What if your data set has 100 points, do you calculate the moving range, and mr bar for just the first 20 points, and use that to draw up the UCL and LCL? or do you take all 100 data points into consideration? At which points do you update the average and mr bar values?
Hi - Mark Graban here... you could use all 100 points if that's all the same system (meaning that the chart is "in control" based on the average and limits that are calculated). There's a bit of an art to choosing the data points for your limits. There are diminishing returns to using more than 25 data points for the average and limits. One approach would be to use the first 20 or 25 data points to calculate the average and limits... and then look at the remaining 75 or 80 data points. Do you see signals or evidence of any shifts (8 or more data points above or below that baseline average that you calculated)? Sorry for the slow reply. You can contact me through my book's website if you have more questions: www.measuresofsuccessbook.com/
In a 20-data point control chart, if any point/s between 6th to 19th went beyond the control limits or shows a special cause of variation, should we recalculate the chart starting from the "last point" where the special cause of variation was seen? OR better to "start" recalculating "only" after gathering more than 20 data points?
Hi Lei - If there are signals in the initial baseline time frame, Don Wheeler teaches that we should specifically NOT remove those data points from the calculations. My advice would be if the data starts showing signs of predictability, I would calculate new average and limits after more data points are established by that new "system." You don't need 20 data points... you can calculate preliminary limits with as few as four data points (per Wheeler again) -- the limits are BETTER and more valid if you have 15, or 20, or 25 data points. The limits get more valid, but there's diminishing returns / gains from having more data points. You can tweak your initial limits as you get more data points, but do NOT perpetually re-calcuate them. Establish the solid baseline and then only re-calculate if you see signals that the system has changed significantly again.
@@MarkGraban Sorry, I don't quite get it. For my first 20-data point chart, data point 6 went beyond the control limits. Now, I have 20 more data points. My question is, at which point should I re-calculate: (a) from point 6 from my 1st control chart i.e. when a special cause of variation was seen, or (b) from data point 21?
@@lly7454 Yes, this video shows just the X Chart. The full technically correct method includes the MR Chart, but there is minimal additional benefit to using both charts together in the real world.
![Leveraging a Strategy to Execution Framework A Journey of Transformation [Webinar] Karen Friedenberg](http://i.ytimg.com/vi/gwCiN6W-xlg/mqdefault.jpg)
![[Preview] Leveraging a Strategy to Execution Framework: A Journey of Transformation #webinar #lean](/img/n.gif)
Thanks in a million. Great stuff.
Nice video and very informative. Thanks
What if your data set has 100 points, do you calculate the moving range, and mr bar for just the first 20 points, and use that to draw up the UCL and LCL? or do you take all 100 data points into consideration? At which points do you update the average and mr bar values?
Hi - Mark Graban here... you could use all 100 points if that's all the same system (meaning that the chart is "in control" based on the average and limits that are calculated). There's a bit of an art to choosing the data points for your limits.
There are diminishing returns to using more than 25 data points for the average and limits.
One approach would be to use the first 20 or 25 data points to calculate the average and limits... and then look at the remaining 75 or 80 data points. Do you see signals or evidence of any shifts (8 or more data points above or below that baseline average that you calculated)?
Sorry for the slow reply. You can contact me through my book's website if you have more questions: www.measuresofsuccessbook.com/
In a 20-data point control chart, if any point/s between 6th to 19th went beyond the control limits or shows a special cause of variation, should we recalculate the chart starting from the "last point" where the special cause of variation was seen? OR better to "start" recalculating "only" after gathering more than 20 data points?
Hi Lei - If there are signals in the initial baseline time frame, Don Wheeler teaches that we should specifically NOT remove those data points from the calculations. My advice would be if the data starts showing signs of predictability, I would calculate new average and limits after more data points are established by that new "system." You don't need 20 data points... you can calculate preliminary limits with as few as four data points (per Wheeler again) -- the limits are BETTER and more valid if you have 15, or 20, or 25 data points. The limits get more valid, but there's diminishing returns / gains from having more data points. You can tweak your initial limits as you get more data points, but do NOT perpetually re-calcuate them. Establish the solid baseline and then only re-calculate if you see signals that the system has changed significantly again.
@@MarkGraban Sorry, I don't quite get it. For my first 20-data point chart, data point 6 went beyond the control limits. Now, I have 20 more data points. My question is, at which point should I re-calculate: (a) from point 6 from my 1st control chart i.e. when a special cause of variation was seen, or (b) from data point 21?
@@lly7454 You could use the most recent 20 data points to calculate an average and limits... sorry for the slow reply.
what kind of control chart is this?
It's an XmR Chart methodology
@@MarkGraban Thanks! But why is there one chart only? Isn't the XmR chart composed of two charts?
@@lly7454 Yes, this video shows just the X Chart. The full technically correct method includes the MR Chart, but there is minimal additional benefit to using both charts together in the real world.
@@MarkGraban I see, thanks alot! Can we also use this XmR chart for "marketing" metrics?
@@lly7454 Absolutely. If it's time series data... you can use the XmR chart (or just the X Chart)