Amajor objective of the workshop was to examine methodologies used in family research to explore how different kinds of studies could be combined to yield a deeper and more accurate picture of family structures, processes, and relationships. In family research, biological and behavioral processes are often inseparable, but significant advances have recently emerged that offer new opportunities for distinguishing and measuring these processes with greater precision. The presentations summarized in this chapter demonstrate both the great potential of incorporating biological measures into family research and the considerable challenges in doing so.
Yet the integration of biological measures into family research can be difficult. The relationships between biological mechanisms and specific behaviors (such as parenting practices) are typically complex. In addition, integrating biological and behavioral research typically requires close collaboration among investigators with different backgrounds, training, and methodological perspectives.
It is important to note here that some domains of family research were beyond the scope of this single workshop. For example, the full range of biobehavioral approaches—including developmental epigenetics, gene-environment interaction, and developmental neuroscience—have all produced large new fields of research with relevance to the study of families in recent years. These are worth more attention, but it was not possible to integrate them into this workshop.
The presentations did review some focused sets of methodologies and
concerns. This chapter looks at three research approaches: family research on the biological stress response system, the effects of family life on child health, and the contributions of econometric studies to causal inference in family research. The research methodologies used in each of these areas are distinct, yet they share certain concerns and approaches that may offer a way of linking disciplines into multidisciplinary efforts.
ASSESSING THE BIOLOGICAL STRESS SYSTEM: CONSIDERATIONS FOR FAMILY RESEARCH
Environmental factors and life experiences affect human development, behavior, and health through their impact on physiological processes, such as activity of the biological stress response system. The activity of one component of this system—known as the hypothalamic-pituitaryadrenocortical (HPA) axis—affects nearly every organ system in the body, with impacts on cognition, emotion, memory, behavior, and health. Darlene Kertes, assistant professor of psychology at the University of Florida, described some of the strategies and challenges in examining the HPA axis in family research. She highlighted the need for methodological development to facilitate integration of multiple levels of analysis, from genes to the social environment.
The activity of the HPA axis is critical to maintaining homeostatic processes and facilitating adaptation to physical and psychological stressors. Two streams of input relay information about both systemic stressors, such as pain and inflammation, and psychogenic stressors, including actual and perceived threats in the environment. Both inputs act on the hypothalamus to trigger the release of corticotropin-releasing hormone. This initiates a biological cascade resulting in the release of glucocorticoids (cortisol in humans) into general circulation. Via feedback loops, cortisol acts to terminate the stress response as well as to sensitize brain regions involved in fear to shape an individual’s future behavioral and physiological responses to threat. Long-term effects of cortisol are achieved by its action as a transcription factor regulating gene expression in target tissues. Thus, the HPA axis is an adaptive system in which life experiences affect responses to future events, with potentially widespread consequences for behavior and health.
Whereas activity of the biological stress system is essential for life, chronic or repeated elevations may have deleterious effects. Disturbances in the HPA axis are linked with impaired growth in children, disturbed immune functioning, altered memory and attentional processes, and altered fear circuits in the brain. Altered activity of this system is also associated with a variety of disorders—psychiatric, gastrointestinal, and cardiovascular, among others (De Kloet et al., 2005).
Because cortisol can be used in both experimental and naturalistic settings, it is studied in a wide variety of family research contexts, Kertes observed. For example, research has shown that cortisol reactivity to a psychosocial stressor differs in the presence of a personal friend or spouse (Kirschbaum et al., 1995). Among girls exposed to maternal postnatal depression, basal cortisol levels at the transition to adolescence predicted future depressive symptoms (Halligan et al., 2007). Children of an alcohol-abusing parent showed altered cortisol reactivity in ways that are consistent with disturbances that predate alcohol dependence (Lovallo, 2006).
Kertes described two studies that document effects of early life experiences on HPA axis activity to illustrate strategies and challenges of studying the biological stress system in family research. The first study described long-term effects of early life adversity on basal cortisol levels in children. This study involved measuring cortisol levels among internationally adopted children, many of whom came from orphanages or other types of institutional care in which there was little opportunity to form relationships with stable caregivers (Kertes et al., 2008). Severe relationship deprivation early in life is known to lead to a pattern of growth delay in which linear growth (i.e., height) is primarily affected. This study showed that deprived care severe enough to impact children’s linear growth predicted subtle alterations in basal cortisol levels years after adoption into low-stress homes. Elevated cortisol levels were most evident in the early morning, at the peak of the diurnal rhythm, with no effect of deprivation-induced growth delay on cortisol levels observed at bedtime.
