Stem: a path to self-employment & jobs?

Benedict, Mary Ellen ; McClough, David ; Hoag, John 等

INTRODUCTION

The Small Business Administration (2009) reported that small business creation powers the US economy. However, Shane (2009) warns that promotion of any and all new ventures is misguided, arguing instead in favor of a more efficient policy that focuses on promoting businesses with growth potential. Further, others argue that a policy to promote all new businesses results in sizeable deadweight loss because generally, starts ups tend to create few jobs (Henley, 2005). It therefore appears that the challenge for policymakers is to support economic development while minimizing the deadweight loss associated with creating new firms. In recent years this challenge has resulted in an education policy targeting disciplines believed to further economic growth.

Our research is motivated by the SBA claim that small business is an engine of economic growth and evidence suggesting that economic innovation occurs in highly technical industries, such as computer software and biotechnology (Audretsch, Keilbach, and Lehmann 2006). In response to these claims and evidence, governments at the federal and state levels have initiated programs to promote physical science, technology, engineering, and mathematics (STEM) education. The macroeconomic benefits from promotion of STEM education are expected in the formation of new firms requiring expertise in STEM disciplines. As noted by Shane (2008), not all firms are created equal. Some firms have greater growth potential and hence create more employment opportunities for society. Accordingly, to examine policy promoting STEM, we ask two questions:

How does the rate of self-employment of STEM graduates compare to non-STEM graduates?

How does firm size associated with self-employed STEM graduates compare to firm size associated with self-employed non-STEM graduates?

Using data from the 2003 National Survey of College Graduates (NSCG), we examine these questions. The NSCG provides data for over 100,000 individuals who have attained at least a bachelor's degree, as well as detailed industry information regarding respondent employment history. In addition the NSCG includes detailed information on the academic field of study and the job status, paid or self-employed, for each respondent. Using the NSCG, we test whether academic study in STEM disciplines promotes self-employment and job creation and therefore supports the educational policy of targeted funding to these disciplines.

PAST RESEARCH ON STEM, EDUCATION, AND SELF-EMPLOYMENT

Why STEM?

In recent decades the skills needed for employment have continued to become more closely related to technology requiring a stronger background in science and engineering. Stine (2009), using US Census Bureau data, reports that the number of workers in science and engineering occupations grew by nearly 800% between 1950 and 2000, which exceeds the 230% growth rate of the labor force and the 490% growth rate of all managers and professionals. The STEM growth rate was more than three times that of the labor force during the 1990s. In addition, evidence of increased demand in STEM occupations is revealed in the lack of unemployment faced by STEM graduates. For example, unemployment in science and engineering occupations was 1.6% in 2006, a year during which the overall unemployment fluctuated between 4.4% and 4.7%. Wadhwa, Rissing, Saxenian, and Gereffi (2007) find that among technology firms founded by immigrants during the period 1995 to 2005, seventy-five percent of the highest attained degrees reflected STEM fields. These findings suggest that efforts, through education, to support STEM should pay off in more jobs.

Advocates for public policy emphasizing STEM education cite data from the Organization for Economic Co-operation and Development (OECD) that shows US fifteen-year-olds doing relatively poorly compared other OECD countries (OECD, 2006). In 2003, US students ranked below the OECD averages in math and science, ahead of only Portugal, Italy, Greece, Turkey, and Mexico on the math literacy scale and Slovakia, Spain, Norway, Luxembourg, Italy, Portugal, Greece, Turkey, and Mexico on the science scale. The US advantage over these countries was not statistically significant, suggesting that the situation could be worse than the ordinal ranking implies. Similarly, advocates cite statistics indicating that the US graduates fewer engineers than India or China. Wadhwa (2005) suggests that these data are incorrect. His research shows that the United States continues to lead India and China, although he concedes that elementary and secondary science and math education can be improved. Regardless of this discrepancy, the STEM disciplines are considered major sources of increased competitiveness due to the potential for technological innovation and job creation. Public policy promoting STEM disciplines is a response to the perception that the United States needs more STEM graduates to be competitive and maintain job growth in the years ahead.

Various organizations seek to direct attention to American competitiveness, more specifically the ability to innovate. In 2003, the National Science Board (NSB) examined national policy for science and engineering. The metric employed by the NSB compares the ratio of natural science and engineering degrees of the 24-year old population by country. The final report indicated that reduced student interest in STEM-related disciplines threatens economic growth and national security. The concern is motivated by expected increases in retirements combined with forecasted growth of related occupations. The NSB reports forecasted growth in science and technology occupations at three times the rate of all other disciplines. It must be noted that the NSB concern reflects a relative rather than absolute decline. Since the mid-1970s, this ratio has increased in the United States but the ratio has increased more among other developed countries. In 1975 the United States was among the leaders with Finland and Japan but by 1999 had tumbled down the list. Despite its own 50% increase, other countries, including Finland and Japan, doubled and tripled their ratio (National Science Board, 2003). Concern over decline in the future competitiveness of the United States motivated a response from affected special interest groups. These groups can be categorized under two labels: firms facing foreign competition and organizations likely to benefit from increased spending to develop STEM graduates.

One such group, Tapping America's Potential (TAP), is a consortium of business organizations formed in 2005 to promote public investment in basic research, public sponsorship of advanced research, and expansion of the pool of science and engineering graduates. The stated goal of TAP is to double the annual number of science and engineering graduates from 200,000 to 400,000 by 2015. A second consortium of business organizations and well-known technology companies sponsored a task force in 2005 to explore US competitiveness. The resulting report expressed concern that the United States awards fewer science and engineering degrees than either Asia and Europe, foreign students are increasingly remaining at home to pursue study in science and engineering disciplines, and the United States has a smaller share of the worldwide total of science and engineering doctoral degrees (the Task Force on the Future of American Innovation, 2005). Concern is justified by reported data that show Europe publishes more science and engineering academic articles than the United States and the US lead in patent applications is shrinking. The task force makes no recommendations but the implication that the United States must increase the number of students graduating with degrees in science and engineering to remain competitive is clear.

