WHO STARTS NEW FIRMS?
LINEAR ADDITIVE VERSUS INTERACTION BASED MODELS
Paul D. Reynolds; Marquette University
The most popular model for attempting to account for individual's involvement in the entrepreneurial process has been a linear additive model. The results have not been very successful and the conception is at odds with the most salient characteristics of most entrepreneur's start-up stories, which emphasize the unique combination of events that lead an individual to start a business. Automatic Interaction Detection (AID) procedures provide a mechanism for exploring the extent to which unique combinations of events may underlay a major event in complex data set. The same data set is analyzed in two ways to illustrate the benefits of this alternative procedure.
Every year millions of new firms are started in market economies around the world. There is a strong suspicion that people that start new firms are unique or in distinctive situations. The efforts, however, to understand how or which individuals decide to become involved in a new firm start-up have not been very successful--or satisfying. There is, for example, very little overlap between the "entrepreneurial stories" in the media and the systematic efforts to explore factors leading to entrepreneurial behavior (e.g. self-employment) in large scale random samples.
A recent comprehensive overview of research on new and small firms (Storey, 1994) provides a summary of four efforts to analyze representative samples to determine the nature of individuals involved in self-employment (a surrogate for entrepreneurship used by Evans and Leighton, 1989; Dolton and Makepeace, 1990; Blanchflower and Oswald, 1990; and Blanchflower and Meyer, 1991). All four studies were based on representative samples in the thousands (from the U.S., U.K., and Australia) and included data related to a range of personal and background characteristics affecting the "probability of entry into self-employment." The resulting multi-variate analyses produced different variation of the linear additive models developed from multiple regression analysis.
A variety of operational definitions and analysis procedures were utilized in the different studies, the reviewer summarized the results by indicating only when a variable included in a linear model was statistically significant, as shown in Table 1. The good news is the absence of conflicting results. That is, there are no cases where one study finds a significant positive effect and the other a significant negative effect. Despite the diverse operational definitions and procedures, there is some consistency in the results.
There are, however, two other problems that are not addressed. One is the extent to which the resulting models help to explain the phenomena under consideration. The concept of "explain" can be considered to have two interpretations. First, in terms of a formal model that can account for or control the variation in the observed phenomena. Frequently referred to as the amount of variance explained by a formal model. This information is rare in these analyses, in part because the concept of "variation" as a feature of a distribution is more appropriate when the variable is continuous, such as predisposition toward self-employment (an attitude) or income from self-employment. It is not so useful when the dependent variable is dichotomous, as whether or not a person is involved in entrepreneurship (or reports they are self-employed).
One analysis was completed on data from 150,275 employed white men 25-60 years old in the United States from 1968-1987 (Evans and Leighton, 1989, pg. 527). This Current Population Survey data includes information on the individuals employment status in the survey week and the previous year. A linear probability model was developed to estimate the probability of becoming self-employed (considered an indicator of entrepreneurship). Variables that were statistically significant in the equation (at least at the 0.05 level) included personal income, personal income squared, net family assets (liquidity), dropping out of college, finishing college, post graduate education, urban residence, married, and military veteran. Not statistically significant individually, but left in the model as adding to the overall significant, were age, age squared, and whether or not the individual dropped out of high school. The resulting model was statistically significant and explained 0.286% of the variance, this left 99.7% of the variance in reported self-employment unexplained by this linear model.
Confronted with this modest level of overall predictive success, it is reasonable to expect the analyst to focus on hypothesis testing--where there is no ambiguity over the relative impact of a independent variable in a model.
A second conception of understanding focuses on the extent to which the formal models appear to be consistent with specific examples of individuals that decide to focus on entrepreneurship (or self-employment). Linear additive models imply that a set of factors all encourage a shift toward self-employment and that when enough of these "forces" are added together, it will trigger a shift in the individual that will produce a decision to be "self-employed." So when experience, gender (maleness), family tradition of entrepreneurship, need for achievement, educational attainment, personal wealth, and the like are all found in the same individual the pressure to be entrepreneurial may be great enough to cause them to cross the threshold to pursue self-employment.
