NCBI Bookshelf. A service of the National Library of Medicine, National Institutes of Health.

National Research Council (US) Committee on Advances in Collecting and Utilizing Biological Indicators and Genetic Information in Social Science Surveys; Weinstein M, Vaupel JW, Wachter KW, editors. Biosocial Surveys. Washington (DC): National Academies Press (US); 2008.

Cover of Biosocial Surveys

Biosocial Surveys.

Show details

17Multilevel Investigations: Conceptual Mappings and Perspectives

, , and .

A century ago, antibiotics were nonexistent, public health was underdeveloped, leisure time was largely reserved for the wealthy, and germ-based diseases were among the major causes of adult morbidity and mortality. Improvements in living standards, public health efforts, leisure and lifestyles, and medical technology have made population health a notable success story in developed nations. Along with this improvement came an increase in life expectancy and a shift in the kinds of illnesses that cause death. In the early 21st century, most people today will avoid or survive the infections that were the major causes of mortality a century ago and instead die late in life from chronic degenerative conditions, such as cancer and cardiovascular disease. The combination of longer lives and increased prevalence of chronic conditions has raised concerns that people will spend their later years sick, limited by physical disabilities, and saddled with costly health care expenses. The biological and social sciences are rallying to address these issues. An important part of this response rests on data from the introduction of biological indicators and genetic information in social science surveys, which permit the investigation of associations across levels of organization that were inconceivable a century ago.

It is important to recognize that complex health states and outcomes tend to be multiply determined and are subject to contextual (e.g., environmental, cultural, social) as well as biological influences. Statistical models now exist that include stochastic error terms at various hierarchical levels of aggregation, which are applicable to the data matrices that span biological and social levels of organization. Our goal here is not to review these statistical models but to provide a more generic discussion of the conceptual issues that arise in the analysis of social and biological determinants of a healthy life span. We suggest that we must move beyond associations to mechanisms to meet the challenge of identifying the social and behavioral factors that influence the likelihood of remaining healthy and functional for the entire life span. The identification of associations and mechanisms depends on the accurate mapping of biological measures (e.g., biomarkers) to social and behavioral constructs in surveys. Such mappings will be aided by experimental or statistical controls for other factors (e.g., medications, time of day, activity level, body mass index) that influence biomarker expressions; attention to contextual variables (e.g., ethnicity) that may moderate the nature of the mappings; and a careful consideration of the sensitivity, specificity, and generality of the mapping in any given investigation. Before delving into these points, however, we describe briefly the nature of the data sets increasingly available to biological and social scientists.


The contributions to this volume demonstrate that scientific and technological advances have dramatically altered the data available to study complex behaviors and healthy aging. Estimates among biologists a decade ago were that 100,000 genes were needed for the cellular processes that are responsible for human behavior and aging, but humans have only a quarter that number of genes (Pennisi, 2005). This finding has fostered a recognition that a gene may have multiple small effects (pleiotropy), that many genes may act in additive and configural fashions to produce small effects both on specific abilities and on general abilities, and that genetic expression can be altered by the social as well as the physical environment in which humans live and work. The advent of single-nucleotide polymorphism (SNP) microarrays permits genome-wide association studies that would have been considered impossible less than a decade ago, and microarrays are on the horizon with which to study many if not all functional DNA polymorphisms in the genome (Butcher, Kennedy, and Plomin, 2006).

In addition to the global analysis of genes (genomics), technologies now exist for large-scale analyses of gene transcripts (transcriptomics), proteins (proteomics), and metabolites (metabolomics) in cells, tissues, and organisms. Among the important advances in quantitative analyses of these data is multivariate genetic analysis, which goes beyond analyzing the variance of each phenotype considered separately to analyze the covariance between them (Butcher, Kennedy, and Plomin, 2006). The number of SNPs and the number of various combinations of SNPs can be very large, however, and the complexity of the mapping problem is magnified by the presence of nonisomorphic intervening steps, which, for example, contribute to variation of phenotypic expression as a function of the physical or social context. Bioinformatics tools such as TELiS, which can be used to examine similarities in signaling pathways (transcription factor binding motifs) by genes that are found to differ between groups of interest, may be used to construct an intermediate level of organization, thereby improving the mapping of genotypes to phenotypes.