Not everyone agrees, however, that science and engineering specialists alone are the solution to the perceived decline in US competitiveness. Consistent with the NSB and TAP, Liberal Education & America's Promise (LEAP) identifies knowledge of the physical and natural world, quantitative literacy, and information literacy as essential learning outcomes (National Leadership Council, 2007). What makes LEAP unique is that the learning outcomes are broadly defined in the context of liberal education. The position of the National Leadership Council is that a liberal education enhances opportunity and recognition through critical and creative thinking combined with effective communication. Newsweek editor, Jon Meacham (2010), makes a similar point that the importance of a liberal education may be greater today than in the past, arguing that the low hanging fruit of innovation has been picked; therefore, ongoing innovation may well depend upon the ability to make connections that might otherwise remain unconnected. His point is that graduates with a (classical) liberal education are capable of making connections that might be missed by those lacking the broad exposure of a liberal education. Meacham and the National Leadership Council do not argue against the need for science and engineering competency; rather, they make the point that science and engineering alone are not likely sufficient.

The America Competes Act (ACT) of 2007 emphasizes the need for more investment in research and education (Stine, 2009). Key provisions include: increased government expenditure on basic research, government sponsorship of advanced research, and government investment in promotion of STEM graduates in order to increase the pool of qualified workers. The underlying assumption in the ACT is that graduates of STEM disciplines disproportionately contribute to innovation and prosperity.

Is Education Associated with Self-Employment?

Many studies have found a positive correlation between human capital variables and self-employment (Aaronson, 1991; Parker, 2004). Generally, additional years of schooling increase the probability of self-employment. Work experience and previous attempts at self-employment are likewise positively associated with self-employment (Wadhwa, Aggarwal, Holly & Salkever, 2009).

In recent years, empirical research has explored the influence of education, specifically educational attainment, in greater detail. Parker and van Praag (2005) treat education as an endogenous variable in a model of self-employment and find that additional years of schooling are associated with lower capital constraints, which in turn increases the probability of transition to self-employment. Koellinger (2008) identifies educational attainment as the key factor explaining variation in the level of innovation among entrepreneurs (Although the focus of our paper is on the educated self-employed, we include a brief overview of schooling and entrepreneurship because we later investigate firm size and type of major.). The empirical analysis distinguishes between entrepreneurs who innovate and those who imitate a successful commercial firm. For Koellinger, higher levels of education establish knowledge of the current state of science and technology and facilitate opportunity recognition. Baumol, Schilling and Wolff (2009) explore the popular notion of inventive entrepreneurs with minimal schooling. Examination of biographies fails to support this perception. Rather, the authors find that as the complexity of technology increases, the educational attainment of inventors advances faster than that of entrepreneurs. This finding seems consistent with the findings of Koellinger (2008) yet potentially challenging to the purposes of the National Leadership Council, if the emphasis is placed on invention. However, the National Leadership Council can argue that the value to society is not the invention but rather the recognition of the application and development of the resulting commercial opportunity.

Not only is education positively associated with self-employment rates, a positive association exists between education and job creation. Studies utilizing aggregate data from the UK find that education is positively associated with job creation by the self-employed (Cowling, Taylor & Mitchell, 2004). Burke, Fitzroy, and Nolan (2009) confirm this finding among men but find less compelling evidence for self-employed women. Using British micro-level panel data covering the years 1991 to 1999, Henley (2005) examines the determinants of job creation by the self-employed and finds that educational attainment among the self-employed workers is positively associated with creating employment in the United Kingdom.

Despite extensive research examining the relationship between education and self-employment, only limited research has examined the relationship between academic major and self-employment. Dolton, Makepeace, and Inchley (1990) reveal that higher self-employment rates are associated with particular academic fields of study. They analyze British survey statistics and find that self-employment rates are highest for recent graduates in law, architecture, and agriculture. Finnie, Laporte, and Rivard (2002) examine Canadian survey statistics and find that self-employment rates are highest among health graduates. The survey reveals high self-employment rates among graduates of the social sciences and humanities category, which dominates the population of graduates.

The principal contributions to the literature of the present study are, first, the use of U.S. data and, second, the emphasis on STEM graduates and the respective relationship with self-employment and firm size. We investigate whether graduates of a STEM discipline are more often self-employed than graduates of a non-STEM discipline. We then examine the relationship between self-employed and paid STEM graduates and firm size.

THE NSCG MAJORS AND SELF-EMPLOYMENT

The NSCG Dataset and Overview of Self-Employment Paths

The National Survey of College Graduates (NSCG) examines the educational and career characteristics of the United States college-level individuals (sestat.nsf.gov/sestat/sestat.html). The National Science Foundation conducts this and other surveys to form the SESTAT system (Scientists and Engineers Statistical Data System). The 2003 NSCG surveyed a random sample of individuals living in the United States, under the age of 76, who had received a bachelor's degree or higher prior to 2000. The public-use sample includes data on 100,042 individuals.

Our analysis includes working individuals in the NSCG who are 35 or older. Additional deletions due to missing data reduced the sample to 63,076, of which 10,919 (18.5%) are self-employed. (1) Fairlie (2004) estimates the self-employment rates ranging from 9.3% in 1979 to 9.8% in 2003. The higher than normal self-employment rate in the NSCG is likely due to the fact that the NSCG is limited only to individuals with at least a bachelor's degree and as noted in the previous section, education is positively correlated with self-employment. Further, Parker (2004) finds that older workers are more likely to be self-employed. Given that our sample is limited to educated respondents 35 years old or older to accommodate completion of formal schooling and determination of career paths, it seems reasonable that the self-employment rate of our sample would exceed the national average.

Consistent with the literature, the final dataset includes information on individual characteristics (age, gender, retirement status, and race) and family characteristics (marital status, number of small children, total number of children, and whether the spouse works). Several variables test the relationship between education and academic field of study on self-employment. The human capital effect is captured with several binary variables representing the type of highest degree (a bachelor's degree is the omitted category in the base regression) and the number of years since the individual obtained the highest degree, used as a proxy for experience. Our main question requires that we control for the respondent's bachelor degree field of study, which requires seventeen different categories of academic majors (a grouping of non-science majors is the omitted category). (2) Appendix 1 provides a detailed description of all variables used in the analysis.

Table 1 presents the descriptive statistics for the full sample of the variables used in the analysis separated by paid and self-employed workers. (3) Means tests reveal that self-employed respondents tend to be white, male, married, and more than two years older than paid workers. When we first define STEM as an individual with any STEM-related degree, we find that 43.7% of those who are in paid work and 42.9% of those who are self-employed have some sort of a STEM degree. (4) The difference is statistically significant at the 10% level of significance. The self-employed are more likely to complete a professional degrees (16 percent) compared to paid workers (2.7 percent).

