Is It the Epidemiologists' Elusive "Fundamental Cause" of Social Class Inequalities in Health?
Is It the Epidemiologists’ Elusive “Fundamental Cause” of Social Class Inequalities in Health ?, Linda S. Gottfredson, 2004.
When groups differ substantially on the average in IQ but not other factors causing individuals to become ill or injured (e.g., non g-related genetic risk or motivation), then even small g-related differences in risk at the individual level can cumulate over persons to produce large group differences in rates of morbidity and mortality. This means that although risk factors that create group differences in health also create individual differences in health, the reverse is not necessarily true. That is, it explains how a single fundamental cause might account for most between-group, but not most within-group, differences in health.
g’s Practical Value Is Pervasive and Increases When Tasks are More Complex
Nearly a century of research (Schmidt & Hunter, 1998, in press; see also Gottfredson, 2002) shows that g predicts on-the-job performance to some extent in all jobs (a corrected correlation of .5, on the average), best in the most cognitively complex jobs (from about .2 in the simplest jobs to .8 in the most complex), best when performance is measured objectively and relates to the most core technical duties of a job, and almost always better than any other type of predictor.
With only one exception — conscientiousness/integrity — none of the less cognitive traits yet measured (temperament, interests, and the like) adds much if anything to the prediction of core job performance, except sometimes in narrow groups of jobs.
g affects job performance primarily indirectly, by promoting faster and more effective learning of essential job knowledge, during both training and experience on the job. Higher levels of g also enhance job performance directly when jobs require workers to solve novel problems, plan, make decisions, and the like. g appears to have increasingly large direct effects when jobs are less routinized or less closely supervised; more fraught with change, ambiguity, unpredictability, and novelty (and hence are inherently less trainable); or otherwise require greater exercise of independent judgment and “innovative adaptation” (Schmidt & Hunter, in press).
In contrast, the simplest, low-prestige jobs, in which g does little to predict performance, tend to be highly routinized, repetitive, supervised, and to call mostly for physical strength and tolerance of unpleasant physical conditions (Gottfredson, 1997b, pp. 100-105).
g Is at the Center of the Causal Nexus for Socioeconomic Success and Pathology in Adulthood
The second fact is that children’s IQ predicts their later socioeconomic success better than does their parents’ attributes. For instance, sons’ education, occupation, and income correlate higher with their own IQ (“true” correlations of .68, .50, and .35) than they do with either their fathers’ education (.43, .35, and .21) or occupation (.48, .44, and .29; Jencks et al., 1972, pp. 322, 337). The same predictive superiority of IQ over family background is found for the social pathologies too (Herrnstein & Murray, 1994; Hirschi & Hindelang, 1977; see also Gordon, 1997). ... IQ differences among siblings produce essentially the same degree of inequality in adult success and pathology among the siblings as do comparable IQ differences among strangers (Murray, 1997, 1998; Olneck, 1977, pp. 137-138).
Relation of g to Functional Literacy in Everyday Life
Many tasks performed in everyday life are the same as ones performed on jobs, from driving and cooking to planning and advising, so we might expect g also to predict the quality of self-maintenance and self-care in daily life — but again, especially when daily tasks are more complex.
The Psychometric Properties of NALS Functional Literacy Mimic Those of g
... the NALS was intended to measure “complex information-processing skills” that “go beyond simply decoding and comprehending text” by sampling a “broad range” of tasks that cover a variety of “universally relevant contexts and contents” and which can be “arrayed along a continuum” of difficulty (Campbell, Kirsch, & Kolstad, 1992). ... early analyses indicated that NALS literacy is actually highly general in nature ... In short, the NALS test of functional literacy appears to measure almost nothing but g, at least in native-born populations.
The Correlations of NALS Functional Literacy with Socioeconomic Outcomes Mimic Those of g
Functional Literacy Measures Highly General Skills and Abilities
The teachability of highly specific skills is therefore no guide to the teachability of a general underlying ability that promotes unassisted learning and the effective deployment of a whole panoply of specific skills in natural settings. In fact, military training programs that were intended to raise functional literacy were able to improve performance on specific work literacy tasks but did little or nothing to improve general literacy (Sticht, Armstrong, Hickey, & Caylor, 1987, ch. 9). Not all individuals learn equally well when exposed to the same instruction, because higher g promotes faster, more extensive, and more complete learning of what is being taught (Schmidt & Hunter, in press; Sticht, 1975).
