Self-Selection into Entrepreneurship Education and Implications for Evaluation

Abstract

Entrepreneurship is a driving force for economic wealth. In past years, interest and investment in entrepreneurship education and training programs have increased substantially. However, research on the impact and success factors of entrepreneurship education remains ambivalent. While some studies find that entrepreneurship programs foster skills relevant for entrepreneurs, others find only weak or no effects. Part of this ambiguity may be explained by self-selection effects, raising the question of whether students interested in entrepreneurship education are upfront different from those who are not. Understanding this is important to choose adequate control groups when evaluating entrepreneurship education. We address this research gap by testing our hypothesis on a sample of 359 non-applicants and 495 applicants of a well-known and successful entrepreneurship program for university students in Germany. The dedicated application process allows us to clearly identify candidates who knew the program and decided against applying. Our results indicate that applicants are more “entrepreneurial” than non-applicants along several dimensions frequently used to evaluate entrepreneurship education. Additional analyses reveal statistically significant a priori differences between male and female applicants. This suggests that researchers must pay rigorous attention to selecting suitable control groups when evaluating the impact of entrepreneurship education.

Introduction

Entrepreneurship is an important driver for economies all over the world. Extensive literature links economic growth and development to entrepreneurship (Acs et al., 2018; Audretsch, 2018; van Praag & Versloot, 2007; Schumpeter, 1934). In recent years, policymakers have focused on the potential of entrepreneurship education in higher education (e.g., European Commission, 2013). In addition, bringing forward entrepreneurs and new ventures are becoming accepted parts of universities’ third mission (Nicotra et al., 2021).

The number of entrepreneurship education programs has increased considerably in the last two decades. Most academic studies examining their impact focus on short-term indicators such as entrepreneurial attitudes, skills, knowledge, perceived feasibility, and entrepreneurial intentions (Nabi et al., 2017). However, recent meta-studies (Bae et al., 2014; Martin et al., 2013) reveal that the research body on the impact of entrepreneurship education provides an overall ambiguous and contradictory picture. Martin et al. (2013) report a small positive relationship on entrepreneurship-related human capital assets but also highlight that studies with lower methodological rigor tend to overestimate the effect. Bae et al. (2014) analyze the relationship between entrepreneurship education and entrepreneurial intentions, finding no significant relationship after controlling for pre-education levels. They highlight the possibility of reverse causation through self-selection effects and argue that “entrepreneurial intentions may not be determined by entrepreneurship education, but rather by prior beliefs before enrolling” (Bae et al., 2014, p. 221). Failing to control for potential self-selection might be one reason why the literature has not yet generated consistent findings.

Some studies investigate compulsory programs to avoid potential self-selection biases (Fayolle et al., 2006; Oosterbeek et al., 2010; Von Graevenitz et al., 2010). However, compulsory courses seem to serve a different purpose than voluntary ones. Von Graevenitz et al. (2010) argue that compulsory courses provide informative signals to participants that allow them to learn about their entrepreneurial aptitude. While not leading to stronger entrepreneurial intentions on average, compulsory courses induce a hidden sorting function by helping students determine whether they are suited for entrepreneurship after the course (Fretschner & Lampe, 2019; Von Graevenitz et al., 2010).

Following this line of argumentation, entrepreneurship education programs with a dedicated application process should be considered distinct from compulsory courses. We can assume that students who actively apply to entrepreneurship education programs likely already see a fit in terms of their career interests and abilities and perceive them as valuable for their professional development. Therefore, this self-selection can already be associated with stronger prior beliefs regarding entrepreneurship before enrolling (Bae et al., 2014; Liñán et al., 2018). Following this idea, it is surprising that many studies do not consider participants’ conscious decisions to select themselves into entrepreneurship courses. This is problematic as post-educational measures may not reflect the outcome of the educational intervention if an inadequate control group was chosen.

This is particularly relevant for entrepreneurship education when the possibility of random assignment is challenged by practical limitations (Cook et al., 1979; Costa et al., 2023). For example, to evaluate the effect of participation in a specific experiential entrepreneurship program at Stanford University, Eesley and Lee (2021) compare the rate of startup founders among program participants with engineering school alumni who graduated in the top 5% of the GPA distribution. At the same time, entrepreneurship rates among engineering alumni are between 10% and 30% for the relevant graduation cohorts. The comparison with this control group, while certainly prestigious, thus likely does not capture the true treatment effect of the program. Similarly, results in pre-/post-test studies might be skewed if simply comparing students who participated in entrepreneurship education during the semester without considering their a priori entrepreneurial aptitude, their interest in entrepreneurship, or their ability (Hahn et al., 2020; Karimi et al., 2016; Sánchez, 2011).

While research has started to address this issue (e.g., Liñán et al., 2018), it is not yet evident along which dimensions students who self-select into entrepreneurship education differ from those who don’t. A particular challenge in investigating this issue is to clearly identify students who seriously considered entrepreneurship education and then decided against it. We utilize the application process for a well-known entrepreneurship education program at a public German university to close this gap. The dedicated application process allows us to clearly identify candidates who knew the program, considered applying and decided against it. We only consider students in the group of non-applicants who created an account on the application platform and later abandoned their application. This sampling strategy ensures that we avoid including students who were unaware of the program but would have applied if they knew about it.

We test our hypotheses on a final sample of 495 applicants and 359 non-applicants. Our results indicate statistically significant differences between applicants and non-applicants along several constructs frequently used to evaluate entrepreneurship education, including entrepreneurial intention, attitudes toward entrepreneurship, and perceived behavior control. In other words, students who decide to apply to entrepreneurship programs have an upfront higher intention to found startups. We also find differences in character traits associated with entrepreneurial personalities and show that applicants have had more exposure to entrepreneurship and entrepreneurship education than non-applicants before applying.

Following the tests of our main hypotheses, we take a deeper look at group differences between genders. The gender gap in entrepreneurship is well documented (Elam et al., 2022), and entrepreneurship education has been proposed as a measure to address it (Pimpa, 2021). Groups with a lower predisposition to entrepreneurship were shown to benefit more from entrepreneurship education (Lyons & Zhang, 2017a), and the effect on entrepreneurial career intentions was shown to be stronger for women than for men (Wilson et al., 2007). However, the few studies that actually measure founding behavior (not just intentions) suggest that women benefit less, if at all (Klinger & Schündeln, 2011; Lyons & Zhang, 2017b). As such, it is interesting to understand to which degree male and female students differ as they select into entrepreneurship education.

Along all analyses, we find statistically significant differences between genders. Female students are less likely to apply. Among applicants, only 25% are female compared to 39% among non-applicants. By taking a second look at differences between genders, we show that female students have lower entrepreneurial intentions and attitudes towards entrepreneurship. However, we do not find statistically significant differences between genders with regard to their perceived behavior control and exposure to entrepreneurship. Our analyses also indicate that differences between genders are smaller than the differences between applicants and on-applicants.