STEM Probit Regressions

We next estimate the probability of being self-employed having controlled for factors associated with self-employment, including academic field of study. The dependent variable, Self-Employment, equals 1 if the individual was self-employed in 2003 and 0 if the individual was in paid work. Table 2 presents the results of a series of probit regressions. The first probit regression controls for any STEM-related academic fields of study; the second probit regression controls for advanced degrees in STEM disciplines; the final two probit regressions control for academic field at the bachelor's and master's levels (Column 3) and the bachelor's and professional levels (Column 4). A likelihood-ratio test indicates an overall good fit to each of the models. (5)

Due to the nature of probability models, the coefficients in Table 2 do not represent the marginal effects of individual variables on the dependent variable. Table 3 presents the marginal effects of the variables in the first two columns. We use the coefficients from Column (1) of Table 3 for most estimates, and Column (2) for the advanced STEM degrees. For the base case, the average probability of being self-employed, 20.3 percent, was calculated using the means of all continuous variables and at zero for all binary variables. The base case is a white male with the average number of children, age, and years since obtaining the highest degree, holding a bachelor's degree in nonscience fields, and the spouse does not work. For continuous variables, marginal effects were calculated by adding one unit to the mean to estimate a new probability, then calculating the difference in average probabilities from the base case. Binary variables were "turned on" and the difference in average probabilities was calculated from the base case. Marginal effects for the personal descriptive and family variables are available from the authors on request.

The variables controlling for personal characteristics and education are in the direction expected. We find that self-employment is positively associated with age, marriage, having younger children and a working spouse. In addition whites and males are more likely to be self-employed. We focus on the human capital-related effects. First, the marginal effect of years since receiving one's highest degree (a proxy for experience) is 0.004 (Table 3, Row 12). This effect suggests that it would take about two and one-half years of experience to increase the probability of being self-employed by one percentage point. Second, the results reveal that advanced degrees generally reduce the probability of self-employment. Compared to respondents completing only a bachelor's degree, completion of a Masters degree reduces the likelihood of self-employment by 4.7 percentage points while completion of a doctoral degree reduces the likelihood of self-employment by 7.4 percentage points. Only those with a professional degree have an average probability of being self-employed (27 percentage points) that exceeds individuals with only a bachelor's degree. A STEM degree lowers the average probability of being self-employed by 1.9 percentage points. However, when controlling for the level of the STEM degree, a professional STEM degree increases the average probability of being self-employed by an additional 5 percentage points. The STEM Masters degree also has a positive effect, although the effect is negligible in this model (0.2 percentage points).

Since it appears that professional and Masters level degrees have some impact on the path to self-employment, we control for academic field of study in the next two probit regressions. Column (3) of Table 2 includes all academic majors with a Masters degree; Column (4) includes only those individuals with professional degrees. We find that a number of STEM-related Masters level degrees have a positive association to self-employment compared with the nonscience majors (the omitted category in this regression). In addition, the professional health degree is positively associated with being self-employed. Note also that the addition of specific degrees does not substantially change the coefficient on the base case variables, with the exception of older children, which is now positive and statistically significant in the professional degree model.

The STEM Path to Self-employment

Rather than provide a very large table for each of the marginal effects for the many variables in these regressions, we present flow charts for selected STEM majors to demonstrate how the choice of major and degree leads to self-employment. Self-employment probabilities are estimated with the coefficients from the regressions in Columns (3) and (4) from Table 2. The first probability in the flow charts is the percent of those in the sample with a particular major (e.g., 5.2% of the sample attained a bachelor's degree in biology). The second probability is conditional on the individual having the major. The self-employment probability is estimated at the means of the variables in the regressions, and zeros for all binary variables, except those variables of interest (e.g., the 31.8% probability of being self-employed in the first flow chart controls for having a bachelor's degree in biology, a professional degree in health, at the means of age, years since highest degree, number of children, white males without a working spouse and not previously retired.)

There are 1,870 individuals (17.3% of the sample) who attained a professional degree. Three STEM-related baccalaureate degrees (biology, chemistry, and health) are associated with a relatively high attainment of a professional health degree and high probabilities of being self-employed. Figure 1 demonstrates that even though the probability of being an undergraduate major in biology, chemistry, or health is low, many of these individuals who pursue a professional health degree are more likely to be self-employed than those who stop at the baccalaureate level. This result is not surprising since doctors, dentists, chiropractors, and optometrists, for example, often set up or join a practice.

Further, those who attain a baccalaureate then a Masters degree in health have a 20.4% probability of being self-employed. Clearly, there are individuals who initiate their path to self-employment during college. Also, if an individual receives only a baccalaureate degree in STEM fields, the probability of being self-employed is relatively low, with the exception of a health baccalaureate degree, which leads to a self-employment probability of 22.2%. It appears that the health degree path to self-employment is not an easy one, but yields high rates of self-employment for those who continue on the path.

[FIGURE 1 OMITTED]

As Figure 2 demonstrates, technology majors, who do not have a high probability of self-employment with only a baccalaureate degree, (6) have a high probability of being self-employed if they attain a Masters degree in a technology-related field (34.4%). Technology majors who go on for a Masters in computer science have an average probability of being self-employed (19.2%). Four baccalaureate majors are more likely to attain a Masters degree in computer science compared to technology or other undergraduate degrees: computer science, mathematics, engineers, and management. None of these paths yield a higher probability of being self-employed than that of the individual with a baccalaureate degree in a technology field.

[FIGURE 2 OMITTED]

In Figure 3, we find that for engineering baccalaureates, none of the paths yield self-employment probabilities greater than the sample mean of 19%. It is possible that engineers tend toward paid work because their human capital is highly rewarded within the typical company structure.

[FIGURE 3 OMITTED]

These above STEM paths are the more likely ones that end in self-employment. Others, such as those where individuals who specialize in biology, chemistry, or the physical sciences, have low self-employment rates and are not reported here. In addition, some non-science majors have high rates of self-employment. For example, psychologists develop practices alone and with others, much like those in the health field. Individuals who attain only a bachelor's degree in psychology have a relatively high rate of self-employment (21.8%), while professional psychologists have a 31.4% rate of self-employment (These results are available from the authors). These self-employment rates are higher than the rates for many in the physical sciences or technology.

