# Honest Gauge Study – Stan and others

Six Sigma – iSixSigma › Forums › Old Forums › General › Honest Gauge Study – Stan and others

- This topic has 23 replies, 7 voices, and was last updated 10 years, 5 months ago by Jonathon Andell.

- AuthorPosts
- February 12, 2010 at 6:30 pm #53265
Stan,

I know you have some pretty deep knowledge about Gage R&R. I just finished slugging my way through an article that Wheeler wrote last year entitled, “An Honest Gauge R&R Study”. I don’t know if you have read it but if you and others have, can you offer your take on his position? Thanks.0February 12, 2010 at 7:10 pm #189348Much ado about nothing. If he wants to get attention, he should at least match what is done for

the calculations for the last 20 years and know what is in Minitab. He

appears to be completely ignorant of both.0February 12, 2010 at 7:38 pm #189350Thanks. It was a tough read for a Friday afternoon.

0February 12, 2010 at 7:59 pm #189351Same old crxx he has been publishing for a couple of decades. His

book on Measurement System Evaluation is a good reference, just

don’t follow his conclusions.0February 12, 2010 at 8:29 pm #189352OK,thanks. Have a good weekend.

0February 12, 2010 at 10:27 pm #189354

Bill McParticipant@Bill-Mc**Include @Bill-Mc in your post and this person will**

be notified via email.I think it is well written and points out the issues associated with the average and range method that is in the AIAG Measurement Systems Analysis manual. His point is simply that the variances are additive, not the standard deviations. Many companies still use the average and range method and wonder why their results are poor – when it fact, they might not be. I would imagine that using the ANOVA method to analyze the Gage R&R results will take care of this issue since the % contribution is based on variances.

0February 12, 2010 at 10:50 pm #189356

SeverinoParticipant@Jsev607**Include @Jsev607 in your post and this person will**

be notified via email.Quite frankly (even though I don’t use the technique he is referring to) I say go ahead and condemn the measurement system. So few GR&Rs actually reflect what occurs during normal usage that you should want to be conservative since your estimate likely undereports the true variation.

0February 13, 2010 at 8:23 pm #189365No one uses the average and range method and they haven’t for over

a decade. That Wheeler doesn’t know this is troubling.Do you actually know anything about the subject?0February 13, 2010 at 10:15 pm #189368

Bill McParticipant@Bill-Mc**Include @Bill-Mc in your post and this person will**

be notified via email.Yes, I do know something about the subject. I would estimate that 75% of the questions I get from customers using gauge R&R are about the average and range method and the questions usually indicate that the customer has no idea what they are doing. They just want something to tell them that their measurement system is good – even when it isn’t. There are questions on the ANOVA method – just not nearly as many. So, there are people still using the average and range method. And AIAG still promotes it and they do have a following.

0February 13, 2010 at 10:29 pm #189369You need to find a better class of customer. I’ll give you an Excel

based add-in that takes care of the ANOVA method. Your customers

are using Excel by now – right?Wheeler’s rant is nonsense.0February 13, 2010 at 11:01 pm #189370

Bill McParticipant@Bill-Mc**Include @Bill-Mc in your post and this person will**

be notified via email.Yes, my customers use Excel and even Minitab – which contains the average and range method. Dr. Wheeler’s paper is well-written. I believe Darth asked you and others for a take on Dr. Wheeler’s position. You give no specifics except that, because he addresses the average and range method, it is much to do about nothing. What in the paper is not accurate or true?

0February 14, 2010 at 4:12 am #189371

TaylorParticipant@Chad-Vader**Include @Chad-Vader in your post and this person will**

be notified via email.Darth, I read it, didn’t get the point. Like Stan says, much ado about nothing. But hey, he has to keep his name out there some how, and I truely believe that is all it is. Wheeler wrote it, therefore its law..bull……..No Real world substance at all.

0February 14, 2010 at 10:40 am #189372Accurate? No misleading.True? Yes, but totally irrelevant.His whole point is standard deviations don’t add. Wow really?

