Why are organizational risk models so insensitive?

Several research projects and programs on system safety engineering and Quantitative Risk Analysis in the last 40 years offered very strong evidences of the crucial role that human and organizational factors (HOFs) play in major accidents. According to this increasing concern toward the relevance of HOFs in limiting safety performance of complex socio-technical systems, considerable research effort has been spent worldwide in the last couple of decades. This resulted in a quite rich literature covering from theoretical bases, to accident investigation methods and application to major disasters, to very sophisticated modeling approaches and techniques of HOFs in Quantitative Risk Analysis. Nevertheless, many of the models and applications described in scientific literature demonstrate very limited sensitivity of the accident event probability to the variation of single or multiple HOFs, or an "often obfuscating numerology"[1]. The paper proposes a critical review of the literature on the modeling strategies and techniques of HOFs, in order to point out major current limitations and to partially explain the argued limited sensitivity of these models. Finally, the paper explores five different critical investigation topics as likely origins of the limitations suffered, offering suggestions on additional research questions and methods able to provide further insight.