Reliance, trust and heuristics in judgmental forecasting

Judgmental forecasting gives light to the use of computers in human decision making. This paper reviews studies in judgmental forecasting focusing on what has been learned from human judgment and human-computer interaction. Available information was analyzed in the framework of three dimensions: reliance and trust on computer suggestions and heuristics employed by forecasters to produce forecasts. Results show that computer's advice disuse is pervasive in forecasting; and the disuse increases with higher task complexity and lower perceived system performance. Explanations and past performance are good candidates to increase trust in computer's advice, but the appropriate format to deliver this information is still controversial. Forecasters usually overforecast but report to prefer underforecast, which can lead to a cognitive dissonance and in turn to conflicting goals in the task. Heuristics research in time series forecasting indicates that forecasters heavily assess their own judgment, which in turn tend to be grounded in last outcomes and an overall evaluation of features like mean, trend and autocorrelation. It appears that heuristics not always lead to harmful biases for the forecast.