In Defense of Goliath

It has long been accepted as a given that small businesses are responsible for producing most of the new jobs in the United States. This decade, news reports have touted the job-creating muscle of small businesses. And in his 1993 State of the Union address, President Bill Clinton tied his promise of targeted incentives for small business to his claim that "small business has created such a high percentage of all new jobs in our nation over the past 10 or 15 years." These long-held assumptions have now come under fire in a report titled "Small Business and Job Creation: Dissecting the Myth and Reassessing the Facts,” written by economics professor Steven Davis of the University of Chicago Graduate School of Business.

Davis and co-authors John Haltiwanger of the University of Maryland and Schott Schuh of the Federal Reserve Board argue that survival rates for new and existing manufacturing jobs increase sharply with employer size. Further, they point out that smaller manufacturing firms and plants show substantially higher gross rates of job creation, but not higher net rates of job creation. The net job creation rate in the U.S. manufacturing sector, they conclude, displays no systematic relationship to employer size.

Why then the claims that small businesses create most new jobs? Davis, Haltiwanger and Schuh argue that misleading interpretations of the data are the basis for these claims. They also assert that previous studies of the job creation process rest on data inappropriate for drawing inferences about the relationship between employer size and job creation. "Most of the evidence [cited to support small business job creation] is fallacious or has serious problems of interpretation," says Davis. "Within the manufacturing sector, the proposition [that small businesses create most new jobs] is empirically false."

Many leading studies of the job creation process rely on the Dun and Bradstreet Market Identifier (DMI) files, and therein lies much of the reason for inaccurate claims of small business success in creating jobs, says Davis. "That data set was driven by Dun and Bradstreet's customers asking for credit checks, or by Dun and Bradstreet tracking firms in particular sectors. It wasn't designed to be a statistical tool, but designed to meet the need for knowledge about certain kinds of customers. A firm might not get into the data base until it had been around for some time, because no one asked about it."

By contrast, Davis and his co-authors based their analysis of job creation and destruction behavior in the manufacturing sector on the Longitudinal Research Database (LRD). Housed at the Center for Economic Studies in the U.S. Bureau of the Census, the LRD contains plant-level data at annual sampling intervals for the U.S. manufacturing sector from 1972 to 1988.

"The statistics we use are based on a random sample of the universe," says Davis. "With these statistics, you're starting from the right universe -- all the businesses in the sector. And you've drawn a big enough sample to make the proper inferences." Further, misleading interpretations of data have contributed to misconceptions about small business job creation, the authors say. A major pitfall for prior researchers, they note, has been the regression fallacy, which can lead to bias whenever employers go through short-term fluctuations in size, or when measurement errors suggest short-term fluctuations in observed size of firms.

The authors support the regression fallacy with statistical evidence that appears to suggest that small firms (those with less than 500 employees) have outperformed larger ones in both years two and three, but in fact all three firms are exactly the same size in year three as they were in year one.

These statistics reflect the fact that after good and bad years, firms tend to regress to their sizes in the long run. In the base years of study, firms classified as large tend to have experienced a recent short-term hike in employment. Because short-term movements tend to reverse themselves, firms classified as large are relatively likely to get smaller the next year. The same is true in reverse: firms classified as small tend to have experienced recent declines in employment and are relatively likely to expand the next year. Therefore, the regression phenomenon (so named because of this observed regression to long-run size) makes for the misperception that small firms routinely create more jobs than larger companies.

"Firms do have transitory movements in employment, and when that happens, even if it's based on measurement error, you have a potential to overstate the growth performance of small firms," concludes Davis. Another reason for the inaccurate claims of greater small business job creation in some studies is a confusion between net and gross job creation.

"While gross job creation rates are substantially higher for small plants and firms, so are gross job destruction rates," the authors write. In other words, small businesses create more jobs than large businesses, but also destroy more jobs.

Further, they found no systematic correlation between net job creation (jobs created minus jobs destroyed) and employer size. But their study did turn up "clear evidence" that job durability -- the likelihood that jobs created will survive at least a year -- increases with employer size. The authors note that their observations about job creation are based on manufacturing sector statistics, and acknowledge they aren't yet able to calculate gross job creation and destruction rates for nonmanufacturing industries.

But, says Davis, "the methodological points in the article are not limited to the manufacturing sector. It matters a lot as to whether you correctly measure size and growth. Other recent studies suggest that it makes a difference not only in the manufacturing sector, but in other sectors as well."