From unemployment to self-employment: The role of entrepreneurship training

In the early 1990s, U.S. policymakers recognized that many unemployed workers were interested in self-employment instead of finding regular jobs, but public employment offices did not offer any services to help them start their own businesses. In 1992, DOL implemented two demonstration programs—the Washington Self-Employment and Enterprise Development Project and the Massachusetts Enterprise Project—which offered entrepreneurship training and monetary assistance to unemployed workers who were collecting UI benefits (Benus et al., 1995). The results of these demonstrations convinced Congress to authorize states to create self-employment assistance (SEA) programs for a 5-year trial period under the North America Free Trade Agreement Implementation Act of 1993. SEA programs offered participants a financial stipend to allow them to devote their time to entrepreneurship activities instead of searching for a job and training and technical assistance to support their business start-up efforts.

SEA programs were permanently authorized in 1998 but ultimately fell short of accomplishing their objectives. The programs’ reach was restricted by a congressional mandate that only UI recipients at high risk of exhausting UI were eligible for SEA.

SEA programs were required to use statistical profiling mechanisms developed under the Worker Profiling and Reemployment Services (WPRS) program to identify eligible participants. Only UI recipients with high profiling scores (i.e., high predicted probability of exhausting UI benefits based on observed characteristics) were eligible for SEA. For a discussion of WPRS profiling mechanisms, see Dickinson et al. (2002).

Policymakers recognized these limitations and shifted their attention to programs that would have sufficient resources and be available to all workers, not only those collecting UI benefits. In 2003, DOL funded the Project “Growing America Through Entrepreneurship” (GATE) demonstration programs to test this new approach. Project GATE programs operated through public employment offices in Maine, Minnesota, and Pennsylvania, and (in contrast to the early 1990s demonstrations and the SEA programs operating at the time) accepted applications from all interested workers. The program offered participants training workshops covering topics related to the business start-up process and business counseling to help them develop and implement viable business plans (Bellotti et al., 2007). While Project GATE programs were operating, DOL issued directives to state workforce agencies to encourage the use of Workforce Investment Act (WIA) funds to support entrepreneurship training activities.

U.S. Department of Labor Training and Employment Guidance Letter No. 16-04, February 2005 (https://wdr.doleta.gov/directives/corr_doc.cfm?DOCN=1684).

Efforts to promote entrepreneurship training for unemployed workers were intensified during the Great Recession. In 2008, DOL awarded GATE II grants to workforce agencies in Alabama, North Carolina, Virginia, and Minnesota to operate programs that replicated the Project GATE design (Davis et al., 2017). In 2012, DOL awarded grants to workforce investment boards in Florida and Virginia to implement entrepreneurship training programs. DOL also encouraged the use of WIA funds for entrepreneurship training activities

U.S. Department of Labor Training and Employment Guidance Letter No. 12-10, November 2010 (https://wdr.doleta.gov/directives/corr_doc.cfm?DOCN=2957).

What research has been done on entrepreneurship training in the U.S. has largely examined programs that operated in periods of moderate unemployment. A government study found that the early-1990s demonstration programs in Massachusetts and Washington helped participants to become self-employed, leading to positive short-term effects on overall employment rates (Benus et al., 1995). Effects on earnings were mixed. The Washington program increased self-employment earnings, but this effect was offset by a reduction in wage and salary earnings, while the Massachusetts program had no effects on either type of earnings.

The most recent experimental evidence derives from the Project GATE experiments (Michaelides and Benus, 2012; Fairlie et al., 2015; Michaelides, 2020). Pooling data from the three Project GATE programs, these studies showed that, at 6 months after program entry, treatment cases had higher self-employment and total employment rates than control cases. Effects mainly occurred for participants who were unemployed at program entry; there were no effects for participants who were employed in salary jobs or who were already self-employed at program entry.

These results confirmed findings of an earlier government study (Benus et al., 2010). This earlier study convinced policymakers that entrepreneurship training should focus only on unemployed workers, which led to the decision that GATE II programs would target only the unemployed.

