1,339 teachers were each shown a student's exercise with a wrong mark already on it, and the ones told an algorithm had set the mark finished further from the fair grade than the ones told a colleague had — but only where the mark was too harsh, with no detectable difference where it was too generous
A preregistered experiment in Greece put the right answer within reach of anyone who read the screen, then varied who was said to have set the mark. The label changed how teachers marked when the mark was too harsh and left no detectable trace when it was too generous, and within the harsh version it reached significance among the younger, the better-qualified, the humanities-trained and the more tech-confident, not their counterparts.