Experimental approach

Nested syllogism of experimentation

A: chance alone is in effect B: design (systematic force) is in effect C: observable implication of chance alone

Either A or B. If A then C. Not C. Therefore not A. Therefore B. (not replaced by exceeding low probability of)

Randomised treatment assignment

Treatments are randomly allocated to units. This is where probability enters.

Qualifying differences

1. Assume there is no difference. (ie. treatment group is not related to
2. attribute)
3. Work out how some test statistic distributed in this case.
4. How does the actual assignment compare to this distribution? (calculate a
5. p-value by counting number greater than or equal to and dividing by total
6. number of permutations)

Too hard to work out all possible permutations, so usually just take random sample. Traditional t-tests (etc.) asymptotically based on this principle.