The ability of humans to alter their behaviour in reaction to predictions made about it is listed but not discussed by Ake (in a recent issue of this Journal) as an objection to the possibility of producing testable general laws in political science [Claude Ake, The Scientific Status of Political Science’, n (1972), 109–15, p. 112]. McLean makes this ability the nub of his argument against the possibility of a comprehensive theory of politics and in favour of partial theories [Iain McLean, ‘Comment on “The Scientific Status of Political Science"’, II (1972), 383–4]. Unless the possibility of checking limited generalizations against subsequent behaviour is completely abandoned, however, behavioural reactions of the type cited by McLean seem likely to occur in response to limited as well as to general predictions and thus to constitute as much of an objection to partial theories as to a comprehensive general theory. Philosophically human free will is often cited as an insuperable obstacle to the development of a causal and statistical social science,1 so the question is central to any general discussion of the scientific status of political analysis. Since political scientists happily make limited generalizations, e.g. about voting and coalition behaviour, and at least loosely check them against subsequent behaviour without suffering apparent difficulties, the practical effects of these philosophic objections seem limited. In order to understand why, discussion must descend from the abstract level of free-will versus determinism to some of the conditions which limit behavioural reactions to previously formulated generalizations and predictions: