Abstract

The mechanistic approach in the cognitive and biological sciences emphasizes that scientific explanations succeed by analyzing the mechanisms underlying phenomena across multiple levels. In this paper, we propose a formal strategy to establish such multi-level mechanistic models, which are foundational to mechanistic explanations. Our objectives are twofold: First, we introduce the novel "mLCA" (multi-Level Coincidence Analysis) script, which transforms binary data tables from tests on mechanistic systems into mechanistic models consistent with those tables. Second, we provide several philosophical insights derived from the outcomes generated by this script and its underlying algorithm. Using illustrative examples, we defend the following claims: 1. Inference methods for generating mechanistic models generally require information on how causal factors are assigned to different levels within data tables generated by multi-level structures. 2. The mLCA script successfully produces appropriate mechanistic models from binary data tables, demonstrating the practical application of the philosophical mechanistic approach in the sciences. 3. The number of solutions generated by mLCA increases significantly as the number of relevant factors grows, reflecting adaptations in causal inference methods to meet the demands of multi-level mechanistic modeling. 4. Any further reduction of solutions, if possible, involves pragmatic considerations, a point that carries profound implications for the broader ambitions of the mechanistic approach. By addressing these points, our paper contributes both to the development of practical algorithmic tools and to a deeper philosophical understanding of multi-level mechanistic modeling.