EPA engine testing requirements (40 CRF part 1065) repeatedly say “consult a statistician” in § 1065.12, Approval of alternate procedures.
We recommend that you consult a statistician if you are unfamiliar with these statistical tests. …
If the population of paired difference is not normally distributed, consult a statistician for a more appropriate statistical test, which may include transforming the data with a mathematical function or using some kind of non-parametric test. …
“If the population of values is not normally distributed, consult a statistician …
Essentially the regulations say to consult a statistician when you’re using an unusual test (something other than a t-test or F-test) or when the simplifying assumption of things being according to a normal (i.e. Gaussian) probability distribution.
Under textbook conditions, standard tests and standard assumptions apply. However, life does not always follow textbook conditions. Real data is often a little wonky. That’s not necessarily a good or bad thing, but it does mean that uncritically applying the most common statistical techniques may not give accurate or even meaningful results. But statisticians are experts in handling the quirks of real-world data. We have more tools in our toolbox t-tests or F-tests.
We can help you understand what the statistical terminology in 40 CRF part 1065 § 1065.12 means. And we can help you design and conduct experiments to satisfy the regulation’s requirements.
We are statisticians, not lawyers, but we do consult for lawyers in matters where law touches on statistics. This page is not legal advice, nor can we give legal advice. But we can give statistical advice.
If you’d like to discuss how to apply advanced statistical techniques to your engine testing data, let’s talk.
Trusted consultants to some of the world’s leading companies
Photo by Chad Kirchoff on Unsplash