The resulting image is not very consistent with the stories told by most that pursue entrepreneurship and start a new venture. These discussions usually emphasize a unique combination of circumstances that lead a person to pursue a firm start-up. While every entrepreneur likes to think their own situation and reactions are unique, their may be some common features in these "conception stories."
But using a linear modeling strategy to find these unique intersections among personal and contextual factors involves identifying specific interactions among many variables, such as age and gender, or gender and educational attainment, educational attainment and regional growth, or what have you. But the number of potential interactions are enormous--approaching infinity. Even if all possible interactions could be identified, even very large samples may not allow adequate tests of all possible alternatives.
Another alternative, to be pursued in this discussion, is a procedure designed to identify potential interactions. This procedure, developed as an alternative to random selection of interactions for linear regression analysis, provides a systematic way to explore the major interactions among the independent variables in representative samples. The results focus--in the following analysis--on the unique situations that lead individuals to pursue, or not pursue, a new firm start-up. Using data from a representative samples of U.S. adults involved in the start-up process, the outcomes from using an Automatic Interaction Detection (AID) procedure are illustrated. The results are more successful--in terms of predictive power--than a linear regression modeling procedure and give a greater sense of the unique set of factors that lead individuals to pursue new firm start-ups. There is more confidence that a situation that would trigger a decision to start a new firm has been identified.
SOURCES OF DATA
The major stumbling block to a precise description of the entrepreneurial or firm start-up process has been locating nascent entrepreneurs. A procedure has been developed that starts with a random selection of households, followed by a random selection of an adult member of the household (Palit and Reynolds, 1993; Reynolds and White, 1993). In the course of a phone interview, this person is asked about their work activity. The questions allow a person to indicate multiple work activities, full-time work and part-time work, full-time work and business ownership, and so forth.
At an appropriate point in the interview, they are asked: "Are you, alone or with others, now trying to start a business?" If they answer yes, they are then asked (1) if they have "given serious thought to the new business" and (2) if a number of different behaviors associated with starting a new firm (such as sought a bank loan, filed for incorporation, leased equipment, hired employees, and the like) have been "initiated" or "completed." They are also asked the month and year all reported actions are initiated. Those that report two or more firm gestation behaviors are considered "nascent entrepreneurs." They are also asked if the new business is, to their knowledge, included in any of the standard lists of businesses: Dun and Bradstreet files, unemployment insurance files, social security files, or the federal tax return listings. This can be used to determine the percentage of the sample that are engaged in the gestation process and, in turn, the proportion of the population that may be considered nascent entrepreneurs.
The results of this procedure for a representative sample of U.S. adults (18 and older) is presented in Table 2. It shows the proportion of respondents who said they were starting a new business and reported initiating each gestation (or start-up) activity. The U.S. sample was developed from the October and November 1993 University of Michigan Survey of Consumers (Curtin, 1982) and 64 actually completed these questions. But after eliminating those reporting (a) less than two behaviors or (b) a positive monthly cash flow prior to 1993, only 40 remained. The items are presented in the order of "popularity" among the respondents, the order of presentation in the interview is indicated in the first column of numbers. About 95% of those reporting they are "now trying to start a new business" indicate two or more start-up behaviors, the average was 6.7 (range from 1-15). The types of businesses are typical for the U.S. as 74% of the start-ups are in construction, retail, or services (Reynolds, 1995), compared to 70% for existing U.S. businesses (U.S. Small Business Administration, 1994, Table 2.1).
It is possible to estimate the prevalence of adults (number/100 or percents) that are nascent entrepreneurs, trying to start new firms. Using various background characteristics helps to explore the personal attributes or contextual factors that encourage individuals to become a nascent entrepreneur and pursue the gestation of a new firm. These are presented in Table 3. It should be noted that low rates of prevalence, almost all are below ten in a hundred or 10%, and samples in the hundreds are associated with very wide standard errors of the mean.