A second study described cortisol reactivity to a variety of novel social and nonsocial events among typically developing preschool-age children. This study tested a potential buffering effect of parenting quality on young children’s HPA axis reactivity (Kertes et al., 2009). There was evidence that children showed heightened cortisol reactivity to social or nonsocial challenges if they had a temperamental (behavioral) vulnerability to reacting to these types of events with fear and inhibition. For children very fearful of social interactions, having a sensitive, responsive parent—even though the parent was not present—buffered their biological responses to novel social events.
Whereas these studies document the impact of early experiences on children’s HPA axis activity, they also illustrate some of the challenges of detecting effects in biomarker data. It is actually quite difficult to elicit a biological stress response among children in an experimental context, Kertes pointed out. Ethical constraints limit the intensity of stressors that can be used, and experiments are terminated if a participant exhibits distress. One immediate and pragmatic solution is to target research ques-
tions aimed at identifying subgroups of individuals for whom particular kinds of stressors are likely to elicit a biological stress response. The study of typically developing children described above illustrates this point. In that study, Kertes et al. (2009) subjected 4-year-olds to a battery of mildly stressful events, including being separated from the parent, interacting with an experimenter that included some body contact, being asked to interact with strange, novel objects, and being approached in a conversation by a stranger. There was no evidence for an overall HPA axis activation among most children to this series of events. Rather, some children showed stressor-specific biological responses that directly related to their individual temperamental vulnerabilities. Children high in social fear showed biological stress responses to the social challenges but not nonsocial ones, and the opposite was true for children high in nonsocial fear. Thus, said Kertes, research questions can be tailored to detect stress responses within the ethical constraints of mimicking children’s everyday experiences.
“Targeted research questions are a pragmatic but limited solution,” said Kertes. The inherent challenge of ethically eliciting a stress response in children has resulted in the development of a large number of protocols with limited or varied effectiveness. Protocols that activate the biological stress response system that are both effective and ethical for use with children or across the developmental spectrum are particularly lacking. Basic science research is needed for standard methods of eliciting and assessing stress responses in research with children and families, with attention to the factors that most consistently elicit a biological stress response (for a discussion, see Dickerson and Kemeny, 2004; Gunnar et al., 2009).
Detecting associations between life experiences and biological measures is further challenged by the varied factors that impact the activity of biological systems. Cortisol levels, for example, are affected by digestion, sleep, exercise, systemic stressors (such as inflammation or pain), caffeine, alcohol, tobacco, endogenously regulated basal activity, and perceived or actual psychosocial stress. Typically, researchers interested in psychosocial influences impose sampling constraints (e.g., on food or drink consumption or sampling days) to minimize the impact of these factors. However, there may be physical or psychosocial stressors specific to certain populations or age groups that may confound results. For example, in grade-school children (particularly boys), cortisol levels differ on days that children participate in structured extracurricular activities like sports, compared with days when they are just in and around the home (Kertes and Gunnar, 2004). “This cautions us against erroneously attributing differences in children’s cortisol to some other variable if we don’t assess or control for it,” Kertes said. In the study described earlier on internationally adopted children (Kertes et al., 2008), elevated evening cortisol levels
previously reported among this population were not apparent when sampling was restricted to exclude days that children participated in sports.
Since limiting sampling for every possible known and unknown confound is impractical, another strategy is to refine statistical methods to disentangle variance that is stable in individuals or is due to some predictor of interest. For example, a structural equation modeling technique termed latent state trait modeling distinguishes variance in a phenotype that is due to stable, trait-like factors from the variance due to situational or state factors. As applied to basal cortisol data, approximately half of the variance in children’s cortisol can be explained by trait factors at both the peak and the nadir of the diurnal cycle (Kertes and van Dulmen, 2010). “This method might potentially allow us to improve our ability to detect subtle relations between environmental or behavioral factors and the stable trait-like component of cortisol in individuals while parceling out other factors that affect day-to-day fluctuations.”
