“We’ll retrain everyone” is doing far too much work

Retraining is politically convenient. Workers gain new skills, employers gain staff, and nobody has to argue about who bears the loss. Unfortunately, the size of the measured benefit is much smaller than the promise usually attached to it.

David Roodman and Anthropic economist Maxim Massenkoff reviewed 56 randomized US studies, alongside European experimental evidence. Offering someone a training place increased employment by roughly two to three percentage points and annual earnings by about $1,000, against a program cost near $13,000. Those are real gains. They are also modest.

Accounting for taxes and benefits, governments recover more than half their spending; the programs roughly break even overall. This is not evidence that every training dollar is wasted.

Some sector-specific programs do much better. They work with employers in industries that need workers and connect participants directly to jobs. Replication has proved difficult, however. Expanding a successful local pipeline is different from creating millions of suitable vacancies.

Zvi’s objection is aimed at the policy shortcut: a mildly beneficial program gets presented as though it settles the displacement problem. It does not. His deliberately blunt verdict on retraining is better read alongside the review’s more measured findings.

There is also a distinction between helping a worker reach an existing job and ensuring enough desirable jobs exist after a major technological change. Training can address the first without resolving the second.

The researchers recommend expanding and rigorously testing promising approaches now. That is a concrete proposal. Assuming those approaches will scale before demonstrating it is a much larger bet.