The current literature provides no evidence on the efficacy of entrepreneurship training during recessions and no formal theoretical models of how the effectiveness of training may vary with the business cycle. This is a conspicuous evidence gap given the policymaker’s support that entrepreneurship training programs received during the Great Recession. Conceptually, demand for training may be higher during recessions, when business start-up costs and individual opportunity costs tend to be low, and thus more unemployed workers than usual may decide to pursue self-employment (Rissman, 2003; Rampini, 2004; Alba-Ramirez, 2006; Mandelman and Montes-Rojas, 2009; Biehl et al., 2014). Also, the socioeconomic profile of the unemployed population is affected by prevailing economic conditions, with relatively more educated and experienced workers becoming unemployed during recessions than during good economic times (Koellinger and Thurik, 2012; Michaelides and Mueser, 2013). It is thus likely that, during recessions, there would be relatively more skilled unemployed workers who might be willing to pursue self-employment and, with the right assistance, succeed. On the other hand, during recessions, start-up financing is scarce for many reasons, including that credit institutions are reluctant to invest in start-ups or because declining home values limit the ability of aspiring entrepreneurs to use home equity to finance their business (Block et al, 2010; Fairlie, 2013; Kennickell et al, 2014). Thus, entrepreneurship training may be less valuable in a weak economy because participants would not be able to secure start-up financing even if training helps them develop a viable business plan and provides them information on various financing options.

Our article provides empirical evidence to fill this literature gap. We investigate and compare the effects of government-sponsored entrepreneurship training during a period of moderate unemployment and a severe recession and assess whether entrepreneurship training is a viable policy option to combat unemployment at different stages of the business cycle.

DOL implemented Project GATE in Maine, Minnesota, and Pennsylvania because these states had SEA-enabling legislation that allowed them to legally support entrepreneurship training activities. The Pennsylvania program operated in Philadelphia and Pittsburgh, the Minnesota program operated in Minneapolis/St. Paul and Duluth, and the Maine program operated in Portland, Lewiston, and Bangor. Program recruitment in each state followed a straightforward sequence. Interested individuals who were at least 18 years old could register at designated public employment offices, by calling a toll-free number or through the program’s website. To boost registration, the programs implemented a marketing campaign that included brochures and posters at program sites, fliers, media events, public service announcements, and networking with local government agencies and community-based organizations (Benus et al., 2010). Registered individuals were invited to an orientation meeting to obtain information about program services. Those wishing to participate were required to complete an application form and sign an informed consent statement that program participation was not guaranteed and that participants would be randomly selected from the pool of all applicants. Program registrations started in September 2003 and were done on a rolling basis through July 2005.

Individuals who completed the application form and signed the informed consent statement (hereafter, applicants) were randomly assigned to the treatment or the control group. Treatment individuals were offered two types of services: training workshops and individual business counseling. Workshops were designed to familiarize participants with different aspects of starting and running a business, such as how to develop a business plan, apply for start-up capital, develop and maintain customer relationships, manage finances, and hire staff. In the individual business counseling sessions, participants had an opportunity to discuss and determine the feasibility of their business idea with an experienced business counselor. Counselors assisted participants who had promising business ideas to develop viable business plans or make modifications to their existing plans. Counselors also provided information on available credit options, technical support to determine eligibility for Federal start-up financing programs, and credit application assistance. On the other hand, counselors encouraged participants with less promising ideas to consider alternative options that appeared to be more feasible based on their interests and background, including to pursue salary employment instead.

As noted in the implementation report of the Project GATE programs (Benus et al., 2010, p. 39): “While [counselors] were instructed not to deny services to participants based on their business ideas or suitability for self-employment, most [counselors] did give participants frank assessments of their ability to pursue entrepreneurship and in some cases discouraged participants whom they believed would not succeed.”

To ensure that training met high-quality standards, DOL used a competitive process to select training providers with experience in providing small-business counseling. Through this process, Small Business Development Centers (SBDCs),

SBDCs were created by the Small Business Administration to provide technical assistance services to nascent entrepreneurs and small businesses. SBDCs offered similar services with those offered by Project GATE and at no charge, and thus, program applicants (particularly control cases) could on their own initiative access free workshops and business counseling at local SBDCs. At the time, SBDCs operated in each of the programs’ areas of operation – one in each Portland, Lewiston, and Bangor (Maine); two in Minneapolis/St. Paul and one in Duluth (Minnesota); and two in each Philadelphia and Pittsburgh (Pennsylvania). For details, see: https://www.sba.gov/offices/headquarters/osbdc/about-us.

