Avoiding algorithm errors in textual analysis: A guide to selecting software, and a research agenda toward generative artificial intelligence
通过评估LIWC、DICTION等四款软件，发现同方法族软件结果高度一致，而不匹配工具会导致显著错误，并据此提出结构化选择指南，帮助研究者提升构念效度，为AI文本分析奠定基础。
• We develop a systematic process to select textual analysis software for complex constructs. • Our study evaluates four software packages using value-based management (VBM) as a test case. • We show that software from the same methodological family yields near-identical results. • We quantify how unsuitable tools distort results despite being established in other fields. • Our framework links AI prompts to theory-driven constructs for valid analysis. The use of textual analysis is expanding in organizational research, yet software packages vary in their compatibility with complex constructs. This study helps researchers select suitable tools by focusing on phrase-based dictionary methods. We empirically evaluate four software packages—LIWC, DICTION, CAT Scanner, and a custom Python tool—using the complex construct of value-based management as a test case. The analysis shows that software from the same methodological family produces highly consistent results, while popular but mismatched tools yield significant errors such as miscounted phrases. Based on this, we develop a structured selection guideline that links construct features with software capabilities. The framework enhances construct validity, supports methodological transparency, and is applicable across disciplines. Finally, we position the approach as a bridge to AI-enabled textual analysis, including prompt-based workflows, reinforcing the continued need for theory-grounded construct design.