Acceptance Criteria for Linearity Test



Acceptance Criteria for Linearity

It is clear that the auditors are now expecting full MSA studies (i.e. bias, stability, linearity) not just Gauge R&R. What are you using for your acceptance criteria for linearity and why? Any insight would be appreciated.

Atul Khandekar

I think 'Goodness of Fit' AND the 'Slope' of Regression Line are the factors to look at. If there is a uniform bias over the range of the instrument, the regression line will be horizontal and if G-of-F is low, the line has no meaning. However, I couldn't find any guidelines as to the actual value at which the G-of-F can be called 'low'.

Again, the slope of the line may well be zero and Goodness of Fit > 0.9, but what value of bias should be accepted?


Al Dyer

Just from a practical point of view,

As the result of a long term linearity study on a height gage, the following hypothetical results are realized.

I base this on the simplified definition that linearity is the measurement of bias over the range of the gage.

From 0 - 1" the bias is .0010
From 1 - 2" the bias is .0015
From 2 - 3" the bias is .0200
From 3 - 4" the bias is .0200
From 4 - 5" the bias is .0015
From 5 - 6" the bias is .0010

What is the acceptable criteria for linearity? I really don't know or care.

My care:

With the documented results I can determine which products/characteristics can be measured with the height gage.

If my targeted dimension is .500 +/- .010 I "would" feel comfortable using the gage.

If my targeted dimension is 2.500 +/- .010 I "would not" feel comportable using the gage.

If my targeted dimension is 5.500 +/- .010 I "would" feel comportable using the gage.

I don't know the correct term, is it the "10% discrimination rule" or the "10 to 1 rule"

Am I making sense? Need help from the gurus!


[This message has been edited by Al Dyer (edited 10 April 2001).]


Hi guys,

someone with a PhD in statistics told me a goodness of fit of about 0.7 is sufficient.

About the slope of the line, I have the same question: how much is acceptable?

Or maybe the linearity analysis is used just to estimate the bias at vaious working values?

Anyone can help?

Top Bottom