QuikSigma: Gauge Repeatability and Reproducibility, Part 2

We continue our lesson from Part 1 in Gauge Repeatability and Reproducibility  by diving right back in to how to evaluate a measurement system using the G R&R tool in QuikSigma.




I’ve loaded some data for us to analyze. These ratings here are our output variable and these are on a scale of one to ten, supervisor’s rating, or set of supervisors ratings, of call center technicians and how well they handled situations. There’s a lot of numbers here, but the one that we’re really interested in is the percent gauge R&R right here in blue, 26.8 percent. That is the standard deviation of the measurement system error divided by the total observed process variation and of course that also includes the measurement system error. We’d like this number to be below ten percent in order to think we have a good system and are capable of detecting changes in the process.

Well, this is much greater than 10 percent anything in the 30-10 percent range we call marginal, so this is the deep edge of marginal. Now there are two factors that we look at, repeatability and reproducibility. That’s our R&R. Repeatability, the 24.2 percent, is our gauge itself. That’s just test-retest error and the reproducibility, is the variation that is from the fact that you’ve got multiple people making measurements and they may have slightly different techniques or methods and this will tell you where you need to concentrate if you have a problem. Now, the number of distinct categories, if it’s fewer than four, you’ve got a problem. This is the number, this is the same as the discrimination ratio on inter class correlation coefficient. It’s the number of times that the measurement variation will fit inside the total process variation and fit inside is kind of in quotes.

Now, there’s another column that we can get over here. This helps us tell if we’ve got enough of a measurement system to detect process change. There’s another thing that we might be interested in, which is the percent tolerance, or the P-T ratio, or the precision to tolerance ratio. To demonstrate that, I’m going to put some specifications in, a 10 and a 4, and then I’ll hit calculate again and now we get a new column. Fundamentally, this column is directed at the question, can we tell good parts from bad? So this is a number over here, the percent gauge R&R, that the production manager would use. This is a number that the incoming inspection or the outgoing inspection would use and as we can see here, we’re really not very adept at telling good technicians from bad ones. We’re way past our thirty percent. So obviously there’s an opportunity here.

Return to Blog Posts >>

Leave a Reply

Your email address will not be published. Required fields are marked *