Is The Great Stagnation Actually Just a 'So-So' Stagnation?

Human capital adjustment attributes lots of GDP growth to extra education, but if this education is mostly signaling, then we're getting more growth from new tech than we thought.

In my last post on literacy and the Department of Education I discovered an astounding stat:

The NCES ran two rounds of a literacy test, one in 1992 and one in 2003. The overall average score on the test didn’t change (276 vs 275 out of 500), but within every educational attainment group scores dropped massively.

Source

High school dropouts got less literate on average because the highest scoring dropouts in the 90s became the lowest scoring graduates in the 2000s as standards were lowered and more students were pushed through into more education. Literacy scores among Graduate degree holders dropped by 13-17 points in a decade. If a graduate degree cannot even teach you how to read, it's probably not having large effects on any other more complex forms of human capital.

So over the past 50 years of rising educational attainment, there has been little to no gain in actual skills. If basic literacy and numeracy do not improve after years of extra schooling, more complex and harder to measure gains do not seem likely either.

Math and reading scores on tests that have been surveyed over longer periods also show essentially zero improvement in the population average scores. The average score on the reading test in 1971 was 255 out of 500, in 2023 it was 256 and math scores went from 266 to 271.

All this suggests that increases in educational attainment since the 70s have been almost entirely credential inflation. People go on to further degrees not because it increases your skills; it doesn't even teach you to read, but because a high school diploma is no longer a reliable signal of skill or conscientious.

What this means for the Great Stagnation

The Great Stagnation is, in many ways, a claim about economic statistics. In particular it is a claim about Total Factor Productivity (TFP). TFP is what’s left over after you count up all the economy’s inputs and all of it’s outputs and see how they change over time. If your output grows without a change in your inputs, we attribute that gain to an increase in productivity, usually due to new technology.

The Great Stagnation is When TFP Stops Growing. Source

To calculate TFP you need accurate estimates of how much inputs have changed, e.g how many hours people worked, how many pounds of steel they used, or how many machines they employed. Economists also try to adjust for is the quality of these inputs. 10 hours of labor from a skilled practitioner will produce more than 10 hours of labor from someone’s first day on the job, and we don’t want to attribute that increased output to new technology.

The most important measure of labor quality that economists use is years of education. Basic economic theory predicts that workers are paid an amount equal to the value of what they produce. Extra degrees confer higher wages for those who hold them, suggesting that these degrees raise worker productivity. Therefore, if the workforce as a whole gets more educated, it must also be getting more productive.

This is a valid extension of basic economic models, but once you allow for uncertainty over worker productivity and information gathering costs, it becomes plausible that education is more about proving existing skills rather than increasing them.

The evidence I showed above confirms this theory. Educational attainment has risen by 3+ years since 1970s but population level reading and math scores have barely changed, and scores within educational attainment groups have fallen. As more people are pushed through lowered standards to a high school diploma, the wage premium to this degree has fallen, suggesting that much of its value was in the differentiating signal rather than skill gains.

If this story is true, and more education hasn’t been raising productivity, then all the GDP growth we thought was due to more skilled workers is actually coming from somewhere else. That means we’ve been undercounting TFP, and technological progress for decades.

By how much exactly? Well, if we take the same SF Fed series of TFP data that made the graph above, and remove the adjustments they make for gains in human capital (based mostly on rising educational attainment), we get something that looks like this.

Note: I remove the HC adjustment starting after 1970 because before this point, literacy scores were improving.

The Great Stagnation is shrunk by about a third!

What does this teach us? Perhaps not as much as it may seem.

For one thing, no amount of fiddling with the TFP statistics will change the physical reality of our world. It won’t invent flying cars or turn around the Henry Adams curve. If we can make The Great Stagnation disappear in TFP statistics but it still remains evident in the physical world, that’s all the worse for TFP statistics.

The shift in TFP growth rate caused by removing human capital adjustments would change the quantitative results of papers like “Are Ideas Getting Harder to Find?”, but the qualitative results would be similar.

The human capital adjustment I removed is mostly based on educational attainment, which is pretty linear, so this adjustment doesn’t change the qualitative trend breaks that jump out of this graph. There’s still a trend break in the early 70s, still a spurt of rapid growth in the late 90s to early 2000s, and still a general slow-down of growth even as research investment increases.

If we only saw the adjusted chart, we'd probably still be scratching our heads wondering WTF happened in 1971.

Still though, all of this does suggest that we need to be more skeptical of educational attainment as a proxy for human capital, and that some significant fraction of the ‘missing’ technological progress since the 1970s may actually just be an artifact of this assumption.