Assessing the differential effect of poll exposure on voting behaviour

近年来，在预测2015年和2017年大选结果以及英国退欧公投结果明显失败之后，政治民调受到了越来越多的审查。有鉴于此，最近英国上议院特别委员会（House of Lords Select Committee）发布了一份报告，呼吁民意调查行业“让自己的房子井然有序”，但值得注意的是，该报告拒绝了在任何时期禁止民意调查公布的提议。这在一定程度上可以用最近的一项发现来解释，即这种失败并没有在很大程度上偏离历史上普遍存在的不准确，但它也被框定在民意调查对公众舆论的影响方面。也就是说，民意测验专家对特别委员会的结论表示欢迎，因为任何基于不准确的预测可能不公平地影响投票结果的潜在民意调查中断，都无法解释这样一个事实，即在没有民意测验的情况下，人们仍然可能受到他们对结果的预期（甚至更容易出错）的影响。然而，考虑到大量且越来越多的研究推定地证明了从众效应的存在，这种观点很奇怪，根据这种研究，选民对民意调查的了解会给领先的候选人带来进一步的优势。因此，鉴于本文提供的证据，民意调查似乎以一种许多人可能认为不民主的方式影响选举结果。然而，我认为，这一论断受到了这一工作中概念上的不足的危害。不仅从众效应是由不同的研究人员松散而不同地定义的，而且很少有人试图确定这种效应的因果机制。从众效应可能只存在于某些个体中，取决于某些特征或环境，例如与身份、党派关系、知识、情感、政治参与或效能有关。甚至可能在控制了正确的替代解释后，这种效应根本不存在。此外，虽然对这种效应的观察性研究一直在努力解决内部有效性问题，但由于缺乏对从众效应的明确理论表达，实验研究忽略了通过提供理论上合理的治疗来最大化实验现实主义的关键要求。因此，我的论文对这种情况进行了评估，旨在通过明确定义正在发挥作用的概念，假设几种机制，这些机制可以基于对现有学术的全面阅读，并通过联合分析对这些机制进行测试，从而重新定位文献。这种方法使我能够通过提供多层次、细致入微和现实的处理方法，与理论密切接触，产生一个可以应用中介和调节分析的输出，以清楚地确定哪些机制在数据中得到证实，并直接控制替代解释。除了重建我们如何更广泛地思考从众效应和民意调查效应的概念贡献之外，我还将提供关于这种效应如何产生（如果它存在的话）的新颖经验见解，并进一步加深我们对如何在政治行为研究中部署联合分析的理解。

Political polling has come under increased scrutiny in recent years following notable failures to predict the 2015 and 2017 election outcomes, as well as the result of the Brexit referendum. Inquiries undertaken in light of this culminated in a recent House of Lords Select Committee report calling for the polling industry to 'get its house in order', but, notably, rejecting the case forbanning poll publication for any time period. This could partly be explained by the recent finding that such failures do not majorly diverge from the norm of historically endemic inaccuracy, but it has also been framed in terms of the influence of polls on public opinion. That is, pollsters welcomed the Select Committee's conclusions on the grounds that any potential polling blackout, founded on the idea that inaccurate forecasts could unfairly sway the vote, would fail to account for the fact that, in the absence of polls, people could still be influenced by their (even more prone to error) expectations of the outcome. This notion is curious, however, given the large and growing body of research putatively demonstrating the existence of a bandwagon effect, according to which voter exposure to opinion polls gives further advantage to the leading candidate. It would therefore seem, given the evidence presented in this literature, that polls affect electoral outcomes in a way many might deem undemocratic. I argue, however, that such an assertion is jeopardised by conceptual inadequacies within this body of work. Not only is the bandwagon effect loosely and differently defined by different researchers, but also very few attempts have been made at identifying the causal mechanisms through which this effect is had. It may be that the bandwagon effect only exists among certain individuals, contingent on certain characteristics or circumstances, related for example to identity, partisanship, knowledge, emotions, political engagement or efficacy. It may even be that the effect does not exist at all when controlling for the correct alternative explanations. Moreover, while observational studies of this effect have struggled to deal with issues of internal validity, in the absence of clear theoretical expressions of the bandwagon effect, the crucial requirement of maximising experimental realism - by providing theoretically sound treatments - has been overlooked in experimental studies. My thesis, therefore, takes stock of this situation and aims to reorient the literature by clearly defining the concepts at play, hypothesising several mechanisms through which these may operate based on a comprehensive reading of existing scholarship, and testing these through conjoint analysis. This method allows me to engage closely with theory by offering multi-layered, nuanced and realistic treatments, produces an output to which mediation and moderation analyses can be applied to identify clearly which mechanisms are borne out in the data, and directly controls for alternative explanations. In addition to the conceptual contribution of reestablishing how we can think about bandwagon effects and poll effects more broadly, I will thus offer novel empirical insights into how any such effect might be had, if it exists at all, and further our understanding of how conjoint analysis can be deployed in studies of political behaviour.