Order hard copy editions of Frontiers of Entrepreneurship Research by mail
MEASURING BUSINESS FORMATIONS AND DISSOLUTIONS:
A NEW TIME SERIES
William J. Dennis, Jr., NFIB Education Foundation
William C. Dunkelberg, Temple University
Terri Dial, Wells Fargo Bank
The number and composition of business formations and dissolutions in the United States has been a troubling statistical issue for researchers, policy-makers and advocates. This paper introduces a new time-series that measures the number of business formations and dissolutions. The series also characterizes the individuals and businesses involved by selected attributes. The data collection technique is an "omnibus" survey that interviews 36,000 households every year. The first interviews in the series were conducted in January, 1995. This paper discusses methodological issues involved, reviews selected results from test surveys and the first two months of data collection, and describes public access to the raw figures.
The number of business formations and dissolutions in the United States over any defined time period remains a mystery.1 No "official" figures exist. Statistical agencies do not even publish an estimate. Indeed, the "wild card" in the official monthly employment total is the formation/dissolution phenomenon (American Statistical Association, 1994). Private sources do not provide much help either. Dun & Bradstreet's New Incorporation series, the most common citation, is at best a crude proxy of formations and its Business Failure series measures just a corner of the dissolutions phenomenon.
Estimates of formations/dissolutions from researchers working with the data vary wildly. Birch (1987), for example, believed that from the 1970s through the mid-1980s, millions of new businesses started each year. Reynolds (1995) estimated 925,000 annually during the early 1990s of which fewer than 200,000 employed someone other than the owner. The Small Business Administration (1993, p.38), using numbers supplied by the Employment and Training Administration at the Department of Labor, calculated formations to be about 870,000 in both fiscal 1991 and 1992. Approximately 80 percent of the those formations were de novo starts and the remainder purchases. Others were more conservative. Armington gauged the number at 300,000 - 400,000 per year during the late 1970s, and Kirchhoff and Phillips provided a similar 400,000 estimate for the later 1980s (Kirchhoff, 1994, pp. 145-146).
Dissolution estimates, when made, generally parallel formation estimates, though at somewhat lower levels reflecting population growth. For example, the Small Business Administration (SBA, p.38), again with Department of Labor data, estimated 835,000 to 838,000 dissolution annually between fiscal 1989 and 1991.
These significant variations stem largely from the data bases employed. Armington, and Kirchhoff and Phillips calculated their estimates from Dun & Bradstreet data and obtained similar results. However, that data set contains well-known liabilities, the most important for current purposes being the under-representation of new and young firms (Kalleberg, 1990). Birch also used the Dun's files, but compensated for the under-representation in his estimate with judgment. SBA's effort was ultimately based on state unemployment insurance files. That data set has its own problems, not the least of which is an ill-defined unit of analysis (Kalleberg, 1990). Reynolds employed a different tack. He went directly to the public through telephone interviews, an approach this research emulates. His idea was to eliminate the intervening data collection entities and directly interview the people making the formation/dissolution decision.
A second issue in quantifying formations and dissolutions is a consistent and meaningful definition of the phenomenon. Definition has always been recognized as a precondition for reliable estimates. However, it has usually been subordinated to contents of the data set. For example, estimates developed from the D&B files generally represent employers, though not necessarily so. Reynolds (1990) for the first time presented estimates specifically for employing businesses, and for employing and non-employing businesses combined. Other important defining characterizations still wait to be drawn, e.g., full-time and part-time operations. Thus, the problem is not only constructing a reliable estimate, but a reliable estimate of what?
THE NEW FORMATIONS/DISSOLUTIONS TIME-SERIES
This paper reviews progress on a new formations/dissolutions time-series. As such it more closely, resembles a report than a paper with hypotheses, methodology, data collection, etc. But even in report form, it raises serious definitional, methodological, and data compilation issues, and offers insights into potential findings from a substantial research effort.
Purpose of the Research
The primary purpose of this research is to measure business formations and dissolutions in the United States on a quarterly basis. The corollary is to determine how those formations and dissolutions2 occur (founded or purchased, and closed, lapsed, or sold), the pervasiveness of multiple ownership, and the significance of the enterprise involved. The secondary purpose of the research is to profile those people who form businesses and the nature of any changes in that population over time. The tertiary purpose is to quantify isolated facts or pieces of information which singularly offer insights, suggest research, or quantify the unquantified.