Refining methods that facilitate the detection of family effects on HPA axis activity is likely to be of growing interest because of the impact of HPA axis activity on emotional and physical health. However, methodological innovation and statistical advances to facilitate analysis of environment-behavior-biological relations need to focus on the array of biological measures of interest to family research. At the physiological level, these include activity of the sympathetic adrenomedullary system, the immune system, and other steroids and peptide hormones as well as sleep disturbance/circadian rhythmicity and indices of brain functioning. All of these interact with the HPA axis in influencing behavioral and health outcomes. Advanced analytic techniques, refinement or standardization of protocols assessing momentary changes or basal activity, and growth of technologies capturing long-term activity with minimally invasive procedures are needed to foster this work. These methodological advances would facilitate the study of family effects on biological changes that influence risk for physical and mental disorders.
Another important conceptual and methodological issue in stress research is that stress biomarkers are often not correlated highly or even at all with behavioral measures of stressful life events or perceived stress. “Researchers are often very frustrated when they start to collect stress biomarkers and discover this fact,” Kertes said. From a methodological perspective, part of the reason for this uncoupling may be measurement concerns with the behavioral measures themselves (Monroe, 2008). Interactions with sex steroid or other peptide hormones may also play a role.
From a conceptual perspective, however, the uncoupling of behavioral and biological measures of stress to some degree is to be expected. When combined, they provide a more complete view of exposure and response. “Biological measures do not replace the need for behavioral
measures,” she said. “Both help us to disentangle stressors or even the same stressor acting on the biological stress pathway in different ways. For example, poverty might impact children’s cortisol via its effect on family stress, but it might also disrupt endocrine systems via the effects of environmental toxins. We need both levels of assessment to identify the mechanism of action.”
As Gilbert Gottlieb argued, events at various levels—environmental, behavioral, physiological, and genetic—constantly interact with one another in a multidirectional way over the life course. As applied to stress research, a stressor in the environment might elicit a change at the behavioral level (Gottlieb, 1992). If it does not sufficiently meet the challenge, it may also elicit a change at the fast-acting physiological level (including the HPA axis). If the immediate physiological response does not meet the challenge, it in turn elicits a change at the genetic level—that is, in gene expression. This suggests that coping resources at one level may prevent a stressor from impacting the individual at other levels. The results from the cortisol study with preschoolers illustrate this point. The 4-year-old children who were behaviorally fearful of social challenges did not show cortisol elevations in response to those challenges if they had a history of exposure to sensitive, high-quality parenting. “This speaks to the need for multiple levels of analysis,” said Kertes.
Methodological advances that promote multilevel research are also needed because family effects on emotional and physical health have multiple modes of transmission. These include direct genetic effects and gene-environment interplay, changes in gene expression initiated by the HPA axis or epigenetic mechanisms, and direct cultural or social modes of transmission. Capturing the joint and interactive effects occurring via multiple modes of transmission will require both collaboration across disciplines and cross-training of individual researchers, Kertes said.
One major methodological challenge to integrating across multiple levels of analysis, particularly when bridging biological and behavioral data, is balancing the need for deep phenotyping of behavior and the environment with the need for sufficiently large sample sizes to detect interactions among the environment, behavior, and biology. This is particularly true for research involving genetics, in which the effect of any given genetic variant is small for complex traits. Although comprehensive genotypic and phenotypic assessment is ideal, another strategy is to balance these various priorities across a program of research rather than an individual research study. For example, HPA axis disturbances are believed to play a role in stress-related mental health problems, including alcohol dependence and major depression. Family and life stress may in part promote these biological changes and emergence of disorder, but genetic risks are also likely to be involved. Gene-identification studies with large sample sizes but limited
phenotyping can identify potential genes of interest, such as those involved in neurotransmission or the biological stress response (Kertes et al., 2011). Top candidates then can be integrated in studies with family, developmental, and/or physiological data to ask meaningful questions about the interplay of genetic risks with psychosocial factors on behavioral or biological functioning.
Methodological development that supports integration across multiple levels of analysis has two key benefits. First, resolving the challenges inherent to integration across disciplines can fuel conceptual and methodological innovation in the disciplines from which they draw. Second, integration of biological data in family research has the potential to personalize preventive interventions, in which modifiable environmental conditions can buffer individuals’ risks for poor outcomes in the face of biologically influenced vulnerabilities.