In 2008, DOL awarded GATE II grants to the Alabama, North Carolina, Virginia, and Minnesota workforce agencies. The only distinction with Project GATE was that GATE II programs accepted applications only from unemployed workers, while Project GATE programs accepted applications from all individuals, regardless of employment status. GATE II operated statewide in North Carolina, Alabama, and Minnesota, while the Virginia program operated in Northern Virginia and Richmond. There were notable implementation differences across GATE II programs. Alabama and North Carolina accepted applications from all unemployed workers (18+ years old); Virginia and Minnesota accepted applications only from older unemployed workers (45+ years old). North Carolina and Virginia used random assignment to determine participation and conducted follow-up surveys to collect outcomes information; Alabama and Minnesota offered training to all applicants and did not collect follow-up outcomes. Thus, our GATE II analyses consider only the North Carolina and Virginia programs.

North Carolina and Virginia replicated Project GATE’s recruitment and service delivery process. Similar to Project GATE, GATE II programs used a mass media marketing campaign to boost enrollment and accepted registrations on a rolling basis from January 2009 through December 2010. Individuals who registered for GATE II were invited to an orientation to receive program information. Those who remained interested in the program were required to complete the application form and sign the informed consent statement. Applicants were then randomly assigned to treatment and control groups. Treatment individuals were offered program services, while those in the control were informed that they were not selected for the program and had no access to program services

Similar to Project GATE programs, GATE II programs implemented checks to ensure that individuals could not reapply for the program and get assigned to the treatment group after they were initially denied services.

Project GATE recruitment began in September 2003 and continued through July 2005, a period of moderate and declining unemployment while GATE II recruitment was done from January 2009 through December 2010, near the depth of the Great Recession. As such, economic conditions differed across states when their respective programs were operating (see Figure A1

in Appendix). Pennsylvania, Minnesota, and Maine had low unemployment and increasing self-employment rates, while North Carolina and Virginia had high (and increasing) unemployment and declining self-employment rates. When programs were operating, earnings for self-employed and wage and salary workers followed positive trends in the Project GATE states and declining trends in GATE II states. There were also notable differences across states over the business cycle, including that self-employment rates were higher than the national rate in Minnesota and Maine and lower than the national rate in Pennsylvania and Virginia. States also differed in the characteristics of unemployed workers (see Table A1 in Appendix).

Table 1 summarizes applicant characteristics based on information from the application form. Among Project GATE programs, Pennsylvania attracted much higher proportions of female, nonwhite, and younger applicants. Minnesota and Maine applicants had similar characteristics, except that more Minnesota applicants had a college degree. The majority of applicants had no self-employment experience and no business plan, although many had work experience in an industry related to their business idea and in a family business. Many applicants had bad or no credit histories and low annual household incomes, suggesting limited access to financing.

Differences across GATE II programs are more pronounced, which is expected given that Virginia recruited only older unemployed workers. North Carolina applicants were younger and less educated. A large proportion of Virginia applicants had self-employment experience and good credit histories, and many more applicants in Virginia than in North Carolina had household incomes of at least $50,000. North Carolina applicants were older and less educated and had lower household incomes compared to applicants in the Project GATE states. Virginia applicants were older and relatively more of them had a college education and good credit than did Project GATE applicants.

Characteristics comparisons between applicants (Table 1) and the unemployed population (see Table A1 in Appendix) indicate that, in all five states, nonwhites, workers 35–54 years old, college-educated unemployed workers, and individuals with $25,000–49,000 annual household income were overrepresented among applicants. These comparisons suggest that unemployed workers with high levels of human capital and labor market experience were more likely to apply for training. Except for the fact that men were overrepresented in Project GATE states but not in GATE II states (and some other minor differences), there is limited evidence of systematic differences in observed applicant characteristics over the business cycle.

Except in North Carolina, all programs randomly assigned about half of the applicants to the treatment; the North Carolina program assigned about 75% (see Table A2 in Appendix, top panel). To assess whether treatment and control groups were balanced in observed characteristics, we estimated regression models in which the dependent variable is an indicator that equals 1 if the individual was assigned to the treatment and zero otherwise. Control variables include those in Table 1 and a constant term. Regression results (see Table A3 in Appendix) show that of the 100 estimated parameters, only one was statistically significant at the 5% level or less and another four were statistically significant at the 10% level. These comparisons indicate that random assignment created treatment and control groups that were generally balanced, with some differences possibly attributed to chance.

