Webinar: A Novel Approach to Calculating Process Capability for Non-Normal Data
Process capability indices have been used for decades to quantify the amount of material not meeting specification. These calculations rely upon several assumptions, one of which is normality of the data. The more the normality assumption is violated accuracy of how much material exceeds specification becomes grossly distorted.
Historically statisticians have provided a variety of techniques to assist. Most of these require transformation of the data (Box-Cox, Johnson, distributional fits, etc.). However, applying transformations blindly without understanding the behavior of the data can lead to equally poor estimates. Unless one is a statistician, applying transformations and making any sense of it is at best confusing and still often not accurate. The percentile method is subject to sensitivity to sample size.
A different approach that is robust to all common problems (except a non-stable process) will be introduced. These common problems include outliers, non-normal data, multiple distributions, 0’s and negative values, threshold values, limitations to sample size, and more. It will allow the practitioner to more accurately estimate the defect levels than other approaches. No statistical software is required; all of this can be done using software such as Excel®️.