Gage R&R - Anova Method - Standard Deviations vs. Variances

R

rderoeck - 2006

#1
I believe the traditional GRR method calculates the percentage R&R incorrectly because they add standard deviations instead of variances. I assume this is the reason that %RR and %PV do not add up to 100. When I compare these percentages to the Anova results (% contributed) they are very different. A recent sdtudt resulted in %RR=10.26 vs. %RR=1.9669 (Anova).

Which method is considered most accurate? I find Wheeler's EMP studies much more intuitive than either of these. Also, does the Anova method compares measurement variance to the study parameters or specifications?

Thanks,
Rich DeRoeck
 
Elsmar Forum Sponsor

Miner

Forum Moderator
Staff member
Admin
#2
Can you attach the data from your recent study? I have never seen a difference as large as you describe.

The ANOVA method will provide slightly different EV and AV results from the Average/Range method because it separates the Operator/Part interaction, but the GRR should be very similar. Some difference will still exist because the calculations are different.

Regarding which method is more accurate, either method will give you reliable results that will enable you to make good decisions.

The MSA method has the advantage of being the de facto standard given the support of the automotive industry. It also provides a numerical measure of the gauge acceptability.

The Wheeler method is based on Shewhart control charts, is statistically sound, visual, and as you stated, is intuitive once you understand the theory. The downside is that you do not have a numerical measure of the gauge acceptability, and few people have heard of it, let alone understand it.

Since both methods agree on how to conduct the study, differing only in the analysis of the same data, you can create a spreadsheet that analyzes the same data both ways.
 

Jim Wynne

Staff member
Admin
#3
:topic: A while back I received a PPAP submission that included an ANOVA GR&R study report. I asked the supplier's QM why he chose ANOVA over average-and-range (which is, as you say, the de facto standard). There was a moment of silence, after which he said that he'd have to check on it and get back to me. He called back a few hours later saying that A) he had been unaware until my phone call that there were different methods, and B) his gage tech told him that he used the ANOVA calculations because the average-and-range results were "too high." :frust:
 
R

rderoeck - 2006

#4
Here's the gage study data:

Part-1 Part-2 Part-3 Part-4

14569.7 14703.9 14158.3 13819.1-Oper A
14703.5 14702.8 14167.5 13816.5
14671. 14707.1 14226. 13817.8

14698.3 14703.3 14159. 13722.9-Oper B
14636.5 14606.1 14428.4 13880.7
14669.4 14706. 14162.6 13700.

14704. 14704.8 14222.4 13815.6-Oper C
14703.3 14704.6 14166.4 13815.3
14636.5 14606. 14428.4 13880.7

14704. 14699. 14158. 13814.-Oper D
14703. 14704. 14158. 13816.
14704. 14702. 14167. 13816.

14702.7 14703. 14159.8 13822.3-Oper E
14702.6 14702.8 14158.2 13822.2
14705.4 14676.3 14159.4 13815.9
 

Miner

Forum Moderator
Staff member
Admin
#5
rderoeck said:
I believe the traditional GRR method calculates the percentage R&R incorrectly because they add standard deviations instead of variances. I assume this is the reason that %RR and %PV do not add up to 100. When I compare these percentages to the Anova results (% contributed) they are very different. A recent sdtudt resulted in %RR=10.26 vs. %RR=1.9669 (Anova).

Which method is considered most accurate? I find Wheeler's EMP studies much more intuitive than either of these. Also, does the Anova method compares measurement variance to the study parameters or specifications?

Thanks,
Rich DeRoeck
I identified your problem. Open the attached file and follow along with the explanation.

First, the difference between your 10.26 and 1.9669:
These are two different indicators of gauge suitability. The 10.26 is % Study Variation and means that the gauge variation is 10.26% of the observed process/part variation. This means that this gauge would be suitable for use in controlling the process or performing statistical studies. The 1.9669 is % Contribution and means that the gauge contributed 1.9669% of the total variation. The % Study Variations are based on StdDev and do not add up to 100%. The % Contribution is based on StdDev^2 (Variance) and does add up to 100%.

Second, The difference between the XBar/R method and ANOVA method:

The XBar/R method calculates the variation for Repeatability and Reproducibility, provides results and stops there. The ANOVA method calculates the variation for Repeatability, Reproducibility and Operator/Part Interaction. Minitab then performs an ANOVA, makes a determination of the significance of the Reproducibility and Operator/Part Interaction. If these components are not significant, Minitab pools them back into the Part-to-Part variation. That is why the two methods give differing results.

Third, does the Anova method compares measurement variance to the study parameters or specifications?
It does not compare "Variance", but does compare the StdDev to specifications. In Minitab, you must enter a tolerance under an Option menu, to see this in the Session Window.
 

