An unexpected influence of widely used significance thresholds on the distribution of reported P‐values