What should new-age research organizations be optimizing for?
What should new-age research organizations be optimizing for?
This post is the first of my new Substack, the Engineering Innovation Newsletter. Every piece in this newsletter shares a common goal: to propose concrete ways forward for those looking to help shape the future of technological research and innovation. For more context, check out my introductory post here.
Enjoy and please subscribe if you’re interested!
The modern research ecosystem emerged largely in the wake of World War II and has sparked a massive increase in funding for basic research, patenting, and the production of scientific papers. However, it hasn’t produced innovations that have increased productivity and the standard of living for the average American at anywhere near the rate of the inventions of the first half of the 20th century. Today, I’ll explain why that might be the case and what there is to do about it.
In this piece, we’ll explore:
How the modern research ecosystem came to be
What the current system does well
What the current system does not do well
Why the computer age had only a moderate effect on productivity
Some reasons to be optimistic
And concrete examples of metrics new research organizations can use to overcome the issues that the current system runs into
How the modern research ecosystem came to be
America was, for decades, viewed by many as the R&D lab for the rest of the world. America's academic research ecosystem is a major part of that; with successful technologies like personal computers, the internet, and countless medical and pharmaceutical innovations arising from basic research from these institutions, it's not hard to imagine why. But all of that success doesn't come for free. We spend a pretty penny funding this research.
Each year, the US Government hands out over $30 billion in research grants via the NIH and just under $10 billion via the NSF. Until recently, this kind of spending dwarfed what any other country was willing to spend. And, as you can see below, the magnitude of this investment has increased dramatically in the decades since 1950.
The NSF was established in 1950, with the NIH coming decades later. Prior to World War II, the government primarily spent money on specific “research projects” that were directly relevant to the government’s interests, such as land surveying. Other areas of pre-WWII research included research related to public health questions, work done by the Bureau of Standards to help US businesses, and the beginnings of defense research during the First World War. The massive increase in funding that we observe in the chart above has helped create the modern research ecosystem, with a large focus on basic research, that we know today.
The push to create an organization like the NSF, which tends to focus on basic research, was a direct outcome of the World War II victory in which American research and development on things like radar and nuclear energy, headed by academics working with the military, was seen as pivotal in winning the war. Vannevar Bush, wartime Director of the Office of Scientific Research and Development and former Dean of Engineering at MIT, is seen as the most instrumental person in pushing for the creation of the NSF. He believed, as many do today, that basic research is "the pacemaker of technological progress". Underlying most technological innovations is some piece of academic research. So, it was believed, if we grow the body of scientific research and insights to draw on, then technological growth should follow.2 3
What the modern system does do
Looking at the raw numbers, our system has done an astounding job of growing the body of research.
The growth rate of patents is skyrocketing. In 1950, the US granted only 43,000 utility patents, compared to an astounding 350,000 in 2020.
The growth in the raw number of pieces of scientific research each year has been even more impressive. These numbers can vary depending on the data source and method of calculating, but by any metrics the growth in yearly publications is massive. The graphic below, built using data from four major academic bibliographic databases, demonstrates that there were only about 27,500 publications in the year 1880. The year 1950 had about 200,000. The year 2018 clocked in with just over 5 million publications.4
The growth in publications is so massive that some academics argue that it is too big — researchers can’t possibly keep up.
What the modern system does not do
The modern system of research has not produced the kinds of innovations that raise wages for regular people. And, while money can’t buy everything, it can buy things like food, housing, spare time, less stress, and is generally a great proxy for standard of living.
GDP has continued to grow in recent decades, and usually, you would expect hourly compensation to grow along with it. In the following figure, you can see that this relationship does tend to be true throughout the early 1900s. However, you can also see that the gap between the two lines below has been growing since about 1970. That’s not good for regular workers.5
Corresponding Growth in Real GDP and Real Hourly Compensation
In previous eras, growth in real income grew just as fast or faster for the bottom 90% as it did for the top 10%. This has not persisted into the modern era. In fact, in the period since 1970 real wage growth for the bottom 90% has tended to shrink.
Growth Rate of Real Income: Top 10% vs. The Rest
Why is this happening? What changed?
In 1987, MIT economist Robert Solow said, “You can see the computer age everywhere but in the productivity statistics.” And while there was a brief where that was not the case, his quip is still largely true today. The current system of research funding has not produced inventions that make ordinary people and workers more productive at anywhere near the rate of the inventions of the early 20th century. It is those productivity increases that, usually, drive up compensation for ordinary workers.
