Statistical methods to test hypotheses about direct and indirect effects from a person-oriented research perspective are scarce. For categorical variables, previously suggested approaches use configural frequency analysis (CFA) to detect extreme patterns (CFA Types/Antitypes) that are responsible for the observed direct and indirect effects. Existing methods rest on complex (log-linear) model comparison strategies and may perform poorly with respect to Type I error protection and statistical power. We, therefore, propose a simplified configural approach to answer the question “What carries a mediation process?” This simplified approach is based on two log-linear models that are needed to estimate (variable-oriented) direct and indirect effects. The first model identifies extreme patterns for the predictor-mediator path, the second model searches for extreme cells in the mediator-outcome path. Joint significance testing can be used to test the presence of mediation. Definitions of Mediation Types/Antitypes are given based on possible Type/Antitype patterns for the binary simple mediation model. In two Monte-Carlo simulation experiments, we evaluate the performance of the simplified approach in a homogenous population (i.e., where all individuals develop homogenously along a variable-oriented mediation mechanism) and a heterogenous population (i.e., where specific configurations, instead of a variable-oriented effect, drive the mediation process). Results suggest that the presented approach performs acceptably with respect to Type I error protection and statistical power. In general, larger sample sizes are preferable to reliably detect mediation-generating configurations. An empirical example is given for illustrative purposes and extensions and limitations of the proposed method are discussed.

从以人为导向的研究角度来检验关于直接和间接影响的假设的统计方法很少。对于分类变量，先前建议的方法使用配置频率分析 (CFA) 来检测导致观察到的直接和间接影响的极端模式（CFA 类型/反类型）。现有方法依赖于复杂（对数线性）模型比较策略，并且在 I 类错误保护和统计能力方面可能表现不佳。因此，我们提出了一种简化的配置方法来回答“什么进行调解过程？”这个问题。这种简化的方法基于估计（面向变量的）直接和间接影响所需的两个对数线性模型。第一个模型识别预测器-中介器路径的极端模式，第二个模型在中介-结果路径中搜索极端单元格。联合显着性检验可用于检验中介的存在。中介类型/反类型的定义是基于二元简单中介模型可能的类型/反类型模式给出的。在两个 Monte-Carlo 模拟实验中，我们评估了简化方法在同质种群（即所有个体沿着面向变量的中介机制同质发展）和异质种群（即特定配置，而不是一个变量导向效应，推动调解过程）。结果表明，所提出的方法在 I 类错误保护和统计能力方面的表现是可以接受的。一般来说，较大的样本量更适合可靠地检测产生中介的配置。出于说明目的给出了一个经验示例，并讨论了所提出方法的扩展和限制。