These results suggest that past studies evaluating entrepreneurship education run at risk of self-selection bias. Future evaluation studies of non-compulsory entrepreneurship education should utilize not only pre-/post-measures but also select adequate control groups instead of convenience groups to avoid self-selection. For example, students who applied but were not admitted can serve as a control group for courses or programs with more applicants than open spots. In cases where participants are selected not randomly but by some measure of fit or qualification, study designs need to control for active selection. One approach for this would be to use applicants’ ranks in a regression discontinuity design to construct a quasi-experiment (Lee & Lemieux, 2010).

Our paper seeks to make three contributions. First, we study the difference between applicants and non-applicants in the example of a non-compulsory university entrepreneurship education program. By utilizing the discrete decision of students to apply or not apply after registering for the application process, we can clearly identify students who considered applying and decided against it. Our results show that students who self-select into entrepreneurship education differ from those who don’t, along several dimensions frequently used to evaluate entrepreneurship education.

Second, we take a deeper look at gender differences across both applicants and non-applicants. Gender research in entrepreneurship and female entrepreneurship have both become topics of increased research interest in recent years (Deng et al., 2021). With female students being less likely to apply, we investigate the differences between male and female students along the measured psychometric constructs.

As a third contribution, we discuss the implications for research seeking to evaluate the impact of entrepreneurship education. While randomized experiments are considered the “gold standard” for identifying causal relationships (Cook et al., 2002), randomization is often not possible due to practical limitations. Quasi-experiments with nonrandomized control groups are often the next best alternative. However, control groups need to be chosen with care. Because of self-selection dynamics, future studies need to pay careful attention to select control groups from the same population of entrepreneurial students to isolate the treatment effect of the educational intervention. Finally, future studies should delineate between compulsory and non-compulsory entrepreneurship education at university when discussing their impact.

Conceptual Framework and Hypotheses

We hypothesize that students who self-select into entrepreneurship education are more “entrepreneurial” than those who do not apply. We operationalize this concept by examining three types of measurement frequently found in entrepreneurship literature: entrepreneurial personality traits, entrepreneurial intentions, and past exposure to entrepreneurship and entrepreneurship education.

The Entrepreneurial Personality

Entrepreneurship research has long shown interest in identifying traits associated with entrepreneurial behavior. The word entrepreneurial is commonly used to describe people characterized by foresight, creativity, and adaptability. Entrepreneurship research often refers to the entrepreneurial personality (Littunen, 2000; Vries, 1977) as a set of personality traits along which founders and non-founders, e.g., employees, differ. While in the working environment, the entrepreneurial personality is often distinguished from that of managers (Chen et al., 1998; Zhao et al., 2010), in the university environment, a distinction is usually made between entrepreneurially inclined students and those who are not (Gürol & Atsan, 2006; Hansemark, 1998; Kolvereid, 1996). As the entrepreneurial process can be divided into several phases, examining the entrepreneur in the early phases of founding a company is particularly interesting, as this is where personality likely has the most significant influence (Frese & Gielnik, 2014; Hambrick, 2007).

Personality traits can, amongst other things, “include abilities […], motives […], attitudes […], and characteristics of temperament as an overarching style of a person’s experiences and actions” (Brandstätter, 2011, p. 223). A recent extensive literature review by Salmony and Kanbach (2021) looked at 95 studies between 1985 and 2020 and identified the Five Factor Model (Big 5), Need for Achievement, Innovativeness, Locus of Control, Risk Attitudes, and Entrepreneurial Self-Efficacy as commonly measured traits. While Salmony and Kanbach (2021) highlight that past literature has come short of clearly delineating between different types of entrepreneurs (e.g., agriculture entrepreneurs, nascent entrepreneurs, students with entrepreneurial interest), they find general evidence that many of these traits are more expressed in entrepreneurs. Thus, we hypothesize that students who self-select into entrepreneurship education have more pronounced entrepreneurial character traits compared to students who decide not to.

Hypothesis 1

Students who self-select into entrepreneurship education programs exhibit more expressed character traits associated with the entrepreneurial personality than students who don’t.

Based on recent literature (Salmony & Kanbach, 2021), we hypothesize a positive relationship between applying to the program and the Need for Achievement, Innovativeness, Risk-Taking Propensity, Internal Locus of Control, Openness, Conscientiousness, and Extraversion. We hypothesize a negative relationship between applying and the External Locus of Control and Neuroticism and no relationship with Agreeableness.

Entrepreneurial Intentions

Along with the growth of entrepreneurship education programs at universities (Kuratko, 2005; Solomon, 2007), researchers have started to study their impact (Nabi et al., 2017). In an extensive systematic literature review on outcomes of entrepreneurship education, Nabi et al. (2017) find that most evaluation studies use short-term measures, such as entrepreneurial attitudes and intention, as outcome variables. However, the reported impact of entrepreneurship education evaluated by using short-term measures across studies is overall ambiguous (Bae et al., 2014; Nabi et al., 2017).

Particularly, entrepreneurial intention and its antecedents are interesting to look at in the context of self-selection as they have served as a theoretical framework for many studies (Bae et al., 2014). Under the Theory of Planned Behavior (Ajzen, 1991), entrepreneurship is framed as planned behavior that is best predicted by observing intentions (Bagozzi et al., 1992). The formation of intention is preceded by three components: (1) the attitude toward the behavior, (2) subjective norms, and (3) the degree of perceived behavior control (Ajzen, 1991; Zhang et al., 2014). Attitudes towards the behavior refer to the degree of positive or negative personal valuation an individual holds towards the behavior. Subjective norms refer to the perceived social pressure to conform to the socially expected mode of conduct. Perceived behavior control reflects perceived situational competence (Ajzen, 1991; Zhang et al., 2014). In the context of entrepreneurship, “perceived behavior control” is often used interchangeably with “entrepreneurial self-efficacy” (Newman et al., 2019).

The relative importance of attitudes, subjective norms, and perceived behavior control is expected to vary across situations and behaviors (Ajzen, 1991). In the context of entrepreneurship, perceived behavior control and attitude toward entrepreneurship were shown to be robust predictors of entrepreneurial intention (Krueger et al., 2000). Subjective norms, however, were shown to be less predictive in the context of entrepreneurship. Multiple studies found no significant direct relationship between subjective norms and entrepreneurial intention (Autio, Keeley, Klofsten, Parker, & Hay, 2010; Krueger et al., 2000; Y. Zhang et al., 2014). For that reason, we do not include subjective norms in our analysis. Instead, we follow the suggestion by Krueger et al. (2000) and Zhang et al. (2014) to examine the effect of entrepreneurial exposure instead. When a person is exposed to an entrepreneurial setting through their parents, relatives, friends, peers, or firsthand experiences, a positive diffusion of their entrepreneurial knowledge and normative behavior should positively affect their entrepreneurial intentions (Zhang et al., 2014).

Building on recent work by Liñán et al. (2018), we expect students who self-select into entrepreneurship education to have higher entrepreneurial intentions than students who don’t. We, therefore, propose the following hypothesis.

Hypothesis 2

Students who self-select into entrepreneurship education programs exhibit higher entrepreneurial intentions than students who don’t.

In line with the theory of planned behavior in the context of entrepreneurship (Ajzen, 1991; Krueger et al., 2000; Liñán & Chen, 2009), we hypothesize that there is a positive relationship between applying to the program and Entrepreneurial Intention (EI) as well its antecedents Attitudes Towards Entrepreneurship (ATE), Perceived Behavior Control (PCB), and students’ Entrepreneurial Exposure.