This argument is best laid out by Robert Gordon, an economist from Northwestern University and former grad student of Robert Solow, in his now-famous book The Rise and Fall of American Growth. The most important metric utilized throughout the book is Total Factor Productivity (TFP). TFP, one of three components that make up growth rates in output, is used by economists to track how quickly aggregate output is growing relative to the growth of labor and capital inputs. Essentially, TFP tracks how much of the total growth rate in outputs per dollar seems to be due to technological improvements.
To make this a little more concrete: if I invent a fancy new machine that can produce 50 Wonka bars as quickly and cheaply as the current machines can produce 25, that is the kind of invention that would make TFP go up. These are also the kinds of inventions that, throughout the early 20th century at least, tended to increase wages for the factory workers, the factory owners, and the inventors. The factory owner got more money due to the increased efficiency, the inventor got money for creating such a great machine, and the workers got more money in practice both because the price of goods like Wonka bars went down and the fancy new Wonka bar machine needed a more-skilled, highly-paid worker to run it.
It would not be ridiculous for somebody to say that the American dream was built on the back of exceptional growth rates in TFP.
The following chart breaks down the average growth rate in output per hour over three notable periods: 1890-1920, 1920-1970, and 1970-2014. As you can see, the yearly growth rates in the period of 1920-1970 were quite exceptional. That is not because of exceptional performance by education and capital deepening, the other two components that contribute to growth in output (which we don’t need to talk about now). The high growth rates in output per hour were almost entirely due to high rates of TFP, the metric that tracks growth in output due to technology.
Average Annual Growth Rates of Output Per Hour (and its components)
You might be thinking: “Well, it seems like our productivity rate was still sky high for the first 20 years under this current system of research—since the NSF was created in 1950. That’s pretty good, isn’t it?” And that would be a very fair question.
It would also lead one to believe that either:
This modern system of research funding did extremely well for two decades and just lost its way at some point, or
Maybe the research ecosystem had nothing to do with the growth rate anyway.
And that might be the case...if there wasn’t a widely accepted piece of evidence among the economists who study this area that explains this two-decade gap. And that piece of evidence is that it generally takes about 20 years, on average, for a technology to become implemented and scaled to the point where it positively impacts productivity (Matt Clancy has a fantastic piece on this one figure alone).6 That would mean that, roughly speaking, our steady stream of innovations that was underlying this exceptional growth in TFP dried up right around 1950…
And, since I am writing for a crowd who tends to be on the cutting edge of certain fields, I imagine many who work in the applied space of their field will be saying, “That doesn’t square with my experience at all! Work in my field tends to build off of academic work that is from the past ten years almost exclusively.” And, if that is you, then it is very likely because you work in a field like computer science, nano-science, computational biology, or material science. Every field has its own average lag from discovery to implementation. Those specific fields tend to have an average gap that is as much as two to three times shorter than the average piece of academic STEM work.7 So, if you work in a field like one of those, it is not surprising that you tend to deal with academic literature that is something like five to ten years old on average. But the average is 20 years.
With this 20-year lag explained, let’s circle back to the TFP point. High TFP growth also tends to take into account inventions that are intuitively good for well-being, sometimes known as standard of living increases. And the 1920-1970 period was like...REALLY good for standard of living increases. If those high TFP growth numbers for the 1920-1970 period seem too good to be true, here are some of the things contributing to them.
Infant mortality rates plummeted. Even small effects on infant mortality tend to have massive impacts on standard of living even in relation to other medical advances since, for each infant life saved, you could generally expect that child to live another 60+ years (depending on the decade).
Plummeting Infant Mortality Rates
We largely conquered the eight contagious diseases that were the most common causes of death for Americans.
Deaths from Eight Most Common Contagious Diseases Vanished
Yearly output grew and grew while average hours worked shrank and shrank. The gray line in the figure below shows that laborers who worked in non-agricultural, private industry worked about 58.5 hours per week in the 1890s and that number shrank to as low as 41.1 in 1950. These numbers continued to shrink thereafter, but at a much more modest rate.
