Extracting causal relationships from data generated by disparate structural experiments

This article presents a framework for inferring quantitative causal relationships from data generated by disparate structural experiments (DSEs). The term DSE is introduced to describe a set of experiments that, individually, aim to answer a unique research question (or set of questions), but the associated data are aggregated and used for a common purpose. In contrast to DSEs, controlled structural experiments (CSEs) are designed with a single goal (or small set of related goals) where the issue of confounding is addressed in the design of the test matrix. Consequently, the variable relationships established from CSE data can be directly interpreted as being causal. On the other hand, inferring causal relationships from DSE data requires addressing the issue of confounding in the analysis stage. The proposed framework leverages the language, principles, and methods from the broad field of causal inference that have been developed over the last few decades. Specifically, the overall approach centers on the matching technique, which was developed to extract causal relationships from observational data (i.e., data generated by an uncontrolled or natural experiment). A case study is presented utilizing a data set generated by DSEs performed on reinforced concrete shear walls. Using the proposed framework, the causal effect of two key design variables on the drift capacity is quantified. The results obtained from the causal analysis are markedly different from those produced by traditional statistical techniques.