Frontiers of Entrepreneurship Research 1995

Frontiers of Entrepreneurship Research
1995 Edition

1995 Abstracts

Navigational Aids

  • Babson-Kauffman Entrepreneurship Research Conference (BKERC)
  • Frontiers of Entrepreneurship Research Topical Index
  • Return to Frontiers of Entrepreneurship Research 1995 Edition
  • Research (General)

  • Table Of Contents

    Order hard copy editions of Frontiers of Entrepreneurship Research by mail

    The tables and graphs in this document are currently unavailable online.  Please see online Order Form at if you wish to order a copy of


    Bruce A. Kirchhoff, New Jersey Institute of Technology
    Patricia G. Greene, Rutgers University


    Although the hypothesis that small firms create the majority of net new jobs is widely believed, a number of researchers have attacked this using empirical data. Since these attacks have been presented as corrections of past methodological errors, their authors may influence policy makers to question the role of small firms. This paper argues that the substance of these attacks is actually a difference in theoretical perspectives rather than methodological differences. Furthermore, past efforts to calculate small firms' share of net new jobs are incorrect for analyzing the dynamics of capitalism. The appropriate methodology for understanding the dynamics of economic growth is analysis of new employment created over time by cohorts of newly formed firms. Recognition of this fact, combined with empirical research supporting this approach, reinforces economic development policies designed to promote entrepreneurship.


    Nearly everyone has heard or read a political speech or a commentary wherein the statement is made: "Small businesses create the majority of new jobs in the economy." This is the small business job creation hypothesis. This hypothesis was created with the work of David Birch in 1979. Prior to 1979, labor economists had analyzed published labor statistics for many years and consistently found that most net new jobs were created by firms in the largest size classes. This analysis was done by counting the number of jobs in each size class in the current time period and subtracting the number of jobs in the same size class in a previous period. Invariably, the vast majority of net job increases occurred in the largest size classes. This method assumes that the firms in each size class in the current period are the same firms as in the previous period. In other words, inter-class movement of firms is negligible. Since no other statistical sources existed to test this assumption, this assumption was unchallenged.

    Then, in 1964, Ijiri and Simon came across a strange phenomena. The distribution of firms by size as determined from published government statistics was a near perfect fit with the "skew" distribution. The reason this was strange is that the skew distribution was widely known to describe dynamic phenomena where the underlying process was one of change, not stability. The fit of the skew distribution suggested that interclass movement among firm size classes was apparently taking place but was unmeasured. The conventional assumption of negligible inter-class movement was in doubt. Yet, no other data was available to clarify this contradiction.

    In the mid-1970s, David Birch received a grant from the Economic Development Administration to study how the movement of firms across state boundaries state employment growth. Recognizing that U.S. government classified statistics could not be used to carry out this research, Birch created a new data base that defined each firm in a base year and measured its individual location and employment behavior in each succeeding year for eight years. Using this new data base drawn from the credit rating files of Dun and Bradstreet, Birch found and reported in 1979 that inter-state movement of firms was a minor part of overall job changes. And, he also discovered that 82 percent of the new jobs created were created by small firms.

    Birch's results stood in sharp contrast to all previous research. Not surprisingly, few economists believed Birch. They argued that the long history of employment analysis could not be toppled by a single piece of research that used a strange and questionable data base. If Birch's findings were correct, the government's published classified statistics were hiding the reality of a dynamic economy wherein interclass movements -- i.e. firm formation, growth, decline and termination -- were hidden from view.1 Furthermore, Birch's findings provided meaning to Ijiri and Simon's [1964] concerns about the underlying dynamics in firm size. Interclass movement was not negligible but instead was a major factor in determining overall employment growth. The small business job creation hypothesis was born.

    Disbelief in Birch's work continues to dominate mainstream macro-economic thinking today. Evidence of this emerges in the surprising absence of theoretical discussion about job creation statistics in economics journals and recent attacks on the small business job creation hypothesis. The recent attacks on the research showing that small firms create the majority of net new jobs serve a useful purpose by questioning the truth of these findings and providing evidence from other data sources. Such critiques need to be examined carefully to determine what information they add to our understanding of the dynamics of capitalist economies and to evaluate if small firms actually do create the majority of net new jobs. But, to carry out a careful review of these attacks, we need to understand the theoretical constructs that underlie the resistance of mainstream macro-economists towards the findings of Birch and others.

    This paper begins with a review of the conflicts between static and dynamic economic theories and then moves to an examination of the differences in methodologies with an emphasis on comparing the job creation conclusions that emerge from applying them. The paper then examines the research from Birch through more recent critics to show how each adopts static or dynamic methods to demonstrate their results. Next, the dynamics that drive capitalism are explored by tracing firm formation and growth. Finally, the paper draws conclusions about the role of job creation research in guiding government policy development.