(As for the white line, do not pay much attention to it and its post-1950 decline. This is largely due to the entry of large numbers of part-time female employees into the labor force among other factors that are not relevant to today’s piece—but are quite interesting.)
Decreased Hours Worked Per Week
“Conveniences” such as running water, plumbing, and electricity went from rare/virtually non-existent to almost ubiquitous over the period. Each one of these was invented in the late 1800s.
Diffusion of Modern Conveniences
Massive time, labor, and health savers such as automobiles, washing machines, and refrigerators also came into being and widespread use in this exceptional time period.
Diffusion of Modern Appliances
The list of game-changing technologies that were invented and disseminated is long and impressive. Some include:
Running water: Protected the family from waterborne diseases and revolutionized the daily routine of the family more than you can imagine. Before this invention, “washday” could take the women of the house as much as an entire day of hard labor. This would happen about once a week. Women were also burdened with carrying out the chamber pots, dirty dishwater, and cooking waste. Baths were seldom taken due to the extreme effort required.
Electrified homes: Turned the home into a power source that could support some of the game-changing “appliances” and “conveniences” in the above charts.
Electrified factories: Turned factories into cheaper, more efficient power sources to power all variety of electric machines that would be invented to serve the interests of factory owners and make workers more efficient and well-paid.
Electric light bulb: Freed many humans from being beholden to the hours of the sun.
Electric elevator: When the electric elevator allowed buildings to extend vertically instead of horizontally, the very nature of land use was changed and how “dense” urban density could be was massively increased.
Automobiles: Cars were more than just a massive improvement over horses. When cars replaced horses as the primary form of transportation, society unburdened itself from the need to allocate a quarter of its agricultural land to produce horse food—not to mention the massive labor force no longer required to remove their waste from city streets.
Mason jars/fridges: In the late 1800s, households spent half of their family budgets on food. The advent of mason jars and fridges not only made the food safer, but also were two of the many major inventions, all from this era, that helped reduce the average household budget spent on food from over 50% to 16% in 1970. And all the while, the variety and quality of the food was increasing.8
Sewing machines: Heavily eased the burden of women in making clothing in the intervening period before department stores and mail-order catalogs became ubiquitous (these stores and catalogs were also invented in this period).
X-rays and modern antibiotics: Also invented and implemented in this time period.
In my opinion, that’s an extremely impressive list. It puts in context charts, like the one below, showing outlier-type growth in output along with the reduction in hours worked and increased standard of living from 1920 to 1970. The difference these inventions made in the day-to-day lives of people is truly remarkable.
Growth Rates of Output per Person, Output per Hour, and Hours Worked per Person
The small effect the computer age did have
The Third Industrial Revolution did have its day in the sun. When Solow went on the record saying one can “see the computer age everywhere but in the productivity statistics,” an economist named Paul David replied in return: “Just wait.”
TFP Growth’s Brief Revival
David believed that rather than computers underperforming their expected impact, we were just living through a lag of the type that we discussed earlier. He turned out to be right, at least to some extent. The growth rate of US aggregate productivity shot up from 1996 to 2004 and then resumed its previous slow rates. Eight years is certainly nothing to scoff at — but it’s nothing compared to the 50 years of compounded growth experienced from 1920 to 1970.9
There is some chance that maybe if we just hold our breath and keep doing what we’re doing, it will turn out that our current system of innovation is actually doing great and we just need to wait another two decades to see it. But if given a 50/50 bet on that, that’s probably not where the smart money is. That 1994-2004 period coincided with a massive push to introduce computers to most offices and workplaces around the country that could help workers perform manual tasks much more efficiently. But now, 15 years later, if you look around an average cubicle farm you’ll see people working with largely the same technology and tools you saw 15 years ago. So it’s not likely that you’ll see newfound, exceptional jumps in productivity from that same tech.
The Third Industrial Revolution brought about substantial change in many aspects of business practice, entertainment, communications, and the collection and processing of information. But it has mainly been felt in this limited sphere of human activity, while the previous era of innovation changed the world around us. As a concrete example, take another look at the charts from above on the diffusion of modern conveniences and appliances. A house today, in 2022, looks significantly more like a house in 1940 than a house in 1940 did from a house in 1900.10 The previous era of innovations truly shaped the world around us in a way that modern tech just hasn’t.
