In Quality Assurance we must normally use samples of data drawn from a system that naturally exhibits variation we can make mistakes even in controlled experiments. The process either is in control or it is not or similarly we have a batch of parts of materials that has been generated by a system that either was or was not in control.
We can decide either to accept or to reject the output. If the process is in control and if, based on our information we would reject the output – then we have made an error called a type I error. We, the producer risk making such an erroneous decision on the basis of the probabilities that are associated with the inherent variability of the system and the sample size. Logically, this risk is called the producer’s risk because if the decision is made it is the producer who absorbs the loss.
Similarly there is a risk that we may accept output as a good product when in fact, the process is out of control. This decision is called type II error and is termed the consumer’s risk. In statistical control models, we can pre-set the probabilities of type I and type II errors.
Philosophically, the concept of the consumer’s risk may be in part the cause of American manufacturers’ misjudgement of what is really acceptable quality. In fact, from a broader competitive perspective both risks are the producer’s and we now recognize that the type II risk may be the most serious of the producer’s risks since it can result in the loss of customers and market share.
From the point of view of control methods, we can apply statistical control concepts by sampling the output of a process to keep that process in a state of statistical control (process control) or by sampling a lot of incoming materials to see whether it is acceptable (acceptance sampling).
In process control, we monitor the actual on-going process that makes the units. This allows us to make adjustments and corrections as soon as they are needed so that bad units are never produced in any quantity. This procedure is a direct application of the statistical control chart, and as with acceptance sampling parallel procedures are available for those situations in which sampling is done by attributes and for those in which measurements are made of variables that measure quality characteristics. Control procedures should be used by suppliers to control their processes prior to shipment. The acceptance sampling procedures are designed to ensure the quality of incoming materials.
Acceptance sampling lets us control the level of outgoing quality from an inspection point to ensure that, on the average no more than some specified percentage of defective items will pass. This procedure assumes that the parts or products have already been produced. We wish to set up procedures and decision rules to ensure that outgoing quality will be as specified or better. In the simplest case of acceptance sampling we draw a random, sample of size n from the total lot N and decide, on the basis of the sample whether or not to accept the entire lot. If the sample signals a decision to reject the lot the lot may be subjected to 100 per cent inspection during which all bad parts are sorted out, or it may be returned to the original supplier. Parallel acceptance sampling procedures can be used to classify parts as simply good or bad (sampling by attributes) or to make some kind of actual measurement that indicates how good or bad a part is (sampling by variables).
The previous material dealing with industrial quality control has clear objectives about what to measure and control and sophisticated methodology for accomplishing these ends. In non-profit organization however, the objectives and outputs seem less well defined and the control methodology seems relatively crude.
The profit motive provides a focus for all kinds of managerial controls, including quality. By contrast, non-profit organizations exist to render service, and their success is judged in those terms. Measuring the quality of the services is difficult in part because the attributes of quality are somewhat more diffuse. The quality characteristics of most of these kinds of services are multidimensional and are often controversial and reducing quality measurement to something comparable to specific dimensions or chemical composition may be impossible.