chris reid
@christopher10  Reputation - 49 Rank - Aluminum
|
Often when I create a control chart I have a large number of data points outside of the control limits. I find this difficult to understand as minitab only has my sample to work with.
Also I really appreciate pointers on a generic approach-
A) If the process is out of control?
B) If the process is in control but has high varience?
Thanks all.
Chris.
|
Chris Seider
@cseider Reputation - 3002 Rank - Titanium
|
A. Yes, points outside your control limits mean you are exhibiting special causes so you are out of control.
B. A process can be in control but have high variation. The SPC looks for special causes related to trends and outside 3 sigma limits (typically). Process capability /variability is independent of evaluation of process control.
|
Prabhu V
@prabhuvspj Reputation - 477 Rank - Aluminum
|
Hi,
Adding to the above point, always an user has to ensure that the user is using the correct control chart for the situation faced by the user.
|
chris reid
@christopher10 Reputation - 49 Rank - Aluminum
|
Hi thank you very much both.
The thing I’m struggling with is if I have 100 data points in a sample how can so many of them be outside 3 sigma in that samle, it doesn’t make sense to me?
)-:
|
Robert Butler
@rbutler Reputation - 2138 Rank - Silver
|
I think you are going to have to provide more details before anyone can offer much in the way of guidance. Offhand it sounds like you are making the wrong program choices with respect to control chart construction. Is there any chance you could post the 100 numbers? If you could some of us could look at what you have and perhaps offer some additional thoughts.
|
Euston
@dabrewajeff Reputation - 20 Rank - Aluminum
|
If, indeed, it is a problem you encounter often, it could be incorrect construction as suggested, but you should check your CC setup parameters under tools>options. Specifically, make sure you are set to the defaults (see help menu if you don’t know) and your 1st test is set for K=3. And is your data normally distributed?
|
Joel Smith
@joelatminitab Reputation - 974 Rank - Copper
|
@christopher10 I doubt you’re going to get much hep without providing the data. My first two thoughts are that you have either (a) an Individuals control chart that has a drifting mean over time or (b) highly skewed data. But without context or seeing data it’s really hard for anyone to provide advice.
|
Darth
@Darth Reputation - 1285 Rank - Silver
|
@dabrewajeff Are you suggesting that the needs to be normally distributed for the control chart to be effective?
|
Chris Seider
@cseider Reputation - 3002 Rank - Titanium
|
@Darth
LOL, you gonna reopen that age old discussion? :)
|
Darth
@Darth Reputation - 1285 Rank - Silver
|
@cseider No, just asking an innocent question and being non-cranky.
|
chris reid
@christopher10 Reputation - 49 Rank - Aluminum
|
Hello all,
Thank you for helping, here is 1 of my examples ( I appreciate i should have 100 data points but I think this will show you what I mean )
|
Joel Smith
@joelatminitab Reputation - 974 Rank - Copper
|
@christopher10 It doesn’t look like the attachment came through…perhaps a different file type? Katie isn’t sdhowing any love towards Minitab and doesn’t have .mpj or .mtw files as an allowed type for attachments! (hint, hint Katie…)
|
Euston
@dabrewajeff Reputation - 20 Rank - Aluminum
|
@Darth I don’t think my question was suggestive, was it? I’m with everyone here….I’d love to touch that data. Anyway, the thought behind the question was the distribution could explain why you would see more points in violation of the limits.
|
Katie Barry
@KatieBarry Reputation - 7991 Rank - Gold
|
@joelatminitab – Sent the question of file allowance to Cyger. I’m not sure if that’s something we have control over, but I’m looking into it!
|
Joel Smith
@joelatminitab Reputation - 974 Rank - Copper
|
Thanks Katie! In the meantime hopefully the data can be provided in another format.
|
Chris Seider
@cseider Reputation - 3002 Rank - Titanium
|
Suggestion….if you have the data in Excel or Minitab…copy the column of data and paste into the comments field and we can just paste into our field.
|
chris reid
@christopher10 Reputation - 49 Rank - Aluminum
|
Hi here is an example
3031
3072
3136
3023
3290
3282
3319
3544
3436
3482
3434
3449
3707
3560
3618
3472
3456
3342
3457
3660
3657
3852
3926
4006
*
3676
3555
3100
2720
3141
3219
3251
3110
2814
2872
3034
3171
3118
3019
3006
3186
3100
3100
3340
3275
3148
3192
3313
3214
3040
2992
2997
3210
2846
2838
3017
3273
Ive been using an I Chart for this, I think the problem is I misunderstand how stdeviation is calculated. I’ll refresh my memory today but any help or comments you can give would be very much appreciated.
|
chris reid
@christopher10 Reputation - 49 Rank - Aluminum
|
p.s thanks for all the help so far.
|
Joel Smith
@joelatminitab Reputation - 974 Rank - Copper
|
@christopher10 Whoa there…you’re definitely out of control for the first 1/3-1/2 of the chart! It’s hard to give much practical advice without knowing the process, but from the data alone my observations are:
1. Through the first 24 points, before the missing value, you clearly have an upward trend where values keep getting larger and larger.
