Why Control Limits are not the same depending on type of exclusion of data points

[zamshabeer]

Hello

I have just read this article on the Minitab blog regarding the 3 options in Minitab for excluding data points from control limit calculations. I tried to post the hyperlink, but the forum rules does not allow me to post the link until I make 10 posts. If you google "Leaving Out-of-control Points Out of Control Chart Calculations Looks Hard, but It Isn't" by Cody Steele | 01 June, 2017, you can access the Minitab blog article. Leaving Out-of-control Points Out of Control Chart Calculations Looks Hard, but It Isn't

My question: Why is the upper control limit different in option 1 when compared to options 2 and 3?

I would have expected the mean and control limits to remain the same irrespective of how you exclude the data points in the calculation. It is getting difficult to explain to my manager and client why these small differences are acceptable.

Thanks
Alakazam

 

[John Broomfield]

Control limits are a function of the data being analyzed. Omit data and the control limits may change. We want to be able to clearly see what data points need further investigation.

This blog suggests that you may see more clearly what data points need further investigation if you omit the data from “known” causes. The data are not permanently omitted but can be hidden temporarily to highlight further improvement opportunities.

BTW, caution is due because of the unknown causes.

To me hiding some data points is a bit like using a moving average to smooth the plot so the trend can be seen more clearly.

 

[zamshabeer]

Control limits are a function of the data being analyzed. Omit data and the control limits may change. We want to be able to clearly see what data points need further investigation.

This blog suggests that you may see more clearly what data points need further investigation if you omit the data from “known” causes. The data are not permanently omitted but can be hidden temporarily to highlight further improvement opportunities.

BTW, caution is due because of the unknown causes.

To me hiding some data points is a bit like using a moving average to smooth the plot so the trend can be seen more clearly.

Hi John

Thanks for your response. I understand that omitting data will impact the control limits.

But my question is slightly different. In the article, it shows three ways that omitted data can be displayed.
1) Not displayed at all
2) Displayed as gaps in the graph
3) Displayed in the graph

In all 3 cases, the data is omitted from the calculations of the control limit.

I would have expected:
1) that in all 3 cases, the control limit to be different to the original data set (this is true in the article - so no issues here)
2) that in all 3 cases, the control limits to be the same. Here is my issue. They are not the same in all 3 cases. I am trying to understand why when in all 3 cases we have omitted the offending points from the calculation.

Hope that distinction is clear.

Thanks
Best regards
Alakazam

 

[Miner]

Only Minitab can say for certain, but what I suspect is happening is:

• Option 1 (Like it never happened) - When these points are excluded, Minitab probably calculates a moving range between the points before and after the exclusion. This additional moving range will impact the control limits.
• Options 2 (Guilty by omission) & 3 (Hide nothing) - Since these points are acknowledged and either shown by a gap or as a red dot, Minitab probably does not calculate a moving range between the points before and after the exclusion.

The only way to know for certain is to ask Minitab, or obtain the raw data and calculate it manually.

While Minitab is a great software package, it has some annoying inconsistencies in terminology and in how later enhancements match up with earlier features.

 

[Miner]

I found the data used for the blog and analyzed it. My suspicions were correct. I recreated their analysis, but included the moving range charts. If you look closely, you can see where the moving ranges were either included or omitted.

 

[Matt Savage]

Hi John

Thanks for your response. I understand that omitting data will impact the control limits.

But my question is slightly different. In the article, it shows three ways that omitted data can be displayed.
1) Not displayed at all
2) Displayed as gaps in the graph
3) Displayed in the graph

In all 3 cases, the data is omitted from the calculations of the control limit.

I would have expected:
1) that in all 3 cases, the control limit to be different to the original data set (this is true in the article - so no issues here)
2) that in all 3 cases, the control limits to be the same. Here is my issue. They are not the same in all 3 cases. I am trying to understand why when in all 3 cases we have omitted the offending points from the calculation.

Hope that distinction is clear.

Thanks
Best regards
Alakazam

It looks like what is described is these methods:

• Filter the data out of the chart. In essence, the data does not exist and never did.
• Filter the data out of the chart but show gaps in the data line to indicate missing data.
• Show the data but add some sort of indicator to identify the outliers and the fact that the outliers were not used.

One common way to deal with outliers:

• Identify the outliers.
• Determine if there is an assignable cause (reason) for the outlier.
• Assuming the outlier is undesirable, take steps to prevent it from occurring again.
• Don’t use that outlier in the control limit calculations since it represents an event that should not occur again.
• Rather than “hide” the outlier, treat it as a special-cause and show it as a “ghosted” point. This gives the reader of the chart information where the data value was and because it looks different, they are likely to better understand that outlier is not used in the calculations.

In the following chart created with SQCpack, I am using financial data to show what I am describing.

 

[zamshabeer]

I found the data used for the blog and analyzed it. My suspicions were correct. I recreated their analysis, but included the moving range charts. If you look closely, you can see where the moving ranges were either included or omitted.

View attachment 28132

Wow!!! amazing!! and thank you so much !!!

So it looks like it's only half omitting those data points when you ask Minitab to omit those data points during the calculation.

Makes me want to just use option 1 which sounds to be more accurate where those data points have been completely omitted from the calcs. Such a shame that this defeats the purpose of the "Estimate" tab and "Omit subgroups" option in the I-Chart options dialog box.

 

[Miner]

The differences are small, and would make little practical difference, but it is another of those annoying inconsistencies I mentioned. You get that when you have different coders writing different portions of the software without a higher level standard uniting those sections.