Cells and Surveys: Should Biological Measures Be Included in Social Science Research? (2001)

Kenneth W.Wachter

Biological indicators are nothing new to social surveys. Two such indicators, height and weight, tabulated by social class, were already being collected during the reign of Queen Victoria by the British Association for the Advancement of Science. Its President for 1885, Francis Galton, invented linear regression in the course of his analysis of heights. Height turns out to be a many-faceted biological indicator of health and standards of living and a sensitive probe of social differentials. The intellectual lineage that leads from Galton through Robert Fogel and his research partners is a major source for the present wide interest in biological indicators among social scientists.

In three other respects, the example of height is a good precedent for biological indicators in general. In the first place, height is the quintessential expression of an interaction between genes and the social and economic environment. Each child’s height is determined to a considerable extent by the genes of its parents, and yet observed differences across time and between groups and populations mainly reflect environmental influences associated with nutrition, physical effort, and disease. In the second place, the study of height is interwoven with the study of the social dimensions of health, successfully bridging the biomedical, economic, and demographic research communities. In the third place, from the very beginning, analysis of data on height has required and given birth to new statistical methods, from the fitting of normal distributions and linear regression onwards. All three observations are central to the new interest in biological indicators in social surveys: the primacy of

gene-environment interactions, the bridging of research communities, and the imperative for new statistical approaches.

As individuals, we know so much, but only so much, about ourselves. Self-reports as the mainstay of social surveys have taken us a long way, but now survey research is stretching itself to move beyond them. One sign of the change is the increasing investment in linking survey data to administrative data, for instance to earnings records or medical expenditure claims. Secure data facilities are being established to protect respondents’ confidentiality while facilitating flexible research. Another sign of the change is the panoply of initiatives treated in previous chapters involving the collection of genetic, anthropometric, endocrinological, cognitive, and physiological markers in conjunction with socioeconomic measures and demographic histories.

This volume is meant, first and foremost, to address the question of what we stand to learn, early on, from the inclusion of biological indicators in social surveys. The future will be changing rapidly and probably unforeseeably, but sensible first steps are within view.

When I think about the immediate value of including genetic and physiological indicators in social surveys, I find that I want to speak in praise of negative results. I fear that over the next decade we are likely to see a crude biological determinism trying to gobble up the social sciences. Envision streams of television and internet announcements of “the gene for X” and “the gene for Y” —the gene for math, the gene for lying, the gene for winning elections, the gene for millionairehood, the double-gene for billionairehood, the gene behind early retirement, early childbearing, early divorce, depression, satisfaction, long life, joy, and luck. One effect of the new genetics on the popular imagination will likely be the reinforcement of beliefs that complex outcomes have simple causes. The appeal of hidden variable theories of behavioral mechanics is doubled when the hidden variables take the name of genes and submit themselves to decoding.

If such claims were all silly, our problem would be media relations and public communication of science. But some such claims will not be silly. There may well be significant genetic determinants of some social behaviors. Sifting the valid from the spurious is a critical role for social science in the coming decade. It is particularly important because some claims may carry with them troubling political and ethical baggage. I am speaking “in praise of negative results,” because the most important priority may be the ability to say on firm evidence that a false claim is false, that a purported effect just isn’t there.

The social science community needs to be prepared to evaluate the efficacy of ostensible determinants of behaviors, basing its investigations on data that are representative at the population level and include appro-

priate control variables. Much of what has come out of genetic searches for genes figuring in diseases derives from studies of highly selected samples of individuals with little in the way of controls for demographic and behavioral covariates. Early announcements of findings generally overstate the strength of the effects that would be seen at the level of whole populations. Such announcements tend to exaggerate the simplicity of the genetic pathways themselves. Social scientists will not be in a position to evaluate claims and speak with authority unless the capability to analyze genetic indicators is already at hand in data sets which contain the relevant sociological, economic, and behavioral responses. If, in each case, we have to wait three years while we go out and collect new data, we will have excluded ourselves from the scientific debate. These stories, when they happen, play out quickly.

Summing up this line of argument, I believe it is a matter of urgency for us to have biological material collected by some of our leading social surveys as a resource ready to allow timely careful testing of new claims. This means we need to solve problems of appropriate safeguards and informed consent sufficiently well to have stored blood that can be used to type for new candidate genes as they come into the limelight of scientific debate. For a variety of reasons, among them imprecisely defined phenotypes and sparse coverage of family members and relatives, social surveys are not usually suited for hunting candidate genes associated with conditions and outcomes. But once candidate genes have been identified by geneticists and epidemiologists, the role of social surveys in confirmatory analysis becomes paramount. This role can only be realized with the creation of the required data resources in advance. Many of us are believers in preventive medicine. We also need preventive social science.

