PhDs Aren't Starting Companies Like They Used To

PhDs Aren't Starting Companies Like They Used To

The burden of scientific knowledge and managerial complexity is crushing our best and brightest.

STEM PhDs aren't starting companies like they used to. Meanwhile, fresh PhD founders earn less than they did 20 years ago.

Cushy tech jobs are looking better and better to our most highly-educated as entrepreneurship loses its relative attractiveness.

Here's why.

Two decades of declining PhD entrepreneurship

STEM PhDs are creating fewer startups than they used to:

"We... show a decline over the past 20 years in both the rate of startups founded and the share of employment at startups by the highest-educated science and engineering portion of the U.S. workforce. The declines are wide-ranging and not driven by any particular founder demographic category or geographic region or scientific discipline."

Here's what the data shows: today, of STEM PhDs working in the private sector, only ~20% start companies, down from 30%+ in the late 90s:


"Figure 1 displays the large decline over the past two decades in the share of founders among PhDs in science and engineering. In 1997, 34 percent reported being a founder of a startup, but by 2017 this rate had declined to 21 percent, a decline of 38 percent."

Further, STEM PhDs are joining startups at lower rates in general, even outside of founder roles. Startups are much less attractive to PhDs than they used to be. This is true across demographic groups (gender, race, location, citizenship, etc.):

"Since 1997, the share of founders in these startups has declined by around 38 percent, not limited to any particular founder demographic or ethnic group or occupation, and this decline is widespread across regions of the United States. The share of workers at startups has followed the same path of decline."

"the employment share of science and engineering PhDs at startups is also falling over time. The downward trend is not driven by any particular category of PhDs. For example, figures similar to Figure 1 for males versus females, whites versus non-whites, California versus the rest of the U.S., and for U.S. versus non-U.S. born PhDs differ in levels but their dynamics are almost exactly the same as those in Figure 1... The dramatically falling share of founders and employment at startups raises the prospect of a drying up of high-tech, high-opportunity startups."

Why the change of heart among our PhD graduates? As the study authors speculate:

"A potential source of this decline is the exponential increase in the amount of scientific knowledge."

As the fields from which these PhDs come have become increasingly complex and extensive over time, the amount of scientific knowledge one must cram in to operate at the cutting edge of the field has exploded. Take any field and you see the same trend — becoming an "expert" is much, much harder than it used to be, and this burden of scientific knowledge weighs on potential founders.

Not only are fewer PhDs starting or joining startups, they're also waiting longer after completing their graduate coursework to found companies:

"Consistent with the burden of knowledge increasing for founders, the years of work experience among founders shows a steady increase. The regression results imply that the average founder had about 14 percent longer post-PhD work experience in 2017 than in 1997."

Here again we see the impact of the burden of scientific knowledge. With so much existing research and scientific knowledge out there to assimilate, PhD graduates are delaying entrepreneurship, taking more time to gather work experience before setting off on their own.

Learn to earn

Even worse, relative to their more experienced peers, newly minted PhDs who do found startups earn much less than they used to, making founding a company an even less appealing proposition:


"Figure 2 illustrates a pronounced decline over time in the earnings of less experienced founders relative to their more experienced peers. Here, we separate less and more experienced founders by the median number of years after PhD (13 years), although other reasonable cut-offs lead to similar results. Between 1997 and 2001, founders with post-PhD experience at or below the median earned more... but by the 2010s, the situation is reversed, with less experienced founders earning on average 30-40 percent less than other founders. Prior work experience apparently is becoming much more valuable for founders over time."

The figure compares the earnings of PhD founders with below-median and above-median work experience and demonstrates that this ratio has fallen meaningfully over the past two decades. Whereas the groups used to have earnings parity, today young PhD founders suffer a significant earnings disadvantage relative to their more experienced peers, making 30-40% less.

Absolute earnings for the highly-educated in general have risen over the same period, so this might not be terrible news. Young PhD founders could still earn more than they used to in absolute terms.

Not so. Earnings for PhD founders with below-median work experience declined in real (inflation-adjusted) terms, from $73K in 1997 to only ​$58K in 2017, a 20%+ pay cut:

"The earnings of less experienced founders declined not just in relative but also in absolute terms. The average inflation-adjusted earnings of founders with below the median post-PhD experience were $72,616 in 1997, whereas 20 years later their earnings were ​$57,517, a decline of more than 20 percent"

The earnings of young STEM PhD founders are their lowest level in more than 20 years.

What if we combine both inexperienced and experienced founders and look at the overall trend in PhD founder earnings? Surely this has trended positively?