Small g-Related Effects Cumulate and Snowball into Bigger Ones
Why one person rather than another misreads a particular bus schedule on any particular day has many causes and is probably little related to individual differences in g. Nor is such misreading, by itself, likely to be particularly consequential. The crucial point, however, is that g’s effects are pervasive and consistent. As gambling houses know well, even small odds in one’s favor can produce big profits in the long term when they remain consistently in one’s favor and other influences are more erratic. Information processing is involved in all daily tasks, even if only to a minor degree, so higher g always provides an edge, even if small. In contrast, other influences (fatigue, advice, etc.) tend to be more volatile and haphazard, and thus likely to cancel each other out over time. As the NALS data illustrate, people with higher literacy (g) tend to perform better on all literacy tasks, whether it is dealing with banks, restaurants, or social service agencies, deciphering one’s financial options, or engaging one’s rights and duties as a citizen.
The Remarkably General Relation between Social Class and Physical Health
Misreading a map or train schedule may be a nuisance, but misreading a prescription label can be a hazard. Such problems are the focus of health literacy researchers, who have independently confirmed that inadequate thinking skills impede effective self-care. They have also shown how seemingly minor inadequacies can cumulate into big health problems. Health psychologist Taylor (1991, p. 310) has pointed to the importance of intelligence for health self-care.
"Factors that influence patients’ ability to understand and retain information about their condition include intelligence and experience with the disorder. Some patients are not intelligent enough to understand even simple information about their case, and so even the clearest explanation falls on deaf ears."
The “Poverty Paradigm” Cannot Explain Social Class Differences in Health
The introduction of Medicaid and Medicare in the United States during the 1960s soon led to the poor making as many physician visits per year as the non-poor, but large class differentials in health remained — even when the poor began to visit physicians at a higher rate than the non-poor (Rundall & Wheeler, 1979, p. 397). Great Britain and other countries that had expected to break the link between class and health by providing universal health care were dismayed when the disparities in health not only failed to shrink but even grew (see The Black Report by Townsend & Davidson, 1982; also Link & Phelan, 1995, p. 86; Marmot, Kogevinas, & Elston, 1987, p. 132; Susser, Watson, & Hopper, 1985, p. 237).
... equalizing the availability of health care does not equalize its use. Perhaps most importantly, less educated and lower income individuals seek preventive health care (as distinct from curative care) less often than do better educated or higher income persons, even when care is free (Adler, Boyce, Chesney, Folkman, & Syme, 1993; Goldenberg, Patterson, & Freese, 1992; Rundall & Wheeler, 1979; Susser et al., 1985, p. 253; Townsend & Davidson, 1982, ch. 4).
Second, greater use of medical care does not necessarily improve health (Marmot et al., 1987, p. 132; Valdez, Rogers, Keeler, Lohr, & Newhouse, 1985). To illustrate, when a large federally-funded RAND-conducted randomized controlled experiment tested the effects of subsidizing health care costs at different levels in six cities across the United States, participants with free care used more medical care than those with only partly subsidized care, but their health was no better after two years. Participants with free care had indiscriminately increased their use of inappropriate as well as appropriate care (Lohr et al., 1986, p. 72). [...]
Third, health depends more now than ever on “private precaution” and “health lifestyle.” The American Psychological Society (APS) noted in its 1996 Human Capital Initiative report on health (American Psychological Society, 1996, pp. 5, 15) that:
"Seven of the 10 leading causes of death have aspects that can be modified by doing the right thing; that is, by making healthy choices about our own behavior….[Mortality] could be reduced substantially if people at risk would change just five behaviors: Adherence to medical recommendations (e.g., use of antihypertensive medication), diet, smoking, lack of exercise, and alcohol and drug use."
[...] The “paradox,” Susser et al. (1985, p. 254) note, is that increased public health efforts at “prevention in many instances ha[ve] widened the disparity in health between the social classes,” perhaps because “new preventive techniques have turned on personal behavior [e.g., not smoking] rather than on social engineering [e.g., controlling infectious disease by providing clean water and requiring immunizations for school entry].”