Diffusion of Modern Conveniences
Diffusion of Modern Appliances
Some (Measured) Techno-Optimism
From the evidence above, we can say with a reasonable level of confidence that the post WW2, modern system of basic research has not matched this general level of TFP increase in any kind of sustained way. But there are many reasonable people who do not see this as a failure of the modern research ecosystem. Their defense is that many of the primary innovations discovered and implemented in the first half of the 20th century were things that could only happen once. Things like the electric light bulb, the electrification of homes, household plumbing, and automobiles can’t be “discovered” for a second time. They were one-time events.
And, while the fact that something can only be discovered once is objectively true, I fundamentally buy into the ethos that there are always new and better things to be discovered and built. It’s just not always obvious what they are.
Not all inventions are created equal in their ability to improve human well-being. And there is no problem with that. However, in constructing scientific research ecosystems and allocating mental resources and money, we should be taking every pain to directly allocate these resources to projects with a high expected ROI to human well-being. The next section lays out some rough ways that those looking to build in the metascience space—or simply those who want to build a company that helps regular people— can do this kind of smart goal-setting.
Concrete Targets to Improve the Situation
If your interest is piqued by the concept of improving total factor productivity, then this next section is for you. The goal of this particular newsletter is to not just provide information on the space of metascience, but to then take the next step and provide builders in the space with concrete ideas on which to start building.
And anybody who has ever built or managed anything knows that it is invaluable to have concrete, ideally measurable goals that you can work to improve day in and day out. And while TFP is a great metric, it can be a bit convoluted. It isn’t the easiest thing to align an organization/funding agency/research firm around.
Thankfully, Robert Gordon is a great social scientist who understands that intuitive explanations and frameworks are just as important as the computations they explain. Building off of some points from his work, the following list could be used as a rough, rule-of-thumb way for an organization that is looking to produce technology and insights that help people to set goals. The major “metrics” that tend to contribute to TFP increases and standard of living increases are inventions that tend to:
Increase people’s ability to buy more/better food
Increase people’s ability to buy more/better housing
Increase people’s life expectancy
Increase people’s time spent on leisure—working less
Increase people’s enjoyment while doing some fixed amount of leisure (this one is the least important)
And the standard of living increase could come because the invention tangibly increased people’s potential earnings, and subsequently the ability to buy some of the above components or just because the tech made them easier to come by. These metrics are one, of likely many, fantastic examples of ways for research organizations to set concrete goals that don’t leave them liable to increasing the number of scientific publications without increasing well-being.
These metrics help us intuitively understand just how things like running water and the automobile had the massive impacts on productivity and standard of living that they did. Running water put possibly as much as a whole workday’s worth of hours back into a housewife’s week on average. When is the last time something was invented that saved you 8 hours per week?
Again, people who say that the automobile can only be invented once are correct. But there are numerous areas of research that could satisfy the above components and where there’s a lot of technological improvement to be made:
I still work about 35 hours per week. That number is a lot closer to the 58.5 hours worked on average in 1890 than it is to zero...
I still need to sleep 6-8 hours to be a functioning human. If someone could find a way to help me fall asleep faster or get more “restfulness” out of a given amount of sleep, then both could be massive improvements in adding to my leisure time, quality of life, etc.
Just pick any of the five metrics from above and you can start imagining ways to improve them massively and ways to measure that improvement. These are the kinds of questions that the brightest STEM minds of our generation should be working on.
While most of this piece has been speaking with the scope of research organizations in mind, there are aspects where certain tech companies can and have been contributing. Impossible Foods and Beyond Meat have the potential to massively mitigate the environmental problems and land use of the meat industry, among many other problems. Investments in longevity such as Altos Labs, if they pan out and are made affordable, can increase life expectancy. And driverless cars can actually save us 8 hours per week, just as running water once did for housewives. The potential impact of efforts like these is much different than those of companies like Facebook or Netflix.
Conclusion
Our stagnant growth rates in output are not solely the fault of the R&D ecosystem. Things like improvements in management practices could also increase TFP. But if you’re reading this Substack, you likely believe in and want to work towards the romantic notion that technology and innovation can create better lives for everyone, not just a rich subset of people. And I do as well.