Developments in tissue and blood assays, ambulatory recording devices, noncontact recording instruments, and powerful and mobile computing devices have also burst onto the scene in recent years. These technologies make it possible to measure a variety of biological parameters in naturalistic as well as laboratory settings and in population-based health research. One such development is the use of drops of whole blood collected on filter paper from a simple finger prick to collect and analyze biological samples that previously required venipuncture (e.g., McDade et al., 2000). McDade, Williams, and Snodgrass (2006) identify over 100 analytes that can now be measured in dried blood spot samples, approximately half of which have particular relevance to population-level health research (e.g., cortisol, CD4+ lymphocytes, C-reactive protein, glycosilated hemoglobin, immunoglobin tumor necrosis factor, Epstein Barr Virus). With the inclusion of these measures in population-based health research, the weak associations that one would predict to exist between multiply determined variables (e.g., stress and C-reactive protein or blood pressure) and the potential influences of moderator variables (e.g., age, ethnicity, socioeconomic status) can be tested. These potential moderator variables may operate through differential reactivity (e.g., certain ethnicities or age cohorts may show salt sensitivity), differential exposure (e.g., certain ethnicities or age cohorts may consume more salt in their diet), or both. Distinguishing between these processes is crucial to moving from the description of associations to the delineation of causal mechanisms.

Recent “epidemics,” such as obesity and cardiovascular disease, cannot be fully explained in terms of genes alone, because major shifts in the human genome require much longer periods of time to unfold. These new health challenges require consideration of environmental exposures (e.g., the deployment of soda machines and fast food options in public schools) and individual differences in response to exposures (e.g., individual consumption patterns, salt sensitivity). Importantly, cultural, economic, political, social, psychological, behavioral, and environmental assessments are becoming more detailed, multidimensional, reliable, sensitive, and temporally rich. Early measures of social and behavioral predispositions were once characterized by general indices with poor reliabilities and validity. Although self-report measures often are viewed with suspicion because respondents may not be willing or able to respond accurately, the proper application of psychometric procedures to scale construction and validation and the inclusion of validating behavioral metrics have produced self-report measures with reliability and validity coefficients that rival or exceed the psychometrics of many physiological assessments (e.g., Burleson et al., 2003).

Experience sampling methods (Larson and Csikszentmihalyi, 1983) and day reconstruction methods (Kahneman, Krueger, Schkade, Schwarz, and Stone, 2004) are the sociobehavioral equivalent of ambulatory physiological recordings and make feasible the frequent sampling of social, psychological, behavioral, and biological states. The introduction of multilevel modeling (MLM) with temporal lags (Hawkley, Preacher, and Cacioppo, 2006) further permits frequent random-interval sampling and longitudinal analyses of environments, behaviors, social status, and biological responses. These new analytic techniques permit more powerful tests of mappings between social and biological domains.

Finally, spatially multidimensional electromagnetic, hemodynamic, and optical imaging devices, coupled with temporally precise electrophysiological methodologies, now make it possible to track changes in brain activity with impressive spatial and temporal resolution. The resulting data structures contain millions of elements that can span multiple levels of organization (Herrington, Sutton, and Miller, 2007). Although not yet appropriate for inclusion in population-based health research, these techniques make it possible to test specific hypotheses about the brain mechanisms underlying a variety of psychological processes, and the transduction of these psychological factors into peripheral biological activities and healthy aging. These laboratory techniques are already being used to test a subset of respondents in population-based studies, and ambulatory versions of electroencephalography are currently being developed.

