Forgetting and the Learning Curve: A Laboratory Study

The industrial learning curve is widely used to predict costs and labor requirements wherever learning is taking place. Little is known, however, about the reverse of this process: the forgetting that occurs during production interruptions. The ability to estimate cost increases due to forgetting would be useful for economic lot size determinations, bidding on repeat orders, estimating the cost of strikes, and so on. Empirical studies apparently have not been published. Field data are difficult to obtain and easily confounded by extraneous variables. Thus a laboratory experiment was undertaken to test selected variables that should (or should not) affect forgetting. A review of relevant psychological literature reveals two key findings: (1) Forgetting may be negligible for "continuous control" tasks but considerable for "procedural" tasks. Bicycle riding is representative of continuous control, while operating a computer is clearly procedural. (2) Forgetting is a function of the amount learned and the passage of time, not of the learning rate or other variables. The experiment employed paid subjects who worked between four and eight hours to learn both a procedural task and a continuous control task (assembling and disassembling a mechanical apparatus). Each subject returned at an assigned time, up to 114 days later, and repeated the task for about four hours. Thirty-one of 35 subjects produced usable data. For purposes of the study, "forgetting" was defined as the excess of actual time over learning-curve-predicted time, summed over the first four trials after the interruption. "Amount learned" was defined as the achieved reduction in time-per-unit before interruption. As hypothesized, forgetting was a function of amount learned and elapsed time, with 71 percent of the variance being thereby explained in a regression equation. Forgetting rate does not appear to be related to learning rate, contrary to conventional wisdom in the learning curve literature. Additional findings are (1) that the relearning rate is not correlated with the learning rate, but may be related to skill factors and the degree of original learning; (2) that subjects have poor insight into their memory states; and (3) that, within the specific task, learning rate is highly correlated with the time taken to complete the first unit. Further research is indicated to investigate whether the forgetting rate may be constant, and therefore applicable across broad classifications of tasks. The determinants of relearning also are a suggested area of investigation. The results of this study apply to individual learning curves, and questions remain concerning the aggregation of individual performance into group performance.