Measuring job and establishment flows with BLS longitudinal microdata

The relatively recent development of longitudinal establishment data sets h as generated considerable excitement in both the academic and the statistic al communities. The descriptive statistics coming out of these data sets il lustrate the large amount of volatility at the individual establishment lev el that underlies the smooth time series of aggregate employment growth. Th is finding not only has stimulated the review and updating of existing labo r market theories, but also has motivated U.S. statistical agencies to prod uce longitudinal job flow statistics from their administrative data sets. T his article describes a new longitudinal database from the Bureau of Labor Statistics that has the potential for enhancing microdata research into top ics such as job creation, job destruction, and the life cycle of establishm ents. The literature on the demand for labor in general and on gross job flows in particular has flourished during the past decade. Perhaps the most importa nt finding discussed is the tremendous heterogeneity in establishment-level employment changes that is evident in the job creation and job destruction statistics underlying net employment growth. For example, using data spann ing much of the 1970s and 1980s, Steven J. Davis, John C. Haltiwanger, and Scott Schuh report that, on average, 5.5 percent of manufacturing jobs were destroyed and 5.2 percent created over a 3-month interval.(1) The -0.3-per cent difference between these two statistics is the average net employment growth per quarter. Despite all that we have learned about the labor market from the aforementi oned literature, the call for better data always resonates. Three aspects o f existing data are often mentioned. First, much of the early work using U. S. data was restricted to the manufacturing sector.(2) Recent research usin g unemployment insurance data from various States, however, has illustrated how job creation and job destruction in manufacturing may not be represent ative of other industries.(3) Second, the existing empirical work on job fl ows, either by choice or by necessity, is largely based upon data that excl ude the smallest establishments. Small plants with fewer than five employee s, for example, are not in the sample frame of the Annual Survey of Manufac tures used by Davis and Haltiwanger,(4) manufacturing plants with fewer tha n five employees from the Census of Manufactures are excluded from Timothy Dunne, Mark J. Roberts, and Larry Samuelson's analysis,(5) and the sample u sed by Patricia M. Anderson and Bruce D. Meyer includes only firms with at least 50 employees.(6) Finally, many of the existing studies use annual dat a, whereas the ideal data for studying grass job flows would be quarterly o r perhaps even monthly. Data at these high frequencies are necessary for an alyzing seasonal. patterns in employment growth or for analyzing the short- run employment growth immediately following business "birth" and immediatel y preceding business "death." The longitudinal database introduced in this article is not subject to any of the three limitations just mentioned. The microdata from which the datab ase is constructed are the unemployment insurance reports that employers in the United States are required to file with the States. These data are ess entially a quarterly census of establishments in all industries, which impl ies that the job creation and job destruction statistics derived from the l ongitudinal database have the potential to be valuable economic indicators published by statistical agencies of the U.S. Government. In the next section, job creation and job destruction are defined, and the relation between these new statistics and those already published by the Bu reau of Labor Statistics is described. Following that, a detailed descripti on is given of the unemployment insurance microdata and the construction of the longitudinal data set. Because it is desirable to distinguish among es tablishments that are expanding, contracting, opening, and closing, special attention is given to the longitudinal linkage algorithm used to minimize the incidence of spurious births and deaths. In the final section of the ar ticle, several tables are presented that highlight job creation and job des truction statistics across industries.