    The dominant macro-economic theory in the U.S. today and for most of the 20th Century is general equilibrium theory. The general equilibrium model's great strength is that it provides a foundation for mathematical expression of events in the market, allowing for prediction of outcomes of events. Thus, it is widely used for macro-economic policy analysis. However, it is a static model as equilibrium is a static condition. It is assumed that any deviation from equilibrium sets in motion forces that will move the market back to equilibrium.

    One assumption in general equilibrium theory, economies of scale, is of special interest to the job creation debate. Combining this assumption with the perfect market assumption of a uniform market price leads to the theoretical conclusion that large firms have lower costs, greater profits, and therefore the greatest profit incentive to expand. Not surprisingly, general equilibrium economists expect that large firms will be the primary creators of net new jobs. The job creation research prior to Birch [1979] yielded results that demonstrated the truth of these assumptions and therefore reinforced economists' beliefs in general equilibrium theory.

    Another theory exists that has not been widely embraced in the U.S., Schumpeter's creative destruction theory. Schumpeter did not believe that capitalism was characterized by equilibrium markets. Instead, he saw markets characterized by a constant flow of innovation driven by entrepreneurs who entered markets with the ambition of using their innovations to create growth. In this way, entrepreneurs create new wealth for themselves and the overall economy while destroying existing market structures by taking market shares from existing large firms [Schumpeter, 1942]. Schumpeterian economists expect to find job generation dominated by firms that begin life as small, new entrants to markets. The expected dynamics are small firms growing as older, larger firms decline. Birch's results confirm their expectations and therefore have a great appeal to Schumpeterian economists.

    Schumpeter's theory lacks the mathematical elegance of general equilibrium theory since the model focuses on dynamics that cannot be modeled in Newtonian mathematics. In the absence of mathematical modeling, creative destruction does not provide predictions of future outcomes of events and therefore is unable to provide assistance to policy formation. Thus, although Schumpeter's theory challenges general equilibrium theory for accuracy in its description of capitalism, it is inadequate to provide policy guidance and widely ignored by the macro-economic mainstream in America.


    When placed in this theoretical context, Birch's results are not merely a new set of statistics from a strange data base. Birch's results are a direct attack upon general equilibrium theory and a revitalization of Schumpeter's view of capitalism. It is no surprise, then, that criticisms of his work are quickly disseminated among mainstream macro-economists and eagerly believed. But, economic researchers are not debating differences in theory. Instead, the debate concerning the small business job creation hypothesis is dominated by arguments about methodologies. Thus, it seems appropriate to take a section here to discuss the two dominant research methodologies, comparative statics and dynamics.

    Comparative Statics

    The traditional methodology used to compute job creation by size of firm is called comparative statics. One compares the employment of a size class in period t1 with the employment of the same size class in period t1+n where "n" is defined as positive integer defining an elapsed time appropriate to the interests of the researcher. The difference in employment observed in such a comparison is assigned to the size class that is measured. The implied assumption is that the firms in a size class in time period t1+n are the same as those in time period t1 and employment increases (or decreases) are attributed to these firms. If any changes occur in the make up of firms in a size class during the interval n, such changes are credited to the ending period size class. The implied assumptions are that the firms in the size class at the ending period caused the changes in employment and that interclass movement of firms is negligible. These assumptions are at the core of the differences between the results obtained when the same data is analyzed with static or dynamic methods. As more interclass movement occurs and accounts for growth in employment, the two methods will increasingly diverge.

    A couple of examples may help here. Assume we are examining a period of time of ten years, say 1981 to 1990. During this period, a firm with 5 employees grows to a size of 605 employees. At the beginning period, 1981, the firm contributes 5 employees to the 1 to 19 size class. By 1990, the 1-19 size class has lost the five employees while the 500 to 999 size class has gained 605 employees. The comparative static calculation will consider that this firm was in the 500-999 size class for the entire period by measuring its total employment (605) in the 500-999 size class in 1990. Thus, although only 600 new jobs were created since 1981, comparative statics will report 605 new jobs in size class 500-999. Total net job change for all classes will still be correct because the 1990 measurement will report a loss of 5 jobs in size class 1-19.

    Now assume a period of decline. A firm with 605 employees in 1981 experiences misfortune and declines to only 5 employees in 1990. Comparative statics will show a loss of 605 employees in the 500 - 999 size class and an increase of 5 employees in the 1-19 size class. Thus, although only 600 jobs were lost, the larger size class records a loss of 605 jobs.

