Qualitative Comparative Analysis: Addressing Methodological Challenges, with particular reference to survey data

拉金的定性比较分析(QCA)是一种创新的研究方法，旨在建立一个结果的必要条件和/或充分条件的构型。它在社会科学中越来越多地被使用，通常是由研究人员根据他们的分析提出政策建议。因此，重要的是它的用户和政策制定者了解它的优点和缺点。我们将评估QCA存在的突出问题，并制定应对措施。QCA反映了拉金的假设，即社会科学中的许多理论都与案例集有关，特别是它们特征的因果属性，单独和结合考虑。案例被视为特征的组合，后者使用集合来描述。例如，考虑一下为什么白人工薪阶层男孩的教育成就较低的理论主张。这些是关于三组案件的索赔：白人、工人阶级和男性。QCA会评估，在多大程度上，成为“白人工人阶级男孩”是取得某种结果的充分或必要条件。它还对这三个因素的存在或不存在的所有其他组合进行评估。QCA有一种形式使用CRISP集合，其中一个案例简单地位于某个集合中或不存在于某个集合中，以及一种使用模糊集合的形式，其中CASE可以在集合中具有部分从属关系，从0到1。模糊集合QCA的发展是为了解决CRISP集合不能很好地描述某些因素的问题。想想“成人期”吧。18岁的人不是简单的成年人或非成年人。在这组成年人中，模糊的成员资格分数可能是合适的，比如0.8。某些结果的充分或必要条件的程度可以用模糊集来评估，就像对清晰集一样，但也会出现额外的复杂情况。QCA是在政治学中为少数情况发展起来的。自那以后，包括我们在内的研究人员将其用于调查数据，取得了丰硕成果。这类数据集最常用的方法是回归分析形式。有了回归，评估充分性和必要性的关系就难得多了。回归在确定假设的独立变量的净影响时，也存在忽视特定案例的背景以及因果关系的构型方面的危险。QCA解决了这些问题。然而，鉴于与更成熟的方法相比，它仍处于发展的早期阶段，它自身存在的问题尚未解决，我们将在三个方面进行工作：采样、技术问题和对QCA结果的解释。抽样问题包括有偏差的样本和丢失的数据如何影响QCA结果。技术问题包括，例如，由于使用模糊逻辑而产生的逻辑悖论，例如在某些情况下，条件既能满足结果，又能满足结果的缺失。另一个重要的技术问题是校准，即为案例分配模糊值(例如，对于18岁的“成年人”，模糊值为0.8)。校准影响哪些条件配置被识别为对结果是充分和/或必要的。解释涉及一些问题，如是否以及如何根据QCA结果做出因果声明，以及QCA发现的多条因果路径是否可能反映了对结果的不充分概念化。我们将进行数学逻辑工作，但也将使用现有的数据集。后一项工作将作为开发更抽象的论点的试验台，提供使用QCA的说明性案例研究。它还将使我们能够评估QCA和传统技术作为分析各种类型和不同目的数据的方法的相对优点，并使我们能够讨论QCA和回归技术的相对优点，作为向政策制定者提供信息的手段，使他们了解调解产生社会期望结果的复杂因果路径。

Ragin's Qualitative Comparative Analysis (QCA) is an innovative research method which aims to establish the configurations of necessary and/or sufficient conditions for an outcome. It is used increasingly within the social sciences, often by researchers who make policy recommendations based on their analyses. It is therefore important that its strengths and weaknesses are understood by its users and policymakers. We will evaluate outstanding problems with QCA, and develop responses to them. QCA reflects Ragin's assumption that much theorising in the social sciences concerns sets of cases, especially the causal properties of their characteristics, considered singly and in combination. Cases are treated as combinations of characteristics, with the latter described using sets. Consider, e.g., theoretical claims about why white working class boys have low educational achievement. These are claims about cases in the intersection of three sets: white, working class and male. QCA would assess the degree to which being a "white working class boy" is a sufficient or necessary condition for some outcome. It also assesses this for all other combinations of the presence or absence of these three factors.QCA has a form employing crisp sets where a case is either simply in or out of some set and a form employing fuzzy sets where a case can have partial membership in a set, scaled from 0 to 1. Fuzzy set QCA was developed to address the concern that some factors are not well described by crisp sets. Consider "adulthood". An 18 year old is not simply adult or non-adult. A fuzzy membership score of, say, 0.8 in the set of adults might be appropriate here. The degree to which conditions are sufficient or necessary for some outcome can be assessed for fuzzy sets, as for crisp sets, but additional complications arise.QCA was developed in political science for small numbers of cases. It has since fruitfully been used with survey data by researchers including ourselves. The most commonly used methods for such datasets are forms of regression analysis. With regression, it is much harder to assess relations of sufficiency and necessity. Regression, in determining the net effects of supposedly independent variables, also carries the danger of ignoring, as well as the configurational aspects of causation, the context of particular cases. QCA addresses these problems. However, given that, compared to more established methods, it is still in its early stages of development, it has unresolved problems of its own which we will address.We will work in three areas: sampling, technical problems and the interpretation of QCA results. Sampling concerns include how biased samples and missing data affect QCA results. Technical concerns include, e.g., logical paradoxes arising from the use of fuzzy logic, such as situations where conditions can be sufficient for both an outcome and its absence. Another important technical issue is calibration, i.e. the allocation of fuzzy values to cases (e.g., 0.8 for an 18 year old for "adulthood"). Calibration affects which configurations of conditions are identified as sufficient and/or necessary for an outcome. Interpretation concerns issues such as whether, and if so, how, causal claims can be made on the basis of QCA results, and whether multiple causal pathways uncovered by QCA might reflect inadequate conceptualisation of outcomes.We will undertake mathematical-logical work, but also work with existing datasets. The latter work will serve as a test bed for the more abstract arguments developed, providing illustrative case studies of the use of QCA. It will also enable us to evaluate the relative merits of QCA and conventional techniques as ways of analysing data of various types and for various purposes, and it will allow us to discuss the relative merits of QCA and regression techniques as means of informing policymakers about the complex causal pathways that mediate the production of socially desirable outcomes.