In sum, integration of physiological processes in family research is important because they serve as mechanisms by which family experiences impact an individual’s response to future events as well as their emotional and physical well-being. Implementation, however, requires careful attention to methodology, and challenges remain. Nevertheless, because family effects are transmitted through physiological and genetic routes as well as through social and cultural routes, multiple levels of analysis are needed to adequately capture the effects of family life on individual behavioral and health outcomes.
INSIDE FAMILY LIFE: MULTIPLE LAYERS OF INFLUENCE ON CHILDREN’S HEALTH AND WELL-BEING
Children’s health is rarely if ever the result of a single factor, said Barbara Fiese, professor of human development and family studies and director of the Family Resiliency Center at the University of Illinois at Urbana-Champaign. It is embedded in a familial, social, and cultural context that changes over time, including parents’ beliefs and practices, neighborhoods, and access to health care, among others. Even something as straightforward as feeding a child becomes subject to the effects of income, media, and peers as a child grows up.
Many daily activities support the health of children, including routines created to support eating, sleeping, and physical activity. More broadly, family health is sustained through planning, open and direct communication, a sense of order and routines, and a belief that challenges in everyday life are manageable (Fiese, 2006). Family health is compromised when planning is absent or thwarted, routines are disrupted, communication is strained, and everyday life challenges consume personal energy.
Multiple factors can be combined in a cumulative risk model to predict childhood health problems. These factors include such things as poverty, parents’ perceptions of discrimination, neighborhood factors, and cultural stress. However, these factors do not reveal much about what happens in a family over time. Also, the focus on a single disease state does not reflect what often happens in real life.
Fiese described several studies involving family life and asthma. The studies were conducted in upstate New York and in Denver, Colorado. They involved approximately 400 Hispanic, black, and white families with a child between ages 5 and 12 with persistent asthma. About 58 percent of the families had two or more adults in the household, and 30 percent of the mothers had a high school education or less.
Asthma is the most common chronic illness of childhood. In any given classroom, 1 child in 10 is likely to have a diagnosis of persistent asthma. It is an expensive disease to treat, but it is treatable. Comorbidities include anxiety, sleep disturbances, and overweight conditions.
The household routines needed to manage asthma include taking medication twice a day, avoiding such environmental allergens as tobacco smoke and pet allergens, engaging in daily physical activity, and getting a good night’s sleep. At the same time, families with asthmatic children have to juggle home and work life, they move and experience job loss, they have babies and get divorced, they have to care for their elders, they experience domestic violence, they have psychiatric illnesses and suicidal ideation, they are involved in gang killings, and sometimes their children die. “All of these experiences have happened to members of the families in our studies,” she said.
Fiese examined three questions during her presentation:
- Are routines associated with children's health and well-being?
- Are different aspects of routines associated with different health outcomes?
- How can the study of household routines inform the study of health comorbidities?
Lung functioning was ascertained through spirometry tests. The study also gathered parent and child reports of functional severity, such as how much the child was wheezing and coughing or waking up in the middle of the night. Daily diary reports included information on nighttime waking. The quality of life of the child and the parent were measured through such factors as how activities were disrupted by symptoms. Comorbidities, such as anxiety symptoms of the child, were ascertained through a structured diagnostic interview, and the study also looked at obesity.
Routines were measured through self-reports, semistructured interviews, questionnaires, and videotapes of family mealtimes. The families ranged considerably in terms of their level of organization and their commitment to routines.
The most basic routine was whether a child had taken his or her medicine. Less than half of the children Fiese studied took their medicine as prescribed. Taking medication can be measured through recall, reports to physicians, or a computerized chip on the bottom of an inhaler that measures not only whether a child took the medicine, but whether it was taken appropriately.
A simple eight-item questionnaire measured the likelihood that parents have routines around taking medication and the amount of burden that they feel in carrying out these medication routines. Results showed that if families have such routines, children are more likely to take their medication (Fiese et al., 2005). The factor most related to quality of life for both the caregiver and the child was whether caregivers reported these routines as burdensome. This was true both for caregivers and children. Children who reported that they worry more about their symptoms and that their symptoms get in the way of having a relatively normal life were more likely to have parents who reported that carrying out routines was difficult.
