无意识偏见在公民对公共部门绩效评估中的复制研究

在过去的一个世纪里，关于公共部门的负面印象一直是公共行政学者关注的焦点。韦伯（1918）早在一百多年前就描述了德国官僚体系中无法被公众理解和认可的成就。类似的观点也出现在Goodsell（2004）和Frederickson（1997）的研究中，构成了Marvel（2015, 2016）提出的“反公共部门偏见”（APSB）的基础。Marvel定义APSB为公民无意识地将贬损性词汇与公共组织联系在一起，无论这些组织的实际表现如何。

在这项研究中，我们对Marvel（2016）的研究进行了复制，通过中国的公民样本（N=1410）进一步验证了APSB对公共部门绩效评估的负面影响。虽然结果再次确认了APSB的存在，但研究也发现，在特定条件下，有利的绩效信息可以改变公民的APSB倾向。这一发现不仅为公共行政学者提供了新的理论视角，也为公共部门管理者在改善公众对公共部门的态度方面提供了实际的指导。

For over a century, negative perceptions of the public sector have been a focal point for public administration scholars. As early as 1918, Weber described the achievements of the German bureaucracy as something that could never be fully understood or appreciated by the public. Similar views were echoed in the works of Goodsell (2004) and Frederickson (1997), forming the basis of what Marvel (2015, 2016) terms “anti-public sector bias” (APSB). Marvel defines APSB as citizens' unconscious association of derogatory terms with public organizations, regardless of their actual performance.

In this study, we replicated Marvel’s (2016) work by using a sample of citizens from China (N=1410) to further examine the negative influence of APSB on public sector performance evaluations. While the results reaffirm the presence of APSB, they also reveal that under certain conditions, favorable performance information can alter citizens' APSB tendencies. This finding not only provides new theoretical insights for public administration scholars but also offers practical guidance for public sector managers aiming to improve public attitudes toward the public sector.

为了验证“反公共部门偏见”（APSB）在不同文化背景下的适用性，我们按照Walker等人（2019）提出的复制框架，重新进行了Marvel（2016）的实验。本研究的复制类型属于“概括和扩展”，我们在原始研究的基础上进行了多项修改，旨在更广泛的背景下测试APSB的普遍性。

首先，我们将实验场景从美国移至中国，以探讨APSB在中国这一独特背景下的表现。中国的社会主义意识形态强调公共部门在国家治理中的优越性，公共部门组织在中国承担着广泛的职责。中国社会对公共部门的信任度在全球范围内名列前茅，与西方社会的“小政府”理念形成鲜明对比。这使得中国成为一个理想的测试场域，用以检验APSB及其对公共部门绩效评估的影响。此外，Hofstede（2001）指出，亚洲文化（以中国为代表）在权力距离、避免不确定性、个人主义与集体主义等方面与西方文化存在显著差异。通过在中国进行实验，我们希望验证APSB的影响是否具有普遍性，从而为相关学术讨论提供新的实证依据。

其次，我们设计了一个新的实验分支，即实验四，用以测试单一类别隐性联想测验（SC-IAT）的结果。此举旨在弥补原始实验中相对IAT测量的不足，从而获得更稳定的对公共部门隐性态度的衡量。我们还将实验中的公共和私营实体替换为中国邮政速递物流公司（CPEL）和中通快递（ZTO Express），分别对应于美国的USPS和FedEx。通过这些调整，我们能够更贴近中国的实际情况，同时保留与原始实验的可比性。

最后，我们提出了一个新的假设，即通过强调社会责任的视频广告能够有效干预并降低公众的反公共部门偏见。与西方社会不同，中国的国有企业在提供公共服务时承担了更大的社会责任，CPEL就是其中的典型代表。我们认为，视觉信息比文字信息更能成功引起人们的兴趣和情感，因此，通过视觉广告来提升公共部门的形象可能会有效缓解APSB。通过这些创新的实验设计，我们希望在验证Marvel（2016）发现的同时，进一步探索如何在中国背景下减少APSB的影响。