STEM graduates and firm size

The previous section suggests that STEM majors, except for those in health fields, are not generally self-employed compared to non-STEM majors. A subsequent question we ask is, are STEM majors associated with small firms typically considered the engine of growth for society? As noted in the introduction, the statement "small business is an engine of economic growth" is a generally accepted principle, often leading to governmental support for small firms. The Bureau of Labor Statistics reports that firms with less than 100 employees accounted for 46% of the employment growth from 1992 through 2005; when adding firms with less than 500 employees to the total, the percentage increased to 65% (Helfand, Sadeghi, & Talan, 2007). We therefore examine whether the STEM degree is associated with small firms.

Ideally, detailed employment data over time would enable one to examine job growth rates. However, the 2003 NSCG dataset offers only categories of firm size by number of employees. Therefore, we examine the relationship between STEM education and firm size, defined as large or small. (7) Because we are also interested in self-employment versus paid employment, we can examine whether a STEM degree is associated with four employment situations:

Let j=0 when an individual is self-employed with a small firm; j=1 when an individual is a paid worker in a small firm; j=2 when an individual is self-employed with a large firm; j=3 when an individual is a paid worker in a large firm. (8)

For our purposes, a multinomial logit analysis is employed to investigate how a STEM degree is associated with the probability of being in one of the four employment situations. The multinomial logit model assumes that the dependent variable represents several categories that have no special order but the categories are associated with a set of individual characteristics. The model is defined as:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]

where [[pi].sub.ij] represents the probability that individual i is situated in employment j. When j=m, the model assumes a referent case, which in our model is the probability of being self-employed with a small firm. Thus, the estimation process results in three different logit regressions, where the probability of being in employment situation j is compared to the probability of being self-employed with a small firm.

The effect of each independent variable is best examined with an odds ratio estimate. The odds ratio examines the odds of one event occurring in relation to another, given a particular regressor. For example, if the odds of being self-employed with a large firm equaled the odds of being self-employed with a small firm for the STEM degree, the odds ratio would be 1. If the odds ratio is greater than 1, then STEM is more highly associated with the categories represented by the numerator in Eq.(1). The [x'.sub.i] represents the individual characteristics that are associated with the employment situation and the [[beta].sub.j]s are parameters to be estimated. We include whether or not an individual has a STEM degree, the age of the individual, and a series of industry binary variables, based on classifications from the NSCG. The omitted industry is public administration.

Table 4 presents the multinomial logit results. The likelihood ratio test indicates that the model fits the data well (We report the statistical significance of the estimated coefficients, but given the large sample size, the high level of statistical significance is to be expected.). The important relationship is that of STEM to the probability of each employment situation. The results indicate that an individual with a STEM degree is 1.56 times more likely to be in paid work in a large firm and 1.55 times more likely to be self-employed with a large firm compared to self-employment with a small firm. The odds ratio of 1.032 in Column (7) indicates that the STEM graduates have about the same odds of being either in a small firm as a paid worker or self-employed. These results indicate that individuals with STEM degrees are not associated with the type of firms normally considered growth engines for society.

The AGE coefficient indicates that growing older is associated with self-employment and small firms. There has been recent research on how age and self-employment are positively related, especially after retirement (Fairlie & Kapur, 2009), thus the result is expected. The odds of being self-employed with a large firm compared to being self-employed with a small firm are greater for those working in mining and gas extraction, manufacturing, transportation, and information services (compared to the referent industry, public administration). The odds ratio estimates are generally less than 1 for all other industries, suggesting that self-employment with a small firm has higher odds of occurring compared to either the odds of paid work in a large firm or small firm for those industries.

A considerable amount of theory and empirical research is dedicated to the role of firm size in explaining variation in wages. Much of this literature seems to offer insight into our findings related to firm size. Barth, Cordes, & Haber (1987) examine firm size and employee characteristics. They assume that monitoring costs increase with higher number of employees thereby assigning a comparative advantage in monitoring to smaller firms. Accordingly, large firms are to inclined hire more productive workers requiring less monitoring. Education, experience, and age are identified as observable variables associated with low monitoring costs. Our findings show that highly educated STEM workers are associated with larger firms as are older workers.

More interestingly for our purposes, an expansive literature explains the observed wage premium that accrues to skilled labor. Griliches (1969) popularized the notion that skilled labor and capital are complementary, which contributes to the greater productivity of skilled labor compared to unskilled labor. Idson and Feaster (1990) note that it is not surprising that small and large employers differ in terms of key variables such as education. Whereas firms are assumed to maximize profits, a hedonic model of choice accommodates worker heterogeneity. Idson and Feaster posit that independently-minded workers will trade income for the independence more likely to exist in a smaller firm. In contrast larger firms will attract and retain workers more comfortable in an interdependent production process. Although intuitively appealing, the choice is complicated by the observed complementarity between skilled labor (human capital) and physical capital. Savoye (1994) reports that larger firms are often characterized by greater production complexity and larger expenditures on research and development. Accordingly, highly educated STEM graduates may very well maximize utility by choosing paid employment with a large firm that combines the physical capital with the human capital of the STEM graduate. In short, the opportunity cost associated with self-employment may be too substantial in light of the paid employment opportunities.

CONCLUDING REMARKS

This paper examines whether STEM majors contribute to self-employment, and whether self-employed STEM majors are associated with larger firms compared to their non-STEM counterparts. The results of this study indicate that only select STEM-related paths lead to self-employment. Notably, individuals with professional health degrees have relatively high probabilities of being self-employed. In addition, individuals with a technology degree who move on to a Masters degree in computer science also have a relatively high probability of being self-employed. Other STEM majors are not typically on the self-employment path. The results suggest that public dollars allocated to the promotion of STEM graduates do not create a lot of new businesses.

We also find that those with a STEM degree are more likely to be employed in a large business rather than small, whether paid or self-employed. This result combined with the previous result suggest that it is not generally likely that added emphasis on STEM will lead to economic development at least through the generation of small firms leading to substantial innovation and growth.