Attachments

R

rderoeck - 2006

#6
With this study there are no product specifications so the study uses the study parameters. However it still allows me to perform an ANOVA so I guess specs are not required. (GagePack software)

So if an auditor asked me what is my % RR for this gage I would say 1.96% or 10.26%?

Rich
 

Miner

Forum Moderator
Staff member
Admin
#7
rderoeck said:
With this study there are no product specifications so the study uses the study parameters. However it still allows me to perform an ANOVA so I guess specs are not required. (GagePack software)

So if an auditor asked me what is my % RR for this gage I would say 1.96% or 10.26%?

Rich
You should say that the GRR is 10.26% of the Study Variation, which makes the gauge suitable for process control.

Some are interested in the ndc (number of distinct categories), which should be greater than 5. In this case, the ndc was 9 and 13 (ANOVA vs. XBar/Range methods), which also makes the gauge suitable for process control.
 
M

mclayton

#8
I believe the traditional GRR method calculates the percentage R&R incorrectly because they add standard deviations instead of variances. I assume this is the reason that %RR and %PV do not add up to 100. When I compare these percentages to the Anova results (% contributed) they are very different. A recent sdtudt resulted in %RR=10.26 vs. %RR=1.9669 (Anova).

Which method is considered most accurate? I find Wheeler's EMP studies much more intuitive than either of these. Also, does the Anova method compares measurement variance to the study parameters or specifications?

Thanks,
Rich DeRoeck
Wheeler attacked these sigma-based rules in favor of variance based rules, and adding recommendations for gages used for SPC purposes (SPC is robust in medium term to gage error, and some non-normality). However, gages used for single step SCRAP decisions are another story, may for economic reasons require more precision.
So the many rules of thumb are all context-dependent and usually require deeper physical knowledge of the real sources of variation, and are often wrong or misleading when data gathering methods are less precise than the core gage capability. Some data aquisition systems mask underlying capability, making the GRR rules look at 20% or more due to rounding in datalogger.

Beyond those issues, REML for unbalanced Anova cases is good method for viewing variance components so one can DO something about the problem, especially if data is plotted using multi-vari plot for engineering and operator discussions. JMP Variability Plot does all of that, for example. But datalogger has to has enough sig figures.
 
K

kaikai

#9
JMP Variability Plot does all of that, for example. But datalogger has to has enough sig figures.
JMP's variability plot can easily analyse GRR data.
Further more, it can analyse EMP type analysis by using 'Loglinear Variance Models'.

EMP type analysis means,
(1)Reasonable Prallelism Between Operators?
(2)Similar Test-retest Error Between Operators?
(3)Similar Operator Average?

'Loglinear Variance Models' can conduct these three Test at the same time.
 
M

marco1960

#10
Hi everybody
I work in a semiconductor company and we are ISO/TS 16949 certified.
We did in the past a study together with a statistician university team to understand R&R methodology and better apply this methodology to our measurement equipment.
We receive a feedback from this team about the wrong calculation of the GRR with Standard Deviation, because if you add the GRR% with the PV% the result is greater than 100%.
SEMATEC that is a Semiconductor companies technical team sent to AIAG a list of comments and interpretation of the MSA manual (including this wrong GRR calculation) asking for feedback, clarifications and changes.
What is not clear to me is the difference between GRR% and contribution%.
How can I explain this to my team?
Thank you for the feedback.
 
