Hypothesis: A testable explanation for observations. Science in general proceeds by an incremental process of refuting null hypotheses. Evidence is then presented persuasively in the form of a pattern that has been calibrated against unmeasured variation.

The null hypothesis is the refutable hypothesis of negligible systematic difference or pattern, in other words that nothing interesting is going on beyond random variation.

The test hypothesis is the alternative hypothesis of pattern, in the form of one or more real treatment effects, as defined by the statistical model.

A statistical test will reject the null hypothesis with probability α of doing so mistakenly (and thereby making a 'Type-I error'), or it will accept the null hypothesis with probability β of doing so mistakenly (and thereby making a 'Type-II error'). Data collection should be designed with a view to maximising the power to detect true effects at a given α, with power defined by the probability 1 - β.

Doncaster, C. P. & Davey, A. J. H. (2007) Analysis of Variance and Covariance: How to Choose and Construct Models for the Life Sciences. Cambridge: Cambridge University Press.

http://www.southampton.ac.uk/~cpd/anovas/datasets/