Control Chart Conundrum
Six Sigma – iSixSigma › Forums › Old Forums › General › Control Chart Conundrum
- This topic has 25 replies, 15 voices, and was last updated 15 years, 5 months ago by
curious george.
-
AuthorPosts
-
January 31, 2007 at 8:20 am #46001
Any wise owls out there can help me with this one, it nearly kept me up all night;
operator A takes 30 readings on a process, a control chart is generated.
operator B takes 30 readings on the same process but at a different site, a control chart is generated.
if the data columns are stacked and a ‘cumulative’ control chart is generated, am I right in saying that this would not be valid, because the pattern and tests for out of control conditions would not be valid.
However, the mean, UCL / LCL and standard deviation from this pooled control chart would be valid for the 60 readings taken as an overview of the ‘whole’ process, is this correct?0January 31, 2007 at 10:27 am #151308
accringtonParticipant@accringtonInclude @accrington in your post and this person will
be notified via email.Yes
No0January 31, 2007 at 1:22 pm #151316
Ken FeldmanParticipant@DarthInclude @Darth in your post and this person will
be notified via email.Key is whether the stacking disrupts the time order of the data not whether the tests are valid or not.
0January 31, 2007 at 1:23 pm #151320so if I time order the data, and then combine the 2 data sets accordingly (i.e. data points will jump from op. A to B etc. as the tests will be done roughly at the same time) to produce a control chart, it is statistically valid even though the readings weren’t taken at the same place?
and if the answer is yes, then my control limits, standard deviation and mean are valid also?0January 31, 2007 at 1:37 pm #151318
Purpose of SPCParticipant@Purpose-of-SPCInclude @Purpose-of-SPC in your post and this person will
be notified via email.The purpose of the control chart is to identify out-of-control conditions in order to bring the process in control, i.e. take action. The purpose is not to estimate control limits per se. The control limits are caclulated to allow you to determine out of control conditions. The control chart does not have the purpose of a survey where you combine samples in order to better estimate the population parameter. A population is static, a process is dynamic. In your case this means that you should keep the two charts separate because your purpose should be to monitor the variation of the two processes. The question becomes: what do you gain from a business point of view by adding the two samples. This is economic control of processes. So half of the formula is statistical, half is economical and action-oriented.
0January 31, 2007 at 1:49 pm #151321Thanks Purpose,
the business point of view is what I’m trying to get to.I might end up with a lot of separate control charts, how do I show how bad the process is ‘overall’?
thx0January 31, 2007 at 2:03 pm #151322
accringtonParticipant@accringtonInclude @accrington in your post and this person will
be notified via email.No and no.
Operator A/ Location1 is an apple; Operator B/ Location 2 is an orange. They’re not the same.
You need one chart for each location. If you want to compare the performance of the two locations, you can plot the charts, in time order, side by side, computing the control limits for each location separately (this is easy – peasy using the MINITAB Stages in Chart Options).
I think you may be missing the point here. What do want to do with the chart?0January 31, 2007 at 2:34 pm #151323
Ken FeldmanParticipant@DarthInclude @Darth in your post and this person will
be notified via email.I agree with the recommendations of the other posters. But, if you want to get an overall picture of the combined processes and they are supposed to be similar, then I would form subgroups taken from both sites and do an Xbar/R chart. While the averages might be OK, the R chart will show extreme variation if the sites are in fact different. So, if you take a sample at approximately the same time from A and one from B and form the subgroup then this might answer your question. On the other hand, you can do two separate control charts to monitor stability between A and B and then do a T test and Test for Equal Variances to see if, in fact, there are signficant differences between A and B with respect to central tendency and variation. That way you get the best of both worlds. You could also show some interesting things by doing a distribution of the combined data and assuming the combined set have to meet a spec you could do some capability analysis as well. You can look to see if the combined distribution is bi-modal or skewed or too wide. Lots of tools available to get to your desired end, don’t restrict yourself to a control chart.