Conduct of the Research
The principal impediment to measuring the number of formations has never been how to accomplish the task -- interview enough people and you will get the answer (Reynolds and Miller, 1994). Rather, the problem has been how to accomplish the task cost-effectively. Since the likelihood of a recent formation in any household is relatively low, thousands of households must be interviewed to obtain enough cases to make a reliable estimate. The number interviews required increases further if the time-series measures activity over comparatively short intervals, e.g., a month or calendar quarter.
Costs mount quickly when attempting to randomly locate and interview a small population. Researchers must pay for each interview in a conventional telephone survey whether or not the respondent qualifies. Since each interview has fixed costs, e.g., the time to reach a respondent qualified or not, the relative number of qualified and non-qualified respondents becomes a major financial consideration.
In this project, researchers must contact at least 2,000 households to obtain 100 qualified respondents. We might be able to purchase non-qualifying interviews for $10 each "on the cheap." But, the cost from an established, reputable firm runs at least 50 percent higher. That translates into $200 to $300 per qualified respondent. Sampling error for n=2,000 with a 95%-5% split is 1.5 percentage points. A 1.5 percentage point error band (95 percent confidence level) compared to a 5 percent qualifying response is substantial. The number of qualified respondents must, therefore, be boosted significantly in order to provide a reliable estimate of formations let alone a profile of owners or businesses involved. To obtain 1,000 qualified respondents in a conventional telephone survey, the cost runs $200,000 to $300,000 per year for interviews and basic tabulations. And that is at the most basic level. Reynolds (1995, p. 17), employing a more complex screening structure and a longer questionnaire than the one used here, placed direct costs at $500 per qualified interview.
The "omnibus" survey technique, also known as a "caravan" survey, largely resolves the cost-effectiveness problem. By purchasing a limited number of questions and sharing the overhead costs with others who also want to ask a limited number of questions, each participant in the omnibus pays only for the questions desired. The effect is to eliminate a large portion of the expense involved with contacting non-qualified respondents.
In this project, we pay for the two screen questions that are asked of everyone in the sample. Only when a respondent passes a screen does the interviewer move to our extended set of questions. Charges for sampling, contact, call-backs, demographic responses, verification and overhead are shared. Average interviewing costs thereby decline significantly. While still expensive, the omnibus survey technique vastly enhances the financial feasibility of collecting data on business formations, particularly in time-series.
Every month the Gallup Organization conducts three omnibus surveys consisting of 1,000 respondents each. Every survey is independent and subject to normal methodological procedures used by Gallup. The three occur over separate intervals during the month. All are household surveys. Random digit dialing is the sampling method employed for each. (A section on methodological issues appears later in this paper.)
The questionnaire used in this research consists of two screen questions -- one each for formations and dissolutions -- and a series of follow-ups should either screen receive a positive response.3 The follow-ups to both screens are parallel. Thus, we obtain symmetric data for each phenomenon. The first section of the questionnaire for qualified respondents focuses on the count of business formations. Experimental interviews found that as many as 25 percent of respondents volunteered that adults in the household had formed more than one business in the last six months. The first follow-ups, therefore, center on the number of businesses formed by members of the household. A month of entry is then obtained for each of the first three businesses formed. These data allow us to screen businesses which were formed more than six months earlier, and to consolidate the count of businesses which were formed in a single month but discovered in various survey months. We subsequently determine if the business was founded or purchased. The final piece of this survey section identifies the adult in the household taking the action.
The second section of the follow-up questions focuses on the individual who formed the business. Since demographic data on the respondent is part of the shared overhead of an omnibus survey, most demographic questions need only be repeated if someone other than the respondent formed the business. Individual demographics include: age, sex, marital status, education, employment status, occupation, household income, race and region of the country. We also posed a question on immigrant status.
The third section focuses on the business. Multiple ownership is the most critical issue since it is essential to the formation measure. The experimental surveys showed most multiple owner situations involved just one owner outside the household. However, the incidence of larger numbers was great enough to include a special question quantifying it. The number of hours worked on the business in the first month of operation and the number of employees other than owners let us distinguish between "significant" businesses and "nonsignificant" businesses. A question on location helps estimate the proportion of home-based businesses. Another on business telephone numbers at formation sheds more light on the use of telephone listings as an estimation for the total business population. Finally, we collect general industry information.
Definitional and Methodological Issues
The unit of analysis for this time-series is the household. The respondent is asked about the business formation activities of any adult in the household, not just his or hers. The choice of the household creates two obvious problems: (1.) respondents are more likely to recall their own activities than activities of another adult in the household, particularly if those activities are modest. Reporting on the activities of others also prevents us from asking questions that only someone directly involved would know, e.g., finance, motivation, or would be too detailed for anyone other than the respondent to answer, e.g., strategy. (2.) Weighting problems arise since adjustment needs to be made based on the composition of each household.
