eling snow, to more formal employment, particularly in fast-food and other retail and service industries, and become more dispersed across job categories. They are also more likely to have supervisory responsibilities and receive more training from their employers as they grow older (Mortimer et al., 1994).
A challenge associated with almost all research on the consequences of working for young people is that of selection effects. That is, young people who work may be different before they began to work than those who do not work and those who work long hours may be different than those who work fewer hours. For example, adolescents who are not interested in school may choose to work longer hours than those who enjoy school. Other differences among the groups may include their past academic performance, their career goals, their families' incomes, their parents' education levels, their motivation, and a host of other factors that are often not explicitly measured. These differences make it extremely difficult to ascertain whether working itself causes any particular outcome (either positive or negative) or whether those outcomes might have occurred whether or not the young people engaged in work.
Because researchers cannot randomly assign young people to the workplace, the committee relied on a careful review of studies that follow young people over time (longitudinal studies) and that take account of the pre-existing differences among youngsters who engage in various work patterns (statistically controlling for differences). These studies measure the statistical correlation between working and various outcomes, that is, the degree to which the occurrence of certain outcomes varies with different work patterns, in the context of pre-existing differences. Although no direct causal link between work and outcomes can be made from correlational studies, a pattern of consistent findings from studies with good statistical controls may be the best information available. Without random assignment, however, selection effects must be considered as an alternative interpretation of results.
Data from several well-designed, nationally representative longitudinal surveys are used in many of the studies discussed below. These include the National Longitudinal Survey of Youth (NLSY),