Investigations of orderly associations in these data matrices, and especially of associations and causal connections across levels of representation, create significant challenges as well as opportunities. Knowledge and principles of physiological mechanisms, biometric and psychometric properties of the measures, statistical representation and analysis of multivariate data, and the structure of scientific inference are all important if veridical information is to be extracted from the confluence of these data matrices. In the remainder of this chapter, we outline a simple model to aid the mapping of elements (e.g., constructs, measures) across levels of representation.


The simplest method of mapping across levels is the correlative approach. There are notable success stories that have employed this approach, such as the Framingham Heart Study, which the U.S. Public Health Service launched in 1948. The early directors of this study, Roy Dawber and Bill Kannel, began with examinations, detailed medical histories, and blood tests of the more than 5,000 Framingham, Massachusetts, residents biennially. Since the inception of the study, new technologies and measures have been added, and epidemiological and data management methods have been incorporated to improve the scientific yield. Originally envisioned as a 20-year study, these researchers and their successors have now followed the health and lifestyles of the residents of Framingham for 60 years. The Framingham study has resulted in the publication of more than 1,200 peer-reviewed scientific articles (Levy and Brink, 2005) and the advancement of understanding of lifestyle factors in the etiology of cardiovascular disease.

By contemporary standards, the number of sociobehavioral and biological measures queried for possible associations was small. Still, the sample size, frequency of assessment, and duration of the longitudinal study were substantial. As Issa (2005) noted, the Framingham study illustrates how well-designed longitudinal studies that follow population-based cohorts can provide information on the etiology and natural history of the course of disease, generating hypotheses that are testable in laboratory research and clinical trials.

It is important to recognize, however, that not every association identified in the Framingham study proved to be robust or informative, and that for every Framingham study, there are others whose scientific yield is disappointing. The associations uncovered in correlative research, especially atheoretical correlative investigations, run a special risk of yielding false discoveries (i.e., nonreplicable associations), and this possibility increases with the number of possible associations that are examined. False discovery rate techniques have been developed to help mitigate this problem, but these techniques do not eliminate it (see Munafo et al., 2003).

The development and adoption of false discovery rate methods represent an advance in dealing with type I error rates, because the cost of near-zero false discovery rates was a high false negative rate. The ratio of the number of missed small discoveries to false discoveries can be substantially greater than one in studies of complex health outcomes, and small associations can carry large economic ramifications once scaled to the level of the population. Therefore, the cost of missing important but small associations (type II errors) can sometimes be greater than the cost of a type I error.

A strength of a correlative approach is the identification of associations that might be replicable and worthy of further study. An important goal of scientific theory is to describe the causal interrelationships among factors, thereby explicating the mechanism responsible for an association. Moving from the specification of associations to mechanisms is therefore an important objective for future research using biological measures in social science surveys. The correlative approach may generate variables (e.g., genes, neurophysiological circuits, demographic or lifestyle factors) or contextual moderators that are candidates for a causal mechanism.

In addition, the correlative approach also may not indicate the nature of the specificity of the association across levels of representation. For convenience, consider the constructs or measures at each level of representation as elements within a domain or set. The mapping between elements across such sets can take one of the following forms (see Figure 17-1):

  1. A one-to-one relation, such that an element in one set or level of representation is associated with one and only one element in another set, and vice versa. An example of a one-to-one relation, discussed below, is the prostate-specific antigen (PSA) assay as a measure of prostate cell activity.
  2. A one-to-many relation, meaning that an element of interest in the one set is associated with multiple elements in another set. An example is the orienting response and its mapping into a phasic heart rate deceleration and skin conductance response
  3. A many-to-one relation, meaning that two or more elements in one set are associated with one element in another set. (This differs from the preceding only when the order of the mapping across levels of representation—for example, social to biological—is specified.) An example discussed below is psychological stress, exercise, time of day, and other factors that can produce increased cortisol activation.
  4. A many-to-many relation, meaning two or more elements in one set are associated with the same (or an overlapping) subset of elements in another set. Not only can psychological stress and exercise, for example, influence cortisol activation but they can also have similar influences on autonomic, catecholaminergic, and immune activity.
  5. A null relation, meaning there is no association between the specified element in one set and those observed in another set.

