An official website of the United States government
Data analysts have known for a long time that estimating the number of jobs created by small businesses is extremely sensitive to the statistical methodology used.1 During the development phase of Business Employment Dynamics (BED) firm-size data, multiple “sizing” methodologies were analyzed, primarily by Cordelia Okolie.2 Ultimately, the BED program decided to use dynamic sizing.
Dynamic sizing is a straightforward measurement methodology that allocates a firm’s quarterly employment growth or loss to the size class in which the growth or loss occurred. Firms are initially assigned to a size class each quarter on the basis of their employment in the previous quarter, but are reassigned to a new size class during the quarter when their employment change indicates that a size-class threshold has been crossed. For example, for a firm growing from 3 to 13 employees, the growth of 10 would be allocated as follows: size class 1–4 would be credited with the growth of 1 employee (the growth from 3 to 4), size class 5–9 would be credited with the growth of 5 employees (the growth from 4 to 9), and size class 10–19 would be credited with the growth of 4 employees (the growth from 9 to13). The methodology of dynamic sizing—also referred to as momentary sizing—was initially proposed by Per Davidsson in two research papers in the mid- to late 1990s.3
Dynamic sizing is based on a measurement process that assumes continuous linear employment growth or loss from one quarter to the next, with the growth or loss allocated to the appropriate size class at the moment it occurred. Thus, for a firm growing from 3 employees in June to 13 employees in September, its growth of 10 employees can be linearly modeled as the growth of 1 employee every 9 days (over a total of 13 weeks from one quarter to the next, for a total growth of 10 employees over the 91 days that make up the quarter during which the growth occurred). If a firm’s employment change could be measured on a daily basis, and if the change occurred linearly within the quarter, then the statistics from this measurement process would be equivalent to the statistics from dynamic sizing with quarterly point-in-time employment data.4
1 This relationship was explained by Steven J. Davis, John C. Haltiwanger, and Scott Schuh, Job Creation and Destruction (Cambridge, MA: MIT Press, 1996) and was confirmed by Cordelia Okolie, “Why size class methodology matters,” Monthly Labor Review, July 2004, pp. 3–12, https://www.bls.gov/opub/mlr/2004/07/art1full.pdf. Okolie used BED microdata in her article.
2 Cordelia Okolie, “Why size class methodology matters.” Monthly Labor Review, July 2004, pp. 3–12, https://www.bls.gov/opub/mlr/2004/07/art1full.pdf.
3 See Per Davidsson, “Methodological concerns in the estimation of job creation in different firm size classes,” Working Paper (Jönköping, Sweden: Jönköping International Business School, 1996), and Per Davidsson, Leif Lindmark, and Christer Olofsson, “The extent of overestimation of small firm job creation—an empirical examination of the regression bias,” Small Business Economics, November 1998, pp. 87–100.
4 Data users interested in learning more about BED firm-size data and the process by which dynamic sizing was selected may consult the article "Business Employment Dynamics: tabulations by employer size" and Okolie, "Why size class methodology matters".