Teaching Innovative Entrepreneurship
Boosting innovation by teaching people to be tech entrepreneurs?
This article will be updated as the state of the academic literature evolves; you can read the latest version here. You can listen to this post above, or via most podcast apps here.
Before we start: Michael Magoon has assembled a big list of substacks related to progress studies that might be of interest to some of you (What’s New Under the Sun is on there!).
A lot of particularly interesting innovation happens at startups.1 Suppose we want more of this. One way we could try to get more is by giving entrepreneurship training to people who are likely to found innovative startups. Does that work?
Entrepreneurship Training In General
There are lots of programs to teach people about entrepreneurship: classes, boot camps, extracurricular programs, etc. Entrepreneurship doesn’t always mean innovation, in the sense we’re thinking about here. Entrepreneurs can start dry cleaners as well as AI startups. But if entrepreneurship training creates more founders of all types, then it’s not unreasonable to think it also helps more people found innovative startups as well as other kinds of new businesses. Fortunately, there is by now a quite large literature that attempts to assess the effects of entrepreneurship training. That literature tends to find it works, at least on average.
For example, Martin, McNally, and Kay (2013) identify 42 studies that assess the impact of entrepreneurship education or training published over 1979-2011. They specifically looked for studies that tried to link education/training to eventual business outcomes of their ventures, or to pre-venture proxies such as intent to start a venture, relevant knowledge and skills, etc. In this sample of 42 studies, six randomized participants into training and compared people who were given the training to those who were not. Another five compared participants to non-participants, but without randomizing them into programs. The rest compared participants before and after training.
Some of their studies find entrepreneurship programs increase the interest of participants in founding a venture, and improves the business outcomes of those who get this training. But other studies find no impact - or even a negative impact. The variation isn’t necessarily that surprising as we’re mixing up all sorts of different programs across many different countries. To sort through all the different results, Martin and coauthors look to see if there is any general tendency one way or another.
To do that, they standardize the results of these studies with a measure of correlation between outcomes of interest and completion of entrepreneurship programs. They then compute the average correlation across all the studies, weighted by sample sizes. In general they find a positive, though small, correlation between training and various outcomes related to entrepreneurship. This persists when they restrict their attention to the subset of studies that use the most rigorous methods of assessment, such as randomization.
The studies covered in Martin, McNally, and Kay (2013) all date to 2011 or earlier. Carpenter and Wilson (2022) perform another review, focusing on studies published in 2015 and later. In their review, they identify ten “high quality” studies. Carpenter and Wilson don’t do a quantitative meta-analysis, but surveying these studies, they argue the results are broadly consistent with Martin, McNally, and Kay’s earlier meta-analysis: these ten studies tend to find entrepreneurship training is usually, but not always, associated with proxies for entrepreneurship, like intention to start a business or relevant knowledge. Not many of these studies consider the business performance of startups, but when they do, they also tend to find those who receive training have more successful ventures.
Entrepreneurship Education for STEM Undergrads
These review articles find that entrepreneurship training has a small and variable effect, but a general tendency to nudge participants towards starting a venture and succeeding a bit more when they do. But most of the programs studied are not specifically targeted towards encouraging the kind of high-impact innovative startups that I’m interested in, in this post. But there are a few studies that seem more directly relevant. Let’s start by looking at two that focus specifically on entrepreneurship training for science and engineering undergraduates. It seems plausible ventures started by them will be more likely to create new technologies.
Souitaris, Zerbinati, and Al-Laham (2007) is one early example, focusing on the effects of entrepreneurship programs on science and engineering students at at two European universities.2 One thing they look at is whether participants in entrepreneurship programs express more interest in founding a venture, after they finish the training.
The study is built on a survey of 250 students, who are surveyed at the beginning and end of the 2002 Spring semester. During that semester, about half the study participants participated in entrepreneurship programming (some of which was mandatory and some of which was voluntary), and about half did not. They then compare how student views (on these surveys) changed over the semester. They find, for example, that intention to start a business over the semester increased more for the students in the entrepreneurship programming, as compared to the control. The effects are statistically significant but small; on a seven point scale (i.e., 1 = not at all interested in starting a business, 7 = very interested), intention to become self-employed increased by an extra 0.1 for people who took entrepreneurship programming during the semester.3 But some other measures hypothesized to be associated with becoming an entrepreneur didn’t change at all, such as a measure of attitudes towards self-employment or actually taking actions to launch a startup.