FIGURE 17-1. Possible relationships between elements in two adjacent levels of representation (domains).


Possible relationships between elements in two adjacent levels of representation (domains). For illustrative purposes, these domains have been labeled “Social” and “Biological.”

Association studies involving elements with a one-to-one relation (absent confoundings and measurement error) produce high correlations, whereas association studies involving elements characterized by a null relation yield an essentially zero correlation. The strength of the association between elements across levels of representation can vary a great deal, however, for mappings (2) through (4), and a many-to-many mapping between two elements across levels of representation can produce correlation coefficients that are quite small, making them difficult to distinguish from a null relation, unless the sample size is large. Thus, the initial establishment of an association between elements across levels of representation through a correlative approach is not sufficient to determine the specificity of the mapping.

Why might it be important to go beyond thinking of associations to the interfaces between levels of representation? First, it is important if we are to move efficiently from association to the specification of mechanisms in our investigations. Second, the nature of the mappings between elements at different levels of representation determines the limits of interpretation one can draw about an association (Cacioppo and Tassinary, 1990).

Consider research in which a biological measure (e.g., salivary cortisol level) is known to correlate with a diagnostic category (e.g., a hypothesized state or condition such as “stress”). This established correlation may then be used to justify an interpretation of differences in the biomarker (e.g., salivary cortisol level) as evidence of differences in the diagnostic category (e.g., stress). This form of inference can be problematic, however. Even if we knew that variations in stress were associated with corresponding variations in salivary cortisol, inferring stress based on cortisol represents an error in interpretation because it ignores the possibility that there are other antecedent conditions that could also produce variations in cortisol. That is, it ignores the specificity of the association or mapping to the construct about which one would like to draw the inference.

It is tempting to suggest that these issues do not apply to genetics because there is no doubt that genes play a causal role in the production of complex behaviors and in age-related changes in these behaviors. To say that genes are causal is not equivalent, however, to specifying which gene, or set of genes, is associated with and causal in a particular phenotypic expression or, for that matter, in specifying the mechanism by which associated genes might influence a particular phenotype. Gottesman and Gould (2003) suggested that the number of genes involved in a phenotype is directly related to both the complexity of the phenotype and the difficulty of genetic analyses, and Butcher et al. (2006, p. 6) concluded that:

Multivariate genetic research consistently points to a single set of generalist genes that accounts for much genetic influence on diverse cognitive abilities…. Although each of the many generalist [quantitative trait loci or QTLs] will involve different molecular mechanisms, a QTL set will be useful in tracing the pleitropic pathways between genes and cognition through the brain to understand how generalist genes have their diffuse effects. These pathways will be complex and determining direct causation will be difficult.

Although difficult, such causal linkages will be more easily resolved if attention is paid to the implications of the many-to-many mapping problem. The mapping between elements across levels of representation may become more complex (e.g., many-to-many) as the number of intervening levels of representation increases.1 Accordingly, the likelihood of complex and potentially obscure mappings increases as one fails to consider intervening levels of representations. Admittedly, it is not always obvious which of several levels of representation might be “adjacent,” except perhaps when levels of representation refer to temporal rather than spatial scope. This caveat that mapping across levels of representation may be fostered by the incremental mapping of elements between proximal levels of organization nevertheless may have heuristic value. For example, endophenotypes such as neurocognitive deficits have proven valuable explanatory constructs between genes and psychiatric diseases (e.g., Gottesman and Gould, 2003; Nuechterlein, Robbins, and Einat, 2005), and in theory the same situation should apply to any mapping that goes from surveys to cells. For this reason, we focus here on the mappings between two adjacent levels of representation. The issues raised about the mappings between adjacent levels of representation can be extended to any number of adjacent levels of organization.

