The Intelligence Paradox: Why the Intelligent Choice Isn't Always the Smart One - Satoshi Kanazawa 2012

Intelligence and Substance Use
Why Intelligent People Drink and Smoke More

Alcohol

NCDS asks its respondents about the frequency and quantity of their alcohol consumption. First, it asks its respondents how often they usually have an alcoholic drink at ages 23, 33, and 42. I use factor analysis to compute the NCDS respondents’ latent tendency to consume alcohol frequently in their adult life. (Latent factors produced by factor analysis have the mean of 0 and standard deviation of 1.)

Then NCDS asks its respondents about the quantity of their consumption of different alcoholic beverages such as beer, spirits, wine, martini, sherry, and “alcopops” (that's British for flavored alcoholic drinks like wine coolers). NCDS asks these questions at ages 23, 33, and 42. Once again, I use factor analysis to compute the NCDS respondents’ latent tendency to consume a large quantity of various alcoholic beverages.

The analysis of the NCDS data shows that, net of sex, religion, religiosity, whether currently married, whether ever married, number of children, education, income, whether depressed, satisfaction with life, parents’ social class, mother's education, and father's education, more intelligent children are more likely to grow up to consume more alcohol in their adult life, measured both by frequency and quantity. Because such a large number of potential confounds are statistically controlled for, it is not likely (albeit technically possible) that the association between childhood general intelligence and adult alcohol consumption can be attributed to another factor.

For example, it's not likely that it is because more intelligent people are more likely to have certain occupations—such as executive or managerial positions that require socialization or negotiation over drinks—that childhood general intelligence and adult alcohol consumption are positively associated, because both education and income, as well as social class at birth, mother's education, and father's education, are controlled for. Interestingly, of these factors, only income and father's education independently increase the respondent's adult alcohol consumption, both by frequency and quantity. Education, social class at birth, and mother's education have no effect on adult alcohol consumption.

Add Health asks four questions about their alcohol consumption: How much they drink, how often they drink, how often they engage in binge drinking (five or more drinks in one sitting), and how often they get drunk. Once again, using factor analysis, I compute Add Health respondents’ latent tendency to consume alcohol.

The analysis of the Add Health data shows that, net of age, sex, race, Hispanicity, religion, marital status, parenthood, education, income, political attitude (liberal vs. conservative), religiosity, general satisfaction with life, whether they are taking medication for stress, whether they feel stress but do not take medication for it, frequency of socialization with friends, number of sex partners in the last 12 months, childhood family income, mother's education, and father's education (in other words, lots of potentially confounding factors), childhood intelligence significantly increases adult alcohol consumption. The more intelligent they are in junior high and high school, the more alcohol they consume in early adulthood.

Once again, given the even larger number of statistical controls in the analysis of the Add Health data than in the analysis of the NCDS data, it is very unlikely that the apparent effect of childhood intelligence can be attributed to something else. Neither income nor education is significantly associated with adult alcohol consumption. As with NCDS, father's education (but not mother's education) increases Add Health respondents’ adult alcohol consumption.

Figures 11.1 and 11.2 show the association between childhood general intelligence (grouped into “cognitive classes”) and adult alcohol consumption, by frequency and by quantity, from the NCDS data. As you can see, the association is perfectly monotonic. The more intelligent NCDS respondents are in childhood, the more they consume alcohol in adulthood. In Figure 11.1 for frequency, “very bright” individuals and “very dull” individuals are separated by nearly a full standard deviation. In Figure 11.2 for quantity, they are separated by four-fifths of a standard deviation. These effects are very large.

Figure 11.1 Association between childhood general intelligence and frequency of alcohol consumption (NCDS)

Figure 11.2 Association between childhood general intelligence and quantity of alcohol consumption (NCDS)

More Intelligent People Are More Likely to Binge Drink and Get Drunk

There are occasional medical reports and scientific studies which tout the health benefits of mild alcohol consumption, such as drinking a glass of red wine with dinner every night. So it may be tempting to conclude that more intelligent individuals are more likely to engage in such mild alcohol consumption than less intelligent individuals, and the positive association between childhood general intelligence and adult alcohol consumption reflects such mild, and thus healthy and beneficial, alcohol consumption.

Unfortunately for the intelligent individuals, this is not the case. More intelligent children are more likely to grow up to engage in binge drinking (consuming five or more units of alcohol in one sitting) and getting drunk.