Eesley and Lee (2020) get similarly ambivalent results in their study of the Stanford Technology Venture Program. This is an entrepreneurship center, founded in 1995, to serve students at Stanford’s engineering college with a focus specifically on high-tech entrepreneurship. Eesley and Lee want to use survey data from Stanford alumni to assess the impact of the program on entrepreneurship by participants: do alumni who participated in Stanford Technology Venture programming go on to found businesses, and are those businesses more successful if they do? Indeed, they do find students who said they participated in this program in alumni surveys were about 10 percentage points more likely to also state they had subsequently worked in a startup.
But of course, that might have nothing to do with what they learned from the entrepreneurship programming, and everything to do with the fact that people who are already interested in entrepreneurship are the kind of people who participate in entrepreneurship programming. If we wanted to tease out the impact of the training itself, we would ideally make the programming compulsory for half the students and block the other half from taking it. That didn’t happen, but Eesley and Lee exploit the fact that the program started in 1995 and was not easily available for students outside the engineering college. They can back out the implied impact of the program by comparing entrepreneurship outcomes between engineering students who graduated before and after the program was created, and by comparing engineering and non-engineering students (who didn’t have access to the program).
When they do that, they now find that the program mostly had no impact on any of their measures of entrepreneurial outcomes. Across 21 different entrepreneurship-related measures, they detect an effect of the program that is statistically distinguishable from zero at conventional levels for two items.4 Otherwise, students who were able to attend the program (by virtue of the years they were at Stanford and the college they belonged to) were not more likely to start ventures than students who did not have access to the program (because they were engineers before the program was offered or not engineers while it was offered).5
Training for Budding Tech Entrepreneurs
So far we’ve got evidence that entrepreneurship training works on average, a bit, when you aggregate over a lot of studies, but also that when we zero in on science and engineering undergraduates in two studies results are not particularly encouraging. That said, given the efficacy of these programs is noisy, we should be cautious about making too much out of two studies.
One potential issue here is that maybe most science and engineering students do not plan to start ventures (probably less of a concern at Stanford, which is in the heart of Silicon Valley). If most students are not going to start ventures, then it will be harder to detect the effects of an entrepreneurship program, just as it’s hard to assess the impact of a vaccine on a disease that almost no one gets. We might expect to see stronger effects if we focus on a subpopulation that has a higher probability of becoming an entrepreneur. This is what Lyons and Zhang (2017) does.
Lyons and Zhang look at the impact of The Next 36 program, an intensive Canadian entrepreneurship program for undergraduates (currently 4 months part time plus 4 months full time) focused on technology entrepreneurship. Over 2011-2015, around 300 people apply for the program per year, and 70 finalists are selected. These finalists go through a second-round screen where they are interviewed by a panel, usually consisting of entrepreneurs and Next 36 founders. Eventually around 36 students are selected per year, based on their interviews, their applications, and program goals to achieve gender and educational background balance.
Lyons and Zhang focus their attention on program finalists, and compare finalists who got in to the program to those who did not. This has the advantage of focusing us on a sub-population who is already interested in becoming an entrepreneur: they applied to the program and were selected as a finalist. That might make it easier to detect the impact of the program, as compared to a population where entrepreneurial activity isn’t common.
But it does introduce a different problem. What if the selection committee is really good at identifying successful future entrepreneurs? In that case, any difference in the entrepreneurial outcomes of the accepted and rejected finalists might be down to the acumen of the selection committee, rather than anything learned in the program. That wouldn’t tell us much about how well we can increase the number of innovative entrepreneurs with training.
Lyons and Zhang do a few things to deal with this. First off, they have data on the applications of all the finalists. So in one analysis, they match each program participant with a rejected finalist with a similar profile: same gender, similar interview scores, similar prior entrepreneurial experience, college majors, and university ranking. For these matched pairs, they then compare the diverging fortunes of the finalist who completed the program to the “twin” finalist who did not. They find finalists who got in to the program were about 20 percentage points more likely to work in some capacity at a startup, compared to an overall average across all rejected finalists of about 45%. This is a bit less than what they find when they don’t match finalists, suggesting the program is selecting people more likely to be entrepreneurs, but not by much. For example, just comparing rejected and accepted finalists without doing any matching, they find finalists who do the program are 24 percentage points more likely to work in a startup, compared to 20% in their matched estimate. They also perform a statistical analysis suggested by Oster (2016) to assess the likelihood that differences between the accepted and rejected finalists that are hidden to Lyons and Zhang but observable to the selection committee might be driving the results. This statistical exercise also suggests selection effects are not driving their results.