    From the above examples it is clear that comparative static analysis during periods of economic growth will credit employment growth to the larger size classes while penalizing the smaller size classes by showing employment losses for firms grown large. During periods of economic decline, comparative statics penalizes larger size classes by exaggerating job losses while giving the smaller size classes credit for jobs they did not create.

    Dynamic Methodologies

    Dynamic methodologies are many, as will be noted in more detail later. However, because this is a new field of research, there are no accepted standards for these methods. Before Birch's [1979] seminal work, there were no dynamic methodologies. Lacking a precedent, Birch simply worked out a methodology that is logically simple and intuitively appealing. Birch defined the size of firm at the beginning of a period. Then, he measured the firm's change in employment and credited that employment to the firm's beginning size class.

    Using the example given above, Birch's method treats the first firm's growth from 5 to 605 employees as a growth of 600 jobs credited to the 1-19 size class, i.e., the firm's size at the beginning of the period, 1981. The second firm's decline from 605 to 5 employees is recorded as a loss of 600 employees charged to the 500-999 size class. This methodology is a source of some statistical problems that are at the core of one of the current criticisms and will be explored at greater length later. For now, it is useful to see the overall effect of these two different methodologies.

    Real Data Example

    Beginning in 1980, the U.S. Small Business Administration created a data base modeled on that built by David Birch. The raw data was drawn from Dun and Bradstreet's Dun's Market Identifier file on a biennial basis (even years). There are eight files for 1976 through 1990. These files are also combined into a single longitudinal file. This latter file can be used to measure job creation over time. These files are called the Small Business Data Base (SBDB).

    With the longitudinal SBDB file, it is possible to apply both comparative static and dynamic methodologies and examine the differences. This comparison of methodologies is shown for ten years, from the end of 1976 through the end of 1986 in Figure 1. The left pie of Figure 1 describes the results of comparative static analysis. Note that the size class of 1-19 employees contributes 13.4 percent of total net employment increases while the largest size class, 500+ employees contributes 52.4 percent of employment increases. If we define small firms as those with fewer than 100 employees, small firms created 31.0 percent of all net new jobs between 1976 and 1986.

    (Figure 1)

    The right pie in Figure 1 describes the application of dynamic analysis of the type used by Birch [1979]. Since this analysis was done on the same data base, the total number of jobs created is essentially identical with that in the comparative static analysis. However, the 1-19 size class now accounts for 26.2 percent of the net new jobs while the 500+ size class has declined to 42.8 percent. As expected, in this period of economic growth, the comparative static analysis placed the employment gains of growing firms into the larger size classes. Dynamic analysis places these employment gains in the smallest size class as the once small firms grow large and enter the larger size classes but have their growth assigned to their beginning size class. The difference between the dynamic share (26.2 percent) and comparative static share (13.4 percent) for the smallest size class indicates that 12.8 percent of the net new jobs created are attributable to firms that were small in 1976 and grew into the larger size classes by 1986. In other words, interclass movement of firms is not negligible but instead it constitutes a significant share of net new jobs. Furthermore, if we again use the definition of fewer than 100 employees, small firms contributed 43.6 percent of net new jobs.

    Another data measurement technique needs to be explained here since it becomes important later in the critiques of job creation research. Static share of employment is measured in a single time period, cross-sectionally, as the percent of total employment provided by firms in each size class. In other words, at any one time, employment is distributed among small and large firms across the economy. The percent of employees in small firms is the small firms' static share of employment. In a typical year, firms with less than 100 employees provide 37 percent of total employment. This is their static share. This is different than the share of net new employment that is measured as changes in employment over time.

    Since both pies in Figure 1 were created from the same data base, the differences shown here are solely due to the methodology applied. Therefore, it is apparent that any criticism based upon methodology needs to be evaluated in terms of the theory that is being tested. Methodology arguments can go on endlessly, as will be demonstrated later, without resolution unless the methodology is evaluated relative to a theory.


    Criticisms of the job generation research based upon Birch's [1979] work have taken two forms. Some criticisms mix comparative static and dynamic methodologies into confusing arguments that fail to persuade. Other criticisms focus on the statistical errors that dynamic methodologies incur due to biases in the measurements of size. We discuss these in this order but before doing so, it is appropriate to briefly review a research article that is often incorrectly cited as a criticism of Birch's work, Armington and Odle's article of 1982.