All programs administered follow-up surveys, asking applicants about the training they had received since program entry (including program training received from the community for treatment cases and similar training received from the community for control cases) and their labor market outcomes at the time of the survey.

The data made available to us for the analyses did not include any administrative data sources, such as state Unemployment Insurance (UI) data and state UI wage records. Thus, our analyses of program outcomes rely exclusively on the survey data collected by the programs.

The fact that many applicants did not respond to the survey and response rates varied by characteristics raises concerns that the treatment–control balance in characteristics may not have been maintained among respondents, which could lead to biased estimates of program effects. Regression models that estimate treatment assignment based on individual characteristics using survey respondents only (see Table A5 in Appendix) show that survey attrition did not cause imbalanced treatment and control samples among respondents, mitigating concerns that estimated effects would be biased. Another concern is that survey attrition caused some differences in characteristics between program applicants (Table 1) and survey respondents (see Table A6 in Appendix). The implication is that the outcomes of respondents may not be representative of the outcomes for the entire applicant population. To mitigate concerns that estimated effects may be affected by survey attrition (aside from having larger standard errors), we estimate effects using regression models that control for individual characteristics.

Figure 1 presents training take-up rates. Most treatment cases—ranging from 64.3% in North Carolina to 87.4% in Minnesota—received at least one type of training (workshops and or counseling). Many of these participated in both workshops and counseling. Many control cases used training options available in the community, but overall control group training take-up rates were lower than the treatment rates. Moreover, treatment cases participated on average in more workshops and counseling sessions than did control cases (see Table A7 in Appendix). Separate analyses of the marginal and joint distributions of counseling sessions and workshops received (see Tables A8 and A9 in Appendix) show that about half (or more) of the Pennsylvania, Minnesota, and Virginia treatment cases participated in at least four workshops, compared with more than a third in North Carolina and about one-quarter in Maine. Relatively more GATE II treatment cases received at least three counseling sessions than did Project GATE treatment cases.

Figure 1

Table 2 presents the effects on training take-up based on model [1]. All programs led to substantial increases in any training receipt ranging from 22.4 percentage points in North Carolina to 42.5 percentage points in Maine. In all states, except Maine, effects were mostly because of increased take-up of both services types; in Maine, the effect was mostly due to increased take-up of counseling only. All programs increased the number of workshops and counseling sessions, with relatively low effects in Maine.

There are two key takeaways from these analyses. One, the five programs increased training take-up (and the quantity of training received), suggesting that government-sponsored programs can be valuable in providing entrepreneurship services to unemployed workers who would not otherwise receive such services. At the same time, there was substantial variation across states in the compliance with the treatment protocol, so the proportion of compliers (measured as the program effect on overall training receipt) ranged widely across programs. Two, there was variation across programs in the level of services received by participants. It appears that Maine participants received a lighter dose of the treatment than participants in other programs, as they participated in relatively fewer workshops. Moreover, GATE II participants exhibited a higher preference for (or GATE II programs provided more emphasis on) business counseling, suggesting that GATE II participants received a heavier dose of the treatment than did Project GATE participants.

Survey responses are used to construct outcome measures at the time of the survey, including three measures of self-employment status: self-employment in own business (indicates self-employment in own business); self-employment as contractor (indicates self-employment as a contractor/freelancer (i.e., did not own a business)); and any self-employment (indicates any type of self-employment). We also construct wage and salary employment outcomes: regular employment (indicates employment in a wage and salary job) and any employment (indicates self-employment and/or employment in wage and salary job). In addition, we measure monthly earnings from self-employment, regular employment, and total (self-employment plus regular earnings). Survey responses are also used to measure the influence that the self-employment activities of applicants had on the employment of other workers: has employees (indicates if self-employed individuals hired employees) and number of employees. Based on the timing of the surveys, outcomes are measured at month 18 for Project GATE programs, month 24 for Virginia GATE II, and month 32 for North Carolina GATE II.

