Setting equivalence limits prior to experimentation

To demonstrate practical superiority, equivalence or non-inferiority, we must inspect the 95 per cent confidence interval for the difference between mean values. The P value is completely irrelevant to all these questions and is potentially misleading. 

Equivalence limits should be set before experimental work begins. It is very difficult to decide objectively where equivalence limits should be placed if you already know the experimental results and what the consequence would be of placing the limits at any particular level. 

For the sake of your own credibility it is also very useful if you can register the decision with some independent party, prior to experimentation. That way, everybody else can be assured that the placing of the limits has not been inappropriately influenced. 

If you push this one too far, it will stand out like a sore thumb, but within reason, you might get away with it. 

However, there is still the question of where we are going to put those pesky equivalence limits. In the first case, we will adjust them nice and close to zero, then even our pathetic effort will cause a practically significant difference. For the latter two, we will invent some reason to set them generously wide and hey presto, we are equivalent or non-inferior . 

---