"Founders' earnings decline on average by about 1.6 percent per year (column 1). However, this is offset by an opposite time trend in returns to experience. The mean number of years after founders receive a PhD is 15.9 years; hence... at the mean work experience, the negative baseline time trend is completely offset."

OK, this takes a bit of unpacking. Earnings for less experienced PhD founders have fallen, while more experienced founders have seen rising earnings. The breakeven point is ~15 years, which is to say founders with fewer than 15 years of post-PhD experience have seen declining real earnings over time, while founders with more than 15 years of experience have seen earnings growth. It just so happens the average PhD founder in the survey had 15.9 years of post-PhD experience, so the two stories roughly balance out.

The good news: overall PhD founder earnings aren't declining! The bad news: PhD founder earnings aren't rising either!


The craziest thing about all this? PhD employees are doing much better:

"In stark contrast to founders, workers' real earnings grow over time, although the increase is relatively small, 0.4 percent per year... The trend toward increasing returns to experience is much weaker among workers than among founders..."

Again, let's unpack. STEM PhDs who work at established companies rather than startups have seen their earnings rise across the board, regardless of experience level:


As discussed earlier, founding a company as a young PhD is less attractive than it used to be. While young PhD founders used to make slightly more than young PhDs at established firms, this relationship has completely reversed, with inexperienced PhD founders now earning significantly less than their less entrepreneurial associates:


In summary, new PhD founders get the worst deal, earning less than both their equally and more experienced peers:


That was all a bit complicated, so before moving on, let's summarize the trends.

  • Previously, experience mattered most within established companies, but over time the "gains from experience" grew much more for founders than workers
  • Experienced PhD founders went from being no better off than their less experienced peers to earning significantly more, whereas the gap between experienced and inexperienced PhD employees stayed relatively stable.
  • Inexperienced PhD founders used to make more than inexperienced PhD workers, but this relationship has flipped
  • Inexperienced PhD founders are now the worst-paid PhDs

The division of tasks

So far we've established:

  1. PhDs are founding far fewer companies, and

  2. When they do found companies, young PhD graduates earn much less than they used to and significantly less than peers working at established companies

Not good. But it couldn't get any worse for our youthful, bright, wide-eyed PhDs could it?

Turns out, it can.

"established firms have an advantage over startups in creating a division of labor in R&D... by introducing more hierarchical layers, reducing knowledge workers' span of control, and allocating more experienced workers to positions with greater managerial responsibility. Further, established firms compensate workers for performing more R&D tasks and supervising more individuals. These developments are not seen among founders. The differences follow from the natural limits imposed by running a small firm with less division of labor and a high amount of multitasking by the founder. The largest firms are even more active in reorganizing job tasks, increasing the depth of hierarchy at twice the rate of all established firms."

Big Tech and other large companies have become more attractive places to innovate for our most well-educated workers. Large companies have better dealt with the accumulated burden of scientific knowledge by enforcing a "division of labor" among employees. Large companies form "knowledge hierarchies", such that no one person needs to know everything, and information can be aggregated up the chain in a rationale fashion. This lets organizations to scale to meet the needs of cutting edge scientific research and development, which increasingly incorporates impossibly vast sums of knowledge.

On the other hand, startup founders do not have armies of willing cadets to offload and delegate to. They are the captain of the ship and also it's most valuable crewmember, so as the burdens of both scientific knowledge and management complex grow, it all gets placed on their shoulders.

The job of a founder has gotten harder — running a startup today requires more management and R&D activities than it did 20 years ago. Quantified, this is about 15% more "tasks", with R&D tasks growing by about 50% and management tasks growing 5%:


"The average number of all tasks for founders increased by about 15 percent from the beginning to the end of our sample, a statistically significant difference... Furthermore, although the number of both R&D and management tasks increased, the increase is more pronounced in R&D, for which it rose by more than 50 percent from 1997 to 2017... As a result, R&D tasks that comprised about 25 percent of all tasks conducted by founders in 1997 increased to 34 percent of all tasks in 2017. As can be seen in Figure 3, this was not accompanied by any decline in management tasks, so the founders had to shoulder the burden of doing more R&D tasks while also running the same or more administrative tasks."

PhD workers at established firms also saw their total tasks grow, but to a lesser degree. R&D tasks for workers only grew 12% over the same 20 year period, and management tasks did not increase at all:


"Workers were also affected by the need to perform more R&D tasks, as the increase in the number of R&D tasks from 1997 to 2017 is about 12 percent. Notably, however, as the figure reveals, workers did not need to handle more management tasks, which remained relatively flat for them."