    Early Criticism

    The first substantive testing of Birch's work was advanced by Armington and Odle [1982], the two researchers who developed the SBDB for the U.S. Small Business Administration. Beginning in 1979, Armington and Odle first created and compared the 1978 and 1980 data files. They cleaned the files, adjusted them to the employment totals from the Bureau of Labor Statistics, and then calculated the employment increases classifying the gains by firm size in the beginning year, 1978. The results showed that 35.8 percent of the net new jobs were created by small firms. This percent was a sharp contrast to the 82 percent reported by Birch [1979]. Since Armington and Odle had discussed and coordinated their procedures with Birch, they were puzzled by the difference. The only explanation they had for this difference was the possibility that Birch had carried out his calculations incorrectly.

    Armington and Odle's results created ammunition for major economic writers who were eager to discredit Birch. However, after six additional years of work with the SBDB, Armington uncovered the relationship between the business cycle and job generation, finding that small firms' share of net new jobs declines during periods of economic expansion and rises during periods of economic recession. The reason is that large firms tend to increase their employment in the later stages of an economic expansion and reduce their employment during recessions.2 The small firms increase employment in both expansions and recessions and therefore show a greater share of total employment gains during recessions (100 percent in 1980-82). Armington and Odle [1982], just by accident, did their first analysis during an economic expansion period when small firms' share was cyclically low. Misinterpretations of Armington and Odle's work continue to be used by critics who assert that Birch's work has been "proven" incorrect.

    Employers Large and Small

    Brown, Hamilton, and Medoff [1990] attacked the overall concept that small firms are the major creators of economic growth in America, focusing their attack on the job creation research. They work from a thorough knowledge of the research in the field, as evidenced by their citations of the literature, but they mix statistics from a variety of sources into a confusing argument. They note that most new jobs in the small firm sector emerge from new firm births. They then argue this source of new jobs is, ". . .an accident of birth - new firms happen to be born small."[p. 24] By this they seem to mean that the small firm sector cannot rightfully claim credit for small firm birth employment. They then correctly note that the other major source of job creation in the small firm sector is a few firms that experience rapid growth. The jobs created by rapid growth of small firms are discounted by them because these firms eventually become large and contribute their employment growth to the large firm sector [p. 26]. They then correctly point out that the small firm sector's static share of employment is relatively unchanged over time. Thus, small firms cannot be great job creators if their static share is unchanged [pp. 24-5].

    But, by definition, successful, growing small firms must leave the small sector and enter the large sector. Such success creates new jobs but firm exit to the large sector leaves the small sector relatively unchanged. Basically, Brown, Hamilton and Medoff's argument seems to be that if you eliminate new employment created by births, and new employment created by the soon-to-become-large firms, small firms are not great job creators. In other words, if you redefine the meaning of small firms as those which are not new and are not prone to growth, you can show that small firms are not important job creators. Their logic is internally consistent, but it is hard to discern any theoretical meaning to their assumed definition small firm.

    On the other hand, from a Schumpeterian perspective, the focus on all small firms is not appropriate. Schumpeter's theory of creative destruction focuses on small firms that achieve creative destruction, thereby growing into large firms. But, it seems clear that Brown, Hamilton and Medoff have isolated (and removed) Schumpeter's creative destroyers as the source of small firm job creation, i.e., births and rapid growers. Thus, in an effort to reaffirm the importance of large firms to economic growth, they have inadvertently acknowledged evidence of creative destruction but systematically ignored it, or credited it to large firms.

    Critique of Statistical Methods-

    Davis, Haltiwanger and Schuh [1993] used the U.S. Bureau of the Census' Longitudinal Data File (LDF) on manufacturing firms to examine job creation in manufacturing for 15 years, 1973 through 1988. They did their analysis using three different methodologies, comparative static (they call "current size"), dynamic (called "base year"), and a new method defined as "average plant size." We begin by reviewing their static and dynamic analysis which is shown in Table 1. They express their results as percentage change in employment rather than number of net new jobs. However, their results show the same effect of methodology. Most of their percentage changes are negative because manufacturing employment in the U.S. declined during these 15 years.

    The left side of Table 1 shows the results of comparative static analysis where the plants are placed in a size class based upon their size at the end of the period, i.e., 1988. Note that the greatest employment loss is 4.5 percent experienced by the smallest size class. The smallest loss is 0.2 percent experienced by the largest size class. These results suggest that large firms create more jobs (lose fewer jobs) than small firms. The right side of Table 1 shows dynamic analysis where plants are placed in a class based upon their employment at the beginning of the time period, i.e., 1973. Here, the greatest percent change in employment occurs in the smallest size class and it is a positive 10.3 percent. In fact, only the two smallest size classes show positive changes in employment. The greatest declines in employment are shared by the five largest size classes, firms with 500 or more employees.