The use of the household rather than the respondent as the unit of analysis raises the population interviewed from 36,000 per year to somewhat over 72,000 per year, and the number of qualified respondents (formations only) from an estimated 600 per year to an estimated 1,200 per year. These numbers are required to reach the research objective given the sampling error we wish to achieve (0.5 percent for the annual estimate and 0.75 percent for the quarterly estimate).
Problems created by using the household rather than the respondent are also less than they might at first appear. Gallup routinely collects the household composition data required to weight the sample, constructs weights, and offers weighted data with every survey. In addition, the number of observations possessing firsthand knowledge can effectively be expanded by collapsing the data over a two year period.
One bothersome issue is recall of less consequential information. Results from experimental interviews suggested that when the recall period extends back one year, respondents were twice as likely to identify themselves as the household adult forming the business as they were to identify another household adult. When the recall period was clipped to six months, the numbers were virtually identical. The final questionnaire employs a six month recall period. However, it is still possible that a respondent might recall an occurrence in which he took part and not a comparable occurrence in which another household adult took part. The potential is particularly great if the experiences were not significant. That raises the question 'is the formation or dissolution of a business, especially a very small business, sufficiently consequential so that a respondent would remember it equally well if he or another adult household member were the participant?' The issue will be discussed subsequently.
The two critical concepts for this time-series are formation and dissolution. What do we mean by formation? What do we mean by dissolution? And then, how do we communicate those concepts to the respondent in order to elicit an accurate response?
Business formation is a complex process rather than a single event (Reynolds, 1995). It occurs over a wide variety of time spans; there are no uniform set of events; and, the events that do occur are in no fixed sequence. But, at some point, the individual is either in business or not. Since there are no events alone or in combination that distinguish a business from a business in gestation, the point of demarcation is dependent on an individual's assessment of his status. So, the screen question on formations revolves around the individual's determination of whether or not he is "open for business." Finally, since purchase of an on-going business is subject to different interpretations about its new business status, we specifically incorporate purchases in the screen. They will be separated from de novo starts later. The result is a screen, "In the last six months, have you or has another adult living in your household, alone or with others, started or purchased a business, one with sales or income?"
Communicating the idea of dissolution is more difficult, particularly in communicating it in a non-threatening manner. The sale of a business is a straight-forward concept. It requires little clarification. Closure of a business generally is also clear, although seasonal businesses, e.g., summer resorts, can cause problems. An inactive business is another matter. For example, what if the business has no sales in the last few months and the owner's attention is elsewhere? Is it still a business or has it dissolved?
Test results from the dissolution screen were not satisfactory. First, we encountered a probable recall problem as virtually the same number reported dissolutions regardless of the six month or twelve month recall period. Further, the number of dissolutions was significantly smaller than the number of formations. The gap is not necessarily fatal as conduct of the test surveys coincided with periods of economic strength and rapid growth in the net number of new businesses. But concern arose when the number of sales did not match the number of purchases. We altered the screen to make it less threatening. Six months was the recall period adopted and the terms "sold," "shut down," and' inactive" were employed. The question in the final draft read, "Are you, or is anyone in your household, an owner or part-owner of any business that may have become inactive, shut down, sold or transferred in the last six months?"
The Gallup Organization typically obtains a response rate in its omnibus surveys of about 50 percent. This leaves considerable room for response bias, though weighting reduces known under-representation of certain demographics. Since we are measuring an uncommon event however, even a small response bias could induce a significant undercount. And, there is reason to believe bias may be present. Prior research shows that new business owners are busier than are other elements of the population, potentially making them more difficult to interview (Cooper, et. al., 1990).
The potential response bias problem will be addressed in three ways: (1.) the average number of "call-backs" required to reach a qualified respondent will be compared to the average number required to reach all respondents every month. If they approximate one another, the presumption is that response bias does not exist. (2.) Since there are about 2.04 adults per household, the number of respondents who report they formed a business should approximate the number who report another adult in the household formed a business. If those figures approximate one another, the presumption is again that bias doesn't exist. However, this test must be viewed with caution as it can examine recall as well as bias. The recall problem could be serious for part-time operations, but should not affect "significant" businesses, e.g., those employing others. The bias check will, therefore, focus on "significant" businesses. (3.) We intend to conduct a "complete-the-frame" study later this year. We will actively pursue non-respondents during a specified period to determine if they have a higher incidence of formations. This effort is critical not only as a test for bias, but to determine the extent and necessary adjustment should bias exist.