So, if you’re in that boat with me, setting up your research organizations or company to improve one of the following might be a fantastic way to make average people’s lives better in the long run:
Increase people’s ability to buy more/better food
Increase people’s ability to buy more/better housing
Increase people’s life expectancy
Increase people’s time spent on leisure—working less
Thanks so much for reading. I’d love to hear your thoughts on this piece or ideas for future ones. Please subscribe or tell a friend if you liked it. It helps me out a lot:)
Additional Citations and Footnotes11 12 13 14 15 16
Rowberg, Richard. Federal R&D Funding: A Concise History. CRS Report for Congress, 1998.
Bush, Vannevar. Science, The Endless Frontier. A Report to the President, 1945
Bornmann, L., Haunschild, R. & Mutz, R. Growth rates of modern science: a latent piecewise growth curve approach to model publication numbers from established and new literature databases. Humanit Soc Sci Commun 8, 224 (2021). https://doi.org/10.1057/s41599-021-00903-w
Gordon, Robert J. The Rise and Fall of American Growth. Princeton University Press, 2016. (This entire section draws heavily on Gordon. He is additionally cited in each graphic.)
Matt Clancy, 2021. How long does it take to go from science to technology?. New Things Under The Sun
Ahmadpoor, Mohammad, and Benjamin F. Jones. 2017. The Dual Frontier: Patented inventions and prior scientific advance. Science 357(6351): 583-587. https://doi.org/10.1126/science.aam9527
Gordon, NIPA Table 2.4.5.
Gordon, Robert J. The Rise and Fall of American Growth. Princeton University Press, 2016.
Gordon, Robert J. The Rise and Fall of American Growth. Princeton University Press, 2016.
Wages are from Production Workers Compensation from MeasuringWorth, price deflator is ratio linked from the PCE deflator 1929–2010, NIPA Table 1.1.9, to the CPI for pre-1929 from MeasuringWorth, GDP is from Balke and Gordon (1989) Table 10, 1869–1928, and NIPA Table 1.1.6, post-1928, and Total Hours are from BLS and Kendrick (1961) Appendix Table A-X.
Average weekly hours are from St. Louis FRED series AWHNONAG, 1964–2013, and are ratio linked from Jacobs (2003) Table 2-6, column 1, 1947–1963, to HSUS series Ba4575, 1900–1947, to Huberman (2004, Table 4, p. 977), pre-1900. Hours per employee are ratio linked from Kendrick (1961) Tables A-X and A-VI, 1870–1948, to BLS CES survey data, 1948–2013.”
Real GDP 1889–1929, Kendrick, Table A-XXII. 1929–2014, NIPA Table 1.1.6. Trend 1870–1928 calculated by linking to Berry (1988) for 1870–1889. Population 1870–1998 HSUS series Aa7, linked to U.S. Census Bureau for 1998–2014. Hours of work 1889–1948, Kendrick, Table A-XXII. 1948–2014 unpublished BLS series for total economy hours. Trend 1870–1928 calculated by assuming that hours per member of the population were unchanged 1870–1889.
Total factor productivity is the geometrically weighted average of the ratio of real GDP to labor input and the ratio of real GDP to capital input, with respective weights of 0.7 and 0.3. Labor input consists of hours from the sources of the above citation multiplied by an index of labor quality, taken from the “educational productivity index” of Goldin-Katz (2008, Table 1.3, column 2, p. 39). The Goldin-Katz index is available for 1915–2005. Our educational index is extrapolated backward from 1915 to 1890 using the Goldin-Katz 1915–1940 growth rate, and it is extrapolated forward from 2005 to 2014 using the Goldin-Katz 1980–2005 growth rate. Capital input consists of the new capital series described in the Data Appendix of Gordon 2016, shown for 1920–1970 in figure A–1 by the line labelled “Add Government Capital.
Morris, Edmund. Edison. Random House, 2019. (not directly used in any way but indirectly inspired several thoughts that went in to the piece)
Interesting analysis and a differentiated take on the emerging field of Progress Studies, I look forward to more posts
I love your focused list of targets for research labs to improve the human condition. But is that really what research labs are best at? Electricity and electrification were a failure of science. Tesla (the guy) and Edison did far more to advance our understanding of ⚡️ than the academy. Antibiotics, sure, but even that was more of a happy mistake than a directed effort. I’m not discounting the role of science to create a platform for innovation, but perhaps it’s the capital allocation (what do people want to fund), the problem set (what do nerds want to work on), and the population (what do people pay for). We need better translation of science into value for humanity, not more science.