Integrating Biological, Behavioral, and Social Levels of Analysis in Early Child
Development: Progress, Problems, and Prospects
Douglas A. Granger and Katie T. Kivlighan Integration of noninvasive, biological measures into behavioral research has increased, but the interpretation of biobehavioral findings in relation to developmental outcomes is rarely straightforward. This commentary highlights the need for specific, theoretically derived hypotheses, multiple measures of behavioral and biological processes, and analytical strategies aimed at explaining interindividual differences in intraindividual change. It is suggested here that the next phase of biosocial research needs to move beyond description and toward development of mid-level theories that will enable researchers to specify, test, and refine hypotheses of how biobehavioral processes interact with social-contextual factors to influence development. These mid-level biosocial models will be necessary to determine whether individual differences in childrens adrenocortical activity confer risk or resilience because of early or cumulative exposure to nonparental care. Developmental scientists have long assumed that physiological processes were critical components of human behavior. However, the nature of many behaviorally relevant biological processes was un- known until recently. It is not surprising that with only a general regard for the influence of biology, the most well-researched models of human develop- ment have focused on issues such as how intrinsic dispositions (e.g., temperament) or characteristics of social environments (e.g., parentchild relationships) affect individual and maturational differences (see Bowlby, 1958; Harris, 1957; Lerner, 1986). Biologists who did study effects of specific physiological processes on development employed basic models to reveal the biological determinants of behavior or the unfolding of the predisposed, matura- tional effects of biology (e.g., puberty) on individual developmental trajectories. Few of these biological researchers took social variables into account, and when they did, the variables tended to be treated as covariates rather than as moderators (Raine, 2002). Advances of the past two decades have enabled new opportunities to integrate biological measures into developmental research. A series of conceptual shifts have occurred that place new emphasis on the contributions of both nature and nurture to indivi- dual development (see Magnusson & Cairns, 1996; Plomin & Rutter, 1998; Rutter et al., 1997). At the cutting edge of this paradigm shift are behaviorally oriented scientists testing biosocial alternatives to traditional models of development. The new bioso- cial perspective employed by these investigators is built on living systems theory (Ford, 1987). Develop- ment is considered a product of individual and interactive influences among genetic, environmental, social, and biological processes overtime (Cicchetti & Dawson, 2002). More specifically, Gottlieb (1992) speculated that biological functions set the stage for behavioral adaptation to environmental challenge. Simultaneously, environmental challenges may induce behavioral change that in turn affects fast-acting (e.g., hormone secretion) and slower responding (e.g., gene expression) biological processes. Biological activities that facilitate a particular behavioral process may be stimulated or attenuated by social forces. The ability to measure biological variables nonin- vasively in saliva has created many opportunities for behavioral scientists to test biosocial models in the context of child development. As the number of studies doing so has increased, it is apparent that interpretation of such findings is rarely straightfor- ward, and key issues outside the realm of basic behavioral training can limit the value of the informa- tion generated. The purpose of this commentary is to highlight some of these special issues so the next generation of studies might more rigorously integrate salivary biomarkers with behavioral and social-con- textual levels of analysis in developmental research. Biosocial Models: Specificity, Predictions, and Speculation Recent studies have clearly described interaction and moderator effects of the type expected by the r 2003 by the Society for Research in Child Development, Inc. All rights reserved. 0009- 3920/2003/7404-0010 Correspondence concerning this article should be addressed to Douglas A. Granger, Behavioral Endocrinology Laboratory, Department of Biobehavioral Health, The Pennsylvania State University, University Park, PA 16803. Electronic mail may be sent to Dag11@psu.edu. Douglas A. Granger and Katie T. Kivlighan, Department of Biobehavioral Health, The Pennsylvania State University. Child Development, July/August 2003, Volume 74, Number 4, Pages 10581063 biosocial perspective. Watamura, Donzella, Alwin, and Gunnar (this issue) revealed that in center-based child care settings, but not at home, salivary cortisol increases across the day. This context-sensitive effect emerges in infancy, peaks during the toddler period, and decreases by early school years. The increase in cortisol across the child care day is associated with socially inhibited behavior (i.e., fear and with- drawal). Booth, Johnson, Granger, Crouter, and McHale (2003) reported that age