12 February 1960. Cogitations

All our ideas are reducible in the long run to the data of sense-experience.

Montesquieu's theory [Copleston, p. 11] is not simply an empirical generalization but deals with ideal types of government. This raises an interesting possibility: if there are theories, e.g. quantum mechanics, in which there is no level of empirically verifiable data but a level of statistical hypotheses that take the place of empirically verifiable data, could one have a theory in which the place of empirically verifiable data was taken by hypotheses of ideal types analogous to statistical hypotheses? And if so, what is the relationship of the hypotheses of ideal types to premises or high-level hypotheses of increasing generalization? Are generalization and idealization the same thing? If not, in what way do they differ? There may be a place for three types of hypothesis that need to be distinguished:

1. high-level, generalized and not empirically verifiable, therefore

2. statistical hypotheses, taking the place of empirically verifiable hypotheses and embodying a probability statement. This involves the question: how does the scientist use these probability statements?

3. ideal-type hypothesis, which says that in a given state (which in fact will never be found) there does exist a state of ought-to-exist-ness of whatever it is that the hypothesis is referring to. For example, take the statement that ‘excessive splitting and projective identification lead to a disintegration of the personality’. ‘Excessive projective identification’ is an ideal state and not an empirical generalization. The hypothesis asserts that in so far as this ideal state is realized, a disintegration of personality ought to be found. Is this a probability statement in a statistical hypothesis? No: but

   Page 124

   I suspect it is a statement that is ‘understood’ when a statistical hypothesis is used, and this is the clue to the problem posed by Braithwaite (Scientific Explanation, p. 124