Given that it’s hard for anyone to spot promising new entrepreneurs (or else venture capital wouldn’t be so risky), it seems plausible to me that accepted and unaccepted program finalists are unlikely to be very dissimilar. If so, Lyons and Zhang’s work shows The Next 36 program is pretty effective at creating technology entrepreneurs. In addition to the increased probability of starting a venture, they find these ventures are more likely to survive, raise external funding, and be classified as technology ventures (rather than being in other sectors, like entertainment or business services).
In fact, if we return to Stanford, we see some evidence that a similar program there also obtains stronger results. In a secondary analysis, Eesley and Lee also examine the effects of the Mayfield Fellows Program, which is an intensive entrepreneurship program offered by the Stanford Technology Venture Program to a select set of 12 undergraduate students per year. Eesley and Lee have data on 268 of these fellows. Unlike Lyons and Zhang, they don’t have data on the runner ups, so instead they compare the fates of the fellows to Stanford alumni who participated in various entrepreneurship courses, or graduated in the top 5% of their class. The Mayfield Fellows are significantly more likely to be involved in entrepreneurship after graduating.
A Speculative Synthesis
So across a variety of studies, we have evidence that entrepreneurship programs in general - the kind that help someone start a business of some kind, whether or not its based around invention - modestly increase entrepreneurship and modestly improve entrepreneurial outcomes. But there’s a lot of noise, and many studies find mixed or negative results. Indeed, when we zero in on two studies looking at entrepreneurship programming for science and engineering students, at best we get a few indicators of impact, alongside many others that show none. However, when we look at two highly selective programs for technology ventures, results seem a lot more compelling (though we have challenges related to disentangling selection from training), at least among the population of students who are already interested in entrepreneurship.
What’s going on? As with all things academic, “more research is needed.” And if you know of more work relevant to this topic, please reach out - this is a living literature review and I’ll update this article. But I’ve written about a few other strands of research related to entrepreneurship, and to close I want to draw out a few connections.
First off, in Entrepreneurship is Contagious and The “Idea” of Being an Entrepreneur, I argued that an important driver of entrepreneurship was whether individuals took entrepreneurship seriously as a possibility for their life. I also argued that social role models were particularly effective channels for communicating or transmitting this idea. This social role model component seems very hard to systematically measure, and maybe helps to account for the variable efficacy of training programs.
Additionally, in those two pieces I argued that once you have the idea that entrepreneurship is a live possibility for you, additional reinforcement of that idea isn’t very impactful. For example - in general, the decision to become an entrepreneur seems influenced by social role models like your peers, coworkers, or mentors. But one group who seem more immune to this are the children of entrepreneurs. Possibly that’s because these folks already thought of themselves as potential entrepreneurs, so they didn’t need to have their beliefs about what’s possibly updated by social role models.
In the context of entrepreneurship training, that might mean that entrepreneurship programs where people opt in are not going to be that effective, because the people who opt in are already considering entrepreneurship as a possibility. Indeed, in those posts I argued that for people already interested in entrepreneurship, learning more about it might make them lesslikely to become entrepreneurs because it can correct some of their misperceptions. I argued this might be why a famous study (Lerner and Malmendier 2013, discussed more here and here) found Harvard Business School graduates who had more ex-entrepreneurs in their classes were less likely to later become entrepreneurs. The fact that entrepreneurship programming might lead people who were initially optimistic about entrepreneurship to reassess it might be another reason the effects are often ambivalent.
Lyons and Zhang provide some additional evidence that entrepreneurship programs might have this effect of warning people off bad ideas for businesses. In an effort to understand what The Next 36 program does for its participants, they give finalists a survey before-and-after the program, and then compare responses to see how participant responses change after completing the program. They find little or no change in the survey responses related to lots of things you might think would matter: things like pitch ability, risk aversion, perceived likelihood of raising $10mn, etc. But participants do rate themselves more capable of evaluating start up opportunities by the end, suggesting participation in the program gave them the tools to better identify bad ideas.