0January 31, 2007 at 3:01 pm #151324Thanks Accrington,
there will actually be several charts for several people at one location, several charts for several people at another. I want to summarise statistically, if possible, how the overall process is currently performing before I introduce a change, then repeat the data collection and summarise statistically the overall improvement.
I get the point about keeping the charts separate, but the stakeholder is going to somehow want to see all the separate charts and data summarised to indicate the business benefit of introducing the change.
This is what I’m struggling to get to, any help appreciated thx.
0January 31, 2007 at 3:41 pm #151325
Jamilah HaronParticipant@Jamilah-HaronInclude @Jamilah-Haron in your post and this person will
be notified via email.Not advisable to stack – they may not be in time order, and the 2 sets of data may have different sources of variation.
UCL and LCL from the combined data are useful if they processes from the 2 sites are stable, i.e key factors are controlled.0January 31, 2007 at 3:52 pm #151326Thanks all,
I think I might be able to answer my own question after reading your replies and more thought – how about using boxplots to combine the data – they show things statistically without corrupting the truth?
0January 31, 2007 at 3:52 pm #151328
Chris SeiderParticipant@cseiderInclude @cseider in your post and this person will
be notified via email.Do not have too much fun with this very powerful, often misunderstood and misundervalued tool.
Remember the basic tenets of SPC charts are to have the data in time based order so stacking the data would most likely be incorrect. Second, one would never consider using one SPC chart on individuals from two different processes. However, if a rational reason existed, you could subgroup one sample from each machine and use an Xbar chart. However, if the items aren’t going to a similar customer or downstream process, I would see no reason for combining the data into one chart.
Good luck….oh, I’m not going to even dwell on the general consensus there should be a degree of process stability before calculating control limits. There are various opinions on being in control before using SPC. I’m in the camp that unless you have real outliers (outside 4 sigma limits), include all of the data and calculate control limits unless there is a cyclical or other pattern in the data.0January 31, 2007 at 4:31 pm #151330
The ForceMember@The-ForceInclude @The-Force in your post and this person will
be notified via email.Define what’s your objective first and identify if the tool fits your objective in terms of the results you want to see.
Need to have two separate charts because 2 different parameters might affect the decision making. Better have a boxplot and two sample hypo test to see any significance or if the other is better.0January 31, 2007 at 4:33 pm #151331
PurposeParticipant@PurposeInclude @Purpose in your post and this person will
be notified via email.you’re really deep-diving into the “war of the worlds” that has been raging between proponents of the sigma score and statisticians like Wheeler (SPC). there are hundreds of threads on this topic on this side, many published papers and even more hot-headed opinions. eventually, you’ll have to come with your solution and justify it based on the goals of your project.
0January 31, 2007 at 4:47 pm #151332
accringtonParticipant@accringtonInclude @accrington in your post and this person will
be notified via email.Are measuring a product or a process characteristic?
If you want to compare different people at different locations, you could use a single control chart,stratified by location and person (I did a simulation, but can’t copy the chart into this e – mail)
You could use box plots, as was suggested previously, but you won’t get any idea of process stability.
If you want to show the stakeholder a single summary of the overall performance (assuming this is a product or service going to a customer somewhere), why don’t you just plot all the data on a histogram?
0January 31, 2007 at 7:50 pm #151349the characteristic I’m measuring is time to perform the process.
I did think of using a histogram to summarise, but while this might be good pictorially, I also wanted some stats to add weight to the summary; what can you put with a histogram, other than min./max., mean, and comment on shape?
regarding stratification on a single control chart, how do you do this on minitab?0February 1, 2007 at 3:56 am #151377By mixing data you have lost the whole purpose of the control chart.
Read Shewhart, Deming and Wheeler for the basics.0February 5, 2007 at 12:39 pm #151576
accringtonParticipant@accringtonInclude @accrington in your post and this person will
be notified via email.Put all of the process data in a single column. Put a unique identifier for location/ person in the column next to the process data.
