Replication market: about half of papers don't replicate, people are good at forecasting what won't replicate, non-replicable studies get cited as much as replicated studies
These numbers include very simple/obvious results. Economics does the best with 2/3rds of studies replicating. A unique weakness of economics is the use of absurd instrumental variables. Education does well, though often the claims are trivial. Results based on interaction effect 10 points less likely to