Standard deviations estimated by range and standard deviations

estimated by samples do exactly the same thing. Is the ANOVA

method a better estimate? Yes, but not that much better. It does

allow you to break out the interaction of part and operator, but

Wheeler doesn’t even make that point. His allowable gauge error is

a joke. No acceptance or rejection of error should be made except

in concert with knowledge of process capability. AIAG standards

are based from 1962 as Wheeler points out. They were at a time

where an estimate using Range was less inaccurate than calculating

the ANOVA by hand. Those days passed in the late 80’s with

Visicalc and Lotus 123 and cheap computing.If you see value in Wheeler’s rant, I am glad I am not your

customer. If you want to pass on their names to me, I could bring

them into the 21st century and accelerate their learning on many

fronts.0February 14, 2010 at 1:07 pm #189374Please identify the correct statement below

a) 15.7 + 17.8 = 23.7, not 33.5

b) 2.4 + 3.2 = 5.6

c) 5.6 + 94.4 = 100

d) 23.7 + 97.2 = 100, not 120.9

e) b and c are correct

These values come from Wheelers article on page 12.

I think his point is that you should use the correct computational methods and if you indicate that A + B comprises a total. Why shouldn’t %GRR + %Part Variation = 100%?

Is 120.9 close enough?

This has nothing to do with the Range versus ANOVA method. I have to chuckle every time I think of the operator*part interaction anyway. Operator A has a fear of the number 5 and every time he measures part 5 he breaks out in a cold sweat and measures it erroneously. The ANOVA method is the preferred method no doubt, but I can’t get over the interaction thing!0February 14, 2010 at 1:16 pm #189375Guess you’ve never seen an interaction. It is great information in

resolving measurement problems – if it exists.His other points are much ado about nothing. Find the papers from

the guys who created this at GM in 1962. They knew about the

problem with addition but felt it was more important to keep it

simple – just convert range to standard deviation (tables had been

available for at least 30 years), or convert range to variance? They

kept it simple in a language that was understood at least by those

who knew SPC.And the rules? They all flow from the measurement “rule of thumb”

of a 10:1 ratio minimum. That where the 30% comes from – do you

understand that?0February 14, 2010 at 1:49 pm #189377Wheeler’s article is also intentionally dishonest.1) His statements of how P/T is to be interpreted is outright false.

Read the MSA manual from AIAG and find the words he says are

there. They are not.2) His NDC to the other metrics. The rules say NDC > 10 is good,

between 4 & 10 is marginal – in agreement with the other

measures. 10 is good,

between 4 & 10 is marginal – in agreement with the other

measures. 10 is good,

between 4 & 10 is marginal – in agreement with the other

measures.3) All of his talk about the measures – the truth is if I know his

Discrimination Ratio (he touts it to be superior), I know NDC and %

contribution and % study.4) his rigorous proof of .675*standard deviation? Anyone ever look

at a normal table? He is using 99% confidence, some choose to use

99.73%.All this said, I think there is tremendous value in reading the body

of work from Wheeler. His views of simplicity and transforming

data are right on target. But, I’ve been running into the Wheeler

measurement zealots for over a decade. The thing I’ll tell you is

they are willing to have a philosophical debate about fixing their

measurement while others go fix their measurement. Want to take

bets on who move quicker to a solution and who has the superior

solution? I have data.0February 14, 2010 at 3:37 pm #189384

SeverinoParticipant@Jsev607**Include @Jsev607 in your post and this person will**

be notified via email.Actually, I’d love to read those papers. Do you have a more specific reference (i.e. article name, database, etc.)? Thus far I haven’t been able to unearth anything.

0February 14, 2010 at 4:53 pm #189389It was referenced in the first AIAG MSA manuals (93 or 94?). I haven’t

looked lately to see if they still do it. Long story short is they knew what they were doing and the 5.15 was

a compromise with development engineers. One of the creators is still

kicking around the ASQ SS circles.0February 14, 2010 at 5:41 pm #189391

Bill McParticipant@Bill-Mc**Include @Bill-Mc in your post and this person will**

be notified via email.Stan, thanks for giving the details of your objections/concerns about his article. It makes it easier to understand where you are coming from on his much to do about nothing. Also, it provides much more information for the readers in this forum.

