AIAG MSA: 4th Edition (Bias Study)

[hazwan2283]

Hi and Good Day to all the Experts in the Cove. My name is Hazwan and i would like to ask a question about the Bias Study.
I am referring to the AIAG MSA 4th Edition guideline book page 89-92 example of a Bias Study.

In that example, they calculated the repeatability standard deviation first and they used this repeatability standard deviation to calculate %EV.
Bias Analysis assumes that repeatability is acceptable which is why their objective is to find out whether is this repeatability is acceptable or not.

The following is the exact sentence quote from the book:
"The repeatability of 0.2120 was compared to an expected process variation (standard deviation) of 2.5. Since the %EV = 100(.2120/2.5) = 8.5%, the repeatability is acceptable and the bias analysis can continue." Now my questions are, what is the threshold limit of the acceptable repeatability value here? Why 8.5% is acceptable? Are they using less than 10% acceptance criteria of GR&R here?.

Sincere apologies if there are any grammatical error(s) above.

 

[Miner]

Unfortunately, the manual never comes out directly and states the acceptance criteria, but it does make a vague reference to GRR, so I think you are safe in making that assumption of 10% acceptance criteria.

 

[Enghabashy]

Some of referenced steps "ANOVA "regarding Bias & acceptance criteria ; for accept Bias or recommend Re-calibration of gauges if the value is critical, example snapshot below also :
1- USE OF STABILITY ANALYSIS1)Obtain a reference standard of a known value (e.g., a gauge block/pin)
.2)On a periodic basis (daily, weekly), measure the standard three to five times. Subgroups should be taken at differing times to include variation due to warm-up, ambient conditions, etc.
3)Enter the data in the Stability worksheet.
4)Interpret the Stability of the gauge using standard control chart analysis
.5)Determine the significance of the Bias, by comparing the t-value of the Bias to the critical t-value. If the t(Bias) is less than the t(critical), the Bias is not significant. If the t(Bias) is greater than than the t(critical), the Bias is significant and should be corrected through calibration.

 

[hazwan2283]

Unfortunately, the manual never comes out directly and states the acceptance criteria, but it does make a vague reference to GRR, so I think you are safe in making that assumption of 10% acceptance criteria.

Hi Miner, i would like to ask 2 questions related to Bias Study to you,..

I noticed that in most of the statistical softwares, be it Minitab, DataLyzer etc i can see that %Bias is always calculated and shown to the user. From my knowledge (quoting from AIAG MSA 4th Edition), in order for us to know whether bias is acceptable or not there are 2 ways,

1. if the p-value associated with the t-statistic bias is less than 0.05, or
2. if zero is between the lower bound and upper bound of bias confidence interval. (i may not phrase it correctly).

So my questions are:
Question 1: why are we calculating % bias here if we are not using it to evaluate the acceptance of Bias?. From my google search, i can't even find the acceptance criteria for this %bias at all...

Question 2: I took the example from AIAG MSA 4th edition whereby they have 15 data (5.8 , 5.7 , 5.9, 5.9, 6, 6.1, 6, 6.1, 6.4, 6.3, 6, 6.1, 6.2, 5.6 and 6) and then i calculated the p-value associated to t-statistic bias. I obtained t-statistic as 0.1217 and the associated p-value as 0.9048
So, if we are using criteria 1 from AIAG MSA which says that bias is acceptable if p-value less than 0.05 which in this case it is not so my earlier conclusion was this bias was clearly not acceptable.


However i further calculated the lower bound and upper bound confidence interval and i got the lower bound as -0.1107 and upper bound as 0.1241. Zero clearly lies between this lower and upper bound which means if we follow criteria 2 the bias study is acceptable.

In the very same AIAG book they mentioned this bias study is acceptable because of criteria 2 but they never mentioned anything about p-value.

So my real question is, does this means every single time when i do bias study i must check for these 2 criterias and if whichever pass that means bias study passed and if both of criteria failed only then i can conclude that this bias study failed?.