As Table 4 shows, prevalence rates sometimes differ a lot by race (diabetes, hypertension) and somewhat by sex (diabetes, heart, hypertension, and stroke). Education (and thus g, its strong correlate) cannot explain differences between the sexes nor many of them between the races, but it seems to operate in an identical manner within all the sex-race groups for any particular disease.
As already noted, improvements in a country’s overall health for any reason are often followed by bigger, not smaller, class disparities in health because health improves at a faster rate among the higher classes (Dutton & Levine, 1989, pp. 34-36; Pappas, Queen, Hadden, & Fisher, 1993; Steenland et al., 2002).
Recent discussions of what the fundamental cause might be have included social support, social connectedness, social anxiety, chronic stress (“allostatic load”), sense of personal control or mastery, experience of control, self-esteem, nutrition, relative deprivation, stigmatization, self-perceived social status, resistance resources, coping strategies, and intrinsic problem-solving capacities (Adler et al., 1994; Clay, 2001; Dutton & Levine, 1989; Link & Phelan, 1995; Marmot et al., 1987; Pincus et al., 1987; and many papers in Adler, Marmot, McEwen, & Stewart, 1999). To my knowledge, none has been shown plausible for explaining the full pattern — especially the generality and occasional reversal — of class disparities across time, place, and disease.
g Meets Key Criteria for Being the “Fundamental Cause”
For example, a 1933 study of 273 predominately white health areas in New York City (Maller, 1933) found that the average IQ of a health area’s school children ranged from 74 to 118. These average IQs correlated -.43 with the neighborhood’s rate of death, -.51 with infant mortality, and -.57 with juvenile delinquents arraigned in court.
A second large study, the Australian Veterans Health Studies (AVHS) Mortality Study, looked at IQ at the individual level. It examined the relation of 57 psychological, behavioral, health, and demographic variables to non-combat deaths among 2,309 Australian soldiers conscripted during the Vietnam War (O’Toole & Stankov, 1992). Prior IQ, which was measured at induction by the Army General Classification (AGC) test, was an important predictor of mortality by age 40. Controlling for all other factors, each additional IQ point was associated with a 1% decrease in risk of death (p. 711).
Another large longitudinal study looked at the precursors of low birth weight, which is of particular concern owing to its strong link to infant mortality (Cramer, 1995). It used the same national data from which the rates of social pathology in Table 2 were derived. Mother’s prior IQ (AFQT score) had a significant relation with baby’s weight at birth, but income had none after controlling for IQ.
1. Stability: Distribution of “causal agent” is stable over time.
Equalizing socioeconomic environments does little or nothing to reduce the dispersion in IQ, as was illustrated by the great variation in intellectual capacities among children born in post-WWII Warsaw despite the city’s Communist government providing the same (integrated) housing, medical care, and other amenities to all inhabitants (Firkowska et al., 1978).
4. Generality: “Causal agent” exerts widespread, similar influence on manifestly dissimilar targets.
Brand’s (1987) review of the diverse correlates of g included physical fitness, longevity, and a preference for low-sugar, low-fat diets and, in the negative direction, alcoholism, infant mortality, smoking, and obesity. Deary and his colleagues (2000b; Deary et al., in press) found that IQ measured at age 11 predicted longevity, cancers, dementia, and functional independence more than 60 years later.
They also found that mothers’ mental ability is a strong independent determinant of glycaemic control in diabetic children (Ross, Frier, Kelnar, & Deary, 2001). There is also considerable work showing that higher g is a key source of psychological resilience among children raised in extremely deprived, neglectful, or abusive environments (e.g., Fergusson & Lynskey, 1996; Garmezy, 1989; Werner, 1995), and that it helps protect adults against post-traumatic stress disorder (e.g., Macklin et al., 1998; McNally & Shin, 1995).
5. Measurability: “Causal agent” is amenable to empirical assessment.
Among the usual indicators of class, years of education is the most g loaded because it correlates .68 with IQ, whereas occupation and income correlate only .50 and .35 with IQ in large, fairly representative samples of men (true correlations; Jencks et al., 1972, pp. 322, 337).
Health Knowledge and its Relation to SES are Highly General and g Loaded
Chronic diseases are the major illnesses in developed nations today, and their major risk factors are health habits and lifestyle.