    (Table 1)

    The switch of job creation between the two sides of Table 1 is due entirely to the switch in methodologies from static to dynamic just as shown in Figure 1. Clearly, Davis, Haltiwanger and Schuh's research with this entirely different data file confirms the role of methodology in defining the job creators in America. And, to the extent that dynamic methodology, i.e., "base year," is appropriate, it confirms the role of small firms in job creation.

    But, Davis, Haltiwanger and Schuh do not accept either of these methodologies as correct. Instead, they state that Birch and the SBA use a methodology that is flawed by ". . .a statistical pitfall known as the regression fallacy or regression-to-the-mean bias." [1993, p. 15]. They explain this pitfall and give an example of how the SBA's results incorporate it. Their explanation is confusing and will not be repeated here because the work of Picot, Baldwin and Dupuy [1994] provides a far better explanation of the problem. "Regression fallacy" is an unfortunate choice of words since no regression was used in any of this research. And their criticism of the SBA's work and data base begins with incorrect assumptions of how the SBA carries out its calculations and concludes with incorrect statements about the data base including citing Armington and Odle's 1982 paper as evidence that Birch's work and the D&B data base are flawed.

    Davis, Haltiwanger and Schuh's widely publicized conclusion is, "We find that large firms and plants dominate the creation and destruction of jobs in the U.S. manufacturing sector" However, they go on to state, "We find no strong or systematic relationship between net job growth rates and either firm or plant size" [1993, p. 24 - 25]. When examined carefully, these are not contradictory conclusions although it would have been helpful if they had published the gross job change numbers to support the first statement. However, the latter statement clearly ignores the right half of Table 1. Evidence to support their conclusion comes from their "average plant size" methodology which they developed to overcome the problem of "regression fallacy" and shown in Table 2. They propose that the least biased measure of plant size is the average plant size calculated as beginning employment plus ending employment divided by 2. The left side of Table 2 shows the results of using average plant size. There is little discernible difference in the percentages between the small and large size classes. The largest percent loss is in the 2500 to 4999 size class. The smallest is in the 5000 or more size class. Other wise, there is no pattern to distinguish small from large firm job creation activity.

    Davis,, use average plant size because they claim it removes a statistical bias that may occur when using the plant size measured either at the end or the beginning of a time period. But, there is no theoretical basis for using average plant size. Neither neoclassical theory nor Schumpeterian theory suggests average plant size means anything and therefore the results are irrelevant. This evidence is a far cry from the widespread misconception that Davis, Haltiwanger and Schuh found evidence that large firms are the great job creators as stated by Davis in the New York Times [Sylvia, 1994]. Nonetheless, if they are correct about the bias in the dynamic methodology used by the SBA, then there is doubt that previous research by Birch and the SBA validly demonstrates that small firms are the job generators.

    (Table 2)

    They applied another method to their data in a later publication [Davis, Haltiwanger and Schuh, 1994] wherein they measure firm size as the average of employment in the base year and the next census year. Censuses are conducted every five years so this is the average employment over a five year period. The results from this analysis appear in the right side of Table 2 and there is no apparent pattern relating firm size to net employment change. Their reason for applying this methodology is to respond to criticisms of average plant size presented by Picot, Baldwin and Dupuy, whose work is discussed in the next section. However, the Census data base with its five year intervals is not equivalent to Picot,'s work.

    Thus, besides identifying a possible source of statistical bias in Birch's and the SBA's methodology, Davis, Haltiwanger and Schuh's research muddles the job generation research with a new statistical methodology that is not based on economic theory but on statistical assumptions. However, it is very important to note that none of their statistical methods creates an analysis that shows large firms create more net job growth than small firms.

    A Canadian Reassessment of the Facts

    Picot, Baldwin and Dupuy [1994] recognize the conflict in methodologies and sought to test alternative methodologies systematically to expose the effect of each on job creation research. In addition to being an insightful analysis of methodology, they make the important general comment that ". . .the outcome variable (the growth in employment) influences the classification variable (the size of firm)" [p. 14]. By this they mean that any change in the method of size classification based on employment automatically changes the outcome of the employment change analysis in a predictable way because of the linkage between the two. In other words, growing firms are increasing employment so classifying firms based on an employment size measure other than their beginning employment, automatically shifts employment growth into larger size classes.

    Picot, Baldwin and Dupuy acknowledge the statistical bias (called "regression to the mean" by Davis,, although there was no regression used in this research) caused by using beginning firm size and describe it precisely by noting that employment changes in a firm consist of two components: a long-run and a transitory component. ". . .many of the changes in the employment levels in firms are transitory and the observed gain (or loss) is reversed in the short-run. There may also be a longer run trend" [p. 10]. Both components can be present in a measurement of firm size. This suggests that the timing of the observation, whether during a transitory peak or trough, determines the size classification accorded the firm, regardless of the long-run employment level.

