Does Cannabis Make Us More Productive?

I should disclose my bias at the outset: I do not smoke cannabis, and I think smoking it is a bad idea. That does not prove anything about economic productivity. It does mean that I should be especially careful not to turn an interesting chart into a conclusion it cannot support.

Still, the United States has conducted a remarkable social experiment. Colorado and Washington approved recreational marijuana in 2012. Alaska and Oregon followed in 2014. California, Maine, Massachusetts and Nevada joined in 2016. By late 2023, adult recreational use had been legalized in 24 states.

As a retired trader, I know what a chart can do. It can sharpen a question—and tempt you into answering it too quickly. So let us start with the question and keep the answer on a short leash.

A 28-point productivity divide

The first chart sets both productivity indexes to 100 on November 6, 2012, the day Colorado and Washington voters approved legalization. By the end of 2025, nonfarm business productivity had climbed to approximately 125.5. Manufacturing productivity had slipped to about 97.5.

This is not a minor sectoral wobble. Manufacturing productivity was already slowing before 2012, but the gap that subsequently opened against the rest of the nonfarm business economy is enormous.

Look under the hood

Labor productivity is simply real output divided by labor hours. To understand the manufacturing decline, we can separate those components and rebase each to the same November 2012 starting point.

The arithmetic is unusually clean:

American manufacturing used roughly 2.6% more labor hours to produce essentially the same amount of real output.

That is more concrete than saying that manufacturing “declined.” Output did not collapse over the full period. Hours rose, but production did not keep pace. The result was lower output per hour.

Are marginal firms being left behind?

Perhaps—but ordinary firm selection should work in the opposite direction. If the least efficient plants close, average productivity should rise. A composition explanation works only if highly scalable or high-productivity activities moved abroad, or if weak incumbents remained because competitive entry, exit and reallocation deteriorated.

Researchers at the Federal Reserve Bank of New York have tested the simpler composition story. Holding industry weights fixed does not remove the slowdown. Both leading and following industries deteriorated, and both leading and following public firms experienced large reductions in productivity growth. The New York Fed appropriately calls the manufacturing slowdown a mystery.

Other plausible explanations include weak capital deepening after the financial crisis, the fading of the extraordinary semiconductor and information-technology productivity boom, offshoring and outsourcing that weakened learning-by-doing, declining business dynamism, and imperfect measurement of high-tech output and imported intermediate goods.

There is also an uncomfortable fact: manufacturing total-factor productivity slowed sharply in the late 2000s. That suggests the problem is not merely insufficient machinery. The efficiency with which capital, labor and intermediate inputs were combined also deteriorated.

So, does cannabis make us more productive?

These data do not say that cannabis caused manufacturing productivity to fall. They also provide no support for the comforting claim that wider recreational use has made the workforce more productive.

What they show is stranger and, to me, more important: during the period in which legal recreational cannabis spread from two states to nearly half the country, nonfarm productivity compounded higher while manufacturing stagnated. By 2025, manufacturing was using more hours to produce almost exactly the same real output as in late 2012.

That coincidence is worth investigating, not exploiting. A serious next step would use state-level data, compare early legalizers with similar control states, distinguish legalization from retail availability, and test whether employment, hours and output changed differently after treatment.

Until then, the honest answer to the title is: we do not know.

But anyone answering “yes” should bring some data.