To examine sleep patterns, the researchers conducted telephone diaries. They called the parents three times during the week and once on the weekend during selected times over the course of a year, gathering a collection of about 500 observations. They looked at four things in collecting the telephone diaries: (1) whether a parent had a negative mood that day, (2) whether a parent was hassled by kids not listening, (3) whether a parent was hassled because plans had to be changed, and (4) whether a disruption occurred in their bedtime routines. Each of these factors was significant in predicting the likelihood that the child would wake up at night (Fiese et al., 2007). The elevated likelihood is not overwhelming, although it is statistically significant. But it is as large as the odds ratios for biological indicators for nighttime waking in response to environmental allergens (such as cockroaches, dust mites, cats).
The researchers also constructed an asthma impact interview to understand how this condition affects family life. In an open-ended interview format, they asked families to tell the story of when their child was diagnosed with asthma and how it affected the child and family life. “We don’t want to hear the story they tell their pediatrician. We want to hear the story that they would tell a neighbor over a cup of coffee. Usually what we get at this point is what we call the head nodding response. Parents say, ‘We know which story you want to hear.’”
The researchers have identified three categories of ways in which
families manage asthma in their daily life: reactive care, coordinated care, and family partnership. In the reactive category, anxiety leads the family to action. The family has not established clear and consistent strategies. In the coordinated care category, a single way to handle all situations has been identified. Typically one or two people are responsible for carrying out doctor’s orders. In the family partnership category, plans are based on multiple sources of information and a shared philosophy, and multiple family members are involved in planning.
These different strategies predicted emergency room use one year after the interview was conducted (Fiese and Wamboldt, 2003). Families in the reactive category were about four times more likely to use emergency room care for their children’s symptoms than families in the coordinated care category and eight times more likely than those in the family partnership category. Families that have less burden in carrying out daily routines and have better medical adherence were less likely to use emergency room care, and they had better quality of life overall for both children and caregivers.
One common comorbid feature of asthma is separation anxiety. When people are anxious or panicked, they can have trouble breathing, and children with asthma are almost three times more likely to have separation anxiety symptoms than those without asthma. Fiese and her colleagues hypothesized that the way in which families interact with each other on a daily basis may mediate this relationship. They looked at interactions during meals, providing a basis for measuring such factors as communication and involvement of parents in children’s lives. They found that families that were able to be responsive during mealtimes, show genuine concern about their child’s daily activities, and manage affect in a positive way were less likely to have children with separation anxiety symptoms (Fiese et al., 2010). In contrast, families who have a child with separation anxiety symptoms have more difficulty getting tasks done during mealtimes, have more problems managing affect, and are less involved with their children.
They found the same relationship when looking at obesity in children (Jacobs and Fiese, 2007). Families that were more organized, regularly managed affect, assigned roles, and showed genuine concern about their children’s activities were less likely to have overweight children. The researchers also made mealtime observations on a second-by-second basis—“which we are calling our DNA prototype of family mealtime”—looking at activity levels, behavior management, and communication, expressed by every family member during a meal. They found that time spent at the meal distinguished families that have a child of healthy weight versus overweight. Children who are overweight spent less time at meals. When these observations were put into a cumulative risk model
that included census tract data, poverty, communication, time spent at the meal, and the scheduling and importance of mealtime, the model demonstrated associations between risk factors and a child’s body mass index and nighttime waking.
Fiese and her colleagues are now translating this work into interventions to promote the relationship between medical adherence and family routines. Targets of the intervention include quality of life, lung functioning, weight status, behavior problems, and health care utilization. For example, public service announcements around the topic of “mealtime minutes” remind families of the importance of mealtime routines, interactions, and time.
This research poses several methodological challenges, Fiese observed. The resources to transcribe, code, and analyze observations and narratives can be in short supply. There can be important differences among families across cultures, socioeconomic status, and life stage. It also can be difficult to capture differences among ages, which is especially challenging with large families. Family size is not necessarily static, with multiple players in a family, including neighbors, cousins, uncles, aunts, grandparents, and babysitters. Disease status may not be clear, and more attention needs to be devoted to comorbidities.
ECONOMIC PERSPECTIVES ON UNDERSTANDING THE IMPACT OF FAMILIES ON CHILD WELL-BEING
One recent example of multidisciplinarity in family science is the increased attention across disciplines to causal inference in estimating family influences. Approaches from economics to estimate unbiased causal estimates in research have been influential in other disciplines. To estimate causal effects using observational data, economists use four main approaches, said Betsey Stevenson of the University of Pennsylvania’s Wharton School.
