When I was taught six sigma, the first time, we focused on the use of the sigma level for the satellite level quality metric. It worked well, but it is difficult to understand what it means with respect to final quality. As a relative metric, one that can show the direction that quality is going, it works fine.
Why use it at all? This metric was introduced as a measure of your process quality, not the product quality. It was based on the rate that defects were being produced. What is called the DPMO (defects per million opportunities) Now this is the count of defects, which a single unit may have more than one, per opportunity. An opportunity was defined as any action or item that can experience or cause a defect. This is the most gamed part of six sigma, the opportunity count.
In an assembled circuit board, the opportunity count could be the sum of the components plus the number of connections on all the components plus the number of solder joints, plus the number of circuits in the board. The opportunity count should be higher on more complex items and smaller for simple products. It accounted for the complexity. We would measure the quality of a product divide it by the number of opportunities to create a defect and the result would be DPMO, or the process quality.
We would estimate the DPMO for all of the primary products, then combine them to estimate the total DPMO for the facility over a period of time, such as weekly. We would use the product DPMO’s to estimate the sigma level of the product, we would then use the facility total DPMO to estimate the facility sigma level. The corporation would then roll up all the site DPMO and calculate a corporate sigma level.
Recognize that this is all about attribute counting divided by opportunity counting that is converted to a sigma quality level. No assumptions of normality or any continuous data was used to calculate the DPMO or the sigma level. Now the sigma level was interpreted as a normally distributed analog for quality.
Since no one actually counts the full defect count, most places used the poisson distribution yield rate to count the defects. You should truly used rolled through put yield (RTY) not normal yield, since it is the count of defect free parts. Using the poisson distribution equation for zero defects, the yield rate = exp(-lambda) where lambda is the average defects per item.
Positives; the sigma level provided a product free measure of process quality.
Negative: you could change opportunity counts to make yourself look better.
The next blog will be the 1.5 sigma shift issue.