A large survey of scientists finds that more than half have tried and failed to reproduce their own experiments.