Most important, the overall prevalence of nascent entrepreneurs for the United States is 3.9%, confirming an earlier survey of Wisconsin adults that found a prevalence of 4.3% (Reynolds and White, 1993). This is, for the U.S. in 1993, one-in-25 adults or about 7.2 million adults. As the average start-up involves 2.2 adults, this represents about 3.3 million firms-in-gestation (FIGs). The success of these start-up efforts and the time required to put the firms in place are the subject of another analysis (Reynolds, 1995).
Statistically significant differences associated with the prevalence of nascent entrepreneurs are related to gender (higher for men); age (highest for 25-34 year olds); educational attainment (higher for post high school); county tenure (length of residence in the county); region of the United States (higher levels in West and Northeast, lower levels in North Central and South); labor force status (highest for the self-employed); martial status (absent for those widowed); and kids in the household (highest for households with one or two kids). No statistically significant differences are found for ethnic status, household income, occupational status, consumer confidence index rating (Curtin, 1982), number of financial reserves (a count of the presence of IRA/Keogh retirement plan, company pension, money market funds, mutual funds, stocks, or certificates of deposit), or the number of adults in the household.
Sources of nascent entrepreneurs, a different question, can be explored by examining the proportion provided by each category associated with each of these variables. This is presented in the right column of Table 3. Some differences are striking. For example, even though the prevalence of men reporting they are trying to start a new firm is twice that of women (5.6% versus 2.7%), because there are more women in the adult population they are responsible for two in five (40%) of efforts to start a new firm. Perhaps more dramatic is the effect of age. While the prevalence of start-ups among those 25-34 years old is over twice the average (9.7%), they are responsible for seven in ten of all reported start-ups (71%). Similarly, those reporting they are self-employed have a high rate of participating in start-ups (12.1 %), but are responsible for one third (36%) of all reported start-ups. In fact, those with full-time, part-time, or self-employment account for about four-fifths (82%) of all start-ups. Starting a new firm is clearly a secondary activity for those already involved in the world of work. Those unemployed account for about one-in-twenty (6%) of the efforts to start a new firm.
MULTIVARIATE ANALYSIS: Logistic Regression Models
The potential for high levels of inter-correlation among these independent variables leads immediately to a multivariate analysis. Logistic regression models, a specialized version of discriminate analysis, have been developed for dichotomous dependent variables and are appropriate for this situation (Norusis, 1992, pp. 1-34). Using the complete sample the results of stepwise logistic regression procedures using SPSS PC V5.0 are presented in Table 4. Note that the results vary depending upon whether a forward procedure, new variables added to increase the fit and predictions of the model, or a backwards procedure, variable eliminated from a full variable model until there is a decrease in the fit and predictions, is utilized.
The results, a set of variables is selected as optimal for a linear additive model, are very similar to Table 1. The models are statistically significantly different from a random model assuming no systematic relationship . In fact, the ability to accurately predict the outcome appears impressive, as both models make correct predictions for 96% of the cases.
This predictive accuracy is somewhat misleading, however, since it is done by predicting that none of the individuals are nascent entrepreneurs--the model predicts that all 1,016 are NOT involved in starting a new firm. Misclassification of 4% that are involved in a new firm start-up results in accurate predictions 96% of the time. Hardly an impressive indication of the value of this modeling procedure for predicting who gets involved in a firm start-up.
