Pareto Diagram Analysis

Pareto analysis provides the mechanism to control and direct effort by fact, not by emotion. It helps to clearly establish top priorities and to identify both profitable and unprofitable targets. Pareto analysis is useful to:

Prioritize problems, goals, and objectives
Identify root causes
Select and define key quality improvement programs
Select key customer relations and service programs
Select key employee relations improvement programs
Select and define key performance improvement programs
Address the Vital Few and the Trivial Many causes of nonconformance
Maximize research and product development time
Verify operating procedures and manufacturing processes
Product or services sales and distribution
Allocate physical, financial and human resources

Pareto analysis may be applicable in the presentation of Performance Indicators data through selection of representative process characteristics that truly determine or directly or indirectly influence or confirm the desired quality or performance result or outcome. Typical observations from Paretodiagram analyses might reveal:

80% of all warranty repairs of a product were attributed to 20% of its parts.
75% of quality defects result from 15% of operations within a process.
10% of the items inventoried represent 70% of the total cost of inventory.

Pareto diagram observations may show the following: (1) that the bars are roughly the same size; (2) that it takes more than half of the categories to determine 60-80 percent; or (3) if the most frequent problem is not the most important. STOP after the first problem is resolved and develop a more discrete set of nonconformance cause parameters to survey, analyze, and diagram.

Process Capability

Process Capability is a determination of whether an existing process is capable of attaining the specified (or desired) performance. Process Capability has a precise, limited definition in a manufacturing context. This determination is based upon observing the history of the process output data. Statistical Process Control is used to determine the expected bounds of the data. These bounds (which are the three standard deviation control limits) are compared to the manufacturing specification for the process. A more general utilization of this concept can be applicable to Department of Energy facilities and processes.

Statistical Control

In order to assess process capability, the process must first be in statistical control. The data that have been charted must be reviewed against their three standard deviation control limits. If no data points are outside of the control limits, and no discernible trends are detected (using the criteria given in the section on Control Charts beginning on Page 2-13), then the process is in control. The original work by Dr. Shewhart emphasized strongly that a process should not be declared in control unless the pattern of random variation has persisted for some time and for a sizable volume of output. He recommended taking at least 25 samples (data points) prior to declaring a process is in control. However, assessments using less data may be made. Using less data for the assessment takes some experience and can become more of an art than a science.

In some cases, a process with a single data point outside of the control limits may still be assessed as in control. To do this requires investigation of the cause(s) for the single data point being unusual. In short, if you can assign a reason for the value of the out-of-control point, and you can reasonably state that such an event will not reoccur, then you should disregard this point, treat it as an outlier, and not include it in the average or control limit calculations.

With the determination that the process is in control, you have also accepted the hypothesis that the process data will continue to behave as the past data have. It will continue to have the same average, and 99.7% of all future data points will fall between the upper and lower (three standard deviation) control limits. All variations that occur in the future data can be assumed to be due to random variation. This has important implications for management of the process, which will be explored later in this section.