What gives? It goes back to the division of labor point I mentioned earlier. Let's start with R&D:

"The number of R&D tasks is rising, but established firms take advantage of a division of labor and knowledge hierarchies"

The growing burden of scientific knowledge increases the number of R&D tasks that founder must occupy themselves with. With little hierarchy or staff to offload this work, founders have no operating leverage. Conversely, large, established firms can shift much of the burden to armies of researchers, more efficiently dealing with the growing stack of research output.

"We find that the number of different R&D tasks has increased more for founders than for workers. And the returns to experience have increased over time for founders but not for workers, highlighting the increasing need for a single person—the founder—to cope with the burden of knowledge in startups. Workers at established firms have, instead, comfortably narrowed their span of control, employed more people indirectly under their control to support their work, and kept administrative duties low. They are also better rewarded for taking on more diverse work and managerial responsibilities than founders. ...established firms have coped more effectively with the increasing burden of knowledge in science by better utilizing the division of labor in innovative work through reorganizing tasks and hierarchies."

Individual workers at established firms have limited scope, while the remaining load gets shared with other team members and subordinates. Further, the firm captures the gains from the successful division of labor and then shares this with employees in the form of better pay for the managers who corral these efforts internally.

"Running a startup might constrain founders' ability to organize its hierarchy efficiently, at least until it has succeeded in growing well beyond its initial size. As science accumulates more knowledge, we would therefore expect PhD founders to have to take on more R&D tasks."

Founders have no ability to pull off similar organizational tricks until their startups reach meaningful scale. Thus, as scientific knowledge has accumulated, they've taken on the additional R&D work themselves.

This has implications for startup management too:

"Also, founders had to deal with significantly more management tasks than workers in terms of levels: about 30-40 percent more at the beginning of the study period, increasing to 50 percent more at the end. The explanation for this difference likely lies in how the two types of firms differ in their organization of work..."

Founders and startups cannot afford the overhead that comes with increasing layers of professional management. This is why startup's delay building executive teams until at least a few years into a startup's lifecycle. Large organizations can and do invest in these management layers, gladly taking on the extra expense:

"... the firm responds by increasing the number of layers... and... [allowing] greater job specialization. Increasing the number of layers of management adds a fixed cost of operations... [Larger] firms are more likely to become hierarchically taller by adding more layers of management... as they can more easily absorb the added fixed cost. Founders at startups appear not to have recourse to this mitigation strategy."

The study's authors quantify these additional layers by tracking two metrics:

  • the number of individuals directly supervised by PhD founders and workers, and
  • the number of individuals indirectly supervised (i.e. supervised by one's own direct reports)

The directly supervised corresponds to the managerial burden that each PhD-holder has (their span of control), while the indirectly supervised tells us how much managerial leverage each person has (the depth of hierarchy). The theory would suggest that workers within organizations should have fewer direct reports over time and more indirect reports, as organizations build out these knowledge hierarchies.

And that's exactly what we see:


  • Workers with PhDs saw the number of employees directly managed decline over time while the number of employees indirectly managed increased over time. In others words, the proliferation of middle management reduces the management burden for any individual worker but increases overall managerial leverage.

  • Founders on the other hand saw no statistically significant trend in their span of control or depth of hierarchy. The organizational structure of startups has not evolved to keep up with the increased burden of scientific knowledge.

"workers [perform] fewer R&D tasks as they age... the span of control decreases for workers at established firms... However... hierarchies deepen over time for workers at established firms... Together, the results suggest that established firms cope with the increasing burden of knowledge on their workers by introducing additional layers of hierarchy, while simultaneously reducing the number of employees who report directly to managers"

Doing *too many* things that don't scale

We need to think harder about making knowledge work... work.

In summary:

"Our findings suggest that if the goal is to restore business dynamism in the high-tech sector, alleviating the burden of knowledge should be front and center in the strategy to attain it."

The declining in PhD entrepreneurship mimics a broader drop in startup formation chronicled elsewhere. But PhD founders are special... and so these trends are especially worrying.

STEM PhDs have founded some of the world's most successful and impactful enterprises. Among them: Google, Intel, VMware, and others.

Ironically, however, today these same companies gobble up the brightest minds we have, discouraging or delaying their own entrepreneurial pursuits.

In Silicon Valley we talk a lot about management practices to help founders and executives "scale." But deeply technical founders face a separate but equally important problem: scaling knowledge.

Thanks to Thomas Astebro, Serguey Braguinsky, and Yuheng Ding, the authors of the study from which much of this essay is derived.