Zeitpunkt Nutzer Delta Tröts TNR Titel Version maxTL Mi 19.03.2025 00:00:35 7.138 +1 724.951 101,6 NerdCulture 4.3.6 1.000 Di 18.03.2025 00:01:11 7.137 -1 724.455 101,5 NerdCulture 4.3.6 1.000 Mo 17.03.2025 00:00:11 7.138 -1 723.936 101,4 NerdCulture 4.3.6 1.000 So 16.03.2025 00:00:11 7.139 +2 723.498 101,3 NerdCulture 4.3.6 1.000 Sa 15.03.2025 00:01:18 7.137 0 723.022 101,3 NerdCulture 4.3.6 1.000 Fr 14.03.2025 00:01:24 7.137 +1 722.556 101,2 NerdCulture 4.3.6 1.000 Do 13.03.2025 00:01:10 7.136 -1 722.135 101,2 NerdCulture 4.3.5 1.000 Mi 12.03.2025 00:00:15 7.137 0 721.608 101,1 NerdCulture 4.3.5 1.000 Di 11.03.2025 00:01:10 7.137 0 721.113 101,0 NerdCulture 4.3.5 1.000 Mo 10.03.2025 00:00:44 7.137 0 720.596 101,0 NerdCulture 4.3.4 1.000
Nick Byrd, Ph.D. (@ByrdNick) · 10/2022 · Tröts: 1.273 · Folger: 1.128
Mi 19.03.2025 11:13
How can performance reviews be (de)biased?
Supervisors' ratings of an (imaginary) employee were biased by last year's rating (regardless of whether #AI generated it).
Listing TWO reasons why the prior rating is too high/low mitigated the #anchoringBias:
https://doi.org/10.1016/j.ijinfomgt.2025.102875
First result: the anchoring bias barely depended on whether the previous year's rating was generated by AI. (Even when the average difference was considered statistically significant, it was so small that it would not have, say, changed the grade points earned in the U.S. grading system).
Second result: the consider-the-opposite intervention caused differences in performance evaluations that would be large enough to make a difference in the U.S. grade point system (and Cohen's d was slightly larger than 1).
[Öffentlich] Antw.: 0 Wtrl.: 6 Fav.: 1 · via Fedica