A school, like a nation, is a complex system existing in discursive rather than physical geographical space (other than as a set of buildings). The meaning of "school" is therefore dependent on actors' perceptions, and these are dependent on factors apparently beyond the school. Research evidence relating to such perceptions comes not as data or phenomena, in the sense of observable behaviour subject to mathematical laws, but rather as phenomenographic fragments: pieces of described experience. Although numbers can be included in such descriptions, they are thus used qualitatively. Overall evaluative judgements about school ("good school"; "effective school") cannot validly be derived from statistical models, however sophisticated, as quantitative data alone cannot capture the indefinable range of experiences relating to any particular school, particularly given a lack of absolute consensus about educational values. Research into the effectiveness of schools can thus have limited predictive validity, although reductionist approaches can certainly reveal strong specific correlations that it is tempting to regard as causes and effects, and that might be seen to endure over a period. From this perspective, given the lack of a firm evidence base, policy-makers with only a researcher's interest in any institution are less well placed to make decisions about particular schools than those "on the ground" with more sophisticated (though still inevitably incomplete) understandings of local complexity. What is important about a school is how it is imagined by those who imagine it, among whom the policy-makers are inevitably a tiny constituency with very limited terms of engagement. Thus, the methodological argument becomes a political argument for decentralisation. (Contains 1 note.)