| T O P I C R E V I E W |
| mbowden |
Posted - 02 Nov 2010 : 09:22:01
Why is it not possible to create an SPC Mean Range Chart using variable data sets? I want to show points for every hour, but one hour I might have 3 data elements, the next I may have 4 data elements, and the next I may have 2 data elements.
I am no statistician, but it seems like this should be doable (at least in a Mean\Range chart), but the documentation says that only an X-Bar Sigma Chart can use variable numbers of data elements.
I noticed that whenever I did attempt to use variable numbers of data elements per point that the summary data appeared to be correct while the category data (the actual data comprising each point) often displayed erroneously.
Can someone shed some light on this? Will using variable numbers of data points ever be possible? |
| 4 L A T E S T R E P L I E S (Newest First) |
| mbowden |
Posted - 02 Nov 2010 : 15:21:57
Thanks for the support. I just ordered the book you suggested as well. |
| quinncurtis |
Posted - 02 Nov 2010 : 13:40:50
The reason we implement a fixed number of samples per subgroup for the Xbar-R chart, is because that is a limitation of the XBar-R chart, as defined by the statisticians who developed it in the first place. We aren't inventing new charts with this software, just implementing standard ones described in the literature. So while we can certainly calculate the mean of an arbitrary number of sample points in a sample subgroup, we don't, because it would no longer be the industry standard Xbar-R chart.
Statistically it gets back to the way the sigma (not mean) of the process is estimated using the range of the values in a sample subgroup. If you vary the number of samples in a sample subgroup, the calculations for the range/sigma calculations become invalid.
Since your number of samples per subgroup are so low (3-5), we can't recommend you use the XBar-Sigma chart. We can only suggest you group your samples so as to meet the fixed samples per subgroup requirement, or use the Individual-Range chart. In both cases you would want to use the Batch Variable control charts, instead of the Time Variable control charts, since your samples are not at fixed time intervals. |
| mbowden |
Posted - 02 Nov 2010 : 13:20:35
Yes, I am referencing the use of a variable number of samples per sample subgroup.
Without being a statistician, I am not completely clear on the reasons why a variable number of elements could not be used on a mean\range chart. I could be totally off base here, but if you consider the following data sets...
1:00 {3,2,1}
2:00 {1,3}
3:00 {2}
they all have a mean of 2. So, it seems like you could display the point "2" for all three times because you would need only divide each subset by the number of elements that you actually have in the subset. What I am trying to understand is whether this is an erroneous assumption on my part or if this is unsupported functionality of the chart.
Anyway, I am not dissing the chart as it is a nice piece of work, I am, however, trying to understand its limitations. In all likelihood I will be purchasing this product for my company if we can make it work for our needs. |
| quinncurtis |
Posted - 02 Nov 2010 : 10:28:42
You are asking about using a variable number of samples per sample subgroup. As the documentation explains, this is only valid with a X-Bar Sigma chart. The special MEAN_SIGMA_CHART_VSS sub type must be used. Find the reference book we use, "Introduction to Statistical Quality Control" by Douglas C. Montgomery for a detailed discussion. You may want to do your own Google search on XBar-R charts with variable sample size and see if you can come with anything other than reasons you shouldn't try and use them
You can't feed in a variable number of samples per subgroup to the other Variable Control charts, because they are written to process a fixed number of samples per subgroup. If you do, you are introducing zeros and null records in places they shouldn't be.
If you are doing 100% sampling, you may want to just use the Individual-Range chart, in batch mode. This will only use one sample per subgroup, but you can vary the number of samples per hour you input. |
|
|