STEM may promote growth via other mechanisms, such as innovations leading to growth through larger business. Perhaps those STEM majors in paid work have the luxury of being innovative without personal risk. Indeed, our data indicates that STEM majors in paid work, especially those with a Ph.D., have a higher estimated number of patents compared to self-employed non-STEM majors. (9)

Shane (2008) indicates that public policy needs to be selective to promote firms that contribute to the economy. Our analysis shows that self-employed STEM graduates tend to be associated with the largest firms. Accordingly, policy promoting STEM disciplines may not necessarily result in more new firms, but STEM graduates are associated with firms that create greater employment opportunities to the benefit of society overall.

This study is a beginning step to better understand whether the focus on STEM is warranted. Our findings suggest that STEM graduates are more likely to work as paid employees in large firms. What role do these employees play? Are they primarily in research areas where the growth of the firm is enhanced? Second, do these workers later become the owners of small firms where their large firm experience is an asset? The next question we need to address is the extent to which technology driven firms are engines of economic development whether small or large.

Appendix 1. Description of Variables
2003 National Survey of College Graduates

Descriptives

 Self-Employed 1 if self-employed (incorporated and
 not incorp.); 0 if paid employment.
 Age Age of the individual.
 Female 1 if female.
 Black 1 if Black.
 Asian 1 if Asian.
 Foreignborn 1 if born outside of the United States.
 Previous Retirement 1 if previously retired.
Family
 Married 1 if married.
 No. Children < 6 yrs. Number of children below the age of 6.
 No. Children >= 6 Number of children age 6 or greater.
 Spouse Works 1 if the spouse works.
Education
 Yrs. Since Highest Years since individual obtained their
 Degree highest degree.

 Highest Degree= 1 if highest degree is a Masters
 Masters degree.

 Highest Degree= 1 if highest degree is a Ph.D., DSc.
 Doctoral EDD. or other doctoral degree.

 Highest Degree= 1 if the highest degree is a J.D.,
 Professional M.D. DDS, or other professional
 degree.

Major Field of Study
 Computer Science Computer and Information Science,
 Computer Systems, any other Computer
 Science Major.

 Mathematics Applied Math, General Math, Operations
 Research, Statistics, Actuarial
 Science, any other Math Major.

 Agricultural Science Animal, Food, Plant or any other
 Agricultural Science Major.

 Biology Biochemistry, General Biology, Botany,
 Molecular Biology, Ecology, Genetics,
 Microbiology, Nutrition, Pharmacology,
 Physiology, Zoology, and other Biology
 Major.

 Environmental Science Environmental Science, Forestry, Other
 Conservation Majors.

 Physical Sciences Astronomy, Physics, Earth Science, and
 other Physical Sciences.

 Chemistry General Chemistry.

 Psychology Educational, Clinical, Counseling,
 Experimental, General, Industrial
 Organizational, Social, and all other
 Psychology Majors.

 Social Sciences Agricultural Economics, Economics,
 Public Policy, International
 Relations, Political Science, Public
 Administration, and Public Affairs,
 Anthropology, Criminology, and
 Sociology, Ethnic Studies, Religion,
 Theology, and Philosophy, Social Work,
 History, Linguistics, Philosophy of
 Science, Geography, History of
 Science, any other Social Science
 Major.

 Engineering Aerospace, Chemical, Architecture,
 Civil, Computer, Electrical,
 Industrial, Mechanical, Agricultural,
 Bio Engineering, Engineering Science,
 Environmental, General, Geophysical,
 Materials, Metallurgical, Mining,
 Naval, Nuclear, Petroleum, and all
 other Engineering Majors.

 Health Professions Audio/Speech, Health Services
 Administration, Health/Medical
 Assistant, Health Technology, Pre-Med,
 Dentists/Optometry, Nursing, Pharmacy,
 Psychological Therapy, Public Health,
 Other Health Majors.

 Technology Computer Programming, Data Processing,
 Professions Electrical Technology, Industrial
 Technology, Mechanical Technology,
 other Technology Majors.

 Management Accounting, Business Administration &
 Management, General Business,
 Managerial Economics, Financial
 Management, Agricultural Business.

 Sales Marketing, & Marketing Research.

 Social Services Social Work and related fields.

 Fine Arts & Drama, Fine Arts, Music, Other
 Humanities Performing Arts Majors, English,
 Liberal Studies, History, Foreign
 Language and other related majors.

 NonScience Any major not listed above.

Source: 2003 National Survey Of College Graduates,
Summary Documentation, PCG03.pdf.

REFERENCES

Aronson, R. L. (1991). Self-employment: a labor market perspective. Ithaca, NY: ILR Press.

Audretsch, D., M.Keilbach & E. Lehmann (2006). Entrepreneurship and economic growth. New York: Oxford University Press, USA.

Barth, J.R., J. Cordes, & S. Haber (1987). Employee characteristics and firm size: are there systematic empirical relationships? Applied Economics, 19, 555-567.

Baumol, W.J., M. Schilling & E.N. Wolff (2009). The superstar inventors and entrepreneurs: how were they educated? Journal of Economics & Management Strategy, 18, 711-728.

Burke, A.E., F.R. Fitzroy & M.A. Nolan (2009). Is there a north-south divide in self-employment in England? Regional Studies, 43, 529-544.

Cowling M., Taylor M. & Mitchell P. (2004). Job Creators. Manchester School, 72, 601-617.

Dolton, P.J., G.H. Makepeace & G.D. Inchley (1990). The early careers of 1980 graduates: earnings, earnings differentials and postgraduate study, research paper no. 78, London: UK Department of Employment.

Fairlie, R.W. (2004). Self-employed business ownership rates in the United States: 1979-2003. Small Business Research Summary No. 243. Washington, DC: Small Business Administration Office of Advocacy.

Fairlie, R. W. & K. Kapur (2009). Is Employer-Based Health Insurance a Barrier to Entrepreneurship? Working paper, UCD Centre for Economic Research, University College, Dublin.

Finnie, R., C. Laporte & M.R.Rivard (2002). Setting up shop: self-employment amongst Canadian college and university graduates. Research paper n. 183, Ottawa, CA: Statistics Canada.

Griliches, Z. (1969) Capital-Skill Complementarity, Review of Economics and Statistics, 51, 465-68.

Helfand, J., A. Sadeghi & D Talan (2007). Employment dynamics: small and large firms over the business cycle. Monthly Labor Review Online, March, 39-50.

