What You’ll Find This Week

It takes roughly 18x more scientists today to keep Moore's Law's doubling alive than it did in the early 1970s. That statistic comes from the most cited productivity paper of the last decade, and for years it's been the reason "ideas are getting harder to find" reads as economic fact rather than a saying people repeat at happy hour.

Two new papers built a 45-year panel covering nearly every firm in America, not just the public ones. They found research productivity rising. This week: why the paper that convinced a generation of executives that the well was running dry only ever measured the shrinking, unrepresentative slice of the economy that still shows up in its dataset.

Here’s what you’ll find:

Don’t Miss Our Latest Podcast

This Week’s Article

In September 2017, four economists, Nicholas Bloom, Charles Jones, John Van Reenen, and Michael Webb, asked a simple question:

Is it getting harder to have a good idea?

Their answer, published three years later in the American Economic Review, was: yes, dramatically so. It now takes roughly 18 times more researchers to sustain Moore's Law's doubling of transistor density every two years than it took in the early 1970s. And the pattern isn't confined to computer chips. Across crop yields, drug approvals, and firm-level R&D in more than a dozen industries, research effort kept climbing while output per researcher kept falling. "Are Ideas Getting Harder to Find?" became one of the most cited papers in the field, and it gave a generation of executives a clean, academically respectable reason to shrink the long-horizon research budget, grounded in the theory that the well was running dry.

Two new papers, both published this year, ran the same question against a bigger sample: the full universe of US firms instead of only the ones that trade on public markets. They found research productivity rising, not falling.

Bloom, Jones, Van Reenen, and Webb didn't rely on one data source. They built the case across multiple, separate case studies: crop yields, semiconductor density, drug approvals, firm-level R&D across more than a dozen industries. Total research effort across the US economy had grown roughly 20-fold since 1930. Research productivity, the amount of output each additional researcher produced, had fallen by a factor of around 41.

Stanford's Graduate School of Business took the finding out of the journal and put it in front of the executives who'd actually have to act on it. To hold economic growth at its current rate, the school found, the US would need to double its total research effort every 13 years. A sample of publicly traded firms showed the same productivity decline in more than 85% of cases. ("Big Ideas Are Getting Harder to Find" | Stanford GSB)

The finding held up because everywhere researchers pointed the lens, they saw the same slowdown. It became the intellectual backbone under a decade of corporate R&D pessimism, not a hunch about innovation labs defaulting to failure, but a peer-reviewed, AER-published answer to the question of why invention felt so much harder than it used to.

Every study behind the "ideas are getting harder to find" consensus leaned on Compustat, the database of financial filings from companies that trade on US public markets. It's the standard dataset for corporate research because it's clean and standardized. It also excludes almost the entire economy, and that exclusion is the subject of both papers published this year.

In May, a team of five economists, Fort, Goldschlag, Liang, Schott, and Zolas, built a 45-year panel tracking patenting activity across the Census Bureau's Longitudinal Business Database, which covers nearly every firm in the country, public and private alike, not just the subset that appears in Compustat. A second, independent paper, by Yoshiki Ando, James Bessen, and Xiupeng Wang, used Census manufacturing microdata covering more than 670,000 establishment-years from 1976 to 2018, and reached the same conclusion by a completely different route. "The Rising Returns to R&D" | Boston University

Two teams, two separate datasets, no shared authors, arrived at the same rebuttal within weeks of each other, which is a harder thing to wave away as one contrarian paper chasing a headline.

Fort and her coauthors' headline finding, in their own words: "average patents per R&D input are increasing, the elasticity of patents to R&D inputs is flat or rising, and there is not systematic evidence of a secular decline in patenting after controlling for research inputs." Across the full universe of firms in their panel, patents per R&D dollar rise to roughly 1.5 times their starting level over the 45-year window. Ando, Bessen, and Wang's manufacturing data points the same direction: both the elasticity of output with respect to R&D, and the marginal return on every R&D dollar spent, have risen sharply since the late 1970s.

Neither paper argues Bloom, Jones, Van Reenen, and Webb made an error. Both point to something narrower and more interesting. The earlier paper measured a real trend accurately. It just measured the wrong population.

The paper's underlying data shows why. Among the full panel of firms, research productivity, measured as patents per R&D dollar, climbs steadily across the entire 45-year window. Among the subset of those firms that also show up in Compustat, the public ones, productivity tracks that same rising trend for the first decade, then reverses hard starting in the 1980s and ends the sample below where it started.

The reversal isn't a coincidence of which companies happened to go public. Compustat's coverage of total US patenting activity fell from 63% to 55% over the period the paper studies, because invention increasingly concentrated in smaller, younger, and non-manufacturing firms, the kind that rarely list on a public exchange at all. The public-company sample didn't shrink randomly. It shrank by systematically losing the exact firms where invention was moving.

Every study that used Compustat as a stand-in for "the economy" was, without anyone intending it, watching an ever-smaller and less representative slice of that economy and mistaking that slice's decline for the whole picture's decline. Bloom, Jones, Van Reenen, and Webb's Moore's Law example survives this correction. Semiconductor R&D really is concentrated in a handful of massive public firms, and that specific case still shows declining productivity. Their broader claim about the whole economy doesn't.

None of this means the last decade of corporate innovation complaints was imaginary. Fort and her coauthors found something else worth sitting with. Even correcting for the Compustat bias, average firm growth, net of the growth attributable to patenting, keeps falling. Companies are generating patentable ideas at a rising rate and turning fewer of them into growth than they used to.

This newsletter has been describing that same gap under different names for over a year: the innovation debt that builds when a company defers hard invention work in favor of visible activity, AI deployment mistaken for AI innovation, efficiency metrics that reward everything except invention itself. Every one of those pieces assumed the raw material, the ideas, was still there, and that the failure sat in the organization. The new data says that assumption had more evidence behind it than anyone realized.

"There are no new ideas" survives as a belief because it explains a real feeling. Growth is harder to produce than it used to be, and invention doesn't seem to pay off the way it once did. Both of those things are true. The ideas didn't run out. The dataset built to track them stopped covering the part of the economy where they'd gone.

The uncomfortable version of that correction, for anyone running a company, is that "we're out of ideas" was always the easier problem to have. It comes with an obvious, if grim, remedy: wait, acquire, or pivot to efficiency. A translation problem, ideas that exist and aren't making it to growth, doesn't come with a remedy that simple. It means the ideas are probably already sitting somewhere inside the organization, and the reason they aren't turning into anything has nothing to do with how many of them exist.