The “first and easiest” thing is to apply a cross-sectional regression analysis, she said. This approach examines the differences among people and tries to identify the causal effects of a single difference while controlling for other differences. This approach has a major limitation because there are often unobserved differences among individuals or groups that interfere with isolating the effects of a single variable.
The second approach is to do a time-series analysis. This technique documents a correlation between variables of interest over time. It works particularly well if there are sharp changes in variables over time, such as a change in policy. However, many things can change at the same time, which is a limitation of this approach.
A third approach is what is called a quasi-experiment. This approach
uses changes in the environment that create roughly identical treatment and control groups for studying the effects of that change. Quasi-experiments can provide estimates of the causal impacts of a particular treatment, but they are better at telling how outcomes change rather than why they change, which can create ambiguity in extending or applying an analysis.
The fourth approach is to use structural modeling. These models use the same data as a regression analysis, but they use theory to constrain the data in an effort to derive understanding from them. The limitations of this approach are that causal impacts can be difficult to estimate and the results are only as good as the theoretical assumptions contained in the model.
Stevenson illustrated two of these approaches—regression analysis and quasi-experiments—in her analysis of the effects of girls’ participation in high school sports on years of schooling completed (Stevenson, 2010). (Her research on Title IX examines, in addition to education, labor force participation, wages, and occupational choice, but for the purposes of the example she limited her discussion to years of education completed.) Students who participate in sports complete more years of schooling. The relevant questions are whether the correlation between sports and schooling is because of the types of people who choose to play sports or whether this is something that occurs because of sports. Answering this question is necessary to consider whether increasing the opportunities for students to play sports would increase their educational attainments.
Stevenson started with data from the National Longitudinal Survey of Youth (NLSY), which has been tracking a cohort of more than 12,000 young people who were between the ages of 14 and 22 in 1979, when they were first interviewed. Her regression analysis included a wide range of independent variables, including personal characteristics, like race, age, IQ, and self-confidence; family characteristics, such as parents’ education and family income; community characteristics; and school characteristics. Some of these independent factors are easier to measure than others, and the ones that cannot be measured can cause bias in the causal estimates if they are correlated with the variable of interest.
After controlling for the race and age of students along with state and urban status, the regression analysis shows that girls who participate in sports acquire about a year’s more education than girls who do not. “That is huge,” Stevenson said. “If we thought that was a causal effect, you should all run out of this room and start sponsoring sports programs.” The effect is about the same for boys who participate in high school sports.
However, as more independent variables are added as controls, the size of the effect shrinks. Adding family characteristics and school charac-
teristics cuts the estimated years of additional education by about a third. Adding student ability and achievement measures, such as student IQ, cuts the effect another third, so that it is now less than half a year. “It turns out smarter kids play sports. For those of you who thought of the dumb jock, that is not true. Smarter kids play sports, smarter kids get more education. Without controlling for IQ, I get big estimates. When I control for IQ they shrink, and now they are at about 0.4 of a year’s schooling.”
Nevertheless, the effects of participating in high school sports never shrink to zero as more and more controls are added. “Every cross sectional regression that has been run, no matter what you control for, you see that kids who participate in sports do better than kids who don’t.”
The question remains whether students who participate in sports are different in ways that cannot be determined from the available data. For example, perhaps those who participate in sports are the type of people who would have stayed in school longer because of an unmeasured factor, such as ambition or energy, that is not contained in the control variables. All possible source of bias cannot be eliminated. But another source of information on the effects of sports on education is available: the quasi-experiment afforded by the passage of Title IX in 1972.
Title IX of the Education Amendments to the 1964 Civil Rights Act declared that “no person in the United Sates shall, on the basis of sex, be excluded from participation in, be denied the benefits of, or be subjected to discrimination under any educational program or activity receiving financial assistance.” It requires that girls be given the same opportunities to participate in sport at boys. “It doesn’t mean equal participation rates, but it does mean that if a girl wants to play and there are boys who are able to play, then either you need to have equal participation rates or you need to be able to make sure that girl can play.”
