Abstract
<p>The Causal Loop Diagram (CLD) method is a structured approach for constructing conceptual representations of how variables may causally relate to one another, developed either individually or collaboratively with a group of experts. However, CLD construction is labor-intensive, time-consuming, and cognitively demanding. Large Language Models (LLMs), with their advanced text processing capabilities and extensive knowledge base, offer the potential to support this process. This paper presents CLDassist, an R package and Shiny app that supports CLD construction by integrating LLMs into the CLD construction process. Researchers can use CLDassist to define a list of putative variables, then query an LLM to evaluate causal links between these variables and provide supporting literature. Rather than replacing expert judgment, CLDassist functions as a thinking partner that helps researchers explore causal relationships and organize causal structures. We describe how LLMs can be used to assist in the CLD construction process and demonstrate empirically that the agreement between CLDassist and human experts is comparable to the agreement observed between experts themselves. In addition to helping researchers generate an initial CLD, CLDassist may reduce tunnel vision in the modeling process by suggesting alternative perspectives and highlighting variables or relationships that may otherwise be overlooked. By combining the established CLD method with the generative capabilities of LLMs, CLDassist provides users with a standardized, multi-stage framework for supporting the early stages of theory formalization.</p>