Latest / The Productivity Podcast with Fexingo: Output, Efficiency, and Long-Term Economic Growth / Why Productivity Growth Has Slowed Since 2005
Transcript
- Lucas: If you look at productivity growth in the United States since the end of 2005, the numbers are sobering. Average annual labor productivity growth — output per hour worked — ran about 2.4 percent from 1995 to 2005. Since then, it's been closer to 1.3 percent. Luna: And that's not just a US story, right? Most developed economies have seen a similar deceleration. Lucas: Exactly. And the puzzle is that this slowdown coincided with an explosion of digital technology — smartphones, cloud computing, AI — the kind of stuff that should have boosted productivity. So either the technology isn't delivering, or we're measuring it wrong. Luna: Which brings us to the 'measurable vs. unmeasurable' argument. Economist Chad Syverson at the University of Chicago has done a lot of work on this. Lucas: Right. Syverson points out that the slowdown is concentrated in industries that are hard to measure. Think healthcare, education, finance, government — sectors where output is difficult to quantify. Meanwhile, manufacturing and retail — where output is easier to count — have continued to see solid productivity gains. Luna: So the headline number might be understating real progress, because more of the economy is in hard to measure services. Lucas: Exactly. In 1980, about 55 percent of US GDP was in what Syverson calls the 'unmeasurable' sector. By 2015, it was around 70 percent. If you're systematically undercounting output in a growing share of the economy, the productivity numbers will look worse than reality. Luna: But some economists argue the slowdown is real. They point to things like declining business dynamism, fewer startups, and slower diffusion of technology from frontier firms to laggards. Lucas: And that's a serious argument. Research by the OECD shows that the gap between the most productive firms and the rest has widened significantly since the early 2000s. The frontier firms — often digital natives — keep getting more efficient, but the rest of the economy isn't keeping up. Luna: So it's not just a measurement problem — there's a diffusion problem too. Lucas: Right. But let's dig into the measurement piece, because it's fascinating. Syverson offers a simple thought experiment: if the productivity statistics are accurately capturing output, then the slowdown implies that the social value of all the technological innovation since 2005 has been smaller than the social value of innovations from, say, the 1990s. That's a hard sell. Luna: Especially when you consider what we can do now that we couldn't in 2005. I mean, I'm holding a device that lets me access most of human knowledge, navigate anywhere, communicate instantly — that has to show up somewhere. Lucas: But here's the tricky part: a lot of that value is consumer surplus — it makes us better off, but it doesn't show up in GDP because it's free or very cheap. Google search, Wikipedia, GPS navigation — these are incredibly valuable, but they're not priced, so they don't appear in output statistics. Luna: Right, and then there's the quality problem. How do you measure the output of a hospital? If a patient survives a heart attack that would have killed them ten years ago, that's enormous value, but the statistician might just count the cost of the procedure as output. Lucas: Exactly. In healthcare, output is often measured by the volume of services — doctor visits, tests, procedures — not by outcomes. So if a new treatment reduces the number of visits needed, measured output goes down, even though health improves. That's a productivity mismeasurement. Luna: So the official numbers might actually be overstating productivity in healthcare, not understating it — if fewer visits means less measured output. Lucas: That's a good point, and Syverson acknowledges that measurement cuts both ways. In some sectors, we may be overestimating output. But on balance, most economists who study this think the measurement issues lead to an understatement of productivity growth, at least in recent decades. Luna: What about software and digital goods more broadly? How do statisticians handle something like a smartphone app that gets updated constantly? Lucas: It's a nightmare. The Bureau of Labor Statistics tries to adjust for quality — they use hedonic pricing models to estimate the value of new features. But it's imperfect. And for free digital goods, there's essentially no price to use. Some economists have proposed adding the value of free digital services to GDP, but it's not standard practice. Luna: There's also the issue of intangible capital. R&D, software, brand equity, organizational capital — these investments are increasingly important, but they're often expensed rather than capitalized in the national accounts. Lucas: Right. Research by Corrado, Hulten, and Sichel suggests that if you properly account for intangible investment, the measured slowdown in productivity growth narrows significantly. In some periods, it disappears entirely. Luna: So the productivity slowdown might be partly an artifact of how we construct the statistics. Lucas: Partly. But not entirely. Even after accounting for measurement issues and intangibles, most estimates still show a genuine slowdown, just a smaller one. And the diffusion problem I mentioned earlier is still a real concern. Luna: So where does that leave us for long-term economic growth? Because productivity is ultimately the driver of living standards. Lucas: It's the fundamental driver. Over the long run, growth in output per person is almost entirely determined by productivity growth. So a sustained productivity growth rate of 1.3 percent versus 2.4 percent means that living standards double every 54 years instead of every 29. That's a huge difference. Luna: And that's why getting the measurement right matters — not just for academic reasons, but for policy. Lucas: Exactly. If we think the slowdown is largely a measurement problem, the policy response is different than if we think it's a real failure of technology to deliver. In the first case, we might focus on improving statistical agencies. In the second, we might need policies to boost competition, R&D spending, or workforce training. Luna: Speaking of policy relevance — and this ties back to what we do here — if you find these kinds of economic puzzles as interesting as we do, and if walking through the economy with us has made something click, it's worth knowing that listener support is what keeps this show ad-free and independent. Lucas: Yeah, absolutely. A couple of dollars a month — the price of a coffee — genuinely makes a difference. It's buy me a coffee dot com slash fexingo, if you've gotten something out of these episodes. Luna: And it's not just about supporting the show — it also signals to us what topics you care about. We pay attention. Lucas: We do. And that feeds back into episodes like this one. Okay, back to the measurement issue — one thing I find really compelling is the idea that the slowdown might be the result of a shift in the type of innovation we're doing. Luna: How so? Lucas: Economists like Robert Gordon argue that the great inventions of the late 19th and early 20th centuries — electricity, the internal combustion engine, running water — had a much larger impact on productivity than digital technology. He calls the digital revolution 'less great' by comparison. Luna: But that seems to discount the transformative potential of AI and machine learning, which are still in early stages. Lucas: Absolutely. And that's the open question. The productivity statistics from the last decade reflect a world before large language models and generative AI became widely used. If those technologies genuinely boost productivity across a broad swath of the economy, we might see the numbers pick up in the next few years. Luna: So we might be at an inflection point similar to the mid-1990s, when the productivity acceleration from IT began to show up in the data. Lucas: Exactly. It took years for the productivity gains from computers to materialize in the official statistics — that's the Solow paradox: 'You can see the computer age everywhere but in the productivity statistics.' And then from 1995 to 2005, the gains showed up. We might be in a similar lag period for AI. Luna: So the story of productivity growth since 2005 might not be a story of stagnation — it might be a story of mismeasurement and lagged effects. Lucas: Right. And the only way to know is to keep watching the data carefully, while also improving how we measure output in a digital, services-heavy economy. Because the stakes are huge.