Getting valid samples and data is just as important as applying any statistical tool.

The people and methods used for taking the samples and data affect the randomness and accuracy of both. Also, the product population is changing as the process changes, sometimes quite radically and often.

It is generally not possible to sample all the mini-populations caused by the people and changes caused by the process.

Use the fishbone diagram to identify the key process input variables (KPIVs) that cause all these mini-populations. Use the problem definition and close analysis of the fishbone to limit your focus.

Generally, the easiest approach to improving a process's output quality is to center the total process or change the process average, rather than reducing the variation. For example, if a diameter of a product is running off center, it is generally easier to get the average back on center than to reduce the process variation.

If the variation is very high, you may have no choice but to attempt to reduce it. However, not only is improving the total process average easier, but the sampling process and valid data issues are minimized. In both cases, use a fishbone diagram to help identify the most stable process (day, shift, operator, product, etc.) to test.

Take a statistically valid sample before and after a change to be confident that the improvement was significant. The formulas for minimum sample size are covered later in the text.

Once the change is implemented, validate the effect on the total process. You can use quality department data for this validation, if the means of collecting data stays consistent before and after the change. If you can't trust the quality department data, you will have to take additional samples from the populations before and after the change to validate that the predicted improvement truly happened.