TEST AND JANUARY-FEBRUARY RESULTS
The NFIB Education Foundation conducted test surveys in August, 1993, and October, 1994. The tests were to determine the approximate number of qualifying respondents per 1,000 interviews and to experiment with words, phrases and even concepts to be employed in the survey instrument. The following tables present selected data from those test surveys and the 6,000 "real" interviews conducted in January and February, 1995. Before reviewing the tables, a note of caution is warranted: the test data are not always directly comparable to January's and February's data. Also, the actual numbers are too small to be more than suggestive. Still, the available data illustrate the survey's potential and are worth reviewing.
Table 1 presents the incidence of business formations and dissolutions by the month of formation/dissolution and the data set employed. Note that in one test survey, the reference period is 12 months; in the other, it is six months -- the reference period finally adopted. Not included on the table were test data with a reference period of "one year," but without a follow-up to determine month and with the unit of analysis the family rather than the household. The latter produced 55 formations (n=1,000) and 25 dissolutions. Those formation numbers are less than other test results, though the dissolution figures are similar.
The data on Table 1 suggest an annual entry rate of about 60-70 businesses per 1,000 households. However, estimating the annual entry rate is vastly more complex than dividing the number of positive responses by the number of households.
Once a respondent indicates that a formation has occurred in a household, we must determine if there has been more than one formation in that household. Table 2 indicates this is not a trivial number.4 We then determine if each formation occurs within the reference period (last six months). Table 1 indicates that the number falling outside the reference period is also not trivial. The next step is to determine the number of "partners"/multiple owners/operators outside the household. The statistical effect of multiple ownership is to raise the probability of locating an owner vis-à-vis a business. Thus, multiple owners must be subtracted, effectively yielding one owner per business. From the resulting total, we subtract the number of formations which are purchases of on-going businesses whether or no they occur during the reference period. The total is divided by the sample size. The resulting fraction is then multiplied by the number of households nationally.
These calculations occur before Gallup has weighted the responses to ensure that respondents reflect the profile of American households.
Without explaining the decision rules, some of which have yet to be finalized, the formula used to calculate the estimate will be:
(1) (N1-IN1)+(((N2-IN2)/2)-M1)+(((N3-IN3/3)-M2)-(NP) * H
N1 = number with one positive response in household
N2 = number with two positive responses in household
N3 = number with three or more positive responses in household
NP= number purchased
IX = number ineligible because they did not occur in reference period
S = sample size
M1 = number with one owner outside household
M2 = number with two or more owners outside household
H = number of households
Initial results suggest an extraordinary number of formations are occurring -- far more than anyone has estimated to date. An early rough approximation lies in the four million range. Dissolutions are fewer. They may approximate three million. However, all businesses are not born equal.
Table 3 breaks out of the initial positive responses to each survey by the number of hours worked in the business per week during the first month of operation. Note that about half worked less than 40 hours per week during the first 30 days. Another 10 percent worked between 40 and 50 hours. Other data gathered from the survey, but not produced here suggests that the only about one in four of these new business employ people other than the owners. Thus, the estimated number of formations depends heavily on how we choose to define a business.
Table 4 presents a demographic profile of the people who formed and dissolved business over the relevant time frames. Nothing in the initial results suggest that the final figures will drastically alter our knowledge of this sociology. The interesting question, however, is change in the profile over time. For example, we know that if series had been instituted 20 years ago, the incidence of women in the population would have been significantly less. How much less is an important issue. Also, the secular change may hide a cyclical component, raising an entirely different series of questions.
Table 5 suggests other possibilities that can spring from this research. The first example addresses the research and practical issue of the ease or smoothness of the transition to and from self-employment. "Negative pushes," the most being a loss of employment, is frequently considered an important stimulus for new business formations. Conversely, recent research suggests that business dissolution is directly tied to other economic alternatives (Cooper, 1992). The time-series probes the work and occupational status of business owners 30 days before formation and 30 days after dissolution. Preliminary results suggest that relatively few are unemployed immediately prior to forming a business and less than 15 percent are unemployed a month after dissolving one. Such transitions are apparently far less rocky than we may have thought.
AVAILABILITY OF THE TIME-SERIES DATA
Wells Fargo Bank and the NFIB Education Foundation will publish the 1995 results for the first year of this new time-series in April, 1996. They will subsequently issue the formation figures on a quarterly basis with an annual report each April and possible periodic reports on particularly interesting issues, e.g., number of home-based formations. Once such announcements have been made, the data will be available for academic research purposes. The cost of the data will be that of reproduction. The principal stipulation is that any paper produced from the data recognizes Wells Fargo Bank and the NFIB Education Foundation as the data's source.
