Capability analysis is an important topic in Six Sigma & Lean Six Sigma. It can be defined as the quantification or measurement of the ability of a process, product, or service to perform to some predefined customer requirement. By convention, there are two primary measures, or metrics, used to estimate capability: short-term sigma level and long-term DPMO, or defects per million opportunities. This chart shows the estimated long-term DPMO for various short-term Sigma levels. Sigma level is often thought of as a “positive” way to describe process capability since it relates directly to the compliance rate or yield of a process.
DPMO, on the other hand, relates to the defects or defectives in a process and thus is often referred to as a defect rate or defective rate. There are several important reasons for measuring the capability of your process, product, or service. First, it allows an organization to estimate the true quality levels for all of its goods and services. Second, it allows different organizations or department to compare their quality levels against each other. Measuring capability also enables an organization to allocate resources based on collected data. Lastly, for attribute output characteristics, a capability analysis will clearly and directly quantify the defect rate. For variable outputs, capability analysis will identify the nature of the problem as a centering issue, a spread-related issue, or a combination of both. As capability metrics indicate how well process outputs meet customer specifications, they are a common thread in all process improvement projects.
Capability analysis can be run on both attribute and variable data. The most common capability metric we’ll use is the sigma level, as mentioned earlier. This statistic is very versatile as it can be used for both variable and attribute measurements. Also, all other capability metrics can be converted to a sigma level.
Completing a successful attribute capability analysis require clear definitions of classifying defects and defectives. Therefore, operational definitions of defects, standardization, and training are all essential. For attribute data, we can use a process’ yield to express the proportion of outputs in compliance with the customer requirements. At its core, attribute capability analysis involves counting “good” units versus “bad” units. When running a variable capability analysis, sigma levels are estimated from process parameters such as the mean, standard deviation, and customer specification limits. The simplest estimate of sigma level is the number of standard deviations that can fit between the mean and the nearest specification limit. Specification limits can be two-sided, meaning there’s an upper and lower limit, or one-sided, meaning only an upper or lower limit exists.
Over time, a process output characteristic will always exhibit some form of variation. This variation can be divided into two sources short-term and long-term. Short-term variation is the variability in an output characteristic when most of the inputs contributing to that output characteristic are held constant. Generally, short-term samples of data will have less variation than long-term samples. A short-term sample will often allow us to estimate what’s known as the “entitlement” of a process under current stable conditions. The entitlement is the best we could ever hope to perform during conditions when inputs are constant and centered. In other words, entitlement is the best the process is capable of performing, given the current design. Long-term variation is variability in an output characteristic when most inputs are allowed to fluctuate in an unconstrained fashion. Long-term samples allow us to estimate the overall performance delivered to our customers, which represents the true process performance.
Although the definitions of short-term and long-term are sound, when measuring fluctuation within our inputs, we often don’t know what the key inputs are or how much they have fluctuated. Instead, this is something we discover during the course of a process improvement project. We can also use time span to determine if our data is short- or long-term. Although the time span will vary from project to project, most practitioners would agree long-term data usually represents data from a year or more, since annual cycles and events that may impact the process will have occurred. However, if the process is not subject to seasonal variation, three to four months may suffice. Short-term data can be difficult to collect without a specially designed study, but many practitioners would say that short-term data can be collected during a quarter or less. The time span in between one quarter and one year will be left to your best judgment—you may wish to label it “medium-term” data.
There are no hard limits for what’s considered short-, medium-, or long-term data. You must consider what time and resources are available for your specific project. Regardless of which time span you use, the important thing to remember is that you must always footnote or annotate your calculations as to which term it is. One final point to consider: the sheer amount of data you have does not determine whether you have short- or long-term data. It’s important to keep a clear distinction between the amount of data collected and the time span in which it was collected.
Perhaps the best metric to use for analyzing the capability of this process is DPMO, or Defects per Million Opportunities—a very common process improvement metric. Calculating this metric is a little more complex, since we must also account for the number of opportunities on each unit. This process is somewhat easier when each unit has the same number of opportunities, as in the case here. When the number of opportunities differs on each unit, such as with purchase orders with varying numbers of line items, counting opportunities per unit becomes much more challenging and may not be worth the effort. Metrics which utilize opportunity counting, therefore, factor in the complexity of a process. To calculate DPMO, we first calculate defects per opportunity or DPO. The DPO is the DPU divided by the number of opportunities per unit.
Check out our Statistical Process Control Training, Lean Training, Six Sigma Training, Lean Six Sigma Training, Continuous Improvement Training or the full range of Training Courses for relevant courses on how to streamline & improve your business processes.