Should we teach using schemas? Evidence from a randomised trial