Thread starter Similar threads Forum Replies Date
J Variation (using ANOVA method) Required for Gage R&R Gage R&R (GR&R) and MSA (Measurement Systems Analysis) 3
Rameshwar25 GRR (Gage R&R) Nested Anova Method Example Gage R&R (GR&R) and MSA (Measurement Systems Analysis) 40
R Gage R&R Anova Method - Study Variance Gage R&R (GR&R) and MSA (Measurement Systems Analysis) 11
C Excel spreadsheet containing Gage R&R with Anova Method wanted Excel .xls Spreadsheet Templates and Tools 22
R Is ANOVA Gage R&R the best method? Reproducibility issues (no inspector effect)? Gage R&R (GR&R) and MSA (Measurement Systems Analysis) 13
N % Tolerance - Type 1 study on the gages, then a gage R&R (ANOVA) Gage R&R (GR&R) and MSA (Measurement Systems Analysis) 4
B How to perform a MSA - GRR (Gage R&R) A&R/ANOVA Vs GUM/VDA Gage R&R (GR&R) and MSA (Measurement Systems Analysis) 1
R P-value and interpretation of ANOVA Gage R & R Gage R&R (GR&R) and MSA (Measurement Systems Analysis) 11
L What is the the difference between the Gage R&R Nested Anova and ISO/REMCO N263? Gage R&R (GR&R) and MSA (Measurement Systems Analysis) 1
S Anova Gage R&R Calculations for 4 Operators Gage R&R (GR&R) and MSA (Measurement Systems Analysis) 2
G GR&R (Gage R&R) Anova versus Xbar Gage R&R (GR&R) and MSA (Measurement Systems Analysis) 4
I Minitab ANOVA Gage R&R Error (vs. EXCEL ANOVA GRR Calculation) Gage R&R (GR&R) and MSA (Measurement Systems Analysis) 5
L Good Excel .xls sheet for Gage R&R that calculates both Anova and Range Excel .xls Spreadsheet Templates and Tools 9
Rameshwar25 Gage R&R Anova - With Interaction or without Interaction? Gage R&R (GR&R) and MSA (Measurement Systems Analysis) 5
B Study Variation and Discrimination Ratio in Gage R&R ANOVA Gage R&R (GR&R) and MSA (Measurement Systems Analysis) 7
E Gage R&R or Anova two way? Using Minitab Using Minitab Software 1
S p-Value(s) for Anova (Crossed) GRR (Gage R&R) Studies Gage R&R (GR&R) and MSA (Measurement Systems Analysis) 7
R Interpreting Gage R&R (2 Way ANOVA) Gage R&R (GR&R) and MSA (Measurement Systems Analysis) 1
E Successful Gage R&R Anova or X/R? Need example Gage R&R (GR&R) and MSA (Measurement Systems Analysis) 6
S Gage R&R Methods - ANOVA, AIAG, 6sigma Xbar/R, GM - Which to Use Gage R&R (GR&R) and MSA (Measurement Systems Analysis) 3
C Sample Size of GR&R (Gage R&R) on Nested Anova Gage R&R (GR&R) and MSA (Measurement Systems Analysis) 4
P Gage R&R Anova Report Gage R&R (GR&R) and MSA (Measurement Systems Analysis) 7
C GR&R (Gage R&R) Anova spreadsheet is telling us that a gage is not working Gage R&R (GR&R) and MSA (Measurement Systems Analysis) 10
T Variables Gage R&R - Minitab v14 - F statistic Calculation - ANOVA - Methods validity Using Minitab Software 2
B ANOVA Gage R&R % Tolerance and Preference Gage R&R (GR&R) and MSA (Measurement Systems Analysis) 2
T Plug Gage Calibration Calibration and Metrology Software and Hardware 1
L Gage R&R TMV Acceptance Criteria Gage R&R (GR&R) and MSA (Measurement Systems Analysis) 4
S MSA for attribute relation gage Gage R&R (GR&R) and MSA (Measurement Systems Analysis) 2
E Zero part to part variation - Gage R&R project Gage R&R (GR&R) and MSA (Measurement Systems Analysis) 15
M Gage R&R and right way to measure Gage R&R (GR&R) and MSA (Measurement Systems Analysis) 16
L Gage RandR on automated equipment. IATF 16949 - Automotive Quality Systems Standard 7
B Gage calibration frequency, ISO and IATF - What are the requirements Calibration Frequency (Interval) 3
D Difference between Test Method Validation and Gage R&R Qualification and Validation (including 21 CFR Part 11) 18
R Determining Uncertainty from Gage R&R Gage R&R (GR&R) and MSA (Measurement Systems Analysis) 1
C Correct Calibration Method for Dial Depth Gage General Measurement Device and Calibration Topics 6
C Gage Block Wringing General Measurement Device and Calibration Topics 3
B Gage R&R with NDC=1 Gage R&R (GR&R) and MSA (Measurement Systems Analysis) 19
S Type 1 Gage R&R or something else? Reliability Analysis - Predictions, Testing and Standards 6
G Should I perform Gage R&R only at the beginning of a new project? Gage R&R (GR&R) and MSA (Measurement Systems Analysis) 6
V Thread Plug Gage Pitch GO Diameter out of spec AQL - Acceptable Quality Level 8
G Gage R&R - Where am I going wrong? Part of a FAIR submission (Aerospace) Gage R&R (GR&R) and MSA (Measurement Systems Analysis) 2
R Gage R&R Excel templates Gage R&R (GR&R) and MSA (Measurement Systems Analysis) 3
K Gage R&R with more than 3 appraisers Gage R&R (GR&R) and MSA (Measurement Systems Analysis) 2
Anerol C Gage R&R Template AIAG 4th Edition IATF 16949 - Automotive Quality Systems Standard 3
T Attributes SPC study - Attributive control (Go gage) Statistical Analysis Tools, Techniques and SPC 5
S Capability or Gage R&R Study for Leak Tester? Reliability Analysis - Predictions, Testing and Standards 15
M Definitive answer on Type 1 vs Type 2 vs Type 3 Gage Study Gage R&R (GR&R) and MSA (Measurement Systems Analysis) 0
V Gage Management and Gage R&R Software General Measurement Device and Calibration Topics 1
D Gage R&R Study on Load Cells - Large chemical blending tanks IATF 16949 - Automotive Quality Systems Standard 1
B Minitab Type 1 Gage Study on True Position Question Measurement Uncertainty (MU) 1

Similar threads

Top Bottom