    Picot, Baldwin and Dupuy then go on to use their data to test the four methodologies used by Davis, Haltiwanger and Schuh, plus one of their own invention, the "prior average size." Prior average size is calculated as employment in the year before employment growth measurement begins (t0) plus the beginning (base) year employment (t1) divided by 2. Their data set is the Longitudinal Employment Analysis Program (LEAP) data file created by Statistics Canada. This file includes longitudinal, annual microdata for all firms in Canada from 1978 through 1992.

    Their results (as shown in Table 4) show that the base year method (dynamic method) and the prior average size method both yield essentially the same results, i.e., small firms create the majority of net new jobs. This is true for the entire Canadian economy and for manufacturing. These results are consistent with Birch's, the SBA's, and Davis,'s base year results. Notice especially that prior average size and base year net job change rates are essentially equal for each firm size class. This suggests that the transitory component hypothesized by Picot,, (and Davis, is not a major factor in determining net job change rates, at least not in Canada.

    The long run average size and "average current size" (equivalent to Davis,'s "firm size"), yield substantially different results than those found in either the SBA's static methodology or Davis,'s methodologies. Here the Canadian economy shows small firms with higher job growth for both size classification methods. Assuming the methodologies are essentially the same, the Canadian economy apparently experiences greater small firm job growth than the U.S. economy. And, this holds true for all segments of the Canadian economy.

    (Table 4)

    This brief review of Picot, Baldwin and Dupuy's work does not adequately express the detailed and complete content of their paper. Table 4 shows only a small portion of their work. In all, they use six alternative definitions of business size class and evaluate these using three different time periods to measure growth. And, they do this for four sectors of the Canadian economy. However, the review given herein does highlight their major contributions to the debate about methodology for measuring dynamics of job generation.

    Overview of Methodologies

    Picot, make it clear that methodology alone can determine the outcome of any test of the small business job creation hypothesis. Several methodologies developed by Davis, show no advantage for small firms and reject the job creation hypothesis. Others, used by Birch and the SBA, or developed by Picot, demonstrate an advantage for small firms and accept the hypothesis. Choice of methodologies depends upon what the investigator wants to prove.

    Strangely, the underlying economic theory is never mentioned in these discussions. For general equilibrium theory, the topic is of no theoretical importance anyway since firm size is the outcome of market forces, not the cause. The job generation hypothesis is merely an annoyance since its outcome does not agree with theory expectations. However, for Schumpeterian economic theory, the topic is important and inadequately addressed by the research reported on small firm's share of job creation. Several researchers have noted that most job growth comes from the formation of new firms. Yet, none of these methodologists test the fundamental principle of new firm formation and growth directly.


    Dynamics of birth, growth, decline and death are best measured with cohort analysis. Two major cohort analyses have been done on job creation. The first, by Kirchhoff and Phillips [1989] used the SBA's SBDB to look at a single cohort of all firms born in 1977 and 1978. Jackson [1995] used cohort analysis to examine firm survival and job creation in the Michigan Employment Security Commission's Longitudinal Data Base.

    New Firms -- Innovation and Growth

    Kirchhoff and Phillips [1989] used the SBDB to select all 814,190 firms born during 1977 and 1978 and followed these firms for six years through 1984. They classified the firms as being in high, medium, or low innovation industries. And, they measured their growth as high, medium or low.

    First, they report that the percent of firms that survive with their original owners does not differ among the innovation classes. Thus, the chances of survival are the same for all levels of innovation activity. Second, they found that the incidence of high growth was 16.9 percent among high innovation firms but only 9.0 percent among low innovation firms. Thus, innovation rate contributes to growth. However, in spite of the higher incidence of high growth rate firms, high innovation rate firms created fewer new jobs than did low innovation rate firms. This is because there are five times as many low innovation firms as high innovation firms. But, only the high innovation class employed more workers after six years than in the base year. This is because firm discontinuances 3 removed more jobs than the surviving firms created in the medium and low innovation rate classes. Overall, at birth, the 814,190 firms employed 4,529,413 workers. In 1984, the 312,662 firms remaining with their original owners employed 3,406,544 employees, a 25 percent employment decline. However, this cohort of firms accounted for 20.6 percent of the net new jobs added to the U.S. private work force between 1976 and 1984. And, they contributed 4.0 percent of total employment in 1984 [Kirchhoff, 1994, p. 192]. Creative destruction was certainly at work.4

    Michigan Cohort Analysis

    Jackson [1995] provides an analysis of multiple cohorts drawn from the State of Michigan employment statistics that are organized into a longitudinal micro data file. He creates a separate cohort for all firms born in each year and then traces all firms in the cohort for survival and employment until the end of the file. Jackson's file has an advantage over other longitudinal microdata files since the employment statistics include information on predecessors and successors in ownership changes. As mentioned earlier, changes in ownership cause the appearance of a death and a birth in both the Census Longitudinal Data File and the SBA's SBDB. Thus, both of these files overstate births and deaths and understate the true employment creation of an age cohort of businesses. Jackson's file more accurately portrays the employment of the businesses, even when they change ownership.

