How complex a learning curve model need we use?

In using learning curves for management control in a typical industrial environment, we seek to identify a number of patterns in the basic data, each of which is an important source of information to be fed into the decision-making machinery. These patterns may beclassified as follows: (a) A trend-line, which in some `best¿ sense, can be used for predicting future output. This trend-line can be influenced by proper design and planning of the product line. (b) `Normal¿ scatter about the trend-line, which constitutes a natural and acceptable variation, and which can be used for setting upper and lower bounds predicted output. (c) `Abnormal¿ scatter about the trend-line, which results in an unacceptable variation. It indicates an avoidable loss in production which can be traced to an assignable cause and hence eliminated by management control. (d) `Deterministic¿ changes in the trend-line. These may be long or short term, and have an assignable cause. An example of a management-induced cause is a planned change in the size or constitution of the direct labour force. To derive a learning curve model which will cope with these four patterns simultaneously is a complex problem. The author believes there are considerable advantages in selecting the simplest model which is adequate for the purpose of efficient management control of a particular enterprise and will review a procedure for doing so. We are, after all, dealing with huge cost savings if we properly plan this activity. Understanding and implementing a simple model derived on the back of an envelope can be often more profitable for management than a sophisticatedcomputerized model, the significance of which is difficult to grasp. The paper concentrates attention on the time constant model, and its variants, as found appropriate to `industry learning¿.