Table 3 summarizes outcome measures for treatment and control cases; the control means represent average outcomes in the absence of the offer of the government program. In the Project GATE programs, 16.8–25.8% of control individuals were self-employed in their own businesses and 6.6–15.4% were self-employed as contractors. GATE II control cases were less likely than those in Project GATE to be self-employed in their own business, indicating that it was harder to start a business during the recession. Differences in monthly average earnings reflect differences in labor market conditions and employment rates across the five states.

Table 4 presents LATE parameters.

See Table A10 in Appendix for corresponding ITT effects.

The North Carolina GATE II program increased any self-employment by 29.9 percentage points, an effect that is attributed to an increase in both types of self-employment. Results show a reduction in regular employment and increases in individual earnings and employment of other workers, although estimates lack statistical significance. In Virginia, we estimate a positive effect on overall self-employment, driven mostly by an increase in independent self-employment. At the same time, training reduced salary employment, salary earnings, and total earnings.

Previous work suggests that individuals with high levels of education and labor market experience (Fairlie and Meyer, 2000; Hout and Rosen, 2000; Reynolds et al., 2004) and those with self-employment experience or experience working in a family business (Hout and Rosen, 2000; Fairlie and Robb, 2007; Edelman et al., 2010) are more likely to possess the skills and know-how to start successful businesses. There is also evidence that financing constraints may explain why many start-up efforts are unsuccessful and many new businesses fail soon after their creation (Blanchflower et al., 2003; Cavalluzzo and Walken, 2005; Asiedu et al., 2012; Loftstrom and Bates, 2013).

It is thus possible that training may dissuade those who lack human capital and entrepreneurship skills from pursuing self-employment, convincing them to pursue regular employment instead. In contrast, individuals with an entrepreneurship background may have a frame of reference that allows them to derive different benefits from training and make better use of business counseling than those who lack this background. Also, workshops and business counseling may be less beneficial for individuals who have limited access to start-up financing, even in cases when they have a promising business plan and the skills to execute it.

Consider, for example, the results in Maine, where training increased both self-employment in own business and employment in regular jobs, suggesting that training helped some compliers to start a business and may have persuaded others to pursue regular employment instead. It is possible that this heterogeneity occurred based on observed factors. Thus, aggregate effect estimates may hide heterogeneity in training effects based on observed characteristics associated with human capital, business background, and financing access. Effects heterogeneity is an important policy concern: if only individuals with certain traits are likely to succeed if they receive training, policymakers and practitioners can improve program cost-effectiveness by targeting workers with characteristics associated with improved training success.

To investigate effects heterogeneity, we estimated LATEs based on gender, race, age, education, self-employment experience, credit history, and household income. We did not find evidence of differential effects based on gender, race, and household income. Here, we discuss results for four characteristics for which there is evidence of heterogeneity in training effects— self-employment experience, college education, credit history, and age.

Note that limited sample sizes do not allow us to estimate with high accuracy effects based on pairwise combinations of characteristics. Separate analyses that attempted such combinations lacked statistical power and are therefore omitted from the paper. Note that, in general, self-employment experience was highly correlated with age (45+ years old) but had low or no correlation with good credit and education (see Table A11 in Appendix).

In Pennsylvania, training increased self-employment in own business and any self-employment for those with self-employment experience but had much smaller or no effects for those with no self-employment experience. The increase in self-employment for experienced individuals (although partly offset by a reduction in regular employment) yielded a 26.0 percentage points increase in any employment. There is limited evidence that effects varied based on education, age, and credit access. Similarly, in Minnesota, training increased self-employment in own business by 49.8 percentage points and any self-employment by 37.6 percentage points for those with self-employment experience. Effects for those with no self-employment experience were much smaller. Training also increased self-employment in own business for individuals with a college degree or with good credit but had much lower effects for those with no college degree or bad credit.

The Maine results show different effects based on observed characteristics, although small sample sizes limit statistical power. Effects on regular employment were large and statistically different from zero for those with no self-employment experience, a college degree, and good credit. These groups experienced large gains in regular earnings. On the other hand, training increased self-employment in own business and reduced contractor self-employment for those with self-employment experience, no college degree, and bad credit. These individuals were also more likely to hire employees to support their business operations.