Taxonomy of Mappings

Tests, assays, or measures more generally have two different but related sets of characteristics. Analytic sensitivity is the ability to detect very low levels of the target analyte, whereas analytic specificity means that detection indicates the presence only of the target analyte. For example, blood sugar levels will vary in a predictable fashion for several hours after ingesting a dosage of glucose. Deviations from the normative values in blood sugar level across time mark a possible problem in metabolism because the blood glucose tolerance test (a procedure for mapping the glucose–blood sugar association) is sensitive and specific as long as the appropriate testing procedures are followed (e.g., fasting prior to the test) to eliminate the other known influences on the observed blood sugar excursions over the course of the test.

The properties of sensitivity and specificity, of course, depend on the elements involved in the mapping.2 For example, only prostate cells produce PSA, and an assay for PSA has very high sensitivity and specificity. But if PSA is mapped to prostate cancer rather than to prostate cell activity, the specificity and sensitivity are quite different. The sensitivity of the assay for PSA for detecting prostate cancer can be low (around 40 percent), as can the specificity (around 60 percent), because high PSA can be produced by active but noncancerous prostate cells. Said differently, the sensitivity and specificity of PSA for prostate cell activity are high, but the sensitivity and specificity of prostate cell activity for prostate cancer are more modest. Distinguishing between the mapping between PSA and prostate cell activity on one hand and prostate cell activity and prostate cancer on the other may make little difference to the physician using a PSA assay to screen for prostate cancer, but it would be important to consider for the researcher who is seeking to understand the mechanism for a measured association between PSA and prostate cancer in a large survey.

A third dimension is the generality of the mapping. In his influential Handbook of Experimental Psychology, S.S. Stevens (1951, p. 20) advised:

The scientist is usually looking for invariance whether he knows it or not. Whenever he discovers a functional relation between two variables his next question follows naturally: under what conditions does it hold? In other words, under what transformation is the relation invariant? The quest for invariant relations is essentially the aspiration toward generality, and in psychology, as in physics, the principles that have wide application are those we prize.

Is the mapping between two elements across levels of representation universally generalizable, or is it moderated by other factors? If it is generalizable without qualification, then the association requires no attention to characteristics of the context or sample population; that is, the mapping would have external validity. Invariant associations were once assumed, but statistical methods are now well developed to test for potential moderators (e.g., Baron and Kenny, 1986), and increasing attention is being paid to the operation of moderator variables. For instance, we raised the issue of moderators above when discussing differential reactivity and differential exposure.

A taxonomy of associations between elements across levels of representation is summarized in Figure 17-2. The initial step is often to establish that variations in an element in one domain are associated with variations in an element in another, thereby establishing an association. An outcome is defined as a mapping in which multiple elements at one level of organization (e.g., biological) are related to an element at another level of organization (e.g., social), and this many-to-one mapping may change across contexts. Initial association studies typically do not address issues of specificity or generality, and the treatment of such associations as invariants is premature.

FIGURE 17-2. Taxonomy of mappings among elements between adjacent levels of representation.


Taxonomy of mappings among elements between adjacent levels of representation.

An invariant relationship refers to a universal isomorphic (one-to-one) mapping between elements across levels or organization (see Figure 17-2). Invariant mappings permit the inference of an element at one level of organization based on the measurement of its isomorphic element at another. A marker is defined as a one-to- one, nonuniversal (e.g., context dependent) relationship between elements across levels of representation (see Figure 17-2). Many medical diagnostic tests, which have sensitivity and specificity only if explicit procedures are followed to eliminate other influences, are examples of markers. As such, inferences based on markers are similar to those for invariants as long as all other elements involved in the mapping are either experimentally or statistically controlled.

Finally, a concomitant refers to a many-to-one but universal association between elements across levels of representation and is similar to outcomes, except that the latter is not universal. Outcome and concomitant mappings enable systematic inferences to be drawn about theoretical constructs based only on hypotheticodeductive logic. Specifically, when two theoretical models differ in predictions regarding one or more outcomes or concomitants, then the logic of the experimental design allows theoretical inferences to be drawn about elements at one level of organization based on the measured elements in another. Strong inferences when dealing with outcome or concomitant mappings are limited to hypotheticodeductive reasoning.

