Simplification errors in predictive models

Organizational and political responses to strategic surprises such as the credit crunch in 2008 and the pandemic in 2020 are increasingly reliant on scientific insights. As a result, the accuracy of scientific models has become more critical, and models have become more complex to capture the real-world phenomena as best as they can. So much, so that appeals for simplification are beginning to surface. But unfortunately, simplification has its issues. Too simple models are so generic that they no longer accurately describe or predict real-world cause-effect relationships. On the other hand, too complex models are hard to generalize. Somewhere on the continuum between too simple and too complex lies the optimal model. In this article, the authors contribute to the ongoing discussion on model complexity by presenting a logical and systematic framework of simplification issues that may occur during the conceptualization and operationalization of variables, relationships, and model contexts. The framework was developed with the help of two cases, one from foresight, a relatively young discipline, and the other from the established discipline of innovation diffusion. Both disciplines have a widely accepted foundational predictive model that could use another look. The shared errors informed the simplification framework. The framework can help social scientists to detect possible oversimplification issues in literature reviews and inform their choices for either in- or decreases in model complexity.