In this study we have addressed a set of issues which have generally not been analyzed previously. We first examined the relationship between adolescent risky behaviors and long-term adult outcomes. Five adolescent behaviors were studied: alcohol use, marijuana use, cocaine use, sexual activity, and delinquency. Multiple outcomes were measured across four different domains: health (alcohol abuse or dependence, drug use), economic (poverty, welfare, employment), family formation (marriage, divorce, fertility), and incarceration.
There is a fairly consistent pattern that engaging in risky behaviors as a teenager is associated with less successful adult outcomes. For many, the relationship is such that the earlier one engages in the behavior, the more likely they face a bad outcome as an adult. The most consistent predictor of a bad adult outcome is age of initiation into sexual activity. Alcohol usage, on the other hand, is perhaps the one teenage behavior least associated with bad adult outcomes. Age of initiation into alcohol usage is, however, associated with adult alcohol abuse or dependence.
Age of initiation into any given risky behavior is highly correlated with the frequency and intensity of engagement in that behavior. However, in our analysis, those who merely experiment are combined with those who have more serious intentions. By not measuring frequency and intensity, we cannot distinguish between the effects of these measures from the effects of early initiation.
These findings represent associations and the relationships should not be interpreted as causal. Many are consistent with what might have been expected, given what is known from other literature. However, few relationships with long-term adult outcomes have been documented. This study provides new findings from a nationally representative survey and contributes to our understanding of the impacts of risk taking by young people. This study lays the groundwork for developing future work to understand these relationships.
The second avenue of investigation in this study was to document the relationship between family environment and adult outcomes. We select three measures of family environment for analysis: family structure, parents' education (as a proxy for socioeconomic status), and parental alcoholism. Most research on the impact of family structure has focused on how it relates to initiation and participation in risky behaviors. Other outcomes studied have generally reflected short-term outcomes such as high school graduation, college enrollment, and teenage pregnancy. This study focuses on a wide variety of long-term outcomes, providing some of the first estimates of such relationships.
Our findings indicate that family structure effects differ by outcome domain measured. Adolescents who reside in intact families at age 14 clearly have the least likelihood of a bad economic outcome and are less likely to spend time in jail. This is less clear for the health and family formation domains. Single mother headed households do not fare as well as intact families in economic outcomes, but compare favorably along other domains. Interestingly, fathers with stepmothers lead to the least likelihood of an alcohol or drug problem as an adult, but also lead to the highest likelihood of having bad economic outcomes or spending time in jail. Single father families have similar relationships implying that fathers are important for reducing the chances of an adult problem with alcohol or drugs.
We found similar associations when examining the relationship between parents' education and long-term adult outcomes. More educated parents are associated with better economic outcomes and less likelihood of going to jail. However, there are contradictory results for the health domains where mothers' education is associated with lower likelihood of adult alcohol problems, but greater likelihood of adult drug usage. Fathers' education had exactly the opposite associations across the two outcomes.
While interpreting any given relationship would be problematic without an underlying behavioral model, we believe our results suggest that the way in which parents help prevent bad outcomes for their children differs across different domains. For economic outcomes, parents can use their resources (financial, networks, etc.) to send their children to college, help them get jobs, and serve as a fallback to prevent financial problems. There is probably a similar mechanism for keeping their children out of jail. However, for non-economic domains, particularly those of alcohol and drug problems, a different set of family processes contributes to an eventual healthy adulthood.
Our third avenue of investigation involved exploring the influence on the pathway from adolescence to adulthood. We restricted our analysis to those who had already chosen to engage in one or more risky behaviors at an early age, i.e. those who had chosen a path more prone to poor outcomes. We then examined the relationships of family structure and parents' education to adult outcomes specifically for this sample. The results are comparable to those found when not restricting the sample. They indicate that these two family environmental measures are associated with changing the paths these early initiators have taken. There is a sense that although the family may not have prevented the youth from starting down a "wrong" path, it can help them from having that choice lead to bad consequences.
These findings perhaps raise more questions than they answer, laying the groundwork for future research. Several aspects of this study can be varied to create a richer analysis. First, we began our examination taking initiation into each risky behavior as given. However, a complete model would treat these decisions as endogenous and model the age of initiation. Such a model would be critical for teasing out the role family environment plays in the choices to engage in adolescent risky behaviors versus its impact after the choice has been made.