I turn now to the kinds of questions that the inclusion of biological indicators in social surveys would help us to address. A prime question, asked from a social scientist’s perspective: How pleiotropic is the world? How typical is it for a given gene or set of genes to be related to multiple phenotypic outcomes? Are the genes that will be discovered to have biological impacts on health also going to turn out to be associated with socioeconomic characteristics and choices? Will such alleles mostly be randomly distributed across subgroups or concentrated in particular subgroups? This is no simple question, considering the large number of dimensions along which population subgroups may be defined.

We are at a point where genetic epidemiologists are beginning to identify genes with alleles that are strongly enough associated with health outcomes to affect mortality rates at national levels. Apolipoprotein E is the prize example so far. Douglas Ewbank’s chapter in this volume cautions us against expecting large numbers of alleles with unambiguous

effects. But studies are already underway looking for alleles that are concentrated among survivors to extreme ages, and they will be bringing to attention a potpourri of genes whose behavioral associations would be ripe for testing. Studies of quantitative trait loci (QTLs) will similarly be identifying candidate regions of the genome associated with biological outcomes whose behavioral correlates will merit study. Will we start discovering a rich web of genetically grounded crosswise connectivity between our biological and our social life-course trajectories? Or will correlations typically wash out? Between the biomedical and the social sphere, has the genome opted for division of labor or entanglement?

Our conceptualization of the overall health impacts of genetic determinants is going to be very different if alleles with strong biomedical effects typically also bring along propensities for a whole set of ancillary behaviors. Genetic configurations may turn out to be associated—causally or through the accidents of genealogy—with risk-taking or health-promoting behavior, with choices about insurance or about human capital investment. They may be associated with propensities for caregiving, with familial stability, or with commitment to the maintenance of kinship networks. Or, in the end, there may be very little here to find. Such a discovery would also be liberating.

Genetic epidemiologists may already have the ingredients for answers to some of the questions about which demographers and sociologists are wondering. The divergent interests of the disciplines make communication tricky. Opportunities to work on common data sets can help create a common language, and the inclusion of biological markers in social surveys is a logical step. An attractive alternative is to include social science modules in future waves of ongoing surveys already replete with biological indicators. The National Health and Nutrition Examination Survey (NHANES) is a promising example. To reap the benefits, conditions for access to the genetic and biological variables would need to be made practicable for specialists outside the health sciences. Social science modules in health surveys would be limited in scope compared to the social surveys that currently collect full employment histories; components of income, assets, and transfers; or family and kinship constellations. Nonetheless, for many purposes, full detail is not essential, and modest social data linked to biological variables would open new horizons. A step-by-step approach is reasonable.

Drawing on ideas expressed by many of the authors represented in this volume, I have my own personal list of issues that I am anxious to see illuminated. Many of these involve anthropometric or physiological indicators rather than genetic markers. First in my mind is an issue of the different meanings of self-reported and measured responses. There is already a lot of work in this area. Survey responses about diagnoses ever

communicated by a physician are being compared to on-the-spot survey measurements of conditions. Self-reports of difficulties in activities of daily living are being compared to measured performance at specified tasks. Self-reports of earnings are being compared with linked earnings histories. Self-reports of expectations about one’s own future survival are being compared against life table projections.

Inclusion of a range of biological markers in social surveys would greatly enhance the range of comparisons available to study. The subject is important because motivation, risk assessment, frustration, or a sense of well-being are as much potential causal factors as cortisol, bank balances, and muscle tone. Furthermore, what we know about ourselves and what we say about ourselves to ourselves and to others is a component of life that is of thematic interest in its own right in this “information age.” What do people hear when they listen to their doctor or watch a pollution special on television? What is the personal probability calculus that people wield for their own decision-making, from youth through age? Coupling self-reports with measured and monitored outcomes is a clear next step toward understanding.

Several of the authors in this book look forward to a time almost upon us when unobtrusive monitoring devices will be able to collect extensive individual data on environmental exposures, activity levels, and physiological responses. I am told that there will be computers woven into my scarf, controlled by the twiddling of my fingers in thin air, and devices in my belt buckle capable of reporting to the survey researcher more about my movements and environmental encounters than I think I ever want to know. Experience with the more modest world of small numbers of carefully pretested biological markers in social surveys will help us prepare to keep our bearings in the face of technological opportunities.

A fairly specific research area in which the juxtaposition of biological and demographic variables will be critical is the interpretation of correlations between educational attainment and outcomes in the realm of health, disability, and mortality. What will help resolve the tension between the camp of researchers who believe that education has content that is causal and the camp who see mainly a drama of selectivity and social classification? Sets of intermediate physiological outcome variables gathered within the framework of longitudinal social surveys may be a key contributor. Current work with the Wisconsin Longitudinal Survey shows the promise of this line of research.

