Why do only two-thirds of famous social science results replicate? It’s complicated