在本研究中，我们严格遵循了Marvel（2016）的实验程序，并根据实际情况做出了一些调整。首先，研究得到了中山大学人文伦理委员会的伦理批准，并在AsPredicted.org进行了预注册。实验参与者是通过NaoDao平台招募的，该平台是中国的一个在线研究平台，提供问卷准备、实验编制、受试者众包、质量控制等多项功能。

尽管Marvel（2016）未报告效能分析，但考虑到出版偏差可能导致效应量估计过高，我们在计算样本量时选择了较小的效应量。基于效应量（0.03）、alpha误差（0.05）和效能（0.9）的设定，G*Power的结果显示每个实验至少需要353名参与者。最终，四个实验的数据在2022年10月至12月之间收集，共有1863名参与者参与。排除不完整的回答后，实验I、II、III和IV的最终样本量分别为353、353、351和353。

值得注意的是，本研究的样本为非随机抽样，类似于其他中国调查公司，如问卷星，因此样本并不具有全国代表性。NaoDao的用户主要是注册用于研究目的的大学生，他们集中在中国的东部和中部地区，且以年轻、高学历、未婚、无子女的人群为主。在实验I和实验II中，参与者在完成IAT后随机接受信息干预，并最终对CPEL的EMS服务进行绩效评分。在复制实验III和IV中，参与者随机分配接受信息干预，然后完成IAT，最后进行绩效评分。

我们使用与原始实验相同的项目、程序和算法来测量复制实验的核心变量。反公共部门偏见通过IAT和SC-IAT测量，使用Greenwald等人（2003）和Karpinski和Steinman（2006）开发的算法计算分数。绩效信息通过随机分配不同形式的信息呈现来进行实验操控。绩效评估通过一个简单的评分问题来衡量：“你如何评价CPEL的EMS的绩效？”参与者以7分制作答，1分代表低绩效，7分代表高绩效。

为了便于将复制研究结果与原始研究结果进行比较，我们使用了与Marvel（2016）研究相同的回归分析方法来检验复制研究的五个研究假设。

In this study, we rigorously followed the experimental procedures outlined by Marvel (2016) and made several adjustments to fit the context. First, the study received ethical approval from Sun Yat-sen University's Human Subjects Ethics Sub-committee and was pre-registered at AsPredicted.org. The participants were recruited through the NaoDao platform, an online research platform in China that offers functionalities such as questionnaire preparation, experiment compilation, subject crowdsourcing, and quality control.

Although Marvel (2016) did not report a power analysis, considering that publication bias may inflate the estimated effect size, we chose a smaller effect size when calculating the sample sizes. Based on effect size (0.03), alpha error (0.05), and power (0.9), G*Power results indicated that a minimum of 353 participants were required for each experiment. Ultimately, data from four experiments were collected between October and December 2022, with a total of 1863 participants. After excluding incomplete responses, the final sample sizes for Experiments I, II, III, and IV were 353, 353, 351, and 353, respectively.

It is important to note that the sample is non-random and not representative of the broader Chinese population, similar to other Chinese survey companies like WJX (Wen Juan Xing). The NaoDao sample pool primarily consists of college students registered for research purposes, mainly from the eastern and central regions of China. In Experiments I and II, participants were randomly assigned to receive information intervention after completing the IAT and subsequently rated the performance of CPEL's EMS service. In replication Experiments III and IV, participants were randomly assigned to receive the information intervention, then completed the IAT, and finally rated the performance.

We measured the core variables of the replication experiments using the same items, programs, and algorithms as the original experiment. Anti-public sector bias was measured using the IAT and SC-IAT, with scores computed using algorithms developed by Greenwald et al. (2003) and Karpinski and Steinman (2006). Performance information was manipulated by randomly assigning different forms of information to subjects. Performance evaluation was measured by a simple rating question: “How would you rate the performance of CPEL's EMS?” Participants responded on a 7-point scale, where 1 represents low performance and 7 represents high performance.

To facilitate comparison between the replication study results and the original study results, we used the same regression analysis as in Marvel's (2016) study to test the five research hypotheses of the replication study.