Type III Error

Turnip Blood

I often tell people that the First Rule of Statistics is "You Can't get blood out of a Turnip." What I mean by this is that if you do not collect data that will answer the question(s) at hand, then you cannot use that data to answer the question. That is, if you want to know if changing the pressure on a process will change the product quality, then you need to change the pressure in the tests. If you want to know if both Temparature and Pressure will affect the product, then you need to change both Temperature and Pressure Independently. That is, you need to run tests at Low Pressure and Low Temperature, tests at Low Pressure and High Temperature, tests at High Pressure and Low Temperature, and tests at High Pressure and High Temperature.

It is amazing how often people expect to use production data where 99% of the time the Pressure is Low and 1% where it is High to determine the effect of Pressure. Ninety-nine cases of Low Pressure and One of High, and why is there "No Statistically Significant Difference"? Or all the tests are at Low Pressure and Low Temperature or else at High Pressure and High Temperature. "Why does the ANOVA say singularity and no degrees of freeedom? I have data. Why doesn't this work?"

I just say, "You can't get blood out of a turnip."

posted by X-Bar @ 10:48 AM