    Because of the inclusion of successors, survival rates reported by Jackson (39 to 59 percent after six years) are much greater than that reported from the other longitudinal microdata bases. Phillips and Kirchhoff [1989] report 38 percent after six years calculated from the SBA's SBDB, and Dunne, Roberts and Samuelson [1988] report five year survival rates of 36 percent calculated from Census' Longitudinal Data File. Jackson's percentage rates are close to work by Kirchhoff [1994, p. 167] that incorporates estimates of ownership change successors.

    Furthermore, Jackson's results show each cohort of new, small firms increases in total employment every year after birth unlike the results reported by Kirchhoff and Phillips [1989]. Of particular interest in Jackson's findings is that after six years, the small firm cohorts (one to 50 employees at birth) show annual rates of change in number of employees (average size) and wages in excess of 1.0, i.e., positive growth. However, in both manufacturing and services, large new firms (greater than 50 employees at birth) show net annual rates of growth of less than 1.0, i.e., declines in employment. The conclusion is that firms that begin life with 1 to 5 and 6 to 10 employees double their employment and almost triple their payrolls by the end of six years. No other size class comes even close to these growth rates.

    The Net Effect of Including Successors

    Comparing Jackson's results to Kirchhoff and Phillips [1989] demonstrates that incorporating ownership change successors into longitudinal analysis makes small firm birth and growth far more important to net new job formation than previously reported. Whereas Kirchhoff and Phillips report decline in total cohort employment after six years due largely to firm departures from the file (either due to deaths or ownership changes), Jackson reports continuous growth in employment when ownership changes are incorporated. Since Census' Longitudinal Data File suffers the same ownership change problem, Census' employment numbers are also distorted in the same way [Kirchhoff, 1994, p. 165-67].

    Interestingly, this distortion due to ownership changes does not affect the share of net new employment of small firms that stay within the small sector. Firms in the small sector that change owners have their employment deducted as a death and added back in as a birth. Thus, ownership changes create a false picture of employment volatility (or churning) but do not effect the net job changes, i.e., births minus deaths. However, when a small firm grows beyond the boundaries of the small firm sector and then changes ownership, the net effect is to subtract employment from the small sector due to a false death and add employment to the large sector due to a false birth. Therefore, ownership changes among new, fast growing small firms tend to bias employment growth towards the large firm sector. Clearly, accurate descriptions of the creative destruction behavior of new small firms can only be attained if the problems of microdata errors due to ownership change are resolved. In the meantime, the frequent mentions of high rates of employment churning should be viewed with caution since some of this, perhaps much of it, is due to measurement error associated with ownership changes.


    We began this paper by noting that there are major differences in theoretical beliefs regarding capitalism and suggested that these differences need to be addressed when assessing the arguments about job creation research. We described the two major theories, general equilibrium theory that suggests that large firms will create most net new jobs and Schumpeterian creative destruction theory that hypothesizes that small firms will create the most net new jobs. We go on to examine the major job creation work in the U.S. to note that most of the conflict among the researchers doing this work has been defined in terms of methodology, i.e., the different definitions of small businesses and/or size classes, or the measurement methods and data bases used to carry out research.

    But, these methodological arguments can continue endlessly as evidenced by the elaborate evaluations of Picot, Baldwin and Dupuy [1994] who consider no less than six different size definitions combined with three different time periods and find different results in each of 18 analyses. There is no clear statement of right and wrong emanating from their inquiry into methodology, and rightly so because the answer to the dilemma is in theory. Methodology is selected to test theory, not vice versa.

    Since we find Schumpeter's creative destruction theory to be compelling, we proceed to analyze those few research studies that use cohort analysis to trace the formation and growth of new small firms. Both Kirchhoff and Phillips [1989] and Jackson [1995] provide ample evidence of creative destruction. But especially Jackson's work demonstrates the vital employment contribution made by new small firms actively growing. Perhaps more importantly, Jackson's work, when contrasted with Kirchhoff and Phillips' work, shows the importance of including ownership changes in longitudinal analysis. Measurements of survival and job creation are greatly affected by ownership changes unmeasured by the SBDB and the Longitudinal Data File.