Entrepreneurship Education and Entrepreneurship Experience

Looking at a compulsory entrepreneurship course for business students, Von Graevenitz et al. (2010) found that entrepreneurial intentions decreased among participating students despite improving entrepreneurial self-efficacy. They explain this effect by arguing that participation in the course leads to valuable information signals for students about what a potential career as an entrepreneur entails and their own entrepreneurial skills. These hidden sorting and alignment functions (Fretschner & Lampe, 2019) help students assess their own aptitude for entrepreneurship and strengthen their conviction about it.

We expect these sorting effects to be stronger when students have less information about entrepreneurship before the fact, such as in compulsory courses. Since our empirical context is a non-compulsory program with a relatively long duration of three semesters, a high workload, and a competitive application process that requires an upfront investment, we expect that applicants take a more intentional approach to deciding whether to apply. We also expect those who apply to have learned about entrepreneurship and their own aptitudes from prior contexts.

Hypothesis 3

Students who self-select into entrepreneurship education programs have more experience with entrepreneurship prior to their application than students who don’t.

Building on previous approaches (Krueger, 1993; Von Graevenitz et al., 2010), we operationalize this concept by looking at three types of prior experience: founding themselves, working in a startup, or participating in entrepreneurship education.

Data and Method

To test these hypotheses, we utilize the application process for an entrepreneurial add-on study program ¹ at a public university in Germany. The program was formally founded in 1998 and has since accepted 25 students per semester. The program’s goal is to “Connect, Educate, and Empower the Innovators of Tomorrow” through a combination of formal coursework, mentorship, interdisciplinary exchange, and access to its alumni network. The program is to be completed next to regular enrollment in an undergraduate or graduate program at the university – hence, an add-on program – and comprises 45 ECTs of coursework.

The program follows a structured curriculum comprising three core modules, elective courses, and an abroad stay. In the three core modules, each lasting one semester, students work in interdisciplinary teams of four to five students and learn how to independently conduct industry trend and scenario analysis, design thinking, digital product development, business modeling, and how to develop go-to-market motions. Importantly, teams do not work on their own startup ideas. Instead, the problem context is provided by industry partners. The goal of the program is to equip participants with the mindset, skills, abilities, and network, enabling them to innovate, possibly in a range of different careers. Doing so, the program adopts a broad framing of entrepreneurship education (Bhatia & Levina, 2020), allowing students to practice entrepreneurial thinking and problem-solving and thereby evaluate whether this would be an appropriate career for them. Overall, the content and components at the program level are comparable to the category of project-based, multi-course entrepreneurship programs, focusing on developing individual skills and mindsets rather than startup creation (Lyons & Zhang, 2017b). In addition to the formalized curriculum, various formal and informal networking events expose participants to an active community of alumni, many of whom are working in entrepreneurial careers.

Despite its workload and small intake, the program is well-known within the local university landscape to have produced remarkable entrepreneurial output. Alumni of the program have co-founded some of the most visible startups in the local entrepreneurial ecosystem. In total, the around 1000 alumni of the program have co-founded more than 250 startups and raised more than $6 billion in funding. ²

Program Context: Application Process

The program has a limited number of spots each semester, so participants are chosen through a competitive application process. The first round of the application consists of a written online application that includes questions on applicants’ academic and professional experiences and motivational letters. In a double-blind review process, each application is scored by multiple people associated with the program. From the initial applicant pool, about 60 finalists are chosen to advance to the next stage of the process. On average, around 300 people apply for each intake, 60 are invited to the second round, and 25 are admitted to the program.

For the context of this study, we utilize the first step of the application process, the online application. This first step comprises the submission of students’ anonymized CVs, academic records, and two written essays. In addition to the around 300 valid applications, an additional 500 accounts are created each semester on the online application platform that end up not submitting their application. In other words, in each application round, there is a pool of students who know about the program, consider applying, and create an account on the application website, only to later abandon their application. This natural event allows us to distinguish between students who decided to pursue entrepreneurial education by submitting their applications and those who decided against it.

Data Collection and Sample Selection

We collected our data with a questionnaire (see Appendix, Table A1), which was distributed after the application period deadline and before applicants received feedback. Completing all questions took between 7 and 10 minutes. We were careful to highlight that participation in the study was voluntary, data was collected anonymously, and participation (or the lack thereof) would not influence the application process. To increase response rates, we raffled five vouchers worth EUR 50 among participants and sent follow-up emails 5 and 12 days after the initial outreach. We collected our dataset in three rounds in May 2021, December 2021, and May 2022. In total, we collected 854 valid responses from students. Of those, 495 are in the applicant group and 359 in the non-applicant group.

We expect this data collection and sample selection to be suitable for testing our hypotheses. First, the study program is well known among students within the local university context as it has produced many successful founders and is advertised in entrepreneurship-related courses at the university before the application deadline. Second, the dedicated application process via an online platform allows us to capture all students who registered an account regardless of whether a final application is submitted. By distributing our questionnaire to all registered users, we can thus be sure that all participants know about the program and have decided to apply or not apply. This approach allows us to avoid including students in the group of non-applicants who would have applied but were unaware of the program. Finally, we only include responses from students that meet the program selection criteria. Thus, we only consider students enrolled in a study program at the Technical University of Munich or the Ludwig Maximilian Universität in Munich.

Research Instrument

We developed a web-based questionnaire to collect multiple constructs. Both applicants and non-applicants received the same questionnaire (see Appendix A, Table A1). The questionnaire measured the following constructs: Entrepreneurial Intention, Perceived Behavior Control, Attitudes Towards Entrepreneurship, Innovativeness, Need for Achievement, Locus of Control, Risk Taking Propensity, and the BIG Five Personality Traits. In addition to the primary constructs, we collected data on ex-ante experiences with entrepreneurship and entrepreneurship education and demographic data. To ensure the validity of the measures, we adapted construct items from previous studies. All questions collected with Likert-Scales (Likert, 1932) were adapted to a 5-point scale, ranging from 1 (Totally Disagree) to 5 (Totally Agree). ³ Constructs measured with several items were combined into an index by calculating the mean from the individual items.

To measure personality traits, we relied on validated scales from prior research. We measured the Need For Achievement with five questions adapted from Steers and Braunstein (1976). We measured Innovativeness with eight items from the Jackson Personality Inventory (Paunonen & Jackson, 1996) adapted by Mueller and Thomas (2001) for the entrepreneurship context. We measured the Locus of Control with four items based on Kovaleva (2012). Using the General Risk Propensity Scale by Zhang et al. (2019), we measured Risk-Taking Propensity with eight items. Finally, we measured the Five Factor Model of Personality (Big 5) with ten items using the 10-Item-Big-Five-Inventory (BFI-10) by Rammstedt et al. (2013). The BFI-10 has a documented high construct and criterion validity for a short scale (Rammstedt et al., 2013). However, it is important to note the limitations of measuring the facets of the BIG-5 with only ten items. Short scales make an economic trade-off between construct coverage and measurement efficiency. Scores derived from such scales typically do not demonstrate the specificity and sensitivity to draw conclusions at the individual level. However, they are suited to be applied at the group level (Ziegler et al., 2014).

