A Tool for Addressing Construct Identity in Literature Reviews and Meta-Analyses1
针对行为科学中构念命名不一致导致的同一性识别难题，设计并评估了六种基于自然语言处理的构念同一性检测器，其中一种组合算法显著优于其他方法，可作为专家进行文献综述和元分析的辅助工具。
The problem of detecting whether two behavioral constructs reference the same real-world phenomenon has existed for over 100 years. Discordant naming of constructs is here termed the construct identity fallacy (CIF). We designed and evaluated the construct identity detector (CID), the first tool with large-scale construct identity detection properties and the first tool that does not require respondent data. Through the adaptation and combination of different natural language processing (NLP) algorithms, six designs were created and evaluated against human expert decisions. All six designs were found capable of detecting construct identity, and a design combining two existing algorithms significantly outperformed the other approaches. A set of follow-up studies suggests the tool is valuable as a supplement to expert efforts in literature review and meta-analysis. Beyond design science contributions, this article has important implications related to the taxonomic structure of social and behavioral science constructs, for the jingle and jangle fallacy, the core of the Information Systems nomological network, and the inaccessibility of social and behavioral science knowledge. In sum, CID represents an important, albeit tentative, step toward discipline-wide identification of construct identities.