-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

- - 14 -

APPENDIX D: TECHNICAL NOTES


Sampling method:   Questions repeatedly arise regarding the meaning of
the numbers determined in the survey and their reliability, particularly
the meaning of a "random sample."

The ordinary connotation of the word random suggests imprecision rather
than the opposite.  If, however, the objective is a practical yet reliable
determination of the average characteristics of a large population of
people or objects, then random sampling is the way it has to be done.

If Time magazine wanted to predict the outcome of a presidential race,
it would first discard the idea of polling all voters, simply on the basis
of impracticality.  Next it might consider getting the opinions of all
subscribers to Time.    Apart from questions of practicality, such a
project would be fatally biased;  the subscribers are not typical,
economically or politically.  In fact, they have also been biased to some
extent by reading the same magazine.   (Used in the statistical sense the
word biased is not a pejorative.)  Thus a very large survey would still
be an unreliable indicator of public sentiment.

How is it, then, that random surveys of a few thousand people can
closely predict the outcome of national elections?   It is because the
surveyors succeed in their deliberate efforts to acheve randomness
(absence of bias).

Another simple illustration from a writer on statistics: a manufacturer of
fireworks obviously wants his product to detonate when ignited.   But he
can't test all his products for this property or the only ones he will
have left will be the defective ones.   This fellow has a special need for
sampling.

He might simply take one day's production, test all of it and assume that
his result applies to the entire year's production.    But now the
possibility of bias creeps in.   The conditions of the raw materials and
processes on that day may not be representative of what happens the rest
of the year.

The best he can do is to select samples to test, at random, all through
the year.  In this way he can avoid the various kinds of bias that might
otherwise creep in.

The meaning of random selection in this case is that he will establish
conditions under which any firecracker is as likely to be selected for
testing as any other.

The establishment of such conditions, and particularly the fraction of
firecrackers he must select for testing in order to achieve a satisfactory
degree of assurance in the reliability of his product, is a subject for
mathematical statistics.   The result can be illustrated from one of the
findings of our survey.

-----BEGIN PGP SIGNATURE-----
Version: PGP Personal Privacy 6.5.8

iQA/AwUBQk8XYbw9MOKEeRC8EQKUuwCg/EFpS7bL1KE7+uRafhX4sgq2IXcAoPYG
NBtVxqq3NFPAmqrwXD6j+R33
=eg4d
-----END PGP SIGNATURE-----