“[T]he most striking finding is the practically unitary character of knowledge: those who are best informed about one subject are likely to be best informed about any other” (Feldman, 1966, p. 166).
Hyman and Sheatsley (1947, p. 413) classified 32% of the public in 1947 as “a ‘hard core’ of know-nothings” that was beyond the reach of information campaigns: “there is something about the uninformed which makes them harder to reach, no matter what the level or nature of the information” (their emphasis). Four decades later, Bennett (1988, p. 486) concluded that, despite rising educational levels, “nearly 30 percent of the public continue to be know-nothings and that they remain concentrated in the same population sectors as in the 1940s.”
... the association of knowledge with social class seems to be stronger when the information in question is more widely publicized by the mass media (Tichenor et al., 1970).
Education-related relative risk was higher for the less educated when the public as a whole was better informed about a disease. For example, in 1955, 48 percent of the public could name at least one symptom of diabetes, 62 percent for cancer, and 69 percent for polio (Feldman, 1966, p. 90). However, the relative risk of persons with 0-8 years of education (compared to those with 9-12 years) not being able to name even one symptom was successively higher for the better known diseases, respectively, 1.7, 3.4, and 4.4 (ORs calculated from data in Feldman, 1966, p. 102). Moreover, the risk gradient for ignorance of the signs of cancer had steepened between 1945 and 1955, from 2.3 to 3.4 for the least educated, as more citizens had learned its signs (Feldman, 1966, p. 121).
Very importantly, Feldman (1966, pp. 140-148) also provided evidence that exposure is not just passive, but that more educated people seek out and attend to more information, which is indicated by their more extensive use of newspapers, magazines, and books (Feldman, 1966, pp. 140-148). ... That adopting health innovations involves active learning rather than just passive exposure accords with evidence that patients adopt birth control (Behrman, Kohler, & Watkins, 2002) and physicians adopt new antibiotics (Burt, 1987) more because of social learning and professional decision-making than mere exposure to social influence.
Health Literacy Predicts Health Knowledge, Health Behavior, and Health
"Over half of the 1.8 billion prescriptions written annually are taken incorrectly by patients ... Because they are used improperly, an estimated 30-50 percent of all prescriptions fail to produce desired results. ... Approximately 10 percent of all hospitalizations and 23 percent of all nursing home admissions are attributed to a patient’s inability to manage or follow drug therapy." (Berg, Dischler, Wagner, Raia, & Palmer-Shevlin, 1993, p. S5)
Worse yet, one study estimated that almost 30 percent of patients were taking their medication in a manner that seriously threatened their health (Roter et al., 1998).
Health literacy reflects mostly g.
... low literacy reflects “limited problem-solving abilities” and began describing literacy as the “ability to acquire new information and complete complex cognitive tasks”(Baker, Parker, Williams, & Clark, 1998, pp. 796-797).
A fourth sign that health literacy is largely g is seen in the strategies that health practitioners use to render health communications more comprehensible to low-literacy patients (Doak, Doak, & Root, 1996).
... educational level is a fallible guide to any particular individual’s literacy level because education through high school represents only years of exposure to learning, not actual accomplishment.
The Army found exactly the same result: its low-literate soldiers read four grade levels below the highest grade they had completed (10.7, on the average; Sticht et al., 1987, p. 45). NALS data also show that a quarter of young adults who left school with 9-12 years of education but no diploma read no better than the average 4th grader (Kirsch & Jungeblut, undated, p. 40).
Health literacy affects health knowledge.
Passing rates on even the simplest tasks tend to be low: 26% of the 2,659 patients did not understand information about when a next appointment is scheduled, 42% the directions for taking medicine on an empty stomach, and 60% a standard informed consent document (Williams et al., 1995).
To be effective, treatment for chronic illnesses such as diabetes and hypertension requires considerable, life-long participation of the patient. Patients with these illnesses presumably receive instruction from their doctors and are motivated to learn how to help monitor, medicate, and otherwise control their illness. Yet patients with low literacy still have shockingly low rates of knowledge about the most basic symptoms of their disease — ones, moreover, that often require them to take immediate action. For example, among 114 diabetics taking insulin daily, fully half of those with “inadequate” literacy but only 6% of those with “adequate” literacy did not know that feeling sweaty, hungry, and shaky is usually a sign that their blood glucose level is low. About 62% vs. 27%, respectively, did not know that if they suddenly feel that way, they should eat some form of sugar (Williams, Baker, Parker, & Nurss, 1998, Table 3).
