Cpk vs Ppk: shortterm vs longterm process variation

When calculating SPC control limits, many find out that even the data they use to calculate those limits already breaks rules within that same reference period. That shouldn't be possible, right? Standard deviation calculations tend to estimate wider variation than what's observed in the sample data.
Well, it is. And it's because of how sigma is estimated: not by calculating the standard deviation over the whole sample, but by estimating it based on the average (moving) range observed.
The same happens with Cpk vs Ppk - with Cpk using that same estimation from range and Ppk calculating over all values in the data set.
That is by design, though, because for SPC and Cpk you want to know short-term variation and specifically try to differentiate between everyday process variations and real process shifts.
#continuousimprovement #sixsigma #spc

[rohanpatil]

Very perfectly explained, Looking forward for such more contents in a world where we only see theoretical or bookish explainations.

[domenicoscarpino]

Great lesson. It took me some time to fully get it, but I eventually got there! This topic is not really very often addressed in a technical way like shown in the video. Tom, you clarified everything. Thank you!

[deanopenn]

Hi Tom, i have seen a scenario where If PP and PPK fail to achieve the minimum requirement e.g. 1.67 but CP and CPK passes. The big factor is the tolerance is so small it doesn't leave much room for any process variation as PP and PPK is long term. What are your thoughts on this due to tight tolerances?

[sshaxy]

Please be careful with you language as you explain these statistics. ppk 1 is not "horrible".. its likes a 0.1% scrap rate which depending on the situation can be EXCELLENT... you also generalize long term short term as "today" vs the "whole year". This can lead to misconceptions and conclusions from your audience.