Henley, A. (2005). Job creation by the self-employed: the roles of entrepreneurial and financial capital. Small Business Economics, 25, 175-196.

Idson, T. & D. Feaster (1990). A Selectivity Model of Employer-Size Wage Differentials, Journal of Labor Economics, 8, 99-122.

Koellinger, P. (2008). Why are entrepreneurs more innovative than others? Small Business Economics, 31, 21-37. Meacham, J. (2010, January 18). In defense of the liberal arts. Newsweek, 54(3), 5.

National Leadership Council for Liberal Education and America's Promise (2007). College learning for the new global century. Retrieved May 8, 2010, from http://www.aacu.org/leap/ documents/GlobalCenturyfinal.pdf.

National Science Board. (2003). Science and engineering indicators. Retrieved May 8, 2010, from http://www.nsf.gov/ statistics/seind02/c2/fig02-27.htm.

National Science Foundation, National Survey of College Graduates 2003, Online at: https://sestat.nsf.gov/ sestat/sestat.html.

Organization for Economic Co-operation and Development. (2006). Program for student assessment, 2003. Parker, S. (2004). The economics of self-employment and entrepreneurship. Cambridge University Press. Parker, S. & M. van Praag (2005). Schooling, capital constraints and entrepreneurial performance. Tinbergen Institute Discussion Paper TI 2004-16/3.

Savoye, B. (1994), "La taille des entreprises, element structurant du systeme productif", Revue d'Economie Industriell, No. 67, 103-19.

Shane, S. (2008). The start-ups we don't need. The American, Washington, DC: American Enterprise Institute, 8893

Shane, S. (2009). Why encouraging more people to become entrepreneurs is bad public policy. Small Business Economics, 33, 141-149.

Small Business Administration (2009). The Small Business Economy. Washington, DC: Government Printing Office.

Stine, D. D. (2009). America COMPETES act: programs, funding, and selected issues. Washington, DC: Congressional Research Service.

Tapping America's Potential. (2008). Gaining momentum, losing ground: progress report 2008. Retrieved May 7, 2010, from http://tap2015.org/resource/ pdf/tap_2008_progress.pdf.

Task Force on the Future of American Innovation (2005). The Knowledge Economy: Is the United States Losing its Competitive Edge? Retrieved May 8, 2010 at http://www.futureofinnovation.org/PDF/Benchmarks.pdf.

Wahwa, V. (2005). About the engineering gap... Business Week. Retrieved Dec. 13, 2010 from http://www.businessweek.com/ smallbiz/content/dec2005/sb20051212_623922.htm.

Wadhwa, V., B. Rissing, A. Saxenian, & G. Gereffi. (2007). Education, entrepreneurship and immigration: America's new immigrant entrepreneurs, PartII. Kauffman Foundation.

Wadhwa, V., R. Aggarwal, K. Holly, & A. Salkever (2009). The anatomy of an entrepreneur: Family background and motivation, Kauffman Foundation.

Mary Ellen Benedict, Bowling Green State University

David McClough, Ohio Northern University

John Hoag, Bowling Green State University

ENDNOTES

(1.) It is important to note that the NSCG employs a sampling method that controls for stratification by groups and nonresponse bias. Thus, SESTAT includes a weighting factor that we use in this analysis. The weighting factor slightly changes the statistical results of the subsequent analysis, but by very little. For example, the unweighted percentage of self-employed is 17.3 percent; the weighted percentage is 18.5 percent. We employ the weights in most of our analyses, but use the unweighted frequency distributions in some tables to avoid confusion between counts and percentages; all results are available on request.

(2.) The NSCG offers thirty-one separate "minor" categories of academic disciplines. Some of these categories contained very small groupings and were therefore combined when necessary (e.g., Earth Science is part of Physical Sciences), while other groups seemed to be natural for agglomeration (e.g., Engineering subdisciplines). We kept as many separate categories for the STEM disciplines as possible.

(3.) Ideally, we would have liked to have added a third category, self-employed and paid. However, the NSCG only reports information on the primary job, which limits us to the dichotomous category of self or paid work.

(4.) The GAO reports that the percentage of degrees awarded in 2003-04 to STEM majors was 27% (GAO, 2006). Our higher rate occurs because the individual can have a STEM degree from the baccalaureate, Masters, Ph.D., or professional level, over a number of years, not at one particular point in time.

(5.) The log-likelihood value is used in a Chi-square test on whether the variables in the model jointly contribute to the explanation of the variance in the probability of self-employment.

(6.) The probability of an individual with a bachelor's degree in a technology field becoming self-employed is 13.6%. Estimation by the authors is available on request.

(7.) There is no common definition of small firm. In some instances, a small firm is defined by revenues and industry; in others it is defined by the number of employees. We opted to define small firms as 100 employees or less simply because the definition has been used by government agencies, such as the Bureau of Labor Statistics.

(8.) Due to data limitations, we cannot state that the individual started the firm, only that the individual currently owns a firm of a certain size. Thus, the estimation process examines only the employment situation in 2003 in order to see whether STEM degrees are associated with small or large firms, paid or self-employment.

(9.) A weighted multiple regression of patents against self-employment status, STEM, degree status interacted with STEM, and controls for baccalaureate degrees indicates that individuals with a Ph.D. in a STEM field and in paid work have an average of .44 more patents than self-employed nonSTEM individuals. This result is not inconsequential since the average number of patents for the sample is 2. Results of the regression are available on request.

