Reducing Contextual Bias In Performance Appraisal

摘要*0011132道格拉斯·H·韦德尔在我们的社会中广泛使用评估工具。绩效评估就是这样一种工具，可以作为重要人员决策的基础，例如分配加薪或推荐晋升或解雇。由于绩效评估强烈依赖于主观判断，它容易受到系统性偏差或扭曲的影响。一种在绩效评估中反复表现出来的偏见是背景对比。这种偏见包括在评估其他工作记录较差的员工后，比评估工作记录较好的员工后，对某人的工作表现给予更高的评价。这种背景依赖是评估错误的严重来源，威胁到评估的可靠性和有效性。本项目探讨了旨在减少或消除绩效评估中的上下文对比的不同程序，从而创造更可靠和有效的评估技术。材料由假想员工的表现概况组成，其中包括他们工作记录的几个方面的信息。在每个实验中，对比偏差是通过让参与者在评估一组材料之前评估不同的材料而产生的。这项研究的目的是制定评估程序，在不同的情况下对核心材料做出相同的判断。这些程序包括操纵类别标签、判断类型和训练方法。实验将测试在多大程度上包括不同类别水平的明确例子将减少语境偏见。其他实验将检验两种替代判断任务的有效性，即比较判断和相似判断，以预测对比效果。最后，另一系列实验挖掘了不同训练方法消除背景偏见的潜力。这项研究的目的是开发可在商业、教育和政府中实施的程序，以产生更可靠和有效的评估。

Abstract***0011132Douglas H. WedellThere is widespread use of assessment tools in our society. Performanceappraisal is one such tool that serves as a basis for important personneldecisions, such as allocation of raises or recommendation for promotion ortermination. Because performance appraisal strongly relies on subjectivejudgment, it is susceptible to systematic biases or distortions. One suchbias that has been repeatedly demonstrated in performance appraisal iscontextual contrast. This bias consists of the tendency to rate a person'swork performance higher after assessing other employees with poorer workrecords than after assessing those with better work records. Such contextualdependency represents a serious source of error in assessment, threatening itsreliability and validity.This project explores different procedures designed to reduce or eliminatecontextual contrast in performance appraisal and thereby create more reliableand valid assessment techniques. Materials consist of performance profiles ofhypothetical employees that include information on several facets of theirwork records. In each experiment, the contrast bias is induced by havingparticipants evaluate different sets of materials prior to assessing a coreset of materials. The aim of the research is to develop assessment proceduresthat will produce the same judgments of core materials across differentcontexts. These procedures include manipulating category labels, types ofjudgments, and training methods. Experiments will test the degree to whichincluding clear examples for the different category levels will reducecontextual bias. Additional experiments will test the efficacy of twoalternative judgment tasks, comparative and similarity judgment, foreliminating contrast effects. Finally, another series of experiments exploresthe potential for different training regimens to eliminate contextual bias.The aim of this research is to develop procedures that could be implemented inbusiness, education, and government to produce more reliable and validassessments.***