Support of the National Science Foundation (grants SBR 9422440 and SBR 9730142) is gratefully acknowledged, as well as the Alfred Sloan Foundation through a grant to the National Bureau of Economic Research project on Industrial Technol ogy and Productivity. This work was carried out at the Boston Research Data Center of the U.S. Bureau of the Census, with data provided by the Center for Economic Studies. We thank Arnie Reznek and Joyce Cooper for their help and cooperation. Tim Dunne also provided early advice on the use of the Longitudinal Research Database. We are indebted to Leslie Papke and Jeff Wooldridge for use of their program to calculate robust standard errors in conditional Poisson models. The work has benefited from insightful comments by Arik Levinson, as well as Wayne Gray, Karen Palmer, Gilbert Metcalf, and participants in seminars at British Colum bia, Brown, Harvard, Mannheim, Wisconsin, National Bureau of Economic Re search, Center for Economic Studies, and Resources for the Future and in presenta tions at meetings of the American Economic Association, National Tax Association, Regional Science Association International, Western Economic Association, and World Congress of Environmental and Resource Economists. The opinions and con clusions expressed in this paper are those of the authors and do not necessarily represent those of the U.S. Bureau of the Census. All papers are screened to ensure that they do not disclose confidential information.