Health literacy predicts health outcomes, even after controlling for social class.
Finally, a study of three urban hospital out-patient populations found that inadequate TOFHLA literacy was more strongly related to self-ratings of poor health (ORs of 1.89, 2.23, 2.55 in the three samples) than was education (ORs of 1.47, 1.53, and 2.13; Baker, Parker, Williams, Clark, & Nurss, 1997). Literacy appeared to be the active ingredient in education because education no longer correlated with self-rated health after controlling for literacy. ... Moreover, the relationship between literacy and self-reported poor health was not related to insurance status or self-reported difficulty in paying for medical care, getting time off from work, or obtaining child care (Baker et al., 1997, p. 1029).
Health self-management is inherently complex and thus puts a premium on the ability to learn, reason, and solve problems.
... health literacy researchers suggest that literacy is a highly general “learning ability” — an “ability to acquire new information and complete complex cognitive tasks” — and that “limited problem-solving abilities” make low-literacy patients “less likely to change behavior on the basis of new information” (Baker et al., 1998, pp. 796-797).
Patients cannot be passive recipients of medical recommendations with which they merely “comply.” Rather, many illnesses require the active participation of patients for proper diagnosis and treatment. We are our own “primary providers” of health care. This is especially true for chronic illnesses such as asthma, diabetes, and hypertension, because they require extensive “self-regulation,” which includes prevention, attack management, and social skills in maintaining social support (Clark & Starr-Schneidkraut, 1994). With asthma, for example, prevention entails “recognizing early signs of asthma, acting on early signs to ward off an attack, identifying and controlling triggers, and taking prescribed medicines properly and on schedule” (Clark & Starr-Schneidkraut, 1994, p S54). “Patients with asthma must deduce when and how best to use medicines [for example, with peak flow monitoring], because drug use in asthma is not just a matter of adhering to an absolute formula the physician provides” (p. S55).
Diabetes, another condition requiring close daily monitoring and adjustments in self-treatment, is even more demanding in this regard (Jovanovic-Peterson, Peterson, & Stone, 1999), especially for insulin-dependent patients, and yet more so for ones using the new forms of “tight control” (Juliano, 1998). These chronic conditions are similar to jobs that require considerable knowledge for good performance but, because conditions keep changing, the jobs cannot be routinized. Like such jobs, chronic diseases therefore require constant judgment in applying old knowledge and the need to spot and solve new problems.
The job of patient is becoming more complex and hence more g loaded.
The explosive growth in new treatments and technologies has created “tremendous learning demands” for anyone with a chronic disease. It therefore portends increasing relative risk for low-g patients.
“Today most illnesses are chronic diseases — slow-acting, long-term killers that can be treated but not cured” (Strauss, 1998, p. 108). They begin developing long before any symptoms appear, which puts a premium on foresight and prevention.
Relation of g and Social Class to Accidental Injury and Death
Injuries from accidents rival chronic disease as a public health problem. ... Injuries are “caused by acute exposure to physical agents such as mechanical energy, heat, electricity, chemicals, and ionizing radiation interacting with the body in amounts or at rates that exceed the threshold of human tolerance” (Baker et al., 1992, p. 4). ... While most accidents represent sudden acute assaults on physical well-being whereas ill-health is often chronic and slow-developing, both often emerge after a slow build-up of hazards.
Accidents Are Not Random, But Emerge from Patterns of Human Action and Inaction
... errors increase when tasks demand higher cognitive abilities. They found that error rates — “human error probabilities” (HEPs) — on work tasks in Air Force and nuclear power plant jobs generally correlated .5 to .6 with the number and level of cognitive abilities that the tasks required. A large study by AT&T estimated that it could reduce employee accidents by 17% and absences due to illness by 14% if it hired from the top 40% of applicants on an aptitude test (McCormick, 2001). And, as noted earlier, the Australian Veterans Health Studies found that IQ was the best predictor of motor vehicle deaths among veterans by age 40.