(For this data, I would only plot the I – chart, given the audience, but you could plot an I – mR chart of you wanted to)
Select Stat>Control Charts>Variables Charts for Individuals>Individuals. In the Variables: dialogue box, select the column which contains the process data. Click the I-Chart Options button. Click the Stages tab. Select the column containing the location/ operator identifier in the Define Stages.. box. Click OK twice.
Good Luck!0February 5, 2007 at 8:24 pm #151606Good luck alright … obviously another mindless SS consultant focussed on numbers rather than purpose.
Don’t mix data from different sources.0February 5, 2007 at 9:17 pm #151608Control charts are really a homogeneity test – therefore, there is no reason why mixed data can’t be used. Control chart should really be used to find sources of variation – so they are not limited to homogeneous data as you suggest.
0February 6, 2007 at 12:44 am #151619
DefinitionsParticipant@DefinitionsInclude @Definitions in your post and this person will
be notified via email.Pete, if you ride the high horse, at least know the technical defintions and be specific about the big words that you are using: “Control charts are really a homogneity test”, “mixed data” and “really be used to find sources of variation”. A little knowledge is very dangerous!
0February 6, 2007 at 10:39 am #151630On the contrary, once again it is you who have consistently chosen to ‘ride a high horse.’
May I suggest you actually go away and tackle a real process instead of a classroom demonstration – one that is not limited to a ‘temporal variation.’
Once you work out how to rearrange your subgroups, with any luck, you’ll be able to find the sources of variation! But I’m not going to hold my breath!0February 6, 2007 at 12:47 pm #151632
accringtonParticipant@accringtonInclude @accrington in your post and this person will
be notified via email.I am not a SS consultant, neither do I consider myself to be mindless. Also, I did not suggest mixing data from different sources. I suggested that the data be stratified by location and operator, and that a single chart plotting each stratum separately be used to compare each location/operator combination.
If this was unclear in my post, please advise why.
Thank you
Accrington0February 6, 2007 at 11:32 pm #151667Control charts are really a homogeneity test
You must be kidding ? Or is this more of Mikel Harry’s theories?0February 7, 2007 at 7:48 am #151677Homogeneity and rational subgrouping seem to be requirements of control charts.
http://www.sei.cmu.edu/str/descriptions/spc.html
“Next, the notions of homogeneity and rational subgrouping need to be understood and addressed. Homogeneity and rational subgrouping go hand in hand. Because of the non-repetitive nature of software products and processes, some believe it is difficult to achieve homogeneity with software data. The idea is to understand the theoretical issues and at the same time, work within some practical guidelines. We need to understand what conditions are necessary to consider the data homogeneous. When more than two data values are placed in a subgroup, we are making a judgement that these values are measurements taken under essentially the same conditions, and that any difference between them is due to natural or common variation. The primary purpose of homogeneity is to limit the amount of variability within the subgroup data. One way to satisfy the homogeneity principle is to measure the subgroup variables within a short time period. Since we are not talking about producing widgets but software products, the issue of homogeneity of subgroup data is a judgement call that must be made by one with extensive knowledge of the process being measured.
The principle of homogeneously subgrouped data is important when we consider the idea of rational subgrouping. That is, when we want to estimate process variability, we try to group the data so that assignable causes are more likely to occur between subgroups than within them. Control limits become wider and control charts less sensitive to assignable causes when containing non-homogeneous data. Creating rational subgroups that minimize variation within subgroups always takes precedence over issues of subgroup size.”
Dan
0February 22, 2007 at 2:57 am #152290
curious georgeParticipant@curious-georgeInclude @curious-george in your post and this person will
be notified via email.I agree, use seperate charts. It sounds like you don’t have data from a rational subgroup [different operators]in each chart.
0 -
AuthorPosts
The forum ‘General’ is closed to new topics and replies.