2. After the missing value, you have a few points in a “settling down” period through about point 30 that show high variation and maybe a lack of control initially.
3. Following point 30 you appear to have good statistical control.
I’m guessing from this data – and I should probably say “know” instead of “guess” – that some change was made to either the process or the data collection at the missing value. Try this:
1. Make a second column that contains 1′s through the missing data point, and 2′s after that until the end of the data.
2. Back in the Individual Chart dialog box (and you should really be doing I-MR to assess variation stability as well) click on I Chart Options and then the Stages tab.
3. Enter your new column for Stages and click OK in each dialog.
This will create a centerline and control limits independently for each part of the data and enable you to more clearly see the process change. I wouldn’t stress over the few out of control points from 26-30 since that is common after a process change…things don’t always “jump” to the new value and sometimes take a few points to stabilize. You may even stage those few points separately to get slightly more accurate control limits in the stable portion.
If it’s something you can share – even if you mask things just a little bit – I’m sure forum visitors would like to know about this process and what change occurred at the missing point!
|
Darth
@Darth Reputation - 1285 Rank - Silver
|
@christopher10 I attached the I/MR Chart that results from your data above. Yes, the process is out of control. One thought might be that the MR chart values are small compared to the actual values and thus the avg. MR used in the I chart calculations might be “compressing” the control limits relative to the magnitude of the actual process values. As you can see at the beginning of the data there is a dramatic upward run followed by an apparent change which dropped the process down. It appears to have stablized now. If we discount the upward data as something happening and the data after that as the new steady state process things might look a little better. You can also “break” the data at the apparent point of change and recalculate the before and after. Mini will let you do that. By the way, the p value of the data set shows that the data is not different than normal. The p value is .168. The s.d. used for the calculations uses the avg. MR as the base and then adjusts using the d sub 2 and E sub two table factors based on sample size. In this case usually a sample size of 2. Hope this helps and that Joel approves.
Attachments:
|
Joel Smith
@joelatminitab Reputation - 974 Rank - Copper
|
I would approve, but I can’t see you behind that mask…
|
Katie Barry
@KatieBarry Reputation - 7991 Rank - Gold
|
@joelatminitab Minitab files should now load without any trouble.
|
Euston
@dabrewajeff Reputation - 20 Rank - Aluminum
|
@darth Agreed on the p value. Still even high p’s(.619/.283 respectively) when analyzed into the 2 stages. Obviously the tighter CL’s on the I/MR is driven by the slow process shift from data point to point as you articulated. @christopher10 I think Darth laid it all out for you. Curious if 1) the data reflects the startup of the process and, 2) if you indeed only collect one data point representing a given point in the process?
|
chris reid
@christopher10 Reputation - 49 Rank - Aluminum
|
Thanks all very much for the advice. All points taken on board and the 2s and 1s tip is great.The data represents orders in delay. As ever I have another question (once I get started I cant stop)
Q- Is there a way I can add an upper and lower spec to a control chart?
Thanks all.
C10
|
Putnam
@Mike86 Reputation - 12 Rank - Aluminum
|
You could annotate it by drawing two lines.
|
Joel Smith
@joelatminitab Reputation - 974 Rank - Copper
|
@christopher10 Well, you could use reference lines, BUT…
This questions has been answered in numerous places so I’ll be brief. Do not put spec limits on your control chart. If you want to assess whether you are meeting specifications, first make sure your data are in control and then use capability analysis. There are multiple reasons that have been documented many times.
|
chris reid
@christopher10 Reputation - 49 Rank - Aluminum
|
Thanks a lot, right here is my last ( i hope ) question.
I’m trying to use a P chart and I get this error message when I select a column
The following column contains values that are too small, the data must be >=1.
The thing is the data is all greater than 1. I cut it from excel so not sure what the issues is. It is very annoying
|
Joel Smith
@joelatminitab Reputation - 974 Rank - Copper
|
Can you share the data? @KatieBarry was so kind as to get Minitab files allowed (best), or your could share the Excel or just paste it here.
|
July 12, 2012Comments Off