Failure to prevent incidents.
In systems under control (Stage 1), individuals necessarily pay only peripheral attention to safety most of the time, because their attention is concentrated on accomplishing the tasks at hand. However, a process that is “under control” (driving a car, mowing the lawn, or playing baseball) is seldom hazard free, so people must always be alert for signs that “something isn’t right” or might veer into danger. ... Many chronic diseases require daily self-regulation to keep body systems within safe limits (blood pressure for hypertension, blood glucose level for diabetes).
The greater the complexity, time pressure, or level of distraction in performing a task, the more difficult it is to maintain focus, monitor deviations, and even know how to keep activities on course. Most industrial accidents happen, as noted earlier, while workers are performing tasks that are complex or non-routine, and thus require them to solve new problems and use less-exercised skills (Hale & Glendon, 1987, ch. 4; Saari, Tech, & Lahtela, 1981).
Pedestrian accidents in U.S. cities have shown the same general pattern, in that most of the pedestrians and drivers involved failed to search for, detect, properly evaluate, or respond appropriately to existing signs of danger (Hale & Glendon, 1987, p. 43).
Failure to contain damage.
Recall that Arvey’s (1986) dominant “Judgment and Reasoning” factor among job demands correlated most highly with the ability to “deal with unexpected situations” (.75), “learn and recall job-related information” (.71), “reason and make judgments” (.69), “identify problem situations quickly” (.69), “react swiftly when unexpected problems occur” (.67), and “apply common sense to solve problems” (.66). These are precisely the skills that the accident prevention process calls upon most.
SES-Related Relative Risk of Accidental Death is General but Patterned by Accident Type
The more personal choice we have in conducting our lives as we see fit, the more our fate depends on our own knowledge, judgment, and foresight — and hence on g. All forms of accident are amenable to some control, even if limited, and therefore — like jobs — they all call for some exercise of g which, even if minor, can cumulate over time.
Generality of SES-related relative risk
Patterns of SES-related relative risk.
Fires are seldom “just accidental,” however. Cigarettes are the most common cause (28%), and children playing with matches account for another 10% (pp. 162-163). Half of adult fatalities in housefires show evidence of high levels of blood alcohol (p. 164). None of these kinds of error would seem to relate to economic status per se.
Relative risk becomes yet steeper for excessive cold (3.1 for people from the lowest income neighborhoods), excessive heat (4.4), and especially exposure/neglect (7.4), where the rates are especially high for persons aged 85+ (not shown). ... Poorer housing might conceivably account for much of the excessive exposure of the poorest elderly, who may often live alone, but risk is also somewhat elevated among infants too, who would never be living alone. Moreover, the SES gradient is especially steep when exposure/neglect is specified as the cause, which directly implicates faulty care. When these forms of accidental death occur among the other, more physically capable age groups, they may often be related to alcohol abuse, which is higher in the lower classes.
By far the largest number of fatalities in this “young male” set is from motor vehicle accidents, where relative risk is 2.4 for the lowest-income group. Differences in seat belt use (and hence the likelihood of surviving a crash) may be a factor, because one study found that adult drivers from high-income areas used seat belts at three times the rate of those in low-income areas, and that the discrepancy was even greater among teenage drivers (Baker et al., 1992, p. 223). Alcohol abuse and other risky behaviors may also be factors, because one third of all fatal crashes occur between 6:00 p.m. and 5:59 a.m. on Friday and Saturday nights and about 80% of males aged 20-55 killed in nighttime crashes have blood alcohol concentrations of at least 0.10 percent (Baker et al., 1992, pp. 244, 254). (A third of adult drownings also involve excessive alcohol [U. S. Department of Health, Education, and Welfare, 1979, p. 9-26].)
The next large set of unintentional deaths affects primarily adult males. ... The six with the steepest risk gradients represent common forms of job-related accidents, and hence are the sorts of accidents — explosions, cuts, electrocution, falling objects, and getting caught in machines — that (together with fires and auto, train, bus, and plane accidents) have been most studied in the accident literature. ... It is important to point out, however, that occupational exposure cannot fully account for the SES differences in relative risk because five of the six fatalities (excluding machinery deaths) occur as often from accidents at home as on the job (Baker et al., 1992, pp. 54, 114-133).