This time-series must maintain certain questions from year to year in order to remain viable. The sponsors are keenly interested in the demographic profile of people forming businesses. Questions providing those data are, therefore, also fixed. However, a few others are subject to annual change, and suggestions for substitute questions are welcome.
The Wells Fargo/NFIB business formation time-series is very much "research in progress." The data have just begun to trickle in. Potential response bias issues must be assessed further and, if present, resolved. Semantic and reporting questions, particularly with respect to dissolutions, remain. Still, this research is highly promising and potentially significant.
Initial appearances indicate that entrepreneurial activity in the United States, as least as defined by business formations, is even greater than generally thought. The number of formations likely exceeds four million per year. A "seething cauldron" of activity is not an inappropriate metaphor for what appears to be happening. However, the number of "significant" formations is much smaller. If the definition of "significant" is the full-time occupation of the owner, the number drops by 50 to 60 percent. If the definition is employing someone other than the owners, it drops by 75 percent.
The converse does not appear to true. The number of dissolutions appears comparatively small -- between two and one-half and three million. The resulting gap between formations and dissolutions can be explained, at least in part, by growth in the business population resulting from a healthy, growing economy. Whether or not the entire gap can be explained in this manner remains an open question.
One conclusion seem warranted this time: the omnibus survey is a promising technique for use in new and small business research. Investigators can obtain reliable and useful data at relatively low cost. Moreover, creative researchers can leverage their data by asking respondents in the initial interview if they would consent to a follow-up interview. The result is an inexpensive means to identify and interview a limited population.
Many prefer to conduct research on either more established or growing businesses. The omnibus survey technique should work almost as well with those populations as with households. The sampling frame would be less satisfactory than households, but the overriding limitation is that no one currently conducts a business "omni." That problem fortunately is temporary. The NFIB Education Foundation in cooperation with the Gallup Organization will initiate a small business omnibus survey later in the year.
American Statistical Association, Panel for the BLS Current Employment Statistics Survey, (1994) "A Research Agenda to Guide and Improve the Current Employment Statistics Survey," January.
Birch, David L. (1987) Job Creation in America: How Our Smallest Companies Put the Most People to Work. New York, NY: The Free Press.
Cooper, Arnold C., Dunkelberg, William C., Woo, Carolyn Y. and Dennis, William J., Jr., (1990) New Business in America: The Firms and Their Owners. Washington, DC: The NFIB Foundation.
_______________, Folta, Tim, Gimeno-Gascon, Javier, and Woo, Carolyn Y. (1992) "Entrepreneurs' Exit Decisions: The Role of Threshold Expectations," Krannert Graduate School of Management, Purdue University, Paper No. 1018.
Kalleberg, Arne L., Marsden, Peter V., Aldrich, Howard E., and Cassel, James W. (1990), "Comparing Organizational Sampling Frames," Administrative Science Quarterly. 35 (1990): 658-688.
Kirchhoff, Bruce A. (1994) Entrepreneurship and Dynamic Capitalism: The Economics of Business Firm Formation and Growth. Westport, CT: Praeger.
Reynolds, Paul (1995) The National Study of U.S. Business StartUps: Background and Progress Report. Presented at the Conference on Dynamics of Employment and Industry Evolution. Mannheim, Germany: University of Mannheim, 19 January.
___________ and Miller, Brenda (1992) "New Firm Gestation: Conception, Birth, and Implications for Research," Journal of Business Venturing, 7, 405-417.
__________ and White, Sammis B. (1992) "Finding the Nascent Entrepreneur: Network Sampling and Entrepreneurship Gestation," mimeo.
U.S. Small Business Administration (1993) The State of Small Business. Washington: Government Printing Office.
1 The authors thank the Wells Fargo Bank of San Francisco for its financial support of this research project. They also express appreciation to Paul Reynolds for his continuing commentary and interest.
2 Throughout the remainder of the paper, the term "formation" includes the term "dissolution" unless a specific differentiation is made.
3 A copy of the questionnaire is available from the authors upon request.
4 The data on Tables 2-5 are not purged of responses that lie outside the recall period.
2 Throughout the remainder of the paper, the term "formation" includes the term "dissolution" unless a specific differentiation is made.
3 A copy of the questionnaire is available from the authors upon request.
4 The data on Tables 2-5 are not purged of responses that lie outside the recall period.
|Return to Babson College
Main Home Page
©1996 Babson College. All rights reserved.
Last updated November 22, 1996 by Cheryl Ann Lopez