We measured Entrepreneurial Intention with three questions adapted from Liñán and Chen (2009). To adapt the item to the university context, we added a time aspect, changing the original item “I have the firm intention to start a firm someday” to “I intend to start a business within the next 5 years” to include a time aspect and sharpen it for application for university students. An index was calculated by averaging the scores of these items. A Cronbach’s Alpha of 0.88 reflects a strong internal consistency following this adaptation. ⁴ We measured Attitudes Towards Entrepreneurship with five questions adapted from Liñán and Chen (2009). We measured Perceived Behavior Control with four questions proposed by Zhao et al. (2005). We measured entrepreneurial exposure by summing up participants’ answers (1 = yes, 0 = no) to the following questions (Krueger, 1993; Von Graevenitz et al., 2010): “Did you found a startup in the past?”, “Did you work at a startup in the past?”, “Did you participate in an entrepreneurship course in the past?”, “Among your family, did someone found a startup?”, “Among your close friends, did someone found a startup?”, “In your social circle, did someone found a startup?“

Results

We find support for H1, H2, and H3. Our results suggest that there are statistically significant differences along all theorized constructs. We fit several regression models to test whether personality traits, entrepreneurial intention, and past entrepreneurship (education) experience predict program application. Even when including all measured constructs and control variables, the estimated OLS regression model (R-Squared = 0.147, p < .001) explains only 14.7% of the variance in the sample, suggesting that students’ decisions to select into entrepreneurship education programs is a complex process that cannot solely be captured by psychometric constructs and demographic variables.

Sample

In total, 854 students answered the questionnaire, 495 of whom were in the applicant group and 359 in the non-applicant group. Table 1 provides an overview of the summary statistics and between-group differences.

Table 1.

Summary Statistics and Difference Between Applicants and Non-applicants.

	Full sample					Applicants		Non-applicants		Difference
	Obs.	Mean	Std. Dev.	Min	Max	Obs.	Mean	Obs.	Mean	Diff.	t-stat.
Demographics
Age	854	23.88	2.79	19.00	38.00	495	23.68	359	24.16	−0.47	−2.45***
Female	854	0.31	0.46	0.00	1.00	495	0.25	359	0.39	−0.14	−4.45***
International	854	0.49	0.50	0.00	1.00	495	0.45	359	0.54	−0.10	−2.80***
Graduate student	854	0.71	0.45	0.00	1.00	495	0.72	359	0.70	0.02	0.72
Total semesters	854	8.95	2.95	4.00	22.00	495	8.94	359	8.95	−0.01	−0.03
Business major	854	0.35	0.48	0.00	1.00	495	0.37	359	0.31	0.06	1.81*
CS/EE major	854	0.30	0.46	0.00	1.00	495	0.32	359	0.27	0.05	1.70*
University degree parents	854	1.29	0.80	0.00	2.00	495	1.30	359	1.28	0.02	0.30
Entrepreneurship parents	854	0.44	0.50	0.00	1.00	495	0.45	359	0.44	0.01	0.27
Past entrepreneurship education
Founded startup prior	854	0.20	0.40	0.00	1.00	495	0.24	359	0.15	0.09	3.18***
Startup employee prior	854	0.52	0.50	0.00	1.00	495	0.58	359	0.44	0.14	3.99***
Prior EE participation	854	1.87	1.79	0.00	15.00	495	2.11	359	1.55	0.56	4.56***
Prior EE - positive	685	3.82	1.13	1.00	5.00	420	3.93	265	3.65	0.28	3.14***
Prior EE - learning	686	3.53	1.13	1.00	5.00	421	3.64	265	3.35	0.29	3.31***
Prior EE - inspiration	681	3.79	1.21	1.00	5.00	416	3.95	265	3.54	0.41	4.41***
Prior EE - people	660	3.65	1.32	1.00	5.00	402	3.80	258	3.41	0.40	3.80***
Personality characteristics
Agreeableness (BIG 5)	854	4.21	0.63	1.50	5.00	495	4.24	359	4.17	0.07	1.61
Openness (BIG 5)	854	3.75	0.77	1.00	5.00	495	3.79	359	3.69	0.10	1.94**
Extraversion (BIG 5)	854	3.66	0.89	1.00	5.00	495	3.72	359	3.58	0.15	2.39**
Conscientiousness (BIG 5)	854	4.05	0.73	1.00	5.00	495	4.14	359	3.92	0.22	4.41***
Neuroticism (BIG 5)	854	2.31	0.96	1.00	5.00	495	2.25	359	2.39	−0.15	−2.20**
Innovativeness	854	3.72	0.51	2.12	5.00	495	3.80	359	3.61	0.19	5.28***
Risk propensity	854	3.51	0.81	1.25	5.00	495	3.59	359	3.40	0.19	3.45***
Need for achievement	854	4.21	0.52	2.00	5.00	495	4.29	359	4.09	0.20	5.55***
Internal locus of control	854	4.14	0.73	1.00	5.00	495	4.18	359	4.07	0.12	2.34**
External locus of control	854	1.98	0.76	1.00	5.00	495	1.90	359	2.10	−0.20	−3.87***
Entrepreneurial intention
EI	854	3.78	1.07	1.00	5.00	495	4.02	359	3.44	0.58	8.02***
ATE	854	4.19	0.80	1.00	5.00	495	4.34	359	3.99	0.35	6.36***
PCB	854	3.79	0.67	1.50	5.00	495	3.91	359	3.64	0.27	5.87***
EEx	854	3.24	1.42	0.00	6.00	495	3.42	359	3.00	0.42	4.27***

Notes: Abbreviations: Entrepreneurial Intention (EI), Attitude Toward Entrepreneurship (ATE), Perceived Behavior Control (PCB), Entrepreneurial Exposure (EEx).*p < .1, **p < .05, ***p < .01.

The results suggest that the groups can be considered homogenous with regards to the following demographic measures: whether they are enrolled in a graduate program (as compared to an undergraduate program), their total semesters at university, whether their parents have a university degree, and whether their parents at some point founded a business themselves. However, we find statistically significant differences between the following demographic measures: Applicants are slightly younger at the time of application (23.68 vs. 24.16 years), are less likely to be female (25% vs. 39%), and less likely to be international students (45% vs. 54%). We also find statistically significant differences at the 10% level regarding students’ study backgrounds. Applicants are more likely to be enrolled in business (37% vs. 31%) or computer science and electrical engineering programs (31% vs. 27%). Given that the program is run by Professors from exactly these backgrounds, this is not all that surprising.

Main Results

Table 1 also shows the between-group differences concerning each of the hypotheses presented above. The results show statistically significant differences between the groups for all hypothesized constructs.

Personality Traits

We find statistically significant differences between applicants and non-applicants with regard to all theorized personality traits, providing support for H1. Table 1 shows the between-group differences. The most apparent differences between the groups can be found with regards to Conscientiousness (4.14 vs. 3.92, t-stat = 4.41), Innovativeness (3.80 vs. 3.61, t-stat = 5.28), and Need for Achievement (4.29 vs. 4.09, t-stat = 5.55).

