Improving the voluntary integration of statistical forecasts and judgment

When regular time series patterns are disturbed by foreseeable special events, judgmental modifications of statistical forecasts may improve accuracy by allowing the estimated effects of these events to be incorporated into the forecast. However, previous research has found that judgmental forecasters use statistical forecasts inefficiently – they make unnecessary changes to reliable forecasts and ignore forecasts that form an ideal base-line for adjustment. An experiment was conducted to test the effectiveness of three simple methods that were designed to improve the use of statistical forecasts by judgmental forecasters: (a) making the statistical forecast the default and requiring an explicit request to change this forecast, (b) requiring the judge to record a reason for changing the statistical forecast and (c) eliciting adjustments to the statistical forecast, rather than revised forecasts. The first two methods led to improvements in the utilisation of statistical forecasts and improved accuracy. The third method was less successful.