Thread by @mattsclancy

True or false:
Science ==> Technological progress.
Here's a thread on recent research trying to answer that question. #ThursdayThreads #ProgressStudies

Many would say answer is obviously "true." But some think technological breakthroughs mostly come from tinkering, and aren't guided by science. Others say ivory tower academics are disconnected from the real world and focus on irrelevant problems.

How to approach the problem systematically? One way is to look to patents. Patents aren't perfect records of invention are still useful. They provide very detailed information on millions of inventions across all sectors of the economy.

Patents are supposed to make citations to relevant "prior art." So a simple way to see if technology builds on scientific research is to see if it cites scientific research. This is exactly what a 2018 paper by Ahmadpoor and Jones does. science.sciencemag.org/content/357/63…

Over 1976-2015 it turns out about 20% of patents directly cite scientific work, and about 10% of science and engineering papers are cited by patents. But this understates the role of science.

Patents (and papers) tend to cite their immediate intellectual forebears, not their intellectual great-great-grandparents. For example, we can imagine an intellectual lineage like:
Riemann Geometry => Einstein's General Relativity => GPS => fitbit step tracker.

A fitbit patent might cite a GPS patent, but it's not going to cite Riemann. So A&J follow citations down the chain to build a deeper citation network.

They count a patent as "connected" to science if it has a scientific paper in its intellectual forebears (i.e., if you keep following it's citations backwards, eventually you hit a paper). Papers are "connected" if any of their intellectual descendants are patents.

By this measure, 65% of patents are connected to science and 80% of science and engineering papers are connected to patents.

(65% might seem low, but it only became normal to cite scientific papers relatively recently so this could be viewed more as a lower bound).

Most papers are 2-3 links removed from being cited by a patent (i.e., they are either the intellectual grandparents or great-grandparents of patents). Most patents are 2 citations away from a paper (i.e., intellectual grandchildren).

Aside: The breakdown follows about what you would expect. The link from math to technology typically has many more steps than biology or computer science.

So that's encouraging. It would be worrying if patents never cited academic research. Nonetheless, citations are imperfect measures of knowledge flows. Do they really indicate the science was necessary for the invention to occur?

One way to answer this is to just ask the firms. Arora, Belenzon, and Sheer (2019) do this (among other things). And yes, firms that report they highly use university research cite a lot more university papers in their patents. static1.squarespace.com/static/593d9b0…

A complementary paper by Acemoglu, Akcigit and Kerr (2016) also indicates there's some real information in those citations. pnas.org/content/113/41…

AAK exploits the fact that the patent office categorizes patents into different technology classes to build a citation network between patent classes. It's a map of what classes of technology are most likely to be cited by other classes.

What's interesting is that if you use data from 1975-1994 to build this citation network map, you can predict patenting in different technology classes based on what's going on in the classes they cite.

For example, if drug patents usually cite organic chemistry patents, then a surge of organic chem patents is correlated with a surge of drug patents a few years later.

That's consistent with the idea that patent citations are pointing to the technology fields that they build on. When there are breakthroughs and advances in field x, then the technology classes that are it's children build on those advances and enjoy a flurry of activity.

Unfortunately, the paper is limited to citations from patents to patents (not papers). But the consensus view seems to be citations of papers are better signals of knowledge flows than citations of patents, so I would be surprised if the same kind of result didn't hold for papers

A final piece of evidence focuses specifically on the biology-pharmaceutical link. Azoulay, Zivin, Li, and Sampat (2015) probably establish the strongest causal evidence that science leads to technology advances. nber.org/papers/w20889

They link medical patents and FDA drug approvals to NIH funded research and identify cases where only a subset of equally meritorious grants (as judged by the reviewers) are funded for idiosyncratic reasons related to how the NIH disperses grants.

This is about as good as you can get without an explicitly experimental design: more or less randomly, some scientific fields and diseases get more research money than others.
What happens? The fields that do more basic research end up with more FDA approved drugs and patents.

So, to go back to the beginning; if we're happy with this evidence, does this mean the science-free tinkering model of technological progress is dead wrong? Not so fast.

I'm sure it still applies to many inventions today (remember - more than 1/3 of patents have zero connection to scientific literature).

And it was probably an even better description of technological advance in the past; Vaclav Smil has argued (via historical case studies) that science only became a really important input into technological advance in the late 1800s. /Fin goodreads.com/book/show/7412…