To understand whether personality traits predict students’ decision to apply to the program, we fit three OLS regression models, adding the measured variables in steps. The results can be seen in Table 2. Model (1) includes only the BIG Five personality model. Model (2) adds the remaining personality traits related to entrepreneurship, and Model (3) the demographic control variables from Table 1. The overall model fit is statistically significant (p < .001) with R-squared values between 0.031 and 0.112. The coefficients of all variables correspond with the hypothesized direction – i.e., a positive coefficient of 0.088 for Conscientiousness indicates that students who score higher on Conscientiousness are more likely to apply. Across the fitted models, not all personality traits are statistically significant – only Conscientiousness, Innovativeness, Need for Achievement, and External Locus of Control are. When adding control variables in Model (3), we find that older students (−0.023***) and female students (−0.155***) are less likely to apply. Graduate students (0.087**) and business majors (0.077**) are more likely to apply.

Table 2.

Regression – Personality Traits and Program Application.

	(1)	(2)	(3)
Agreeableness (BIG 5)	0.019	−0.006	0.002
	(0.71)	(−0.21)	(0.06)
Openness (BIG 5)	0.029	−0.007	−0.009
	(1.30)	(−0.31)	(−0.37)
Extraversion (BIG 5)	0.024	0.004	0.011
	(1.22)	(0.19)	(0.55)
Conscientiousness (BIG 5)	0.088***	0.062**	0.071***
	(3.72)	(2.47)	(2.87)
Neuroticism (BIG 5)	−0.025	−0.003	0.016
	(−1.40)	(−0.17)	(0.85)
Innovativeness		0.116***	0.124***
		(3.08)	(3.33)
Risk propensity		0.031	0.037
		(1.39)	(1.61)
Need for achievement		0.081**	0.074**
		(2.11)	(1.96)
Internal locus of control		0.002	−0.010
		(0.09)	(−0.39)
External locus of control		−0.050**	−0.041*
		(−2.20)	(−1.80)
Controls	No	No	Yes
Observations	854	854	854
F statistic	5.386	6.079	5.544
Prob > F	0.000	0.000	0.000
R-squared	0.031	0.067	0.112

Notes: *p < .1, **p < .05, ***p < .01.

Entrepreneurial Intention

We find statistically significant differences between applicants and non-applicants with regard to entrepreneurial intention and the theorized antecedents, providing support for H2. The difference between applicants and non-applicants in Entrepreneurial Intention is the highest among all measured constructs (4.02 vs. 3.44, t-stat = 8.02). Across both groups, Entrepreneurial Intention and Attitude Toward Entrepreneurship (4.34 vs. 3.99, t-stat = 6.36) can be considered relatively high (all items are measured on a 5-point scale where 5 is the highest value), indicating that the program succeeds in attracting students interested in entrepreneurship. A comparison of entrepreneurial intentions among business administration students from 2004 (Franke & Lüthje, 2004) and 2010 (Von Graevenitz et al., 2010) confirms this notion. ⁵

To understand how far entrepreneurial intention and its antecedents predict students’ decision to apply to the program, we fit three OLS regression models, adding the measured variables in steps. The results can be seen in Table 3. Model (1) includes only Entrepreneurial Intention. Model (2) adds the measured antecedents, and Model (3) the demographic control variables from Table 1. The overall model fit is statistically significant (p < .001) with R-squared values between 0.070 and 0.119. The coefficients of all variables correspond with the hypothesized direction – i.e., a positive coefficient of 0.122 for Entrepreneurial Intention indicates that students who score higher on Entrepreneurial Intention are more likely to apply. Across the fitted models, only Entrepreneurial Intention and Perceived Behavior Control are consistently statistically significant. When adding control variables in Model (3), we find that older students (−0.026***), female students −0.090**), and international students (−0.068**) are less likely to apply. Graduate students (0.086**) are more likely to apply.

Table 3.

Regression – Entrepreneurial Intention and Program Application.

	(1)	(2)	(3)
Entrepreneurial intention (EI)	0.122***	0.088***	0.078***
	(8.02)	(4.18)	(3.72)
Attitude toward entrepreneurship (ATE)		0.017	0.012
		(0.60)	(0.42)
Perceived behavior control (PCB)		0.073***	0.085***
		(2.66)	(3.11)
Entrepreneurial exposure		0.017	0.025*
		(1.42)	(1.74)
Controls	No	No	Yes
Observations	854	854	854
F statistic	64.360	19.264	8.758
Prob > F	0.000	0.000	0.000
R-squared	0.070	0.083	0.119

Notes: *p < .1, **p < .05, ***p < .01.

Entrepreneurship Education and Entrepreneurship Experience

We find statistically significant differences between applicants and non-applicants with regard to their experience with entrepreneurship education and entrepreneurship providing support for H3. Table 1 shows that among applicants, 24% have previously founded a startup compared to 15% in the group of on-applicants. Focusing on past experiences with entrepreneurship, we find that applicants have participated in more entrepreneurship education courses (2.11 vs. 1.55, t-stat = 4.56) and perceived them more positively with regard to several dimensions.

To understand whether these past experiences predict students’ decision to apply to the program, we fit three OLS regression models, adding the measured variables in steps. The results can be seen in Table 4. Model (1) includes past experiences with entrepreneurship education. Model (2) adds first-hand experience with a startup (founding or working in a startup), and Model (3) the demographic control variables from Table 1. The overall model fit is statistically significant (p < .001) with R-squared values between 0.024 and 0.042. The coefficients of most variables concerning the perception of past entrepreneurship education are close to 0. Across the fitted models, only whether prior entrepreneurship education was perceived as inspirational and previous work experience in a startup are consistently statistically significant. However, the latter is only at the 10% confidence level. The regression results thus suggest that students who found past entrepreneurship courses inspirational and have worked in startups are more likely to apply. When adding control variables in Model (3), we find that older students (−0.019**), female students (−0.145***), and international students (−0.076*) are less likely to apply. Graduate students (0.084*) are more likely to apply.

Table 4.

Regression – Past EE and Program Application.

	(1)	(2)	(3)
Prior EE participation	0.019*	0.013	0.016
	(1.66)	(1.15)	(1.28)
Prior EE - positive experience	−0.008	−0.005	−0.008
	(−0.30)	(−0.16)	(−0.29)
Prior EE - learning value	0.000	−0.002	0.002
	(0.01)	(−0.05)	(0.07)
Prior EE - inspiration	0.055**	0.053**	0.058**
	(2.04)	(1.97)	(2.18)
Prior EE - interesting people	0.017	0.017	0.015
	(0.93)	(0.90)	(0.81)
Worked in startup prior		0.073*	0.075*
		(1.85)	(1.92)
Founded startup prior		0.038	0.023
		(0.81)	(0.50)
Controls	No	No	Yes
Observations	854	657	657
F statistic	20.806	4.749	4.080
Prob > F	0.000	0.000	0.000
R-squared	0.024	0.035	0.042

Notes: *p < .1, **p < .05, ***p < .01.

