Learning Curves: Accuracy in Predicting Future Performance

Abstract

Many repetitive construction field operations exhibit a phenomenon known as the learning or experience effect. A learning curve is generated when the time or cost required to complete one cycle of an activity is plotted as a function of the cycle number. For practicing construction engineers and managers, the greatest potential value of learning curves lies in their ability to predict future performance, instead of fitting historical data. This paper presents a new method for using learning curves to predict the time or cost to complete the remaining cycles of an activity in progress, to assess the accuracy of this method, and to compare the accuracy of this method with the standard forecasting technique used in construction cost reporting. Using the proposed method, the accuracy of predicting the time or cost required to complete an ongoing activity improves dramatically for about the first 25–30% of the activity and then levels off to within 15–20% of the actual value. Compared to the standard method using the cumulative average, the new learning curve method is shown to be more accurate. The analysis quantifies the trade-off between accuracy of predicting future performance and the timeliness and potential value of such a prediction.