Measuring Small Business Employment over the Business Cycle
Many analysts have tried to understand why the pace of job growth has been so slow since the end of the Great Recession. This issue has focused attention recently on the hiring behavior of small businesses during the recovery. It turns out that simply measuring the hiring practices of small businesses can be difficult.
To examine trends in small business employment, we first analyzed data from the Business Employment Dynamics (BED) series of the Bureau of Labor Statistics (BLS). These data represent a quarterly census of all establishments under state unemployment insurance programs and characterize about 98 percent of all employment on nonfarm payrolls. We looked at three classes of firms: small (1-49 employees), medium (50-499 employees), and large (500 or more employees).
Prior to the recession, large firms accounted for 44 percent of overall employment, medium-size firms 30 percent, and small firms 26 percent. Comparing shares of employment losses for these different sizes of firms during each of the three most recent recessions, we can see that the smallest firms played a less significant role in the 2001 recession than in the other recessions, while employment losses for large firms were relatively muted in 1990. In the last recession, employment losses were proportional to the employment shares of each group, so it appears that all groups were hit evenly.
Comparing shares of gains in employment during the three most recent expansions, we focus on the share of gains going to firms above and below 500 employees (a commonly used demarcation between large and small firms). The most recent recovery looks very similar to the two preceding recoveries. Smaller firms accounted for about 70 percent of employment gains during the recovery phases of the last three cycles, well above their overall employment share. However, if we focus on firms under 50 employees, we can see that the smallest firms make up a less significant share of the gains in employment during the current expansion compared to the earlier ones.
The time-series patterns of net job creation look fairly similar across firm size. This is especially true in the last cycle, where both employment losses and employment gains moved in concert across the size classes.
One limitation of the BED data set is that it is available only with a time lag of about eight to ten months. For example, as of May 14, 2012, the most recent BED data set is from the third quarter of 2011. There are more timely data on private small business hiring. One prominent example is the Automated Data Processing, Inc. (ADP) national employment data set. These numbers represent about 344,000 U.S. businesses and 21 million U.S. employees, and are obtained as an anonymous subset of the approximately 500,000 business clients who process their payrolls through ADP.
The patterns of net job creation by business size look strikingly different in the ADP and BED data sets. The ADP data show that smaller businesses lost markedly more employment than larger businesses during the recession and have added significantly more employment during the recovery. Indeed, the ADP data indicate almost no growth in payroll employment for large companies during the recovery, which stands in sharp contrast to the patterns reported in the BED data.
Part of this difference is due to definitional differences. The ADP is constructed to be representative of the national distribution of establishments, while the BED is reported at the level of firms. An establishment is typically a distinct business location—a store, hospital, mine, or manufacturing plant. It may be an individual firm—a mom-and-pop grocery store—or it may be an outlet of a large retail chain. A large firm can potentially own many distinct establishments of various sizes. A firm in the BED data is a tax entity and includes all establishments that file under a specific tax ID.
The implication of this definitional difference is that the firm-size distributions are very different in the two datasets. While in the BED firms with more than 500 employees accounted for 44 percent of employment in 2007, their share was only 17 percent in the ADP. To be sure, ADP is not matching to the size distribution of firms, but rather to the size distribution of establishments. We can see this by comparing employment shares in the Quarterly Census of Employment and Wages (QCEW)—an establishment survey—to ADP employment shares. They are nearly identical.
Given the fact that the BED has a much greater share of employment in the large size classes, it is not surprisingly that large firms contribute more to employment gains and losses over the cycle in the data. Our sense is that the BED definition comes much closer to how one typically thinks about firm size and is likely a better data source for assessing the relative contribution of large and small firms to employment change. Thus while the ADP is available for analysis sooner, we feel that the BED data are more appropriate to look at for employment growth by business size.