 

[Bev D]

The difference between the p value and confidence intervals ability to ‘detect’ a statistically significant result is well known. It comes from how the statistic is calculated. Modern statisticians including statistical societies now recommend the use of confidence intervals rather p values as they are more useful. (Confidence intervals give some indication of the size of the difference and the sample size used in the study where p values do not. HOWEVER both only indicate statistical significance and not the SIZE or IMPORTANCE of the difference. In your case it is the size of the bias. There are many ssytems that will exhibit a bias between two methods or systems. What this important is the size of the bias vs the specification limits how they were enter mined and the severity of a part that might be accepted when it is slightly out of specification. This comes from your knowledge of the product and how the specifications were determined. The confidence intervals - or p values - are only a first step in this assessment. If there is no statistically significant bias then you can stop the assessment for the effect for the bias since there is none. If there is a statistically significant bias then you must assess it’s effect - no standard can or should provide you with any ’bright line’ guidance on what is acceptable. I would also remind you that if you get one answer from the p value and another from the confidence interval then the size of the bias is most likely pretty small (unless your sample size was incredibly small, like 10 parts rather than 30 parts). I would also remind you that percentages are notoriously misleading - they are a fake statistic that rely more on the denominator than the numerator.

I used to use a tag line here: “the manipulation of mathematical formulas is no substitute for thinking”.

 

[Miner]

Hi Miner, i would like to ask 2 questions related to Bias Study to you,..

I noticed that in most of the statistical softwares, be it Minitab, DataLyzer etc i can see that %Bias is always calculated and shown to the user. From my knowledge (quoting from AIAG MSA 4th Edition), in order for us to know whether bias is acceptable or not there are 2 ways,

1. if the p-value associated with the t-statistic bias is less than 0.05, or
2. if zero is between the lower bound and upper bound of bias confidence interval. (i may not phrase it correctly).

So my questions are:
Question 1: why are we calculating % bias here if we are not using it to evaluate the acceptance of Bias?. From my google search, i can't even find the acceptance criteria for this %bias at all...

Question 2: I took the example from AIAG MSA 4th edition whereby they have 15 data (5.8 , 5.7 , 5.9, 5.9, 6, 6.1, 6, 6.1, 6.4, 6.3, 6, 6.1, 6.2, 5.6 and 6) and then i calculated the p-value associated to t-statistic bias. I obtained t-statistic as 0.1217 and the associated p-value as 0.9048
So, if we are using criteria 1 from AIAG MSA which says that bias is acceptable if p-value less than 0.05 which in this case it is not so my earlier conclusion was this bias was clearly not acceptable.


However i further calculated the lower bound and upper bound confidence interval and i got the lower bound as -0.1107 and upper bound as 0.1241. Zero clearly lies between this lower and upper bound which means if we follow criteria 2 the bias study is acceptable.

In the very same AIAG book they mentioned this bias study is acceptable because of criteria 2 but they never mentioned anything about p-value.

So my real question is, does this means every single time when i do bias study i must check for these 2 criterias and if whichever pass that means bias study passed and if both of criteria failed only then i can conclude that this bias study failed?.

All of the AIAG reference manuals were written by teams of people who took responsibility for writing different sections of the manuals. Since there was not a single person writing an entire manual, you do find inconsistencies and omissions due to differences in opinion/approach, or assumptions as to what another person may have covered already.

Regarding Q1, software companies are not statisticians, so when they see that the GRR module used %GRR, they think they need to calculate a %Bias as well. Just because the software provides a value does not mean that value is important.

For Q2, you must have a 1st printing of the MSA 4th edition manual. It had a misprint on the bottom page 88 that said "less than alpha." I have the same printing but have an Errata Sheet from AIAG that corrects this to read "more than alpha." This would resolve your issue. NOTE: there may be rare, borderline situations where you might still get conflicting answers.

 

[hazwan2283]

All of the AIAG reference manuals were written by teams of people who took responsibility for writing different sections of the manuals. Since there was not a single person writing an entire manual, you do find inconsistencies and omissions due to differences in opinion/approach, or assumptions as to what another person may have covered already.

Regarding Q1, software companies are not statisticians, so when they see that the GRR module used %GRR, they think they need to calculate a %Bias as well. Just because the software provides a value does not mean that value is important.

For Q2, you must have a 1st printing of the MSA 4th edition manual. It had a misprint on the bottom page 88 that said "less than alpha." I have the same printing but have an Errata Sheet from AIAG that corrects this to read "more than alpha." This would resolve your issue. NOTE: there may be rare, borderline situations where you might still get conflicting answers.

thank you very very much Miner. *a bow down from me*