Eesley and Lee also provide some confirmatory evidence that entrepreneurship programming might dissuade the entrepreneurially-curious off unpromising ventures. In addition to their study of the Stanford Technology Ventures program, their study also examines an entrepreneurship program in Stanford’s (graduate) business school in addition to the program for engineers. Using the same analysis strategy, they find that entrepreneurship program significantly reduces the probability business students become entrepreneurs - though the ones that do achieve better outcomes.
So one reason the results of entrepreneurship training isn’t stronger might be because for students not seriously interested in entrepreneurship, one important factor for whether they begin to take it seriously might be if they meet a role model like them, which is hard to measure and program for. And for the students already interested, they might actually learn some of their entrepreneurial ideas are not very strong, dissuading them from becoming (bad) entrepreneurs. The end result is a mushy effect on the probability of becoming an entrepreneur. But maybe a mushy result is what we want, especially if that’s because the people dissuaded from entrepreneurship would have been more likely to fail.
(This is not to say there aren’t also a lot of other factors - I just can’t tie those ideas as well to previous work on New Things Under the Sun.)
What do we make of the results on more selective programs? If highly selective programs do have more success (a conclusion I’m hesitant to hold firmly without more studies), one possible reason might be because of their inclusion of more intensive mentoring. The Next 36, for example, pairs participants with mentors who are usually experienced entrepreneurs and business leaders. The Mayfield Fellows program also offers alumni mentoring opportunities. In Teacher Influence and Innovation, I reviewed a variety of evidence that training under exceptional innovators exerts a significant influence on student interests and outcomes. Maybe it’s the same for innovative entrepreneurship?
We get a glimmer of evidence on this from Wallskog (2022), a paper I discuss in more detail here. Wallskog isn’t looking at entrepreneurship training; she’s instead looking at social role models. She shows that people who have more ex-entrepreneurs in their workspace are more likely to subsequently start businesses, and also more likely to say role models were an important factor in that decision. She also finds that people who worked with more successful ex-entrepreneurs had slightly better performing businesses than those who worked with less successful ex-entrepreneurs. That’s at least one indicator that entrepreneurial mentorship (by talented entrepreneurs) can improve performance, as we see in other forms of innovation.
But of course mentorship is not the only possible explanation. It may also be down to the fact that intensive programs do a better job building a students network. In fact, when Lyons and Zhang compare before-and-after survey results for their Next 36 participants, they find that the biggest differences are related to networks. Participants say they know more potential investors and founding partners after the program than before. Lyons and Zhang also find accepted finalists have many more connections on LinkedIn than rejected finalists, which is a nice check on the survey evidence.
Taken all together, I’m cautiously optimistic. In aggregate, entrepreneurship training seems to have effects we would expect, though perhaps smaller than we would like. But one reason the effect might be smaller than we want is because entrepreneurship training is actually good at teaching people about what bad startups look like, dissuading some people from diving in. That’s a case where a weak result is not such a bad thing to see, since it mixes in some positive encouragement for people not originally considering entrepreneurship at all, and some negative encouragement for people on the verge of starting weaker businesses. Meanwhile, if we zero in on highly selected samples that are probably more likely to succeed, there’s some evidence these programs increase both the rates of entrepreneurship and the success of these entrepreneurs. Possibly that’s because these programs connect budding entrepreneurs with mentors, investors, and potential future partners.
Thanks for reading! As always, if you want to chat about this post or innovation in generally, let’s grab a virtual coffee. Send me an email at matt.clancy@openphilanthropy.org and we’ll put something in the calendar.
See my posts The Size of Firms and the Nature of Innovation and Big Firms Have Different Incentives. See Kolev et al. (2022) and Kerr, Nanda, and Rhodes-Kropf (2014) for other proposals for why startups are special.
Not named, but in London, UK, and Grenoble, France.
Bear in mind this small effect is over just one semester though; the effects might be larger over a full undergraduate term.
Specifically, participants who go on to found ventures, found their ventures more quickly if they participated in the program.
As an aside, Eesley and Lee do find much more statistically significant results for an entrepreneurship program for students at Stanford’s business school (not generally available to undergraduates). For business school students, the availability of entrepreneurship courses increased the performance of new ventures, but decreased the probability people would become entrepreneurs themselves. Eesley and Lee argue it helped people form more realistic expectations, while also giving them useful skills.