Gender Differences

Across all analyses, we find statistically significant differences between male and female students. Most notably, among non-applicants, 39% of participants are female compared to only 25% among applicants. Gender differences in entrepreneurship research have increasingly become a topic of interest (Deng et al., 2021), with several studies also addressing students (Dabic et al., 2012; Packham et al., 2010; Petridou et al., 2009; Wilson et al., 2007). While these studies examined entrepreneurship-related gender differences in general student populations (Dabic et al., 2012), MBA students and adolescents (Wilson et al., 2007), elective courses at university (Petridou et al., 2009), and short enterprise education courses (Packham et al., 2010), our results add to the literature by reporting differences between male and female students along several psychometric constructs in the context of self-selection into entrepreneurship education programs.

Table 5 shows the demographic properties and mean values of the measured constructs split by gender across the full sample, the applicant group, and the non-applicant group. When looking at applicants and non-applicants separately, we generally find more pronounced gender differences in the group of applicants. For example, female applicants are, on average, older than their male counterparts (24.27 vs. 23.49 years), they are more likely to be enrolled in a graduate program (83% vs. 68%), have studied longer (9.77 semesters vs. 8.67 semesters), and comprise a larger share of international students (62% vs. 39%). Overall, gender differences appear to be more expressed among applicants. We also observe that female students are less likely to be enrolled in a study subject related to computer science and engineering across both the applicant group (12% vs. 39%) and the non-applicant group (15% vs. 34%).

Table 5.

Summary Statistics and Difference Between Genders.

	Full sample (N = 854)			Applicants (N = 495)			Non-applicant (N = 359)
	Male (N = 593)	Female (N = 261)	t-stat	Male (N = 373)	Female (N = 122)	t-stat	Male (N = 220)	Female (N = 139)	t-stat
Demographics
Age	23.65	24.40	−3.62***	23.49	24.27	−2.98***	23.93	24.51	−1.73*
International	0.44	0.61	−4.64***	0.39	0.62	−4.60***	0.51	0.59	−1.41
Graduate student	0.68	0.77	−2.43**	0.68	0.83	−3.10***	0.69	0.71	−0.52
Total semesters	8.69	9.53	−3.86***	8.67	9.77	−3.77***	8.72	9.32	−1.78*
Business major	0.33	0.39	−1.80*	0.35	0.43	−1.44	0.28	0.36	−1.55
CS/EE major	0.37	0.14	6.99***	0.39	0.12	5.56***	0.34	0.15	4.04***
University degree parents	1.33	1.22	1.83*	1.35	1.15	2.41**	1.29	1.28	0.07
Entrepreneurship parents	0.44	0.45	−0.22	0.44	0.48	−0.74	0.45	0.42	0.39
Past entrepreneurship education
Founded startup prior	0.23	0.13	3.46***	0.26	0.16	2.23**	0.18	0.10	2.10**
Startup employee prior	0.53	0.49	1.09	0.59	0.55	0.79	0.44	0.45	−0.10
Prior EE participation	1.92	1.77	1.11	2.13	2.04	0.44	1.56	1.53	0.16
Prior EE - positive	3.77	3.94	−1.71*	3.87	4.11	−1.97**	3.58	3.77	−1.22
Prior EE - learning	3.47	3.66	−2.01**	3.60	3.77	−1.38	3.22	3.55	−2.30**
Prior EE - inspiration	3.73	3.93	−2.00**	3.89	4.14	−1.92*	3.41	3.73	−1.98**
Prior EE - people	3.60	3.76	−1.42	3.76	3.96	−1.37	3.30	3.58	−1.50
Personality characteristics
Agreeableness (BIG 5)	4.21	4.20	0.21	4.24	4.23	0.08	4.16	4.17	−0.13
Openness (BIG 5)	3.75	3.76	−0.22	3.77	3.89	−1.51	3.72	3.65	0.80
Extraversion (BIG 5)	3.59	3.83	−3.63***	3.66	3.91	−2.74***	3.46	3.76	−2.97***
Conscientiousness (BIG 5)	4.01	4.14	−2.56***	4.09	4.30	−3.13***	3.87	4.00	−1.60
Neuroticism (BIG 5)	2.19	2.58	−5.67***	2.16	2.50	−3.50***	2.23	2.65	−4.12***
Innovativeness	3.72	3.70	0.55	3.78	3.85	−1.34	3.63	3.58	1.02
Risk propensity	3.57	3.39	2.92***	3.62	3.50	1.55	3.47	3.30	1.87*
Need for achievement	4.21	4.20	0.31	4.28	4.33	−1.02	4.10	4.09	0.18
Internal locus of control	4.16	4.07	1.71*	4.20	4.15	0.69	4.11	4.00	1.23
External locus of control	1.97	2.01	−0.57	1.88	1.97	−1.20	2.14	2.04	1.16
Entrepreneurial Intention
EI	3.91	3.47	5.58***	4.08	3.84	2.37**	3.63	3.15	3.97***
ATE	4.30	3.93	6.35***	4.40	4.14	3.68***	4.14	3.75	4.04***
PCB	3.80	3.77	0.71	3.88	3.98	−1.47	3.67	3.59	1.12
EEx	3.29	3.13	1.57	3.45	3.33	0.82	3.03	2.95	0.52

Notes: Male/Female columns show the respective mean values. Abbreviations: Entrepreneurial Intention (EI), Attitude Toward Entrepreneurship (ATE), Perceived Behavior Control (PCB), Entrepreneurial Exposure (Eex).*p < .1, **p < .05, ***p < .01.

Personality Traits

Across the different groups, we find consistent and statistically significant differences of the Big Five dimensions Extraversion, Conscientiousness, and Neuroticism. Along each dimension, male applicants score lower than their female counterparts. Among non-applicants, the difference in Conscientiousness loses its statistical significance. We find no differences for Agreeableness and Openness. The only other personality traits where we observe statistically significant differences between male and female students are their risk propensity and Internal Locus of Control. In both cases, female students scored slightly lower than male students when looking at the entire sample. However, their statistical significance diminishes when looking at applicants and non-applicants separately.

These observations, to a degree, align with findings from literature. Past research investigating the difference between genders of the BIG Five dimensions found that women scored higher than men on Extraversion, Agreeableness, and Neuroticism. Differences in Openness and Conscientiousness were only found when looking at subdimensions, and the strength of the effects was moderated by age and ethnicity (Weisberg et al., 2011). It is, however, important to note that gender differences vary between cultures (they are more pronounced in cultures where traditional sex roles are minimized) and that gender differences are overall small relative to the individual variation within genders (P. T. Costa et al., 2001).

While studies in literature report “strong evidence” for gender differences in risk aversion (Charness & Gneezy, 2012), we observe that the statistical significance diminishes when only looking at applicants. Comparing the mean score of male non-applicants (3.47) and female applicants (3.50), the latter scored slightly higher. Existing literature reports little to no difference between genders with regard to Locus of Control (Feingold, 1994; Sherman et al., 1997).