Table 1: Means and Standard Deviations The Self-Employed
& Paid Work College Graduates

 Total Paid Self- T-stat
 Sample Workers Employed
 n n n

Personal Descriptives

Self-Employed 0.19
 (0.55)
Age 48.92 48.51 50.72 -25.70 ***
 (11.88) (11.43) (13.51)
Female 0.42 0.442 0.365 18,83 ***
 (0.70) (0.694) (0.694)
Black 0.07 0.080 0.033 18.00 ***
 (0.36) (0.379) (0.259)
Asian 0.08 0.084 0.083 0.48
 (0.39) (0.387) (0.401)
Other Races 0.02 0.025 0.021 2.31 **
 (0.22) (0.218) (0.211)
Foreign born 0.16 0.164 0.161 0.79
 (0.52) (0.517) (0.538)
Previously Retired 0.04 0.044 0.051 -6.73 ***
 (0.29) (0.043) (0.334)

Family

Married 0.79 0.787 0.820 -7.95 ***
 (0.57) (0.572) (0.560)
No. Children =- 6 YRS. 0.20 0.200 0.190 2.34 **
 (0.77) (0.764) (0.783)
No. Children >= 6 YRS. 0.73 0.735 0.687 4.62 ***
 (1.43) (1.422) (1.470)
Spouse Works 0.59 0.590 0.592 -0.41
 (0.69) (0.687) (0.716)

Degree

Yrs. Since Highest 21.04 20.34 24.13 -37.48 ***
 Degree (14.07) (13.665) (15.052)
Highest Degree= 0.30 0.316 0.209 22.91 ***
 Masters (0.64) (0.649) (0.593)
Highest Degree= 0.07 0.078 0.045 12.74 ***
 Doctoral (0.36) (0.375) (0.300)
Highest Degree= 0.07 0.044 0.160 -46.34 ***
 Professional (0.35) (0.287) (0.527)

STEM

Any STEM Degree 0.44 0.437 0.428 1.77 *
 (0.70) (0.693) (0.721)
Masters in STEM 0.33 0.326 0.340 -2,93 ***
 (0.66) (0.655) (0.691)
PH.D. in STEM 0.31 0.311 0.331 -4.03 ***
 (0.65) (0.647) (0.686)
Professional 0.31 0.311 0.338 -4.56 ***
 Degree in STEM (0.65) (0.642) (0.685)

Computer Science 0.03 0.024 0.011 5.97 ***
 (0.25) (0.222) (0.250)
Mathematics 0.04 0.041 0.032 4.25 ***
 (0.27) (0.275) (0.257)
Agricultural Science 0.01 0.095 0.015 -5.30 ***
 (0.14) (0.136) (0.178)
Biology Biology 0.06 0.062 0.078 -6.83 ***
 (0.35) (0.334) (0.393)
Environmental 0.01 0.006 0.003 1.52
 Science (0.10) (0.103) (0.096)
Physical Sciences 0.03 0.026 0.023 2.15 **
 (0.22) (0.224) (0.219)
Chemistry 0.03 0.030 0.033 -1.72 *
 (0.24) (0.238) (0.260)
Psychology 0.06 0.058 0.063 -2.15 **
 (0.33) (0.325) (0.354)
Social Sciences 0.11 0.109 0.120 -3.59 ***
 (0.44) (0.435) (0.474)
Engineering 0.08 0.085 0.079 2.09 **
 (39) (0.340) (0.394)
Health Professions 0.07 0.071 0.072 -0.70
 (0.36) (0.358) (0.378)
Technology 0.03 0.024 0.033 -5.86 ***
 Professions (0.22) (0.211) (0.260)
Management 0.1717 0.166 0.209 -10.91 ***
 (53) (0.520) (0.592)
Sales & Marketing 0.03 0.025 0.038 -7.38 ***
 (0.23) (0.220) (0.278)
Fine Arts & 0.14 0.139 0.166 -7,75 ***
 Humanities (0.49) (0.483) (0.543)
Other NonScience 0.07 0.069 0.066 1.18
 (0.36) (0.363) (0.363)

Data Source: National Science Foundation, The 2003 National
Survey of College Graduates, weighted for stratification
and nonresponse bias. Standard deviations are in
parentheses. T-statistics test the difference between the
means of paid and self-employed workers. A negative sign on
the t-test indicates that the average is larger for the
self-employed. ***=statistical significance at the 1 percent
level, ** = statistical significance at the 5 percent
level, and * = statistical significance at the 10 percent
level of significance.

Table 2: Probit Analysis of Self-Employment and College Graduates

 Post-Bac Post-Bac
 STEM ONLY STEM by Degree Degree
 N = 63,076 Deg. Level MA Professional
 N=63,076 N=19,795 N=4,183
 (1) (2) (3) (4)

 -1.452 *** -1.452 *** -2.232 *** -1.858 ***
Constant (0.035) (0.035) (0.076) (0.328)

Personal Descriptives

Age 0.006 *** 0.006 *** 0.010 *** 0.009 ***
 (0.001) (0.001) (0.002) (0.003)

Female -0.145 *** -0.148 *** -0.112 *** -0.141 ***
 (0.009) (0.009) (0.020) (0.034)

Black -0.454 *** -0.454 *** -0.388 *** -0.439 ***
 (0.020) (0.020) (0.042) (0.067)

Asian -0.026 -0.026 -0.041 -0.018
 (0.019) (0.019) (0.037) (0.063)

Other Races -0.076 *** -0.076 *** 0.052 -0.505 ***
 (0.029) (0.025) (0.056) (0.106)

Foreignborn 0.074 ** 0.074 *** 0151 *** -0.240 ***
 (0.014) (0.012) (0.028) (0.049)

Previously 0.082 *** 0.083 *** 0.130 *** -0.212 ***
Retired (0.020) (0.020) (0.035) (0.082)

Family

Married 0.013 0.013 -0.038 0.038
 (0.014) (0.014) (0.029) (0.049)

No. Children 0.060 *** 0.060 *** 0.100 *** 0.001
< 6 Yrs. (0.008) (0.008) (0.017) (0.027)

No. Children -0.001 -0.001 -0.029 *** 0.064 ***
>- 6 Yrs. (0.005) (0.004) (0.010) (0.015)

Spouse Works 0.037 *** 0.037 *** 0.046 ** 0.072 **
 (0.011) (0.010) (0.022) (0.035)

Human Capital

Yrs. Since 0.015 *** 0.015 *** 0.019 *** 0.010 ***
Highest (1) (0.001) (0.001) (0.003)
Degree

Highest -0.181 *** -0.179 ***
Degree (0.011) (0.0111)
= Masters

Highest -0.301 *** -0.294 ***
Degree= (0.020) (0.020)
Doctoral

Highest 0.763 *** 0.762 ***
Degree= (0.015) (0.015)
Professional

Major Field of Study

STEM -0.068 *** -0.084 ***
 (0.009) (0.015)

MASTEM 0.006
 (0.038)

PHDSTEM -0.146 *
 (0.084)

PROFSTEM 0.1661
 (0.091)

Highest Degree
 0.277 ***
Computer (0.050)

Math 0.117
 (0.143)

Agriculture 0.226
 (0.143)