Title IX led to a major increase in girls participating in sports. Prior to Title IX, less than 5 percent of girls played high school sports compared with 50 percent of boys. After Title IX, about 30 percent of girls played high school sports and about 50 percent of boys did.
While this changes yields some potentially useful time-series evidence, the useful quasi-experiment comes from exploiting differences across states over time. Differences across states emerge because the percentage of boys who participate in high school sports varies widely by state. In states where boys’ participation is high, more girls need to be given opportunities to participate in sports to be in compliance with Title IX.
By analyzing the change in girls’ sporting opportunities generated by the interaction of the passage of Title IX and the variation across states in boys’ pre–Title IX sports participation rates, Stevenson was able to assess whether girls’ outcomes related to education were changed in a way that
is related to the growth in sporting opportunities generated by Title IX (and in particular, in a way predicted by the preexisting level of boys’ participation). The quasi-experimental approach is to identify a treated and untreated cohort. The treated cohort were those attending high school after Title IX went into effect in 1978, and the untreated cohort were those finishing high school before Title IX passed in 1972.
Combining differences across generations with the differences across states creates what economists call a “differences-in-differences” estimator. It combines time-series and cross-sectional analysis in an experimental setting, thereby controlling for cross-sectional differences and time-series differences, in which the cross-sectional differences are stable over time. It is still possible that some states increased girls’ sports participation more than others because of other factors, such as a school board superintendent who worked very hard at it. But this can be controlled through what are called intention-to-treat or instrumental variables that isolate the exogenous part of the policy change.
This technique shows that female educational attainment rises with the opportunity to play sports. States with a 10 percentage point greater increase in the statewide female athletic participation rate had an overall increase in educational attainment of 0.039 years, an increase in the probability of some postsecondary education of 1.3 percentage points, and an increase of 0.8 percentage points in the probability of obtaining at least a college degree. Since Title IX raised female participation by around 30 percent, these results would be multiplied by more than three to get the aggregate effects. Meanwhile, female educational attainment rose by about 0.7 years over the time period being analyzed. As a result, Stevenson concluded that increases in sports participation caused by Title IX explain about 20 percent of the increase in women’s education over the time period being analyzed. Similar analyses can be applied to the participation of women in the labor market and entrance into previously male-dominated jobs.
Documenting this effect does not mean that every girl should be forced to participate in sports, Stevenson observed. Some may benefit more from playing sports, and some may benefit less. Title IX, by increasing opportunities, allowed girls to self-select whether or not to participate. It remains to be known whether all girls would benefit from participating in sports.
During the discussion period, the presenters were asked how they would use an increase in research funding to extend their work. Barbara Fiese responded that she would integrate more sophisticated biological markers into her investigations. Such markers could be used with all of the
members of a family to look at variations in the family unit over time. “I think that would be incredibly fascinating.”
Another enhancement would be to integrate the investigation with interventions and the response to interventions. It is difficult to do lengthy qualitative observations in intervention science, yet more narrative approaches can capture the richness in family situations.
A final addition would be integrate and translate research results into public arenas. For example, “how can we use this information to inform public service announcements, where we reach a broader audience, and how can we use this information to cast a wider net to communities at large?”
Darlene Kertes said that some of the issues in family research are similar for behavioral and biological measures. As with behavioral measures, attention needs to be paid to developing protocols that can be assessed longitudinally. A second point was that it is important to consider both data collection and consenting methods that are flexible and adaptable. It is difficult to predict what technologies might be available 10 years from now to analyze biological specimens. One challenge is therefore collecting biological specimens that allow for potential future use. Another is establishing best practices for consenting participants in a way that is ethical (particularly for minors) but allows for analyses to be conducted using knowledge and technologies that will be developed in the future.
Hirokazu Yoshikawa asked whether the projects that incorporated biological measures brought together people trained in specific methods or engaged in cross-training to combine the behavioral and the biological approaches. McMahon responded that his work has involved complementary studies proposed to different funders that historically have favored one kind of research over another. To carry out the work, he assembled a group of faculty with different areas of expertise. Although people were trained for each study, the quantitative methods were kept separate from the qualitative methods.
Fiese said that her research has had one team work on the narrative coding, one team work on the observational coding, and one team work on structured interviews. “But I am leaning more toward trying to integrate some of the training within individuals so that they can be a little more flexible because I am seeing this as an added value in their future careers.”
This page intentionally left blank.