Why size class methodology matters in analyses of net and gross job flows: net and gross job flow statistics by size class are produced with data from the Business Employment Dynamics program; alternative methodologies for defining size classes yield sharply different pictures of employment growth

Monthly Labor Review, July, 2004 by Cordelia Okolie

One of the most interesting and often asked questions in empirical economics is whether small businesses create the most jobs. Answering this question requires longitudinal establishment microdata and is an ideal application for the new Business Employment Dynamics data series produced by the Bureau of Labor Statistics. Although it is often argued that small businesses are the fountainhead of job creation and the engine of economic growth, this view is not universally accepted, largely because of differences in the methodology used to construct net and gross job flow statistics. Using different methodologies, this article calculates net and gross job flow statistics by size class, with the aim of showing how alternative methodologies can produce sharply different portraits of employment growth.

Methodology issues

Three methodology issues influence the calculation and interpretation of business employment dynamic statistics by size class: (1) how establishments should be classified into size classes in the construction of net and gross job flow statistics, (2) the appropriate measure to use in the denominator in the calculation of net and gross job flow rates, and (3) whether there are differences in the statistics if the establishment or the firm is the unit of analysis. (1)

Defining size classes. With cross-sectional microdata, defining size classes for establishments is straightforward. For example, an establishment with 3 employees is classified into the category "1 to 4 employees," and an establishment with 11 employees is classified into the category "10 to 19 employees." By contrast, defining size classes with longitudinal microdata is more difficult. For instance, if an establishment grows from 3 employees in the previous quarter to 11 employees in the current quarter, in which size category does it belong?

In the gross job flows literature, there are three methodologies for defining size classes: (1) in base sizing, establishments are classified into size categories on the basis of their size in the previous quarter; (2) in end sizing, establishments are classified into size categories on the basis of their size in the current quarter; and (3) in mean sizing, establishments are classified into size categories on the basis of their average size during the previous and current quarters. In the earlier example in which an establishment grows from 3 employees in the previous quarter to 11 employees in the current quarter, the base-sizing methodology would classify that establishment into the "1 to 4 employees" category, whereas the end-sizing methodology would classify it into the "10 to 19 employees" category. The mean-sizing methodology would classify the establishment into the "5 to 9 employees" category, because the average size during the two quarters is 7 (3 11, divided by 2).

The methodology of classifying establishments into size categories can have large effects on business employment dynamics statistics. For establishments that are growing and that move from one size class category to another, base sizing results in statistics which indicate that employment growth is coming from smaller establishments, whereas end sizing results in statistics which indicate that employment growth is coming from larger establishments. Similarly, for establishments that are contracting and that move from one size class category to another, base sizing results in statistics which indicate that employment decline is coming from larger establishments, whereas end sizing results in statistics which indicate that employment decline is coming from smaller establishments. Economists refer to this statistical phenomenon as the "regression fallacy" or "regression-to-the-mean" bias. (2)

Calculating rates. Another methodological issue is the question of how to compute rates of net and gross job flows. That is, should previous-quarter employment, current-quarter employment, or an average of the two be used in the denominator of the rate? An example will help illustrate the difference between the methods. Suppose employment increases from 1 to 2 and then declines back to 1. A conventional growth rate that uses previous-quarter employment in the denominator would yield a 100-percent increase followed by a 50-percent decrease. Even though the employment changes in levels sum to zero (a one-employee increase followed by a one-employee decrease), the percentages do not sum to zero. In fact, using previous-quarter employment in the denominator results in the sum of the percentages being greater than zero; the sum would be less than zero if current-quarter employment were used in the denominator. In contrast, if average employment were used in the denominator, the growth rate in this example would be a 67-percent increase [(2- 1)/1.5 = 0.67] followed by a 67-percent decrease. The example illustrates the fact that using average employment in the denominator results in rates that are equal in magnitude, but opposite in sign. (That is, the rates are symmetric.)