As the fields of personnel economics and organizational economics have become more visible in recent years, more economists, practitioners, and policymakers have become interested in the internal workings of firms. Fortunately, at the same time as interest in these areas has grown, new data sets have emerged that provide consistent personnel data from a wide variety of firms. This paper provides an example of how newly available data can be used to analyze internal labor markets and suggests how such data can be used to address other issues. Basic questions in personnel economics include how firms set wages and how people move between jobs (within and across firms). Answering these questions is essential to assessing the relative importance of theoretical models as explanations of the nature of employment relationships. These models include agency theory, matching, and search theory, among others. Historically, most attempts to study these models were limited to data sets that are drawn from a random sample of individuals with no identification of firms, such as the Current Population Survey (CPS). While much can be learned from such studies, much of the inference is indirect and the data may suffer from inconsistent or inaccurate self-reported data. An alternative strategy, used by, for example, Lazear (1992), George Baker et al. (1994), and Kenn Ariga et al. (1999), is to procure detailed personnel information from a single firm and use it to study the policies at that firm. While these papers were successful at providing details of the individual firms, they leave open the question of how widely the results generalize, especially given that the results are not consistent even across these three papers, which are based on different firms. An important step in getting past the limits of CPS-style and individual-firm data is to find data sets that provide employee details for numerous firms. Such data sets have been created in the United States, France, Sweden, and other countries. As John M. Abowd and Francis Kramarz (1999) show, these data sets take many forms and, like the data that preceded them, have varying strengths and weaknesses. They have already been used, according to Abowd and Kramarz (1999), in over 100 studies of more than 15 countries. That paper provides details on many of these studies, as well as comparing some of the features of the various data sets. Although the U.S. data have many virtues, they lack job information. A key advantage of the Swedish data used here is its detailed and accurate job classifications. This makes it possible to determine whether job openings are filled internally or externally and to follow employees as they change jobs. The data include many firms, a long panel of years, and accurate wage data, allowing the study of the relative importance of firms and jobs on wage changes and levels. The main results of this paper are as follows. First, the Swedish firms studied fill a significant † Discussants: Henry Farber, Princeton University; Lawrence Katz, Harvard University; Derek Neal, University of Chicago.