Add Health asks its respondents specific questions about binge drinking and getting drunk. For binge drinking, Add Health asks: “During the past 12 months, on how many days did you drink five or more drinks in a row?” For getting drunk, it asks: “During the past 12 months, on how many days have you been drunk or very high on alcohol?” For both questions, the respondents can answer on a six-point ordinal scale: 0 = none, 1 = 1 or 2 days in the past 12 months, 2 = once a month or less (3 to 12 times in the past 12 months), 3 = 2 or 3 days a month, 4 = 1 or 2 days a week, 5 = 3 or 5 days a week, 6 = every day or almost every day.

As you can see in Figure 11.3, there is a clear monotonic positive association between childhood intelligence and adult frequency of binge drinking. “Very dull” Add Health respondents (with childhood IQ < 75) engage in binge drinking less than once a year. In sharp contrast, “very bright” Add Health respondents (with childhood IQ > 125) engage in binge drinking roughly once every other month.

Figure 11.3 Association between childhood general intelligence and frequency of binge drinking (Add Health)

The association between childhood intelligence and adult frequency of getting drunk is equally clear and monotonic, as you can see in Figure 11.4. “Very dull” Add Health respondents almost never get drunk, whereas “very bright” Add Health respondents get drunk once every other month or so, just as frequently as they engage in binge drinking, which makes sense, since binge drinking almost necessarily and by definition would make most people drunk.

Figure 11.4 Association between childhood general intelligence and frequency of getting drunk (Add Health)

In a multiple ordinal regression, childhood intelligence has a significantly positive effect on adult frequency of both binge drinking and of getting drunk (ps < .00001), controlling for age, sex, race, ethnicity, religion, marital status, parental status, education, earnings, political attitudes, religiosity, general satisfaction with life, taking medication for stress, experience of stress without taking medication, frequency of socialization with friends, number of sex partners in the last 12 months, childhood family income, mother's education, and father's education. I honestly cannot think of any other variable that might be correlated with childhood intelligence than those already controlled for in the multiple regression analyses. It means that the effect of childhood intelligence is not confounded with any of the variables already included in the equations. It is very likely that childhood intelligence itself, not anything else that may be confounded with it, increases the adult frequency of binge drinking and getting drunk.

Note that education is controlled for in the ordinal multiple regression analysis. Given that Add Health respondents in Wave III (when their drinking behavior is measured) are in their early 20s, it may be tempting to conclude that the association between childhood intelligence and adult frequency of binge drinking and getting drunk may be mediated by college attendance. More intelligent children are more likely to go to college, and college students are more likely to engage in binge drinking and to get drunk. The significant partial effect of childhood intelligence on the adult frequency of binge drinking and getting drunk, net of education, shows that this indeed is not the case. Childhood intelligence itself, not education, increases the adult frequency of binge drinking and getting drunk.

In fact, in both equations, education does not have a significant effect on binge drinking and getting drunk. Net of all the other variables included in the ordinal multiple regression equations, education is not significantly associated with the frequency of binge drinking and getting drunk. Among other things, it means that college students are more likely to engage in binge drinking, not because they are in college, but because they are more intelligent.

Tobacco

NCDS measures its respondents’ tobacco consumption by asking how many cigarettes a day they usually smoke at ages 23, 33, 42, and 47. I compute their latent tendency toward tobacco consumption by performing factor analysis.

To my surprise, and contrary to the prediction of the Intelligence Paradox, the analysis of the NCDS data shows that, net of the same control variables as above in the analysis of alcohol consumption, more intelligent British children are less likely to grow up to consume tobacco in their adult life.

Add Health measures its respondents’ tobacco consumption by asking on how many days they have smoked cigarettes in the last 30 days and how many cigarettes a day they smoked in the last 30 days. I compute their latent tendency toward tobacco consumption by performing factor analysis.

In sharp contrast to NCDS, and consistent with the prediction of the Intelligence Paradox, Add Health data confirm the prediction. Net of the same control variables as before in the analysis of alcohol consumption, more intelligent children grow up to consume more tobacco in their early adulthood.

The divergent effect of childhood intelligence on adult tobacco consumption is clear in Figures 11.5 and 11.6. Figure 11.5 depicts the monotonically negative association between childhood intelligence and adult tobacco consumption among the NCDS respondents in the United Kingdom. The more intelligent they are before the age of 16, the less tobacco they consume in their 20s, 30s, and 40s. Figure 11.6 shows the largely positive association between intelligence and adult tobacco consumption among the Add Health respondents in the United States. The association is not monotonically positive, but still “normal,” “bright,” and “very bright” Add Health respondents consume more tobacco than their “very dull” and “dull” counterparts.

