Computing CPK for bimodal data, either large readings or zero (majority)..

[#1]

Hi,

I have data points from measurements of a certain defect. The readings are either zero(majority) or high readings.

For example: How do i get the CPK if the USL is 100? Desirable is the zero reading.

60.1
76.8
76
68.3
43.9
60.3
42.8
66.1
49.4
62.1
49.8
66.5
61.2
58.8
59.1
60.3
67.2
68.5
54.6
61.0
64.2
57.7
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0

[#2]

I suggest that you use the Distribution Platform, to help you determine the best way to calculate the Cp/Cpk.  If you startup the Distribution Platform, and then under the red triangle, go to:

    Continuous Fit==>All

it will help you determine which distribution best fits your data, and then under that distribution, it will give you the Cp/Cpk.

[#3]

jabenir,  There is one fundamental question you have to answer before calculating Capability Indices:  Is your process in statistical control?  If the process is not in a "reasonable degree of control", then Capability Analysis is meaningless.  As Dr. Wheeler says in one of his books:  "What can be said for the capability of a process not I statistical control?  Not much!"

Other questions that need to be answered BEFORE a Capability Analysis:  Is the data you presented in time order?  If not, what is the order and what does a Runs Chart and/or Process Behavior Chart look like?  Can you determine defect rates for discrete periods of time with rational sub-grouping?