and stage of pubertal development were important for understanding the expression of testosterone-related problem behavior. Testosterones positive relation with risk behavior and negative relation with depression were also conditional on the quality of parent child relations. That is, as the quality of parentchild relationships increase, testosterone-related adjustment problems were less evident. Sethre-Hofstad, Stansbury, and Rice (2002) reported that in the context of a preschoolers exposure to a novel and potentially challenging task, individual differences in maternal sensitivity predicted the degree of attunement in mother and childs cortisol responsiveness. These studies clearly illustrate that social forces moderate the expression of biobehavioral relationships in children. Biosocial research has succeeded in describing interesting phenomena, but as Raine (2002) noted, the establishment of these associations is only the beginning. Watamura et al. (this issue) noted that the factors associated with age, time, and context-related differences in salivary cortisol change across the day could not be determined from their data, nor did they have evidence that the small increases in cortisol observed affect child development in any important way. Thus, several key questions remain unanswered: Are these changes in salivary cortisol a consequence of intrinsic stable individual differ- ences in socially inhibited behavior? Are these effects shaped by unique features of the child care day or determined by factors associated with social landscapes more generally? Do individual differ- ences in these cortisol-behavior-context associations confer risk or resilience? The mechanisms under- lying these interaction effects must be explained before we can begin to deduce their meaning for child development. We forward the notion that the implications of biosocial observations are often vague because our working models lack the specificity necessary to guide organization and interpretation, except at the most abstract level. Conversely, the depth of our knowledge regarding specific hormone-behavior- context relationships is often too shallow to enable us to shape and refine the biosocial perspective. As a consequence of both of these issues, studies often lack theoretically driven specific or directional hypotheses. Furthermore, discussion of biosocial ob- servations in the literature often falls short because investigators tend to either (a) digress to focus on micro-level detail of the molecular or cellular biol- ogy of the hypothalamic- pituitary-adrenal (HPA) axis (e.g., glucocorticoid neurotoxicity) inappropri- ate for level of analysis applied in the study (e.g., salivary cortisol), or (2) create a scaffold to support the importance of the observations by speculating well beyond the basic nature of the observed relationships (e.g., suggesting nonpharmacological change in the levels of cortisol over the child care day could affect brain development). To extend our understanding of these complex effects we believe it essential that we test specific theoretically derived predictions, refine biosocial theory on the basis of empirical observation, and be tentative with conclu- sions until we can eliminate competing rival hy- potheses. A Critical View of the Bio in Biosocial Models Involving the HPA Axis As behavioral scientists, we champion rigorous assessment using multiple methods. It is surprising, then, that behaviorally trained investigators often employ only a single measure (e.g., salivary cortisol) of the activity of a biological system, such as the HPA axis, in their biosocial studies. The emerging body of salivary cortisol literature, and admittedly our own work, has described cortisolbehavior relationships that are not always consistent. Theor- ists attribute the apparent inconsistencies to differ- ences in the nature of participant populations, individuals previous experiences, as well as real and perceived aspects of social context (Stansbury & Gunnar, 1994). Yet, despite the inclusion of social- cognitive and contextual variables in some of the most recent research, only a small percentage of the variance is typically explained. We forward the possibility that nearly sole reliance on salivary cortisol to assess the nature and function of the HPA axis in child development research may be worth reconsideration. More than 50 years of literature has refined our understanding of the HPA axis as a complex interconnected network of glands that communicate via soluble chemical messengers and self- regulate through positive feedforward and negative feedback loops (Nelson, 2000). When stimulated, the HPA axis releases a range of hormones into the bloodstream From Another Perspective: Biosocial Model and Early Child Development 1059 that have complementary, opposing, and synergistic functions. Our literature search reveals that dehydro- epiandrosterone and its sulfated ester (collect- ively referred to henceforth as DHEA[s]) may be important to include in studies that focus on HPA axisbehavior relationships. In its native form or as a precursor of other steroids, DHEA(s) affects a remarkable diversity of biologic actions (for reviews, see Majewska, 1995; Wolf & Kirschbaum, 1999). Like cortisol, the adrenal gland secretes DHEA(s) in response to adrenocorticotropic hormone (ACTH). DHEA(s) levels follow a diurnal rhythm similar to, but not as pronounced as, cortisol. Levels are high in the morning, decline 29% before noon, and decline 6% across the afternoon (Granger et al., 2003). During acute stress, DHEA(s) levels increase in response to activation of the HPA axis. DHEA(s) is likely to be coreleased with cortisol in response to events and psychological states typically studied in relation to cortisol secretion. Whereas cortisol is often reported to be associated with learning and memory impairment, social inhibition or internalizing behavior problems, and immunosuppression (Chrousos & Gold, 1992; Mc- Ewen & Sapolsky, 1995), research suggests that DHEA(s) has opposite effects. DHEA(s) is positively associated with aspects of learning and memory, and externalizing problem behavior, and it is capable of protecting neurons against the toxic effects of cortisol (see Majewska, 1995; Wolf & Kirschbaum, 1999). Unlike cortisol, there are dramatic develop- mental differences in DHEA(s) levels (De Peretti & Forest, 1976; Parker, 1999). Levels in serum are detectable through the first year of life, but then decline and remain very low until maturation of the adrenal gland in middle childhood (i.e., adrenarche). Starting at about 6 years of age, DHEA(s) levels begin an increase that continues each subsequent year through 20 to 30 years of age and then decline (Parker, 1991). Individual differences in DHEA(s) levels during middle childhood are considered to reflect the early stages of the transition through puberty (McClintock & Herdt, 1996). In females of all ages, and in prepubertal boys, the peripheral conversion of DHEA(s) is the major pathway for testosterone production (Parker, 1991). On average, females have higher levels of DHEA(s) than males (Granger, Schwartz, Curran, & Zakaria, 1999). The risk for initiation and progression of diverse pathological processes, potentiated by prolonged cortisol hyperactivity, is hypothesized to be directly related to the ratio of cortisol to DHEA(s) released by the adrenal. Thus, individuals with the high- est cortisol and lowest DHEA(s) would be at the highest risk (Hechter, Grossman, & Chatterton, 1997). Increases in DHEA(s) levels after exogenous ACTH stimulation are highest at ages when basal DHEA(s) levels are highFduring the first year of life and from middle childhood through young adulthoodFsuggesting possible developmental dif- ferences in the secretory products of the adrenal during the stress response. That is, during infancy and again during the elementary-school-aged years the HPA axis may be capable of secreting high levels of both cortisol and DHEA(s), whereas during the toddler period the HPA axis would secrete more cortisol than DHEA(s). Age and sex differences in the levels of DHEA(s) highlight the possibility that the effects of this hormone would be related to important individual and developmental differences in HPA axisbehavior relationships. With respect to the findings of Watamura et al. (this issue), the developmental peak of the observed increase in salivary cortisol across the child care day was observed during the toddler period when the ratio of cortisol to DHEA would theoretically be high. It would seem critical to know whether the adrenal production of DHEA(s) during this period is or is not adequate to counterbalance the potential nega- tive effects of cortisol hypersecretion. Without corresponding information about DHEA(s), the meaning of individual differences in cortisol secre- tion across the child care day seems more difficult to interpret. This introduction of DHEA(s) and discussion of its relationship to cortisol illustrates that the bio in our biosocial models may not always be operation- alized in a manner consistent with the dynamic and integrative nature of systems such as the HPA axis. The next generation of models designed to guide investigations of HPA axis behavior relationships will require comprehensive representation of both behavioral and biological processes. Exploration of ways to integrate multiple measures of key biologi- cal systems into biosocial research seems worth- while. Statistical Power, Analytical Strategies, and Assumptions of Linearity A review of studies applying the biosocial perspec- tive in child development research reveals short- comings related to analytical strategy and design. First, although most researchers suggest in their introductions and discussions that there are multiple sources capable of affecting variation in salivary hormone levels and reactivity (e.g., cortisol in particular), few studies are designed with sufficient 1060 Granger and Kivlighan statistical power, assuming effect sizes in the .20 to .30 range, to examine multiple independent and interactive effects. Second, almost as if by de- fault, the pervasive analytical strategy has been to test group differences using between-subject or repeated-measures ANOVA designs. This practice is inconsistent with the biosocial assumption that there are multiple interacting sources that affect individual differences in hormonebehavior rela- tionships. The analytical strategies of recent studies that have focused on explaining individual differ- ences in intraindividual change using profile or person analysis (see Cicchetti & Rogosch, 2001; Klimes- Dougan, Hastings, Granger, Usher, & Zahn- Waxler, 