Linearity and Bias Study - Bias Study fails - Seeking advice

A

Anatta

Hi all,

I need help from MSA experts.

One of our engineers did an GR&R study. The spec requirement for the test process is only 2 decimal points and the tester is giving reading on 3 decimal points, which is fullfill the rule of thumb of 1/10.

Porblem is, when doing linerity and bias study, the reference value is 0.20 and the tester is giving out reading of 0.199 for a few points and yet it fails bias study. Sounds like there's no room for tolerence and there's no way to improve the testers.

Also, the slope is only 0.00002774 and yet the p value is 0.034, which shows there's a significant slope that contribute to failing linearity study. I am using Minitab which computes using the AIAG method, can someone advice how should we improve from here or is there anyway we can accept the MSA study?

Attached is the MInitab project output pasted inside the excal file and the raw data.

:confused: Please enlighten. :truce: :thanx:
 

Attachments

  • Linearity and bias study.xls
    44 KB · Views: 953

Miner

Forum Moderator
Leader
Admin
Sometimes you need to distinguish between the concepts of statistical significance and practical significance.

In this situation, your gage shows a negative bias at lower values. Is this statistically significant? Yes. Do you need to do anything about it? Maybe.

You will need to review the calibration of the gage to verify that it meets your calibration acceptance criteria. But do you need to discard the gage simply because it fails this test? No.

If you are using the gage as an inspection gage, your question is entirely valid. The bias, though statistically significant, is of no practical significance in this situation. Be aware of it, but make an informed decision. If, however, you are using the age for statistical applications such as SPC or DOEs, then it does matter and should be addressed.
 

BradM

Leader
Admin
Anatta,

From your spreadsheet, your reference mA value is to 1 decimal place, and your measure is to 1-3 decimal places. I have a $1000.00 (US dollars) multimeter that gives me six decimal place resolution with mA (HP multimeter). Roughly, that gives me like four decimal places to work with. I'm concerned about the level of resolution with your mA readings.

Any calculation/inference is only as good as the data. Saying, if you are concerned about linearity, you will need to maximize the quality of the data to get a linearity calculation that is worth anything.

Very little (if any) data will be perfectly linear. Be sure you establish a reasonable tolerance for linearity. Otherwise, as astutely mentioned by others, you may getting error equivalent to type I error, finding significance when there is actually no significance.

You state that it fails the bias study. How are you assessing this? You may be too stringent in the linearity assessment. I am not familar with Minitab, so I don't know how the software works.

You also state there is no way to improve the testers. Something in your test process has to be a variable that can be improved upon. Yes?
 
A

Anatta

what we meant by the tester can't be improve also brought to constraint of increasing the data sensitivity captured. The management in our company is unwilling to invest more money to upgrade the tester.

In Minitab, it's bias and linearity results is evaluated from the P value.

For linearity, hipothesis null will be slope = 0, thus is P must be more than 0.05 in order not to reject hipothesis null.

For Bias, Bias = 0, P value must also be larger than 0.05.
 
A

Anatta

another question is, if we really go to the other more practical approach to access the MSA study, what are the guidelines?

well, we need to put in a documented procedure on how we accept or reject a linearity study. thus, can we justify our own acceptance criteria rather than follow AIAG approach? What will the impact like if we have our own criteria.

For eg, if slope = 0 to accept linearity, can we determine that if the slope is less than a certain value, the linearity is still consider pass? Like maybe slope < 0.5? :truce: :thanx:
 

Miner

Forum Moderator
Leader
Admin
I would create a two step process in your procedure. The first step being the bias/linearity study. If the p-values are not significant, accept the gage. If the p-values are statistically significant, the second step would be to perform a risk assessment. Look at the practical significance of the linearity slope/bias and make an assessment of whether the gage may still be suitable for use in a specific application.

In your case, there was a statistically significant issue with bias at the low end of the range. However, the amount of bias was of no practical signifiance if the gage is used for inspection, particularly at the mid/high end of the range.

A justifiable risk assessment would say that we recognize the bias/linearity, but have decided that there is no risk to using this gage under these conditions.
 

BradM

Leader
Admin
Excellent advise from Miner.

Given that you cannot improve your resolution, you will want to be cautious about what value you place on the linearity assessment from the software package.
 
Top Bottom