Data has been gathered for a specific analytical test method to start SQC. The issue is that when the data is tested for normality, using Anderson Darling, the null hypothesis can not be rejected.

-I have attached a photo to provide more information about the data.

My question is, does this "non-Gaussian-nes" mean that the test method contains some sort of bias?

PS- Sorry if there are any issues with the form of my question; please let me know if you need additional information.

enter image description here


As far as using your data for standard quality control methods which require that you have Gaussian data, the results of your tests confirm that you are currently unable to reject the null hypothesis and that the data would be acceptable (regarding that requirement).

Most tests also require that the data is in statistical control, as evidenced by a Shewhart Chart or . Things may also change as you currently only have 20 points of data.

While any test is bound to have some type of bias, the Anderson-Darling test is fairly reliable. You can test with Shapiro-Wilk, or the K-S test (sorry, I don't recall how to spell their names, but it is a common test) can also be used. There is also an option for a normal probability plot within that will draw the confidence interval along with the plot.

  • $\begingroup$ Thank you for your response. So what would happen if the collected data, that is going to be used to create the SQC, does not follow a gaussian distribution? $\endgroup$ – Ochemwiz Mar 22 '17 at 1:19
  • $\begingroup$ At that point, you would need to looking at options to either transform your data or use some of the other tools designed for non-Gaussian data. There are plenty of tools out there, and Minitab has a good selection of tools to start with. $\endgroup$ – Tavrock Mar 22 '17 at 1:22
  • $\begingroup$ But wouldn't a non-gaussian data set indicate some form of bias during the data collection? For example, these points are being collected by 4 people at different times over many weeks. Could it be possible that someone is performing the analysis of the quality control material differently? $\endgroup$ – Ochemwiz Mar 22 '17 at 1:45
  • $\begingroup$ Currently, your data is normal, or rather there is not enough evidence to reject that null hypothesis. If you suspect your data collection method, you can conduct a measurement system analysis (msa) to know for sure. $\endgroup$ – Tavrock Mar 22 '17 at 1:50

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.