25.01.2013 Views

popper-logic-scientific-discovery

popper-logic-scientific-discovery

popper-logic-scientific-discovery

SHOW MORE
SHOW LESS
  • No tags were found...

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

430<br />

new appendices<br />

either if δ is comparatively large (roughly, if δ ≈ 1/2) or—in case δ is<br />

small—if n, the sample size, is a large number. We therefore find that<br />

P(e, h) − P(e), and thus our functions E and C, can only be large if δ is<br />

small and n large; or in other words, if e is a statistical report asserting a good fit<br />

in a large sample.<br />

Thus the test-statement e will be the better the greater its precision<br />

(which will be inverse to 2δ) and consequently its refutability or content,<br />

and the larger the sample size n, that is to say, the statistical<br />

material required for testing e. And the test-statement e so constructed<br />

may then be confronted with the results of actual observations.<br />

We see that accumulating statistical evidence will, if favourable,<br />

increase E and C. Accordingly, E or C may be taken as measures of the<br />

weight of the evidence in favour of h; or else, their absolute values may<br />

be taken as measuring the weight of the evidence with respect to h.<br />

8. Since the numerical value of P(e, h) can be determined with the<br />

help of the binomial theorem (or of Laplace’s integral), and since<br />

especially for a small δ we can, by (6), put P(e) equal to 2δ, it is<br />

possible to calculate P(e, h)—P(e) numerically, and also E.<br />

Moreover, we can calculate for any given n a value δ = P(e)/2 for<br />

which P(e, h)—P(e) would become a maximum. (For n = 1,000,000,<br />

we obtain δ = 0.0018.) Similarly, we can calculate another value, of<br />

δ = P(e)/2, for which E would become a maximum. (For the same n,<br />

we obtain δ = 0.00135, and E(h, e) = 0.9946).<br />

For a universal law h such that h = ‘P(a, b) = 1’ which has passed<br />

n severe tests, all of them with the result a, we obtain, first, C(h, e) =<br />

E(h, e), in view of P(h) = 0; and further, evaluating P(e) with the help<br />

of the Laplacean distribution and d = o, we obtain C(h, e) = n/<br />

(n + 2) = 1 − (2/(n + 2)). It should be remembered, however, that<br />

non-statistical <strong>scientific</strong> theories have as a rule a form totally different<br />

from that of the h here described; moreover, if they are forced into this<br />

form, then any instance a, and therefore the ‘evidence’ e, would<br />

become essentially non-observational.* 3<br />

* 3 One might, however, speak of the degree of corroboration of a theory with respect to a<br />

field of application, in the sense of appendices i and *viii; and one might then use the method<br />

of calculation here described. But since this method ignores the fine-structure of content<br />

and probability, it is very crude, as far as non-statistical theories are concerned. Thus in<br />

these cases, we may rely upon the comparative method explained in footnote 7 to the

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!