Judgement or models: The importance of task differences

Abstract

Over twenty years research on cognitive biases and limitations has built a strong case for managers not to use their judgement if a suitable formula or model is available. Much of this work was built on research demonstrating that in an environment where the judgemental cues are well established, pre-specified and not auto-correlated and where there is little contextual information, that a model based on expert judgement (a model-of-man) is more accurate than the expert himself. However, much practical decision making takes place in a setting where cues are not well established, where auto-correlation is present, and where there is general contextual information. This study investigated decisions made in such a setting, namely time series forecasting. Forecasts were estimated judgementally for 111 real life time series. These estimates were then used to construct a model-of-man and the forecasts from this model compared in accuracy with the original judgemental forecasts. This study found the model-of-man not to be superior to man when assessed in terms of forecast accuracy thus demonstrating a commonly occurring task setting in which the dominant research result in not true.