An alternative procedure can be used to give this exploratory procedure a more severe test. This is to randomly divide the 96% of those that are not nascent entrepreneurs into groups of 40. These are then combined with the single group of 40 nascent entrepreneurs and the analysis procedure is completed for each smaller group, 24 distinct comparisons in this case. As the probability of successfully predicting if a person in each group is a nascent entrepreneur is 50%, this gives the procedure a more challenging set of predictions. When this is done, using the forward stepwise procedure, the average correct predictions are 75%, a 50% improvement over the 50% expected by chance. The variables chosen for the 24 linear models varies slightly: age has a negative impact in 21 of the 24; self-employment a positive impact in 20; not living in the north central region in 7; fewer kids in the household in 5; working in a domestic or service occupation in 5; presence of financial reserves in 4; shorter residence tenure in the county in 4; lower levels of consumer confidence in 3; living in the western region in 3; higher levels of household income in 3; male respondents in 3; and so forth. Two variables have inconsistent patterns, homemaker and never-married appear with both positive and negative signs, reflecting the diversity to be expected in these small randomly selected comparison groups.
In either case, the results are hard to interpret in terms of the circumstances that trigger the decision to work on a new firm start-up. It is clear that being younger and currently self-employed (running a business of one's own) seems to have a major impact. But many young people and self-employed are not starting businesses. Perhaps more important (as indicated in Table 3), two-thirds of the start-ups are reported by those that are NOT self-employed.
MULTIVARIATE ANALYSIS: Automatic Interaction Detection Analysis
It is quite clear that a number of factors are related to the decision to start a new firm. A more complete understanding will require attention to the possible interaction effects among a large number of independent variables: gender, age, educational attainment, household income, labor force status, residential tenure, occupational status, and the like as presented in Table 3. The number of possible interactions is substantial--in the hundreds--and the modest research literature on this first transition in the start-up process is not a very useful source of hypotheses.
This is exactly the situation that led to the development of the Automatic Interaction Detection (AID) technique. This procedure is designed for the analysis of representative samples to identify the major combinations among multiple independent variables that affect a single dependent variable (Sonquist and Morgan, 1964; Sonquist, 1970). The procedure involves two stages. First, initial analysis (such as found in Table 3) is used to create nominal or ordinal independent variables. A hierarchy of analysis is then completed. With the SPSS CHAID (CHi-square Automatic Interaction Detection; Magidson, 1992) at each stage the independent variable that provides the most statistically significant relationship to the dependent variable is used to divide the sample--one subgroup for each independent variable category. Each subgroup is then examined to determine which of the remaining independent variables will provide the most significance divisions among the further subgroups. This procedure is continued until no further statistically significant divisions of the sample are possible. Different independent variables may be involved in different paths of the analysis hierarchy. The procedure does not assume a linear impact or any particularly form of interaction. The second stage involves using a one-way analysis of variance utilizing the resultant sub-groups to determine the extent to which subgroup differences are statistically significant.
This procedure was carried out with this U.S. sample, and is presented in Exhibit 1. Age is clearly the dominate factor affecting decisions to start a new firm. The impact of age, however, is not monotonic but curvilinear, with the highest proportion, almost 10%, occurring among those 25-34 years old, with one third the rate among those 18-24 or 35-54 years old. Nascent entrepreneurs are virtually non-existent among those 55 and older. The impact of other factors varies with the age of the respondent, indicating the significance of these interactions. Among the youngest adults, 18-24 years old, the presence of other adults in the household is critical. Among the oldest respondents, 55 and up, the availability of financial reserves is critical. Among the mid-life adults, their occupational status has a major effect, with self-employment substantially increasing the probability of attempting to start a new firm. But for those young adults, 25-34 years old with full or part time work, educational attainment is critical, for almost one in eight that have completed high school are also attempting to start a new business. Among those in later middle years, 35-54 years old, and not self-employed, those new to a county are more likely to be starting a new firm, particularly if they are men.
While two of the variables with great emphasis in the univariate analysis (Table 3) or the linear models developed from logistic regression (Table 4) have a major impact in this search for interaction--age and self-employment--it is also important that a number of others do not. For example, gender does not appear in this model, partially because those 55 and over, who are not trying to start new businesses, are dominated by women. Two-thirds (67.4%) of those 55 and over are women; one-third (38.1%) of all women in the sample--and the adult population--are 55 and over. In all other age categories the genders are relatively evenly balanced.
