Executive Summary Future Research in the NSF Social, Behavioral and Economic Sciences Submitted by Population Association of America October 15, 2010 The Population Association of America (PAA) is the premiere professional, scientific society for over 3,000 behavioral and social scientists who conduct research on the implications of population change. PAA members include demographers, sociologists, economists, health scientists, and statisticians. Population scientists rely on the National Science Foundation (NSF) for its support of large-scale longitudinal surveys, such as the General Social Survey and Panel Study of Income Dynamics. Population scientists also pursue NSF support for their own research projects and centers. The organization s recommendations reflect a desire that the NSF Social, Behavioral and Economic Sciences (SBE) Directorate will maintain and expand its investment in data infrastructure while also funding targeted, novel areas of research, over the next decade. The PAA recommends the NSF SBE Directorate pursue future research in five major areas: Challenge Area #1: Human Capital Investment and the Geography of Families Challenge Area #2: New Data for Studying American Families Challenge Area #3: Causal Inference in Demography Challenge Area #4: Behavioral Epigenetics and Epidemiology Challenge Area #5: The Study of Behavior Change This work is licensed under the Creative Commons Attribution-NonCommercial- ShareAlike 3.0 Unported License. To view a copy of this license, visit http://creativecommons.org/licenses/by-nc-sa/3.0/ or send a letter to Creative Commons, 171 Second Street, Suite 300, San Francisco, California, 94105, USA. 1
SBE 2020: Future Research in the Social, Behavioral and Economic Sciences Comments from the Population Association of America October 15, 2010 Thank you for the opportunity to comment on future research directions for the Social, Behavioral and Economic Sciences (SBE) Directorate at the National Science Foundation (NSF). The following recommendations are based on input from members of the Population Association of America (PAA) (http://www.populationassociation.org/). PAA is the premiere professional, scientific society for over 3,000 behavioral and social scientists who conduct research on the implications of population change. PAA members include demographers, sociologists, economists, health scientists, and statisticians. Their diverse range of research interests include population aging, disability, adolescent health, family formation and dissolution, population forecasting, poverty, health disparities, immigration and migration, mortality, and fertility. Population scientists rely on NSF for its support of large-scale longitudinal surveys, such as the General Social Survey and Panel Study of Income Dynamics. These surveys are essential, seminal resources population scientists use universally. Population scientists also pursue NSF support for their own research projects and centers. Our recommendations reflect a desire the NSF SBE Directorate will maintain and expand its investment in data infrastructure while also funding targeted, novel areas of research, over the next decade. Further, inherent in all of our recommendations is a hope NSF will maintain and expand its support of comparative international social and demographic research. Challenge Area #1: Human Capital Investment and the Geography of Families Why should theoretical models be developed and data collected to improve knowledge about the determinants and consequences of family geography? From investments in children's human capital to caregiving for aging parents, family issues have salience for research and public policy. No academic discipline has a monopoly on the study of families. Demography, economics, psychology, and sociology are all implicated. Most exciting work is explicitly interdisciplinary, and much research not explicitly interdisciplinary often uses insights from other disciplines. Typically, social science research focuses on households, often blurring or ignoring the distinction between households and families. Households are often the sampling units for data collection, making information about non-resident family members unavailable. Yet many important family relationships are across households, and who lives with whom, and who lives near whom, is endogenous. Investments in children's human capital are made by families and by the state (e.g., public schools.) Research on investments in children's human capital is rapidly progressing. Heckman and his collaborators, including both economists and personality psychologists, 2
have expanded the notion of human capital beyond cognitive skills to include health and socio-emotional skills. (See Heckman, James J. 2006. "Skill Formation and the Economics of Investing in Disadvantaged Children." Science, Vol. 312, 30 June 2006, 1900-1992.) One objective of Heckman's research program is estimation of the relationship between "inputs" (e.g., of parental time) and "outputs" such as cognitive and socio-emotional outcomes for children. This exciting interdisciplinary work deserves support. The degree to which families invest in their children's human capital varies. We need a better understanding of why families invest in children's human capital. The anomalous behavior may be of families that invest a lot rather than of families that invest a little. Especially in economics, studies of investments in children's human capital have focused on children reared in "traditional nuclear families." Sociologists generally look beyond traditional nuclear families and consider single-parent families. Yet even sociologists tend to focus on parents investments. Other family members investments (e.g., grandparents) are generally ignored. Many children, especially from low SES families, grow up in close proximity to their grandparents and other relatives. Close proximity facilitates time transfers (e.g., for childcare by grandparents), but we know virtually nothing about the effects of family proximity on cognitive and socio-emotional outcomes for children. Economists usually treat migration and location decisions as if they were made based primarily on labor market variables (e.g., wages; unemployment rates). Other family members (e.g., parents, siblings) are assumed no role in these decisions and, hence, family geography emerges as the almost accidental by-product of these decisions. We know little about factors determining the proximity of young adults to their parents. There is, however, some evidence suggesting that family considerations play a substantial role in location/migration behavior. Family proximity affects mothers' labor force participation, long-term care of disabled elderly and, perhaps, the cognitive and socio-emotional development of children. By encouraging new theoretical models and improved data collection, the SBE Directorate will enhance understanding of the determinants and consequences of family geography Challenge Area #2: New Data for Studying American Families Are existing datasets inadequate to study the changing American family? An emerging "grand challenge" is the ability to capture accurately evolving demographics of the American family. Without more sophisticated data collection over the next decade, researchers will be hard-pressed to analyze changing dynamics of the American family accurately and comment appropriately on the implications of these changes. Demographers agree a representative panel of the American family, designed to study family dynamics, does not exist currently. The National Survey of Families and Households (NSFH) is very dated and has some design problems, limiting its usefulness for studying contemporary family issues. The National Survey of Family Growth (NSFG) 3
