Model combination and adjustment

Abstract

Inside and outside views attempt to predict outcomes by examining a phenomenon’s causal structure or by using reference-class forecasting, respectively. Outside views are most useful when the reference class has a similar causal structure to what is being predicted, while inside views are more likely to be accurate when the phenomenon’s causal structure is well understood and there are few similar phenomena that can be used to make predictions. In cases where neither approach is likely to be highly accurate, the author recommends using a combination of model combination and adjustment, in which multiple outside views and reference classes are used to make a preliminary prediction that can be fine-tuned with inside information. – AI-generated abstract.