0February 14, 2010 at 6:48 pm #189392Cool, so now tell us why you think it is relevant.

0February 14, 2010 at 7:27 pm #189394

Bill McParticipant@Bill-Mc**Include @Bill-Mc in your post and this person will**

be notified via email.Primarily because it is in line with what I have done and taught over the years. I was introduced to the mechanics of SPC way back in 1983. In reality, I have never liked Gage R&R studies because they are simply a snapshot in time of a measurement system – whether you are using the average and range method or ANOVA. Unless the measurement system is consistent and predictable, you can’t be sure of getting similar results at a later time. To know that, you must track the measurement system over time – at least the critical measurement systems.I started in the process industries where the measurement systems were often poor. All our critical tests were monitored by running a standard or control on a regular (usually once per shift) and plotting the results on an X-mR chart. The first objective was simply to get the measurement process into control. If it was not, we treated the measurement system as if it was mechanically broken. We had to find the reason for the out of control point. Once in control, we would estimate the measurement system variance from the average range. This variance includes all the operators who run the test. This measurement system variance was compared to the total process variance obtained from a range chart kept on the variable in production. If the measurement system was responsible for less than 10% of the variance, we concluded that it was acceptable. We could also look at differences between operators using this data and comparing the results using control charts. If it was above 10%, the measurement system needed to be improved. So, I liked seeing his first cut at 10% of the total variance. So, is his paper relevant? Of course it is. If someone takes this approach, they can find out what their measurement system is doing. You do have a point about the operator-part interaction, but I really don’t see that too often. But his approach is better than the average and range method. Although, I agree ANOVA gives the full picture and allows you to take whatever ratios you want to determine if the measurement system is good.Again, thanks for your input. I will have to take a closer look at his probable error information.

0February 15, 2010 at 3:51 am #189398

SeverinoParticipant@Jsev607**Include @Jsev607 in your post and this person will**

be notified via email.Everytime Wheeler writes about a Watershed I want to throw him into one. I find his articles on “how not” to do things much easier to sit through than his articles on “how” to do things. Fortunately, most of his stuff has enough mixture of the two that they become palatable.

0February 15, 2010 at 1:00 pm #189402I have seen interactions, but in the context of the GRR with Factor1 = Operator and Factor 2 = part, they are “interesting” to explain!

It is usually a case where the measurements are not taken in a random order. An operator makes a setup error and measures the same part 3 times in a row with the same set-up error. Very serendipitous to say the least! A botched GRR reveals information about set-up issues. When fully randomized, I haven’t seen issues with interactive effects.

I am not sure about all the watershed stuff!0February 15, 2010 at 6:01 pm #189413

Jonathon AndellParticipant@Jonathon-Andell**Include @Jonathon-Andell in your post and this person will**

be notified via email.I wish you were right, but I run into a lot of what I call the send-in-your-box-top kind of Black Belts still using the average & range method. If only the GM folks in the 1960’s had opted for abacus instead of pencil & paper, we might not be in this mess.Lots of good discussion here. A few additional comments, perhaps pointing out the obvious, but hey that’s what I do.1. Thank you to those who remind us that R&R is but a snapshot in time. Measurement is a process, and experiences variation through time. The “R” chart gives us some clues as to how that process varies through time, but there’s a risk. Most rational sub-group sampling schemes – even those that are designed thoughtfully – are designed to capture process variation rather than measurement variation, and that is as it should be. If we really want to understand measurement variation through time, we might want to consider its own control chart, and hopefully something a whole lot simpler than a full-blown R&R study.2. Many times the R&R is at the start of a DMAIC project. Suppose initial R&R shows that measurement variance is 20% of total variance. If the DMAIC project reduces the process standard deviation by only 25%, the same measurement variance is just over 30% of total variance! What had started out as a marginally acceptable measurement system no longer can meet the needs of the improved process. I have a pretty cools spread sheet that shows those numbers.

0 - AuthorPosts

The forum ‘General’ is closed to new topics and replies.