0.49 and 0.31. These findings help explain how IQ and the genetics associated with it matter for longevity through having a more efficient nervous system that can better process information and has superior “physiological integrity”. The paper also acknowledged that the correlation between lower IQ and earlier death is especially strong in the lowest IQ quartile (Deary & Der, 2005).
Looking at other significant health outcomes in association with intelligence, an examination of hazard ratios associated with common diseases after controlling for likely confounders, including socioeconomic status, reveals a similar pattern: a one standard deviation advantage in intelligence is associated with a 30% reduction in risk of respiratory diseases, a 27% reduction of risk for coronary heart disease, and a 21% reduction of risk for strokes (Calvin et al., 2017). Moreover, increases in levels of intelligence are associated with decreases in instances of death by injury, as well as decreases in smoking-related cancers, type II diabetes, and digestive diseases (Gottfredson & Deary, 2004; Gottfredson, 2004). Meta-analyses find that low-intelligence correlated negatively with individual health outcomes in every stage of life, anticipating afflictions from obesity in childhood to dementia in old age (Arden, Gottfredson, & Miller, 2009; Batty et al., 2008; Belsky et al., 2013; Der, Batty, & Deary, 2009; Wrulich et al., 2013).
An inverse association has also been reported between mental health and intelligence, with a number of psychiatric disorders correlating inversely and significantly with measured cognitive ability. It has been shown that even after adjusting for a wide range of confounders, lower intelligence is prevalent among those with bipolar disorder, conduct disorder, and specific phobia: Adolescents suffering from past-year disorders possess a mean intelligence that is roughly one standard deviation below the mean intelligence of healthy individuals (Konen, 2009). Moreover, across the spectrum of disorders, higher severity is closely linked with lower levels of fluid intelligence (Keyes, 2017). The utilization of GWAS has even revealed associations between cognitive impairments and the polygenic architecture of certain diseases, like Schizophrenia (McIntosh et al., 2013). It is worth noting that there are some exceptions to the generally positive correlation between health and intelligence. Karpinski et al. (2018) recently conducted a study of members of the American Mensa Association, providing a survey through which these high-IQ individuals (98+ percentile) were asked to self-report the “prevalence of both diagnosed and/or suspected mood and anxiety disorders, attention deficit hyperactivity disorder (ADHD), autism spectrum disorder (ASD), and physiological diseases that include environmental and food allergies, asthma, and autoimmune disease” (Karpinski et al., 2018) When compared to the national average, the results of this survey revealed relatively high odds-ratios that were statistically significant. These findings form the basis of a preliminary “hyper-mind, hyper-body” hypothesis, that suggests extremely intelligent individuals possess mental “overexcitability” that possesses the potential to manifest negatively as anxiety and related disorders. Nevertheless, as Karpinski notes, “the hyper brain/hyper body theory is new and as such a number of studies will need to be carried out to better understand its strengths and limitations” (Karpinski et al., 2018)
I understand being skeptical of statistics, and even science in general. It’s healthy. Yet unless we want to operate under the premise that everything in life is equally uncertain, evidence supplied by statistics and science will have to be accepted with a critical eye. Taking this evidence as the jury in the case of Charity Science Foundation vs. Iodine Deficiency, iodine deficiency is the clear winner. IQ may be a social construct, but it is real and it is measurable. Not only does it predict positive life outcomes for individuals, it predicts positive life outcomes for societies as a whole. And in this light, you’d have to be mad to place iodine deficiencies in the category of “causes that no longer deserve consideration”. Salt iodization costs roughly 5 cents per person per year — spending .05% of a dollar per iodine deficient individual could curb the loss of billions of IQ points, and improve the lives and societies of millions in the process. If that isn’t doing good better, I guess I don’t know what is.