In this series we have been looking at the principles of DMAIC. Last issue we decomposed Measure; this issue we will look at step three, "Analyze". In issue 15 we had examined the collection of DMAIC and how they link, now we are gaining knowledge regarding each principle, as individuals.
Step Three: "Analyze"
Previously we have examined Define and Measure. As you may recall, the Measure Phase was to collect the expert opinion (tribal knowledge). Everything in Measure is organized to collect all the possible X / Y relationships and to prioritize them. The prioritization depends solely on the opinion of the experts and is accomplished through a system of forced ranking.
This ranking process includes process maps, prioritization matrixes, and FMEA. The result of this rather exhaustive process is the identification of the Red X's (the most germane X / Y relationships). The Analyze Phase brings statistical logic to the project in order to validate the expert opinion.
Define, Measure, Analyze are for the express purpose of delivering the most useful X's to the Improve Phase. In other words, to make sure you are digging in the right place. The Six Sigma Methods have been carefully organized to deliver the right X's for active manipulation during improvements.
The last thing the team does, during Measure, is to plan the data collection and measurements necessary to validate the X's that are moving forward to Analyze. Additionally, this must include measurement system analysis (MSA). Clearly, measurement systems can contribute uncontrolled variation to the studies and confound the results.
The success of the Measure Phase relies strongly on proper MAS and planning. We can continue to think of this process as moving down through the funnel. As we move down the funnel we discard X's and validate others. This is all part of the assurance that when we begin active manipulation we are acting on the key components.
Many statistical tools are available for the Black Belt's use during "Analyze". The important thing to remember is "no active manipulation" during "Analyze". This means only passive data analysis. That is to say, "study the as is condition".
Many of the statistical methods used fall under the general category of Multi Vari Studies. Below is a brief description of just a few of the methods available to the Black Belt.
Regression and Correlation:
This is a very powerful method for understanding the X / Y relationships. It provides a mathematical model of cause and effects relationships.
Statistical Process Control (SPC):
During the Analyze Phase SPC is used in a passive manner in order to study process behavior over time.
This is important if a specification is involved. It is necessary to see how the process behavior links to the requirements.
Many times it is important to compare two or more variables. Without the aid of statistical significance testing, such as ANOVA, it would be impossible to truly understand differences.
Fraction factorial designs are used to identify the stronger X / Y relationships and to reduce the model even further. This will deliver fewer X's to the Improve Phase and allow for effective optimization designs.
Test for Normality and Stability:
Data that is not normal (symmetrical) and / or not stable (exhibiting special causes) is generally not suitable for further analysis.
This has been a small snapshot into the statistical logic that can be brought to bear. Remember, "Data properly tortured will talk". Profound understanding of the X / Y relationships is essential in order to affect true breakthroughs during the Improve Phase.