Second, although our regressions control for initiation into several behaviors, they do not explicitly account for multiple risk taking. The idea of multiplicity of risk factors is one discussed in the literature but not often tested. From the Problem Behavior Theory perspective, adolescents do not usually engage in one risky behavior, but rather the problems form a cluster of behaviors comprised of drinking, bingeing, smoking, marijuana use, delinquency, and premarital sexual intercourse (Donovan, et al., 1991). An important next step in our research may be to form a risk behavior composite index and examine its predictive ability in terms of adult outcomes.
Third, we have treated family structure at age 14 as exogenous. While some living situations are not reflective of the child's behavior (e.g. divorce), other situations may be the outcome of the child's behavior. For example, a boy who has become involved with drugs and delinquency may be difficult for a mother to handle so she sends him to live with his father. Therefore the endogeneity of family structure needs to be accounted for.
Fourth, we have used family structure at age 14. It was necessary for us to simplify our analysis and age 14 seems to reflect the time when adolescents are beginning to make decisions about whether to engage in various risky behaviors. However, the decision to engage in such behavior, and its impact, is probably the result of a more dynamic process. For example, changes in family structure that occur early in the child's life may have a different impact than those that occur later. And the number and type of changes may have differential impacts. A complete dynamic structural model would be necessary to account for the connection between risk taking and family changes.
Fifth, we have examined the relationships of family structure and parents' education on certain adult outcomes. These, however, are surely proxies for underlying family processes that we are not measuring such as parents' relations with each other for residential parents and with absent parents and parent-child interactions for residential and absent parents. Past research has shown the key role of certain family process variables as risk and protective factors for children whose parents have substance use disorders (Johnson et al., 1995). Some of these measures include family discord, hostility, care giving quality, family cohesion, supportive parenting and monitoring. Recent releases of new data sources such as Add Health and the NLSY97 provide more information on these family processes than was previously available. It will be quite some time, though, before we have observed long-term adult outcomes in those samples.
Sixth, recent cohorts may not behave similarly to the teenagers of the early 1980s. For instance, although adolescent drug usage has been slowly rising in the 1990s, the rates are considerably lower than in the early 1980s. There is evidence that alcohol use at early ages has become steadily more common in the last several decades, but early marijuana use has shown a different trend. Johnson and Gerstein (1998) found that the percentage of 15 year olds who had ever used alcohol or marijuana was higher for the 1961-1965 birth cohort compared to the 1956-1960 birth cohort (25 percent versus 20 percent, alcohol; 17 percent versus 12 percent, marijuana), the two groups which roughly comprise the sample used in this study. Those born in 1971-1975 showed higher rates of alcohol usage at age 15 at 33 percent but no increase in marijuana usage at 13 percent. As risk taking behavior changes across cohorts, it is an open question whether there is a change in the relationship between the behavior and adult consequences.
Various data sets can be used to compare inter-cohort risk taking. However, longitudinal data sets of more recent cohorts such as Add Health, the NLSLY97, and the National Survey of Adolescent Males have not existed long enough to provide long-term outcomes. As these data sets mature, research will be possible to determine if long-term relationships have changed across cohorts.
Seventh, a great deal could be learned about the impact of family structure and family process if one had longitudinal data on multiple generations. Most data sets have only a limited set of information about the respondents' previous or next generations. One exception is the Children of the NLSY79. This data set includes a set of assessments and interviews with the children of the female respondents in the NLSY79. Begun in 1986 and administered biennially since, the Children of the NLSY79 collects information to assess the physical, social, and cognitive development of these children. In 1994, sufficient numbers of these children had reached age 15. At that time, a full interview with these youths was conducted, similar to the one their mothers had received. This sample has become known as the Young Adult sample. The NLSY79 with its accompanying child supplement and young adult interview provides us with extremely rich detail on the lives of children and their parents. In each case information on the mothers exists for many years, extending back to the mother's own adolescence. With data collected through 1998, there are not only lengthy longitudinal data on mothers, but there are now multiple observations for a substantial number of youths. These data will provide an opportunity to study intergenerational relationships in adolescent risky behavior.
Finally, it is clear that research needs to differentiate among different outcome domains. We found important differences in the relationships of family environment across the different outcome domains studied. The underlying family process affecting the various outcomes is undoubtedly different. Introducing to a model such things as parents' income and assets, parent-child interactions, and contributions and involvement by absent parents would enrich our ability to understand these distinctions.