Publication bias in the social sciences since 1959: Application of a regression discontinuity framework