When a new effect or association is found not to generalize to specific contexts or individuals, concerns are typically expressed about the methodological differences between the studies. Such a finding raises several important questions, including whether the original association is replicable and, if replicable, whether the diminution in effect size is attributable to measurement issues (e.g., reliability, construct validity) or to the operation of a moderator variable. Careful attention initially to the psychometric properties of all measures, regardless of their level of organization, to ensure their reliability and validity (including construct validity) therefore warrants attention in the design and analysis of studies going from cells to surveys.

In sum, many health states and outcomes can be multiply determined. To the extent that this is the case, investigators who assume rather than establish an invariant relationship between elements in the social and biological domains are at risk for predictably faulty interpretations. Investigators who incorporate cortisol measures in their social science surveys to indicate variations in stress, simply because stress and cortisol are correlated, are unlikely to contribute much to scientific understanding. This is because other factors that influence cortisol (e.g., time of day, time since consuming food) will be unrecognized and uncontrolled (see Adam, 2006). However, if the investigator next asks what is the specificity of this association, other antecedent conditions that influence cortisol are more quickly recognized, and contexts in which these other antecedents can be controlled experimentally or statistically can be developed to allow strong inductive inferences about the state of an individual's stress based on cortisol levels. That is, the sensitivity and specificity of the mapping of biological elements into macro levels of representation may be context dependent, and attention to these issues improves the quality of inductive inferences.

The term “biomarkers” does not distinguish among the various mappings that are possible between biological (e.g., hormonal, genotypic) and social (e.g., individual difference, phenotypic) representations. The taxonomy we have presented—biological outcomes, concomitants, markers, and invariants—may offer greater specification of these mappings and their properties and provide a useful framework within which to view these associations.


Interdisciplinary research that crosses biological and social levels of organization raises issues about how might one productively think about concepts, hypotheses, theories, theoretical conflicts, and theoretical tests across levels of organization. Abstract constructs, such as those developed by social scientists, provide a means of understanding highly complex activity without needing to specify each individual action of the simplest components, thereby providing an efficient means of describing the behavior of a complex system (e.g., “healthy aging”). Chemists who work with the periodic table on a daily basis nevertheless use recipes rather than the periodic table to cook, not because food preparation cannot be reduced to chemical expressions but because it is not cognitively efficient to do so. Reductionism, in fact, is one of several approaches to better science based on the value of data derived from distinct levels of organization to constrain and inspire the interpretation of data derived from others levels of organization. In reductionism, the whole is as important to study as are the parts, for only in examining the interplay across levels of organization can the underlying principles and mechanisms be ascertained. The goal of this chapter has been to outline a simple model to aid thinking about elements from different levels of organization.

In sum, the identification of associations and mechanisms in the complex multilevel data sets increasingly available depends on the accurate mapping of biological measures to social and behavioral constructs in surveys. Such mappings will be aided by experimental or statistical controls for other factors (e.g., medications, time of day, activity level, body mass index) that influence biomarker expressions; attention to contextual variables (e.g., ethnicity) that may moderate the nature of the mappings; and a careful consideration of the specificity and generality of the mapping in any given investigation. We hope the proposed formulation aids in this effort.