There is some possibility that genetic markers, hormonal assays, or related measurements might prove valuable as instrumental variables in statistical attempts to untangle arrows of causation. Such uses would involve variables whose values were unknown or only known after the fact to an individual. A related idea was highlighted by Robert Willis at

the workshop leading to this volume. Obligations to inform survey participants of the outcomes of biological measurements turn surveys more explicitly than ever before into interventions. If they are carefully and appropriately designed, longitudinal studies in the course of which test results are communicated to participants could aid understanding of how specific pieces of personal medical knowledge affect subsequent behavior.

One strand of research already in progress that may be a good guide to future opportunities is the set of studies focused on the concept of allostatic load, described briefly with references in the chapters by Crimmins and Seeman and by Weinstein and Willis. Allostatic load is a term for the long-range cumulative effects of the body’s physiological accommodations to stress, which can include changes that are adaptive in the short term, but which may also be injurious in the long run. Some researchers have operationalized allostatic load using a composite variable constructed from various biological measurements including systolic and diastolic blood pressure, cholesterol ratios, and cortisol levels. Viewed biomedically, the concept is far from simple. Evolution has equipped organisms with a wide variety of homeostatic mechanisms to cope with stresses whose presence may be bound up with biological success. Cumulative effects may be positive as well as negative, reflecting health rather than depletion. So far, however, as used in demographic applications, allostatic load is predominantly associated with waning health and gradually diminishing prospects for survival.

Going beyond specific applications, the concept of allostatic load suggests a certain philosophical stance about the nature of aging, a view summed up for me in lines from Matthew Arnold’s poem “The Scholar Gypsy”:

Studies of allostatic load are among the most successful forays so far into the world of biological indicators in longitudinal social surveys. Certain features of this work may be important considerations for the field as a whole. One is the demand placed on the development of new statistical and demographic methods. As a measure of cumulative stress, allostatic load should be determined by cumulative processes. New statistical models are being devised which take whole sequences of life-course events or experiences as predictor variables, in place of the one-by-one predictor variables familiar in linear regression. Patterns of challenge and recovery

rather than the presence and absence of specific positive or negative factors are treated as determinants in these new models. The approach makes heavy demands on sample size, but it has the advantage of coming closer to our intuitive sense of the cumulative longitudinal processes of development and aging.

The construction of the prevailing operational measure of allostatic load is itself interesting from a statistical point of view. It is an example of “dimensionality reduction,” a leading principle of multivariate statistical analysis. The measure of allostatic load is a composite index, a nonlinear function of ten variables. Instead of using all ten variables separately, a single index is created. The formula for this index was developed from theoretical considerations. Statistical methods are also available for constructing such indices automatically by searching for combinations of a given set of variables which capture as much as possible of the multidimensional variability in the data significant for some particular purpose defined by the researcher.

In the case of allostatic load, dimensionality reduction takes us from ten variables down to one index. It could take us from thousands of variables down to a handful. Statisticians are rapidly developing nonlinear search methods for dimensionality reduction taking advantage of available computing power. Such methods will only give useful results if some strong underlying relationship is actually present, hidden by its highly multidimensional character, and if the researcher can define what constitutes significant variability in a strategic way. The technology for checking thousands of genetic markers for each individual in a longitudinal survey at low cost may soon be ready. New techniques for obtaining other biological indicators may also provide thousands of variables. The challenge then will be to invent analytic strategies that can make meaningful use of the overwhelming riches of data. Sample sizes in longitudinal surveys will not grow in pace with the numbers of variables. Just as the opportunities of data on heights transformed the field of statistics a century ago, so the opportunities provided by biological indicators in social surveys will demand whole new forms of dimensionality reduction and nonlinear statistical analysis.

The words “allostatic load” prompt one more reflection. There is an allostatic load on every social scientist. Cumulative stress bears down on all of us. As biological indicators come to be included in social surveys, old kinds of expertise will become obsolete, new kinds of expertise will become mandatory, circles of collaboration will have to expand. Stress will increase.

It is pointless to provide for including biological indicators in social surveys unless we can provide for the researchers who can make intelligent use of them. The changes are too rapid for us to wait to grow a new

generation with the needed skills and interpersonal connections. How do we ease transitions by established researchers into this interdisciplinary territory? Training is part of the picture, but not the whole of the picture. We are talking about life-course transitions for researchers, with an interplay of intellectual, psychological, social, and physiological challenges. These are themselves examples of the kinds of processes we seek to study with biosocial surveys. As we study allostatic load, we need to learn to manage allostatic load and, borrowing Matthew Arnold’s words, renew the “elastic powers.”