Entrepreneurial Intention

When looking at the Entrepreneurial Intention model, we find statistically significant differences between male and female students for Entrepreneurial Intention (EI) and their Attitude Toward Entrepreneurship (ATE) but not for their Perceived Behavior Control (PCB) and Entrepreneurial Exposure. These differences are consistent and statistically significant across all groups. For example, female applicants report lower Entrepreneurial Intention (3.84 vs. 4.08, t-stat = 2.37) and Attitude Toward Entrepreneurship (4.14 vs. 4.40, t-stat = 3.86) than their male counterparts. However, it is to be noted that female students who applied to the program scored higher or similarly high on Entrepreneurial Intention (3.84 vs. 3.63) and Attitude Toward Entrepreneurship (4.14 vs. 4.14) than male students who did not apply. Interestingly, we do not see statistically significant differences in Entrepreneurial Exposure (EEx) and Perceived Behavior Control. In both cases, applicants, regardless of gender, score higher than non-applicants. While not statistically significant, female students even scored slightly higher in Perceived Behavior Control than male students among applicants (3.98 vs. 3.88).

Previous literature reports higher Entrepreneurial Intention (Dabic et al., 2012; Wilson et al., 2007), higher Perceived Behavior Control, and higher Attitudes Towards Entrepreneurship (Packham et al., 2010) among male students (Wilson et al., 2007). However, Wilson et al. (2007) also find that entrepreneurship education has a higher impact on women. Among applicants, female and male students participated in a similar number of entrepreneurship courses before their application, which in part might explain the similar scores in Perceived Behavior Control.

It is to note, though, that among the group of applicants, the mean scores for Entrepreneurial Intention, Attitudes Towards Entrepreneurship, and Perceived Behavior Control among both female and male students all rank high when compared to previous studies (Franke & Lüthje, 2004; Maresch et al., 2016; Von Graevenitz et al., 2010). These results suggest that while gender differences in the mean exist for Entrepreneurial Intention and Attitudes Towards Entrepreneurship, the variance of individual differences within genders is substantial. It also shows that while enrollment rates differ between male and female students, the program attracts entrepreneurial students regardless of gender.

Entrepreneurship Education and Entrepreneurship Experience

Female students are less likely to have founded a startup before their application across all groups. The observed difference is consistent at ten percentage points. For example, 16% of female applicants founded prior to the application compared to 26% of male applicants. We observe no difference between genders when looking at prior employment at a startup or prior participation in entrepreneurship education.

Previous work indicates that enrollment rates in entrepreneurship education at universities are typically male-dominated (Bae et al., 2014), which is also reflected when comparing the share of female students among applicants (25%) and non-applicants (39%). However, when looking at the previous participation in entrepreneurship education among applicants, we do not observe any statistically significant differences. In other words, when looking at past behavior of applicants related to entrepreneurship education, both female and male students appear to be similar.

Interestingly, female students across all groups rate their experiences of past entrepreneurship education higher than male students. Except for whether they had met interesting people through participation in entrepreneurship education, these differences are all statistically significant. Previous work lets us speculate about the underlying drivers for these differences. For example, Wilson et al. (2007) show that female students benefit more from entrepreneurship education than their male counterparts. Findings by Petridou et al. (2009) indicate that the upfront motivation for participating in entrepreneurship education might differ between genders. Among others, female students expressed a stronger interest in acquiring knowledge and developing skills compared to male students (Petridou et al., 2009). With different ex-ante expectations between male and female students, their ex-post assessment might differ even when joining the same courses and programs.

Discussion and Conclusion

We present the first study to explicitly investigate the effect of entrepreneurial constructs and personality characteristics on self-selection in entrepreneurial education. We provide several contributions to the academic discourse and find significant differences between students who self-select into entrepreneurship education and those who don’t.

Study designs in entrepreneurship education research have been repeatedly critiqued (Bae et al., 2014; Rideout & Gray, 2013; Yi & Duval-Couetil, 2021) for, among others, a lax selection of control groups (Yi & Duval-Couetil, 2021). This has been a rather pervasive phenomenon, rather than the exception: A recent systematic review of “empirical studies attempting to measure changes in outcomes as evidence of the impact of entrepreneurship education” (Yi & Duval Couetil, p. 1694) finds that 80.3% of identified studies relied on convenience sampling, 57.4% missed a control group, and 31% did not provide any comparison (Yi & Duval-Couetil, 2021).

Our work confirms that self-selection into entrepreneurship education is not just a theoretical argument (Bae et al., 2014; Liñán et al., 2018), but a practical reality. When evaluating non-compulsory entrepreneurship courses or programs, self-selection needs to be controlled for. Results of past studies, which have not done so, are likely biased through a “reversed causal influence of entrepreneurial intention on entrepreneurship education” (Bae et al., 2014, p. 238).

The findings presented in this paper strengthen our current understanding of the phenomenon by providing evidence from a selective entrepreneurship program, showing that differences between applicants and non-applicants exist not just when looking at intention but also actual behavior to enroll in entrepreneurship education.

Future studies evaluating entrepreneurship education must seek to avoid self-selection bias to capture the true effect of the educational intervention. Mere pre-/post-study assessments might produce misleading results if no or inadequate control groups are chosen (Yi & Duval-Couetil, 2021). To maximize scientific rigor, experiments with a random assignment between the treatment and control groups should be used. The random assignment allows researchers to assume that between groups, individuals are homogenous. As a result, randomized experiments allow the isolation of particular variables as predictors for particular outcomes, i.e., the testing of causal relationships (S. Costa et al., 2023). If a random assignment is not possible, control groups should be selected from the same audience as participants to be truly comparable. In the context of non-compulsory programs with limited spaces, the group of applicants who could not secure a spot should be considered as a suitable starting point for the construction of the control group. ⁶

Avoiding Self-Selection Bias in Evaluation Studies

Scholars can generally follow two approaches to avoid bias through self-selection. ⁷ Either they can avoid selection altogether by looking at compulsory courses and programs, or they can choose a control group from the same population as participants, i.e., other applicants who did not receive the treatment. Since differences in pedagogies, methods, and audiences are likely to produce differential outcomes (Nabi et al., 2017), the impact of compulsory and non-compulsory courses should be considered in separation.

Compulsory Entrepreneurship Education

Earlier work attempted to avoid the issue of self-selection by examining compulsory courses (Fretschner & Lampe, 2019; Oosterbeek et al., 2010; Von Graevenitz et al., 2010). However, current research indicates that compulsory courses may have a different effect compared to ones that students actively choose to participate in. For example, Von Graevenitz et al. (2010) report a negative effect of participation in a compulsory entrepreneurship course on participants’ entrepreneurial intentions. In contrast, recent studies looking at specific experiential entrepreneurship education programs find a positive effect on entrepreneurial outcomes (Eesley & Lee, 2021; Lyons & Zhang, 2017b). Compulsory courses appear to provide information signals to students who, in turn, adjust their aptitude for entrepreneurship (Von Graevenitz et al., 2010). Following this line of argumentation, compulsory entrepreneurship courses enable participants to better understand their ability and preference (Bae et al., 2014) towards future entrepreneurship and entrepreneurship education (Liñán et al., 2018) and thus enable informed self-selection into additional entrepreneurship courses. As students who self-select into non-compulsory courses perceive them to be more useful, they are likely more engaged during the course, which might spark different learning outcomes. In other words, because their audience differs, compulsory and non-compulsory courses likely lead to different outcomes. Consequently, findings from evaluations of compulsory entrepreneurship education likely do not generalize to non-compulsory entrepreneurship education.