The second phase of the analysis, comparing the 12 subgroups developed in the first stage, is presented in Table 5. When the groups are ranked on the basis of the prevalence of nascent entrepreneurs and a one-way analysis of variance is completed, the results are statistically significant beyond the 0.0000 level and about 12% of the variance is accounted for. This is substantially more than the analysis of self-employment with the Current Population Survey data from 150,000 respondents, described above, that left 99.7% of variance unexplained.
The ranking presented in Table 5 demonstrates the extent to which efforts to start new firms are concentrated in unique groups in the adult population. Seven in ten of new firm start-ups (69%) are provided by one-sixth (17%) of the adult population; those 25-34 that are self-employed, unemployed, or students (group A) or with employment and more than a high school degree (group B). If those 35-54 reporting self-employment (group C) are added, five-sixths (83%) of start-ups are provided by less than one-quarter (23%) of the adult population. Equally dramatic is identifying that 50% of the adult population that fail to generate any efforts to start a new firm: those 55 and over with no financial reserves; those 35-54 not self-employed and with over 10 years in a county; 25-34 year olds that are homemakers or disabled; and those 18-24 living in households with one or two adults.
The unique contribution of the three top groups (A,B, & C) to the creation of new firms justifies more careful attention to their distinctive features. In each group a minority are involved in a new firm startup (nascent entrepreneurs, NasecE) and the majority are not (Other). They are compared on the basis of the major independent variables in Table 6.
Most of the differences in Table 6, due to small sample sizes, are not statistically significant. The first three variables merely confirm what the CHAID analysis provided: Groups A and B are restricted to those 25-34 years old, group C to those 35-54 years old; Group B is composed of those reporting full- or part-time employment, Groups A and C those that report self-employment but Group A also includes those unemployed or students; those in Group B reported educational attainment beyond high school.
But these three groups of nascent entrepreneurs appear to be somewhat different. The sixteen in the A group, those 25-34 in the labor force but not working for others, are 40% Black, confident about their economic future, less likely to have lived their whole life in the same county, have lower household income and financial reserves, 70% are not married, 70% have no kids in the household, and half live with no other adults. They would appear to be optimistic, independent, and with a career emphasis on managing and starting new firms--a commitment to the entrepreneurial life.
In contrast, the thirteen in the B group, who are the same age but with full-time or part-time jobs and education beyond high school are more likely to be men, have high or low confidence in their economic future, higher levels of household income and more financial reserves, more likely to have white collar work, be married or living with another adult but without children in the household. They would appear to be pursing a new firm either as a sideline to their existing work or as a hedge against potential problems with their day job--entrepreneurship as a hobby or a hedge.
The six in the C group, composed of those 35-54 that are self-employed, seem more diverse, for they may be pursing a new firm as they find a new opportunity or as a hedge against problems with their existing business activity. All white and mostly men, they are long term residents of their counties with diversity in both their financial situations and confidence in the economy. All have been married but now 40% live without another adult and usually without children. This group probably includes both those who, when they were younger, had a commitment to an entrepreneurial career as well as those for whom new firms were a hobby or hedge (Group A or B individuals at a later stage in their life).
While tentative, this overview seems to provide a more complete understanding of the factors affecting the decision to pursue a new firm start-up.
This analysis helps illuminate the significance of at least two variables given a great deal of attention in the economics literature related to new firm start-ups--unemployment and personal liquidity. The assumption that increases in unemployment leads to greater entrepreneurship (reported self-employment) is generally based on aggregated analyses, has focused on the relationship between the unemployment rate and rates of new firm births (Reynolds, Storey, and Westhead, 1994; Storey, 1991; Storey, 1994). While it is true that statistically significant relationships can be found between measures of unemployment and firm birth rates, and that unemployed (as an individual characteristics) is statistically significant in linear additive models associated with transitions into self-employment (see Table 1), it is clear that the majority of efforts to start new firms are initiated by those that are NOT unemployed. As discussed above in relation to Table 3, over 80% of those trying to start a new firm are engaged in full- or part-time work or are self-employed.