    The greatest concern we have is that the methodological arguments of the last few years may distract public policy makers from efforts to give greater attention to the small business sector in economic development policy. It is important to note that no research, except when the purest static methodology is applied, has shown that large firms create a disproportionate share of net new jobs in the U.S. or Canada. Furthermore, cohort analysis demonstrates unequivocally that new small firms do create a disproportionate share of net new jobs. This latter research provides strong evidence supporting Schumpeter's theory -- equilibrium markets are few and creative destruction drives economic development. Public policy must focus on encouraging new firm formation and growth. Foremost among these is destroying barriers to new firm entry into existing markets and to encourage resource mobility, both labor and capital, to encourage new firm formations and growth in all parts of the world economy.


    Armington, Catherine, and Marjorie Odle, 1982, "Small Business--How Many Jobs?" The Brookings Review, Winter, pp. 14-17.

    Birch, David L., 1979, The Job Generation Process, Unpublished report prepared by the Massachusetts Institute of Technology Program on Neighborhood and Regional Change for the Economic Development Administration, U.S. Department of Commerce, Washington, D.C.

    Brown, Charles, James Hamilton, and James Medoff, 1990, Employers Large and Small, Cambridge, Mass.: Harvard University Press.

    Davis, Steven J., John Haltiwanger, and Scott Schuh, 1993, "Small Business and Job Creation: Dissecting the Myth and Reassessing the Facts," Working Paper No. 4492, National Bureau of Economic Research: Cambridge, MA.

    Davis, Steven J., John Haltiwanger, and Scott Schuh, 1994, "Small Business and Job Creation: Dissecting the Myth and Reassessing the Facts," Business Economics, Vol. 19, No. 3, July, pp. 13 - 20.

    Dunne, Timothy, Mark J. Roberts, and Larry Samuelson, 1988, "Patterns of Firm Entry and Exit in U.S. Manufacturing Industries," Rand Journal of Economics, 19(4): p. 495-515.

    Ijiri, Yuri, and Herbert A. Simon, 1964, "Business Growth and Firm Size," American Economic Review, 54 (March); pp. 77-89.

    Jackson, John E., "Firm Size and the Dynamics in a Market Economy," 1995, unpublished working paper, Institute for Social Research, University of Michigan: Ann Arbor, MI.

    Kirchhoff, Bruce A., and Bruce D. Phillips, 1989, "Innovation and Growth Among New Firms in the U.S. Economy," Frontiers of Entrepreneurship Research, Babson College: Wellesley, MA, p. 173-188.

    Kirchhoff, Bruce A., 1994, Entrepreneurship and Dynamic Capitalism, Praeger: Westport, Conn.

    Phillips, Bruce D., and Bruce A. Kirchhoff, 1989, "Formation, Growth, and Survival: Small Firm Dynamics in the U.S. Economy," Small Business Economics, 1: pp. 65-74.

    Picot, G., J. Baldwin, and R. Dupuy, 1994, "Have Small Firms Created a Disproportionate Share of New Jobs in Canada? A Reassessment of the Facts;" a paper presented at the Canadian Economics Association Meetings, June, Calgary, Alberta, Canada.

    Schumpeter, Joseph, 1975 (1942), Capitalism, Socialism, and Democracy, New York: Harper and Row.

    Wagner, Joachim, 1994, "The Post-Entry Performance of New Small Firms in German Manufacturing Industries," Journal of Industrial Economics, 42, No. 2, pp. 141-154.

    1 Federal law requires government statistics to be published in a form that assures the security of individual entity data. Birch used a private data source wherein the security of individual entity data is not compromised.

    2 A comparison of the findings of Birch and Armington and a full explanation of this effect can be found in: Kirchhoff, 1994.

    3 Discontinuances include firm deaths and ownership changes. When a new owner acquires an existing firm, the change in ownership appears as a death (old firm) and birth (new ownership firm). This complicating factor exists in both U.S. longitudinal statistical data files to varying degrees and the degree of error caused by this is not known. Only the Michigan State Employment longitudinal file has a measure of ownership changes.

    4 Other researchers are recognizing the value of cohort analysis. Most notably, Joachim Wagner uses cohort analysis to report six year survival rates of 57 to 65 percent among new German manufacturing firms started between 1979 and 1982. See: Wagner, 1994. In private correspondence, Wagner has indicated that his data includes adjustments for ownership changes.

    Return to Babson College
    Main Home Page
    Table of
    ©1996 Babson College. All rights reserved.
    Last updated November 22, 1996 by Cheryl Ann Lopez