Biology 0.119
 (0.073)

Environmental 0.269 *
Science (0.142)

Chemistry -0.015
 (0.130)

Physical 0.118
Sciences (0.089)

Psychology 0.503 *** 0.722 **
 (0.043) (0.320)

Social Science 0.312 ***
 (0.072)

Engineering 0.237 ***
 (0.044)

Technical 0.744 ***
 (0.062)

Health 0.531 *** 0.604 **
 (0.046) (0.300)

Management 0.366 ***
 (0.029)

Education -0.412
 (0.392)

Social 0.374 *** -0.552 *
Service (0.041) (0.344)

Sales & MKT 0.644 ***
 (0.059)

Arts & 0.344 ***
Humanities (0.039)

NonScience 0.471
 (0.298)

Log- -56270 *** -56267 *** -13408 *** -5368 ***
likelihood

Data Source: National Science Foundation, The 2003 National
Survey of College Graduates. Standard errors are in
parentheses. T-statistics test the difference between the
means of paid and self-employed workers. *** = statistical
significance at the 1 percent level, **=statistical
significance at the 5 percent level, and statistical
significance at the 10 percent level of significance.
Columns (3) and (4) also include controls for seventeen
detailed baccalaureate degrees.

Table 3. Marginal Effects of the Main Variables

Variable Marginal
 Effect

Age 0.002
Female -0.039
Black -0.104
Asian -0.007
Other Races -0.021
Foreign born 0.022
Previously Retired 0.024
Married 0.004
No. Children < 6 Yrs. 0.017
No. Children >= 6 Yrs. 0.000
Spouse Works 0.011
Yrs. Since Highest Degree 0.004
Highest Degree= Masters -0.047
Highest Degree=Doctoral -0.074
Highest Degree=Professional 0.270
STEM -0.019
MASTEM 0.002
PHDSTEM -0.045
PROFSTEM 0.050

Data Source: National Science Foundation, The 2003 National
Survey of College Graduates. Marginal effects estimated by
the authors using Table 3, Column 1 coefficients for all but
the advanced degree effects, which come from Column 2. The
average probability for our base case was calculated at the
means of all continuous variables and at zero for all binary
variables. For continuous variables, marginal effects were
calculated by adding one unit to the mean to estimate a new
probability, then calculating the difference in average
probabilities from the base case. Binary variables were
"turned on" and the difference in average probabilities was
calculated from the base case.

Table 4: Multinomial Logit Estimation of the Probability
of Employer Size & Self/Paid Employment

 N Large & Odds Large
 Paid Ratio & SE

 (1) (2) (3) (4)

Intercept n.a. 7.802 *** 0.651 **
 (160) (0.295)

STEM n.a. 0.444 *** 0.440 ***
 (22) 1559 (44)

Age n.a -0.049 *** 0.952 -0.032 ***
 (0.001) (0.002)

Agriculture/ 272 -7.792 *** 0.000 -2336 ***
Forestry, Fishing (0.227) (0.371)

Mining/ 917 -2.966 *** 0.052 321
Gas Extraction (0.193) (0.326)

Construction 1,028 -6.170 *** 0.002 -1644 ***
 (0.158) (0.292)

Manufacturing 8,019 -3.271 *** 0.038 0625 ***
 (0.153) (0.273)

Wholesale Trade 1,494 -4.814 *** 0.008 -0'698 ***
 (0.158) (0.156)

Retail Trade 2,335 -4.982 *** 0.007 -0'758 ***
 (0.161) (0.152)

Transportation 790 -3.531 *** 0.029 0638 **
 (0.173) (0.296)

Information 1,698 -3.501 *** 0.030 0.236
Services (0.163) (0.288)

Finance, -4.490 *** -0604 **
Insurance 3,775 (0.151) 0.011 (0.274)
& Real Estate

Commercial 10,144 -5514 *** 0.004 -1.375 ***
Services (0.149) (0.271)

Education 15,032 -1.598 *** 0.202 -1'220 ***
Services (0.156) (0.308)

Health-Related 6,077 -4.856 *** 0.008 -2'009 ***
 (0.150) (0.280)

Social Services 874 -4.701 *** 0.009 -1662 ***
 (0.066) (0.355)

Entertainment 1,005 -5532 *** 0.004 -1.228 ***
 (0.157) (0.289)
 -4922 *** -1778 ***

Personal Services 1,458 (0.159) 0.007 (0.342)

Log-Likelihood 30021.36 ***
 Ratio

 Odds Sinall Odds
 Ratio Paid Ratio

 (5) (6) (7)

Intercept 4.966 ***
 (0.165)

STEM 0.032
 1.552 (25) 1.032

Age -0.036 ***
 0.968 (0.001) 0.964

Agriculture/ 0.097 -4.636 *** 0.010
Forestry, Fishing (0.195)

Mining/ 1.378 -2831 *** 0.059
Gas Extraction (0.216)

Construction 0.193 -3626 *** 0.027
 (0.160)

Manufacturing 1.868 -2.776 *** 0.062
 (0.158)

Wholesale Trade 0.498 -3.045 *** 0.048
 (0.276)

Retail Trade 0.469 -3630 *** 0.027
 (0.276)

Transportation 1.894 -2.954 *** 0.052
 (0.188)

Information 1.265 -2'605 *** 0.074
Services (0.170)

Finance, -3.469 ***
Insurance 0.547 (0.156) 0.031
& Real Estate

Commercial 0.253 -3.632 *** 0.026
Services (0.152)

Education 0.295 -1.057 *** 0.347
Services (0.159)

Health-Related 0.134 -3'762 *** 0.023
 (0.155)

Social Services 0.190 -2'144 *** 0.117
 (0.167)

Entertainment 0.293 -3576 *** 0.028
 (0.161)
 -1912 ***

Personal Services 0.169 (0.159) 0.148

Log-Likelihood
 Ratio
Data Source: National Science Foundation, The 2003 National
Survey of College Graduates. Standard errors are
in parentheses. The likelihood ratio tests the overall fit
of the multinomial model. *** = statistical significance at
the 1 percent level, ** = statistical significance at the 5
percent level, and statistical significance at the 10
percent level of significance. The odds ratio examines the
difference in the log of the odds between the category
represented in the column and self-employed with a small
firm. The Small/Large cutoff is 100 employees.