Adam Smith apparently suggested the only difference between the local shopkeeper and the "great merchant" was the availability of capital (Shorrocks, 1988, pg. 256); leading to the liquidity constraint hypotheses--that lack of financial resources will constrain entrepreneurship (Jovanovic, 1982). There seems to be little doubt that this is a statistically significant factor (Evans and Leighton, 1989; Evans and Jovanovic, 1989). One analysis has found that reporting of substantial inheritances on Internal Revenue Service estate returns increases the probability that subsequent personal returns of the beneficiaries will include, for the first time, a business expense schedule, indicating new business activity (Holtz-Eakin et al, 1994). Although these analyses have found statistically significant effects of increased liquidity on reported self-employment, the amount of explained variance has been modest. The conditions under which the availability of funding triggers self-employment has not been identified.
The analysis completed above finds that two indicators of financial well being (household income and the availability of financial resources, plays a minor role in the decision to initiate a new firm start-up. But the impact of financial reserves appears to have opposite effects on the group A and B entrepreneurs considered in Table 6. Those nascent entrepreneurs in Group A have more modest financial reserves in relationship to their comparison group while those in Group B have more significant financial reserves than their comparison group. Given this result, it is no wonder that the empirical support for the "liquidity effect" has been modest.
Overall, the use of the automatic interaction detection procedure has identified unique situations where participation in new firm start-ups in enhanced. The results have improved understanding in two ways. First, the variance accounted for by personal and contextual factors has increased substantially, from less than 1% to over 10%. Second, the combination of factors defining those groups where nascent entrepreneurs are most likely to ge found produces an interpretation that is more closely related to most descriptions of the start-up process.
One caution is in order. The AID procedure, designed to facilitate the "search for structure" has the same drawbacks as other similar techniques, stepwise multiple regression techniques or cluster analysis or factor analysis. The results may be idiosyncratic if the sample is idiosyncratic. To reduce the risk that a distinctive sample may provide a misleading interpretation, the procedure can be applied several times to different samples or to subsets of the same sample. This was not possible in this situation because of the small size of the sample of nascent entrepreneurs, 40 out of a representative sample of over 1,000.
This is, as should be clear, a preliminary analysis of a research program in development. Nonetheless, several conclusions seems justified.
First, it is technically feasible to study the pre-organization or start-up or entrepreneurial process in some detail. While there are some costs--each nascent entrepreneur in the random samples the U.S. adult population represents about $250 in direct costs--they are not so high as to preclude further research. These costs are, compared to the billions in public funds devoted to encouraging new firms and entrepreneurship, rather modest.
Second, participation in the entrepreneurial process is a major activity for those in the U.S. labor force. (Participation in other countries is less clear at this time.) If the 1 in 25 estimate is accurate, then over seven million U.S. adults are involved in the entrepreneurial process at any given time--this is greater than the annual number that get married or have children. Just as with marriage and childbearing, however, new firm start-ups are concentrated among young adults. Further, these individuals are not--as some have suggested--driven, amoral economic sociopaths wreaking havoc on society (Baumol, 1990) or "misfits cast off from wage work" (Evans and Leighton, 1989) but mainstream adults seeking new option in their work careers.
Third, preliminary evidence suggests that the processes leading to the initiation of a start-up reflect complex interactions among personal, life course, and contextual factors. They seem to fall into a small set of well defined groups. Useful understanding or explanations is unlikely to be provided by linear, additive models.
Finally, public policies oriented toward enhancing the number of new firms or encouraging entrepreneurial activity may be the most effective if directed toward a small number of well defined groups, primarily those 25-34 years old. Further research, including follow-up contact with the firms in gestation, will be required to determine the success of these start-ups.
