The libraries on this site haven’t been update for quite a while. The correct answer should be 0.46. We have updated the library to fix this error in the IR Range calculation.
A great tool for getting to grips with Process Behaviour Charts. However, the assertion that the XmR Chart requires process data to be (approx.) Normal is incorrect. Dr. Donald Wheeler showed that limits based on the Empirical Rule work equally well for all manner of distributions. https://spcpress.com/pdf/DJW200.pdf
Thanks for the comment. I would agree that for the standard +-3 Sigma tests for control rule violation, the normality of the data is less of an issue, and certainly not required. The central limit theorem helps here. But this necessitates a non-obvious decision about the proper sample size, and the requirement that the sample subgroups be compiled into rational subgroups. Most of the expert articles on the subject, including the one by Wheeler, don’t address the issue of the Named Control Rules tests (WECO, Nelson, Juran, Hughes, etc), which a large number of the people who use SPC software also use. It seems counterintuitive that control limit tests that use and 1-sigma and 2-sigma control limit values are not going to affected by data that is skewed into some sort of top heavy distribution. If you know of any articles on that topic I would be interested.
Wheeler has written many articles on the use of Process Behaviour (Control) Charts and the importance of Rational Subgrouping. He also notes that the use of multiple “signal detection tests” only increases the likelihood of a false alarm which, in turn, diminishes the economic value and utility of the tool. Wheeler’s advice (based on decades of practical experience) is to look for the exceptional signals (points outside the limits) and for unexpected sustained shifts in the process (the so-called ‘run of eight’). Software applications make it easy to perform continual data analysis, but fixing errant processes is the best use of our time. Context is everything, but I would suggest that spreading one’s gaze across a range of process measures is better than fixating on a single one. Keep up the good work
I agree with your comment. Simply put, adding more control rules to a chart increases the likelihood of false positives. If you treat violations of all control rules with equal importance, you end up spending your time chasing the less important ones at the expense of the most important +-3 sigma rules. That said, the reality is that a large segment of SPC practitioners are fanatics about control rules (WECO, Nelson, Juran, Hughes, etc) and even a Phd in statistical science is not going to change their views. It’s a question of giving users what they want as opposed to what you think they need.
I tried Individual-Range Chart – 2 (Interactive)
input : 10.1,10.3,10.9,10.4,10.8,9.8,10.3,9.9,10.3,10.2
output: Range=0.51
Is this correct?
The libraries on this site haven’t been update for quite a while. The correct answer should be 0.46. We have updated the library to fix this error in the IR Range calculation.
A great tool for getting to grips with Process Behaviour Charts. However, the assertion that the XmR Chart requires process data to be (approx.) Normal is incorrect. Dr. Donald Wheeler showed that limits based on the Empirical Rule work equally well for all manner of distributions. https://spcpress.com/pdf/DJW200.pdf
Thanks for the comment. I would agree that for the standard +-3 Sigma tests for control rule violation, the normality of the data is less of an issue, and certainly not required. The central limit theorem helps here. But this necessitates a non-obvious decision about the proper sample size, and the requirement that the sample subgroups be compiled into rational subgroups. Most of the expert articles on the subject, including the one by Wheeler, don’t address the issue of the Named Control Rules tests (WECO, Nelson, Juran, Hughes, etc), which a large number of the people who use SPC software also use. It seems counterintuitive that control limit tests that use and 1-sigma and 2-sigma control limit values are not going to affected by data that is skewed into some sort of top heavy distribution. If you know of any articles on that topic I would be interested.
Wheeler has written many articles on the use of Process Behaviour (Control) Charts and the importance of Rational Subgrouping. He also notes that the use of multiple “signal detection tests” only increases the likelihood of a false alarm which, in turn, diminishes the economic value and utility of the tool. Wheeler’s advice (based on decades of practical experience) is to look for the exceptional signals (points outside the limits) and for unexpected sustained shifts in the process (the so-called ‘run of eight’). Software applications make it easy to perform continual data analysis, but fixing errant processes is the best use of our time. Context is everything, but I would suggest that spreading one’s gaze across a range of process measures is better than fixating on a single one. Keep up the good work
I agree with your comment. Simply put, adding more control rules to a chart increases the likelihood of false positives. If you treat violations of all control rules with equal importance, you end up spending your time chasing the less important ones at the expense of the most important +-3 sigma rules. That said, the reality is that a large segment of SPC practitioners are fanatics about control rules (WECO, Nelson, Juran, Hughes, etc) and even a Phd in statistical science is not going to change their views. It’s a question of giving users what they want as opposed to what you think they need.