experiments are the most appropriate guides for policy formation. Nonetheless, they are very seldom carried out. Lee and Miller (1997) and The New Americans (National Research Council, 1997:Chap. 7) are the most comprehensive longitudinal calculations available.

Such calculations are complex and necessarily involve many assumptions and projections. Cross-sectional calculations are a simpler alternative to the longitudinal ones, but they are rarely justified by a clearly formulated conceptual experiment (see National Research Council, 1997, Chapter 6 for a discussion of these issues). One such experiment might be: How would the fiscal situation change if all the immigrants in the United States in some year were to vanish? We could attempt to answer this question by calculating the difference between all the taxes paid by immigrants and the marginal cost of all the benefits received by them. We would ignore any cumulative effect of the presence of the immigrants in the past, for example on government debt. Such a cumulative effect would be relevant in the longitudinal calculation, but not for this conceptual experiment.¹ We call this the "immigrant only" approach, because no U.S.-born descendants are included in the calculation. This approach is quite common in cross-sectional analyses of fiscal impacts.

Note, however, that this calculation takes no account of the costs and taxes for the U.S.-born children of immigrants, many of whom impose heavy public-sector costs by attending public schools and colleges. Although these children are U.S. citizens, and not immigrants, their presence is a direct consequence of their parents' immigration. For this reason, it appears appropriate to include their fiscal impact, which is typically done by analyzing the taxes and costs of benefits for all immigrant households or families. These are defined as households or families with an immigrant head or adult member. How would the fiscal situation differ in some year if all immigrant households (including U.S.-born members) were to vanish in that year? We call this the "immigrant household" formulation. This approach is also quite common in cross-sectional analyses of fiscal impacts.

However, there is a serious bias in this type of study, because it counts the U.S.-born children of immigrants only while they are young and living with their parents and while society is investing heavily in their education. As soon as they leave home and get jobs, they no longer live in immigrant households, and their contributions are lost from sight. But if the children are to be included while young and costly, they should also be included when they are older and paying taxes (as argued, for example, in MaCurdy et al., in this volume). Furthermore, these children may themselves have children, who also should be included. Thus, we propose a conceptual experiment in which all the immigrants vanish in some year, along with all the descendants of those immigrants,