Non-Compulsory Entrepreneurship Education

To evaluate non-compulsory entrepreneurship education programs and courses, scholars should choose adequate control groups (Yi & Duval-Couetil, 2021). As the presented results show, applicants can be considered up-front more entrepreneurial than students not applying. To avoid bias through self-selection, scholars may use applicants who applied for the program or course but were not accepted as a control group (Lyons & Zhang, 2017b). This strategy seems feasible in practice as many programs have limited spots. From a research design perspective, the most rigorous approach would be to randomly select participants from the population of applicants (Cook et al., 2002; S. Costa et al., 2023). However, in practice, there is likely an interest in achieving a high program quality and choosing the students that are most suited for the program.

This consequently raises the issue of active selection. ⁸ If participants are selected by some metric of merit, it is likely that those who score higher would fare better with regard to the intended outcomes, even without the treatment. This effect needs to be controlled for, when using the pool of rejected applicants to construct a control group. If a ranking of applicants exists, regression discontinuity designs (RDD) may offer an elegant solution to construct a quasi-experimental evaluation of the local average treatment effect at the capacity threshold while controlling for selection (Lee & Lemieux, 2010). RDD further provides the advantage that they are closer to the “gold standard” randomized controlled trials compared to other evaluation methods like matching on observables, regressions, and instrumental variables (Lee & Lemieux, 2010). Similar approaches have been used to evaluate the impact of accelerator programs (Gonzalez-Uribe & Leatherbee, 2018; Hallen et al., 2020). If no ranking is available other, albeit weaker, identification strategies can involve restricting the control group to the best applicants not accepted to the program or matching individuals on relevant variables, such as age, gender, nationality, and other relevant covariates (see for example, Lyons & Zhang, 2017b).

Why Do Students Apply?

Our results show that students who self-select in entrepreneurship education differ from those who don’t, along several psychometric constructs associated with entrepreneurship. However, the presented regressions explain only between 3.2% and 9.1% of the variance between applicants and non-applicants. So why do students decide to (not) enroll in entrepreneurship education?

With the data available, we cannot give a definite answer to that question. However, we can speculate about alternative effects at play. Bae et al. (2014) and Zhang et al. (2014) hypothesize about two mechanisms based on which students select into entrepreneurship education – ability or preference. Students’ perception of their ability can be proxied by their Entrepreneurial Perceived Behavior Control (PCB) and the preference by their Attitude Toward Entrepreneurship (ATE). We generally observe that applicants score higher in both dimensions compared to non-applicants (see Table 1), indicating that both effects are relevant. In contrast, our regression (see Table 3) indicates that Perceived Behavior Control (PCB) is the stronger predictor when controlling for covariates.

However, students with a high entrepreneurial intention, that is, they are eager to found, might decide against applying for an entrepreneurship education program for a number of other reasons. They may evaluate the required time investment of participating in the program against the expected returns, i.e., in how far they benefit from the program compared to alternatives. For students who already have a clear idea in mind, joining such a program might not be as attractive as pursuing the idea directly. The pedagogical setup of the program might be another relevant factor (Nabi et al., 2017). First, the type of entrepreneurship education (e.g., technology entrepreneurship, social entrepreneurship, etc.) may influence students’ decisions. Second, the duration of the program may play a role. Students who are in the final semester of their studies are likely less inclined to apply for a program that takes another three semesters and would prolong their studies. Third, the required workload, especially when conducted next to their regular studies, may discourage students who are struggling to keep up at university or who need to work next to their studies from applying despite an appetite for entrepreneurship.

Limitations

Given our method, the presented differences between applicants and non-applicants should not be interpreted as causality but as correlation. Additional limitations should be considered when thinking about the generalizability of the results to other entrepreneurship programs. In particular, we hypothesize that the magnitude of effects increases with program duration and intensity (in this case 45 ECTs and semesters). With growing program demands, students have to consider greater opportunity costs of participation (i.e., less free time, forgoing employment or internship opportunities, or reduced time to study). Thus, we would expect that only those students with a larger conviction to pursue entrepreneurship decide to apply. Hence, the results presented in this study might generalize only to comparable program ⁹ and not to typical one-semester-long elective courses embedded in undergraduate and graduate study programs. Future research may attempt to measure the magnitude of self-selection effects for such courses and programs.

First, the data was collected from the application process of one entrepreneurship education program targeting university students. While this avoids endogeneity – e.g., variation in entrepreneurial climate at different universities (Sancho et al., 2021) – and ensures comparability among participants, it limits generalizations regarding other contexts. Specific entrepreneurship programs are often very different, sometimes even so that the “content of syllabi of courses developed by entrepreneurship scholars differs to such an extent that it is difficult to determine if they even have a common purpose” (Henry et al., 2005, p. 103). Future work examining self-selection into entrepreneurship education in different contexts is required to confirm the generalizability of the results.

Second, the pedagogical context of the program should be given consideration (Nabi et al., 2017). We examined an entrepreneurship education program targeting individual university students from different disciplines interested in innovation, technology, and entrepreneurship. The relatively long program duration of three semesters combined with a relatively high workload in addition to students’ main study program likely attracts only a subset of highly ambitious students. This is also reflected in psychometric constructs, such as the high Need for Achievement scores. Thus, when relating these results to different forms of entrepreneurship education, it is important to consider the program’s character, its focus on individuals instead of startup teams, and the required workload. Self-selection effects into programs or courses that have a lower entry barrier and require less commitment, i.e., shorter program duration or lower workload, might be less pronounced. To understand the magnitude of self-selection effects, future work may replicate this study in different pedagogical contexts.

While our data collection allowed us to capture a group of students who considered an application to the program and decided against it, we still miss the group of students who might have been aware of the program but did not consider applying in the first place. Arguably, creating an account is already an initial show of self-selection. Thus, there exists a group of students that is even less interested in entrepreneurship education than the group of non-applicants examined in this study. A comparison of the Entrepreneurial Intentions of non-applicants with existing studies (Franke & Lüthje, 2004; Maresch et al., 2016; Von Graevenitz et al., 2010) indicates that non-applicants rank above the typical student population. This would imply that the actual differences between the true population of non-applicants and applicants could be even larger than the observed effects. While indicative, this comparison must be taken with a grain of salt. A direct comparison is tricky because of the different surrounding university environments (Sancho et al., 2021) and the different scales used. A representative survey would be interesting to establish a comparable baseline for future research.

Conclusion

In this article, we present evidence that students who self-select into entrepreneurship education differ along several psychometric constructs associated with entrepreneurship from students who do not. With this work, we confirm and strengthen prior work, particularly by using actual behavior to distinguish between both groups. By investigating multiple constructs frequently used to evaluate entrepreneurship education, we show how students interested in entrepreneurship education differ prior to applying to entrepreneurship programs. We highlight the risk of self-selection bias in most entrepreneurship education evaluation studies and discuss how to address implications for researchers and educators.