Baumol, William J. 1990. "Entrepreneurship: Productive, Unproductive, and Destructive." Journal of Political Economy. 98(5, part 1):893-921.
Blanchflower, D. G. and D.B. Meyer. 1991. "Longitudinal Analysis of Young Entrepreneurs in australia and the United States." National Bureau of Economic Research. Cambridge, MA: Working Paper 3746.
Blanchflower, D. G. and A.J. Oswald. 1990. "Self Employment in the Enterprise Culture." In R. Jowell, S. Witherspoon and L. Brook (Eds). British Social Attitudes: The Seventh Report. SCPR, Gower, Aldershot.
Curtin, Richard. 1982. Indicators of consumer behavior: The University of Michigan Surveys of Consumers. Public Opinion Quarterly 46:340-362.
Dolton, P.J. and G. H. Makepeace. 1990. "Self-employment amongst graduates." Bulletin of Economic Research. 42(1):35-53.
Evans, David S. and B. Jovanovic. 1989. "An estimated model of entrepreneurial choice under liquidity constraints." Journal of Political Economy 94(4):808-827.
Evans, David S. and Linda S. Leighton. 1989. "Some Empirical Aspects of Entrepreneurship." American Economic Review 79(3):519-535.
Holtz-Eakin, Douglas, David Joulfaian, and Harvery S. Rosen. 1994. "Sticking it Out: Entrepreneurial Survival and Liquidity Constraints." Journal of Political Economy 102(1):53-76. Jovanovic, Boyan. 1982. "Selection and the Evolution of Industry." Econometrica 50(3):649-670.
Magidson, Jay. 1992. SPSS/PC+ CHAID: Version 5.0. Chicago, IL: SPSS Inc.
Norusis, Marija J. 1992. SPSS/PC+ Advanced Statistics: Version 5.0. Chicago, IL: SPSS Inc.
Palit, Charles and Paul Reynolds. 1993. A network sampling procedure for estimating the prevalence of nascent entrepreneurs. Proceedings of the International Conference on Establishment Surveys. Alexandria, VA: American Statistical Association, Pp. 657-661.
Reynolds, Paul D. 1995. The National Study of U.S. Business Start-Ups: Background and Progress Report. Mannheim, Germany; University of Mannheim, Conference on Dynamics of Employment and Industry Evolution. 19-21 January.
Reynolds, Paul, David J. Storey, and Paul Westhead. 1994. "Cross-national Comparisons of the Variation in New Firm Formation Rates." Regional Studies 28(4):443-456.
Reynolds, Paul and Sammis White. 1993. Wisconsin's Entrepreneurial Climate Study. Milwaukee, WI: Marquette University Center for the Study of Entrepreneurship. Final Report to Wisconsin Housing and Economic Development Authority.
Shorrocks, Anthony. 1988. "Wealth Holdings and Entrepreneurial Activity." in Kessler D. and A. Masson (Eds) Modeling the Accumulation and Distribution of Wealth. Oxford, UK: Clarendon, Press, pp. 241-258.
Sonquist, John A. 1970. Multivariate Model Building: The Validation of a Search Strategy. Ann Arbor, MI: University of Michigan Institute for Social Research.
Sonquist, John A. and James N. Morgan. 1964. The Detection of Interaction Effects. Ann Arbor, MI: University of Michigan Institute for Social Research Monograph No. 35.
Storey, David J. 1991. "The Birth of New Firms--Does Unemployment Matter?" Small Business Economics 3:167-78.
Storey, David J. 1994. Understanding the Small Business Sector. London, UK: Routledge.
U.S. Small Business Administration, Office of Advocacy. 1994. Handbook of Small Business Data. Washington, DC: U.S. Government Printing Office.
|Return to Babson College
Main Home Page
(c)1996 Babson College. All rights reserved.
Last updated November 25, 1996 by Frank Lafleur