In North Carolina, training had much higher effects on self-employment for individuals with no self-employment experience than for experienced individuals. Training increased overall self-employment for experienced individuals by 42.3 percentage points, an effect that was due to increases in both types of self-employment. This effect was largely offset by a reduction in regular employment. Training seems to have been more effective for individuals under 45 years old—in this case, training increased self-employment by 61.8 percentage points, any employment by 35.1 percentage points, and the likelihood of hiring employees by 12.5 percentage points.

The Virginia results reveal effects not identified by the aggregate results. Training increased overall self-employment for those with no self-employment experience by 32.8 percentage points, an effect that was largely offset by a reduction in regular employment. Training also yielded small positive effects on self-employment earnings and large negative effects on total earnings for inexperienced individuals. There is also evidence that training may have increased self-employment rates and reduced earnings for individuals with a college degree or good credit history, although estimates for these groups lacked statistical significance.

Although a full cost–benefit analysis is beyond the scope of this study, we outline some of the benefits (and costs) for compliers and discuss considerations for assessing the costs and benefits from the society’s perspective. From the compliers’ perspective, an important program benefit is that they were able to access free training through the programs, so that they enjoyed the benefits of training (which they would not have received in the absence of the program) without incurring the associated pecuniary costs. Benefits for compliers would also include any improvements in earnings resulting from training participation. Effects on earnings are ambiguous. We estimate positive effects on self-employment, regular, and total earnings in Pennsylvania and North Carolina and negative effects in Minnesota and Virginia. In Maine, we estimate large and positive effects on regular and total earnings. Note that, on average, treated individuals in the three Project GATE programs were able to achieve self-sufficiency, with the average annualized total earnings in Pennsylvania ($20,676), Minnesota ($31,080), and Maine ($27,588) exceeding the federal poverty limits.

The federal poverty limit in 2008 ranged from $10,400 for a one-person household to $21,200 for a four-person household (source: https://aspe.hhs.gov/2008-poverty-guidelines-federal-register-notice). In 2008, around the time when the surveys begun, average total earnings (self-employment plus salary earnings) for employed workers was $42,829 in Pennsylvania, $43,828 in Minnesota, and $35,984 in Maine (source: Authors’ tabulations of the American Community Survey).

Self-employed workers may derive much greater levels of job satisfaction than those employed in regular jobs because they value the day-to-day activities, flexibility, and financial independence associated with self-employment (Douglas and Shepherd, 2002;Benz and Frey, 2008; 2009; Fuchs-Schündeln, 2009). Program applicants revealed a strong preference for self-employment, so compliers who became self-employed because of the program may have experienced substantial utility gains. We would thus conclude that compliers in Pennsylvania and North Carolina, who achieved higher self-employment rates and higher earnings as a result of the program, experienced large welfare gains. By the same token, positive effects on regular earnings in Maine may have offset any welfare losses caused by the reduction in self-employment. Similarly, the small earnings decline in Minnesota may have been offset by the increase in overall self-employment. It is uncertain whether the increase in self-employment in Virginia was sufficient to offset the large decrease in earnings.

Overall, our findings suggest that compliers in at least four of the five programs may have experienced positive welfare gains. A more complete analysis would take into account the personal costs of participating in the programs (including pecuniary costs such as travel costs and lost wages and nonpecuniary costs, such as loss of leisure) and other potential benefits deriving from self-employment, such as the employment of family members and business tax write-offs.

Assessing the costs and benefits of the program from the society’s perspective is complicated by lack of appropriate data and other considerations. Program benefits from the society’s perspective would include any welfare gains for participants as well as any savings in UI and other social assistance benefits because participants were able to secure employment. Another potential society benefit is that the self-employment activities of compliers led to the creation of new jobs and the employment of other workers. The Maine and North Carolina results provide evidence that such spillover employment effects are possible. Also, to the extent that the self-employment activities of compliers created new markets and revenue streams, then society benefits would include the expansion of the tax base. Society’s costs include the cost of recruiting and providing services to participants, estimated at $1,258 in Pennsylvania, $1,027 in Minnesota, and $2,342 in Maine (Benus et al., 2010). GATE II costs have not been made public. Society’s costs would also include any displacement effects occurring because training helped individuals to start businesses at the expense of interested unemployed workers with similar business ideas and no access to training. If displacement effects exist in this context, we would expect them to be relatively more important during recessions.