  1. Adam EK. Transactions among trait and state emotion and adolescent diurnal and momentary cortisol activity in naturalistic settings. Psychoneuroendocrinology. 2006;31:664–679. [PubMed: 16584847]
  2. Baron RM, Kenny DA. The moderator-mediator variable distinction in social psychological research: Conceptual, strategic, and statistical considerations. Journal of Personality and Social Psychology. 1986;51:1173–1182. [PubMed: 3806354]
  3. Burleson MH, Poehlmann KM, Hawkley LC, Ernst JM, Berntson GG, Malarkey WB, Kiecolt-Glaser JK, Glaser R, Cacioppo JT. Neuroendocrine and cardiovascular reactivity to stress in mid-aged and older women: Long-term temporal consistency of individual differences. Psychophysiology. 2003;40:358–369. [PubMed: 12946110]
  4. Butcher LM, Kennedy JKJ, Plomin R. Generalist genes and cognitive neuroscience. Current Opinion in Neurobiology. 2006;16:1–7. [PubMed: 16563732]
  5. Cacioppo JT, Tassinary LG. Centenary of William James's Principles of Psychology: From the chaos of mental life to the science of psychology. Personality and Social Psychology Bulletin. 1990;16:601–611.
  6. Cole SW, Hawkley LC, Arevalo JM, Sung CY, Riose RM, Cacioppo JT. Social regulation of gene expression in human leukocytes. Genome Biology. 2007;8(9):R189. [PMC free article: PMC2375027] [PubMed: 17854483]
  7. Gottesman II, Gould TD. The endophenotype concept in psychiatry: Etymology and strategic intentions. American Journal of Psychiatry. 2003;160:636–645. [PubMed: 12668349]
  8. Hawkley LC, Preacher KJ, Cacioppo JT. Multilevel modeling of social interactions and mood in lonely and socially connected individuals: The MacArthur social neuroscience studies. In: Ong AD, van Dulmen M, editors. Oxford Handbook of Methods in Positive Psychology. New York: Oxford University Press; 2006.
  9. Herrington JD, Sutton B, Miller GA. Data-file formats in neuroimaging: Background and tutorial. In: Cacioppo JT, Tassinary LG, Berntson GG, editors. Handbook of psychophysiology. third. New York: Cambridge University Press; 2007.
  10. Issa AM. Factoring risks to the heart. Science. 2005;309:1679.
  11. Larson R, Csikszentmihalyi M. The experience sampling method. New Directions for Methodology of Social and Behavioral Science. 1983;15:41–56.
  12. Levy D, Brink S. A change of heart: How the Framingham heart study helped unravel the mysteries of cardiovascular disease. New York: Knopf; 2005. [PubMed: 20704932]
  13. McDade TW, Stallings JF, Angold A, Costello EJ, Burleson M, Cacioppo JT, Glaser R, Worthman CM. Epstein-Barr virus antibodies in whole blood spots: A minimally invasive method for assessing cell-mediated immunity. Psychosomatic Medicine. 2000;62:560–568. [PubMed: 10949102]
  14. McDade TW, Williams S, Snodgrass JJ. What a drop can do: Expanding options for the analysis of blood-based biomarkers in population health research. Paper presented at the annual meeting of Population Association of American; March 31; Los Angeles, CA. 2006.
  15. Munafo MR, Clark TG, Moore LR, Payne E, Walton R, Fint J. Genetic polymorphisms and personality in healthy adults: A systematic review and meta-analysis. Molecular Psychiatry. 2003;8:471–484. [PubMed: 12808427]
  16. Nuechterlein KH, Robbins TW, Einat H. Distinguishing separable domains of cognition in human and animal studies: What separations are optimal for targeting interventions? Schizophrenia Bulletin. 2005;31:870–874. [PubMed: 16150960]
  17. Pennisi E. Why do humans have so few genes? Science. 2005 July 1;309:80. [PubMed: 15994526]
  18. Platt JR. Strong inference. Science. 1964;146:347–353. [PubMed: 17739513]
  19. Stevens SS. Handbook of experimental psychology. New York: Wiley; 1951.

The exception to this statement is when mappings among elements across adjacent levels of organization is one-to-one, but such mappings are atypical.

Context here is conceptualized as the constraints limiting the elements that are operating in a given measurement environment.



The exception to this statement is when mappings among elements across adjacent levels of organization is one-to-one, but such mappings are atypical.


Context here is conceptualized as the constraints limiting the elements that are operating in a given measurement environment.

Copyright © 2008, National Academy of Sciences.
Bookshelf ID: NBK62444
PubReader format: click here to try


  • PubReader
  • Print View
  • Cite this Page
  • PDF version of this title (7.4M)

Related information

  • PMC
    PubMed Central citations
  • PubMed
    Links to pubmed

Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...