2001; Susman, Schmeelk, Ponirakis, & Gariepy, 2001), multivariate hierarchical modeling (see Adam & Gunnar, 2001; Cohan, Booth, & Granger, 2003), growth curve modeling (see Granger et al., in press), and latent-state trait modeling (e.g., Shirtcliff, Granger, Booth, & Johnson, 2003) are noteworthy. Third, the adult literature suggests that relationships between hormones (e.g., testosterone in particular) and behavior may best be character- ized using U-shaped rather than linear functions (Booth, Johnson, & Granger, 1999). It is not clear whether nonlinear relationships between hormones and behavior have been extensively tested in the child development literature. Perhaps the findings are null and are just not reported. Alternatively, this literature may have largely ignored what endocri- nologists have known for years: Hormone secretion and binding to receptors is not infinitely linear. If the basic physiology of how hormones affect target tissues is not linear, it seems counterintuitive that behavioral scientists have employed analytical stra- tegies assuming linear models of hormonebehavior links. Implications for the Study of Early Child Care In the most recent report, the National Institute of Child Health and Human Development (NICHD) Early Child Care Research Network (this issue) revealed that more time children spent in nonpar- ental care across the first 4.5 years of life predicts at- risk levels of externalizing behavior problems as well as assertiveness, disobedience, and aggression at 54 months of age and in kindergarten. This modest effect remained after controlling for a wide variety of social factors and processes including quality, type, and instability of nonparental care; potential effects of nonparental care on parenting; and maternal sensitivity and family background. A few studies suggest the biosocial perspective has the potential to predict unexplained variation in the long-term effects of early nonparental care on behavioral outcomes. Gunnar, Mangelsdorf, Larsen, and Herstgaard (1998) showed that maternal separa- tion influences childrens threshold for adrenocor- tical activation in response to challenge and that childrens adrenocortical activity rises across the child care day in relation to childrens concurrent social behavior (Dettling, Gunnar, & Donzella, 1999; Tout, de Hann, Kipp- Campbell, & Gunnar, 1998; Watamura et al., this issue). A robust pattern of findings also reveals a link between high levels of childrens externalizing problem behavior and low cortisol levels. This association is reported in studies of normal, at-risk, and psychiatric groups, and is stable across middle childhood (see Shirtcliff et al., 2003). Unfortunately, given the state of our knowl- edge, we cannot differentiate whether individual differences in the change in adrenocortical activity across the child care day are transient or stable, whether they are determined by early or cumulative exposure to nonparental care, or whether they confer risk or resilience with respect to the relevant behavioral outcomes, such as externalizing problem behavior. Clearly, sufficient evidence exists to suggest there may be potential value in the addition of noninva- sive measures of childrens adrenocortical activity in large-scale prospective studies of the effects of early rearing environments on child development. The data generated could be key in differentiating the direction of effect between adrenocortical activity and psychosocial adjustment, and behavior pro- blems, as well as the predictive validity of individual differences in adrenocortical-context specific effects on important developmental outcomes. However, before extensively studying biosocial relations in child care settings, it seems critical that we further develop our understanding of hormonebehavior relationships in the context of the family (Booth, Carver, & Granger, 2000). Our data on biosocial relationships in the family are remarkably thin, and these normative relationships may reveal important hints about the parameters most important to model in the child care context. Future Directions The next generation of biosocial research needs to go beyond simple description of interaction effects and research the fundamental mechanisms and pro- cesses underlying them (Raine, 2002). The search for these mechanisms, however, should not over- focus on discovery of the biological determinants of From Another Perspective: Biosocial Model and Early Child Development 1061 behavior. We underscore the need to develop our understanding of how behavioral, cognitive, and social variables shape individual differences in biological processes. It also seems likely that before our knowledge of child development can be ex- tended to new limits using the biosocial perspective, theoretical advances are needed to help guide and interpret the findings of future research. The appro- priate measurement tools and statistical strategies necessary to move this endeavor toward explaining and understanding the nature of biosocial phenom- enon now exist. The immediate need is for the development of mid-level theories that will enable us to specify, test, and refine hypotheses regarding the integration of biological processes with behav- ioral and social-contextual levels of analysis in developmental research.