Figure 11.5 The association between childhood intelligence and adult tobacco consumption (NCDS)

Figure 11.6 The association between childhood intelligence and adult tobacco consumption (Add Health)

Why Does Intelligence Affect Smoking Differently in the US and the UK?

I'm not sure what accounts for the divergent results from NCDS and Add Health when it comes to the effect of childhood intelligence on adult smoking. However, mine is not the only study which shows such varied results. Other studies¹⁶ have also shown that more intelligent Brits are less likely to smoke, while more intelligent Americans are more likely to smoke.

In my study, the two data sets are different in two major respects. First, NCDS is conducted in the United Kingdom, while Add Health is conducted in the United States. Second, NCDS respondents were born in March 1958, while Add Health respondents were born between 1974 and 1983. Further research is necessary to determine whether it is the cultural differences between the two (otherwise very similar) nations or the generational differences between the NCDS and Add Health cohorts that produce the strikingly divergent results when it comes to the effect of childhood intelligence on adult tobacco consumption.

Among the possible differences between the US and the UK, the public anti-smoking campaign has been far more aggressive and blatant in the UK than in the US. For example, in the US, each pack of cigarettes carries the Surgeon General's (relatively tame and clinical) warning (“Smoking causes lung cancer, heart disease, emphysema, and may complicate pregnancy”) in small print, on the side of the package. In the UK, the warnings are much more blatant and graphic (“Smoking kills,” “Smokers die younger,” “Smoking may reduce the blood flow and causes impotence,” “Smoking can cause slow and painful death,” “Smoking clogs the arteries and causes heart attacks and strokes,” “Smoking when pregnant harms your baby”) in extremely large print, on the front of the package. Note that death is never mentioned explicitly in the Surgeon General's warning in the US, but is frequently mentioned in the UK warnings. And they mention something much worse than death from an evolutionary perspective: impotence (for men) and harm to the baby (for women), and thereby implied lack of reproductive success.

When I saw the warning “Smoking kills” for the first time in 2003, on a pack of cigarettes that my LSE colleague was smoking, I thought it was a joke. It looked like a gag item that one might buy at a novelty store in a shopping mall, like Spencer's or Hot Topic. I didn't realize that it was for real until after I saw other packs of cigarettes with similar warnings later.

Because government warnings and public campaigns (as well as the written language as their medium of communication) are themselves evolutionarily novel, more intelligent individuals may be more likely to respond to such warnings than do less intelligent individuals. This is just one of the possible reasons why intelligence may have such starkly opposite effects on smoking in the US and the UK.

To be honest, I don't find this a particularly convincing answer myself. I feel relatively certain that the national difference in the effect of general intelligence on smoking is robust and real, not a methodological artifact, because different studies using different data sets and methodologies all confirm it. But I don't like my own explanation for it. I feel there is a better explanation out there; I just don't know what it is.

Drugs

At age 42 only, NCDS asks its respondents whether they have ever tried 13 different types of illegal psychoactive drugs: cannabis, ecstasy, amphetamines, LSD, amyl nitrate, magic mushrooms, cocaine, temazepan, semeron, ketamine, crack, heroine, and methadone. Via factor analysis, I compute NCDS respondents’ latent tendency to consume illegal drugs.

The statistical analysis of the NCDS data shows that, net of the same control variables as before, more intelligent children are more likely to grow up to consume more illegal drugs than less intelligent children. The higher their general intelligence before the age of 16, the more illegal drugs that they try before the age of 42.

Figure 11.7 shows the association between childhood intelligence, grouped by “cognitive class,” and latent adult tendency to consume illegal drugs. Just as with alcohol consumption, there is a monotonic positive association between childhood general intelligence and adult consumption of illegal drugs. But the effect of childhood general intelligence on adult consumption of illegal drugs is not as large as its effect on adult alcohol consumption. “Very bright” and “very dull” NCDS respondents are separated only by about one third of a standard deviation.

Figure 11.7 Association between childhood general intelligence and frequency of illegal drug consumption

Add Health asks its respondents about their consumption of five different illegal drugs: marijuana, cocaine, LSD, crystal meth, and heroine. Via factor analysis, I once again compute Add Health respondents’ latent tendency to consume illegal drugs.

The statistical analysis of the Add Health data shows that the effect of childhood intelligence on adult consumption of illegal drugs, while positive as predicted by the Intelligence Paradox, is not statistically significant. So Add Health data do not provide unambiguous support for the prediction of the Intelligence Paradox with regard to illegal drugs as do the NCDS data.