is one of the leading datasets to study family dynamics, but it is not a panel and, thus, inherently limited. The Current Population Survey (CPS) marriage and fertility history supplements were discontinued. As a consequence, family demographers must use outdated datasets designed for other purposes. No existing data sets can be used to study newly-emerged aspects of the American family, including: increasing incidence of cohabitation and movements in and out of that union type; emerging blended families that include children from multiple marriages; growing intergenerational linkages from prime-age families to their adult children and to their increasingly surviving parents; and, increasing instability of household type (married, single, divorced, cohabiting) experienced by children over the course of their childhood. The frontiers of demography are studying these topics, yet the data are inadequate. Neither advancement in methods nor philosophy can circumvent the need for better data. If NSF is interested in advancing the discipline of demography, it should support the creation of a new national-panel on the dynamics of the American family. Challenge Area #3: Causal Inference in Demography Why does demography need its own methodology for causal inference? The field of demography has evolved in many different directions over the past decades, but one significant direction is in making causal inferences about the determinants of demographic outcomes like marriage, cohabitation, fertility, migration, urbanization, and other classic population variables. Like other social science disciplines, demography has not, historically, focused on rigorous methods of establishing causality, tending instead to assume causality when it seemed intuitively reasonable, or else settling for "proximate" determinants (namely, closely preceding events, even if not truly causal). However, most social science disciplines (economics, political science, public policy) as well as applied statistics have become increasingly concerned with causality. In statistics, Rubin and his coauthors have conceptualized causality in terms of counterfactual outcomes (i.e., what would have happened to a person if causal event X had not occurred). In economics, the work of Heckman, Imbens, Angrist, and others has clarified the counterfactual framework in applications to economic behavior and outcomes, and practical methods have developed for determining causality empirically. However, causality establishment in demography has not emerged for four major reasons: A large fraction of the applications in disciplines like economics have been to studying the causal impacts of public programs, laws, policies, and related variables. Demography, while occasionally interested in these kinds of questions, 4
is more interested in the fundamental determinants of demographic outcomes (e.g., trends in the American family). Many applications in statistics and other social disciplines have been "black box" in character--the "effect" of X on y is established, but no inferences can be made about the mechanisms or channels by which X affects y. Demographers, because of their interest in fundamental causes, are inherently interested in mechanisms or channels. Many methods in statistics involve explicit randomization (i.e., experiments). This approach is not possible in demography. Many applications in economics seek to maximize internal validity at the expense of external validity by studying narrow "natural experiments" on special populations or in special circumstances, which are unlikely to be generalized. Demographers are interested in population-level statistics and need to generalize to the population level. A fresh, directed attempt is necessary to establish methods for causality determination in demography. Research is necessary to adapt existing methods to demographic questions, since other disciplines methods are not applicable, and to develop new methods. Further, illustrative applications are needed to test causality determination in specific areas like marriage, fertility, and other outcomes. SBE is positioned to unite demographers with statisticians and methodologists from other social disciplines, and groups of practitioners interested in specific demographic questions, and advance this methodology for demographic research. Challenge Area #4: Behavioral Epigenetics and Epidemiology Will behavioral epigenetics revolutionize both genetic and social epidemiology? In a well-cited paper, 2004, Szyf, Meaney, and others demonstrated that social interactions can change the way genes function. They showed that rat pups who received the least amount of nurturing from their mothers also had the highest levels of methylated DNA at loci on the genome that are linked to glococortoid reception. These rats demonstrated a greater vulnerability to stress because they were physiologically limited in their health stress response due to blocked (methylated) DNA. If these same mechanisms exist in human beings, then these research implications are profound for many reasons: a. This research would provide a physical record of social environmental influences on human bodies. Accurately accounting for differences in genotype and epigenetic processes will reduce error variance and strengthen estimates of environmental influence. 5
b. If the social environment limits (or enables) the function of certain genes then misspecification (or lack of attention) to the complexities of the social environment may impede researchers from finding genetic loci linked to a number of different phenotypes. To date, little research has extended these results to human beings. Methylated DNA is tissue specific and many mechanisms of interest to social scientific communities are neurologically oriented; taking tissue samples from a living brain has some obvious limitations for most ongoing studies. Some work has examined post-mortem brain tissue and the results are consistent with the aforementioned animal studies. Large sample sizes (an expensive endeavor given the current costs of epigenotyping ), population- based sampling techniques (rather than case control studies limited to one phenotype), and environmental variation are required to evaluate properly how the environment influences epigenetic processes. Environmental variation is critical because it is associated w directly with detecting effects. Further, this factor begs the question, what is the environment? To answer, we need consensus on the following questions: At what level do we conceptualize the social environment (e.g., individual, peergroup, family, residential area, workplace, schools, etc.)? What domains of the social environment are the most important (e.g., cultural norms, institutional resources, environmental exposures such as pollution, etc.) Existing studies (see the Phenx project) are organizing common measures of the environment for genome-wide association studies, but to date there is little consensus. A number disciplines could collaborate on this topic, including: medicine, public health, demography, sociology, physical anthropology, social and genetic epidemiology, biostatistics, applied mathematics, and bioinformatics. Challenge Area #5: The Study of Behavior Change Can basic behavioral science elucidate what factors influence people to adopt positive behaviors, change behaviors and/or maintain behavior change? Research advances that could positively affect the health and well being of individuals and society are ineffective if never adopted. For example, research demonstrates consistently the benefits of exercise yet, few people participate in regular physical activity. Individuals know the environment benefits when communities recycle and conserve energy however, these socially oriented behaviors are not adopted universally. By encouraging interdisciplinary research collaboration among, for example, sociology, demography, and psychology, SBE could help unlock the underlying mechanisms affecting maintenance of behavior change. 6