ven in the appreciation of the morality of a people.[206]
Generalisation rests on a vague idea that all facts which are contiguous
to each other, or which resemble each other in some point, are similar
at all points. It is an unconscious and ill-performed process of
sampling. It may therefore be made correct by being subjected to the
conditions of a well-performed process of sampling. We must examine the
cases on which we propose to found a generalisation and ask ourselves.
What right have we to generalise? That is, what reason have we for
assuming that the characteristic discovered in these cases will occur in
the remaining thousands of cases? that the cases chosen resemble the
average? The only valid reason would be that these cases are
representative of the whole. We are thus brought back to the process of
methodical sampling.

The right method of conducting the operation is as follows: (1) We must
fix the precise limits of the field within which we intend to generalise
(that is, to assume the similarity of all the cases), we must determine
the country, the group, the class, the period as to which we are to
generalise. Care must be taken not to make the field too large by
confusing a part with the whole (a Greek or Germanic people with the
whole Greek or Germanic race). (2) We must make sure that the facts
lying within the field resemble each other in the points on which we
wish to generalise, and therefore we have to distrust those vague names
under which are comprehended groups of very different character
(Christians, French, Aryans, Romans). (3) We must make sure that the
facts from which we propose to generalise are representative samples,
that they really belong to the field of investigation, for it does
happen sometimes that men or facts are taken as specimens of one group
when they really belong to another. Nor must they be exceptional, as is
to be presumed in all cases when the conditions are exceptional; authors
of documents tend to record by preference those facts which surprise
them, hence exceptional cases occupy in documents a space which is out
of proportion to their real number; this is one of the chief sources of
error. (4) The number of samples necessary to support a generalisation
is the greater the less ground there is for supposing a resemblance
between all the cases occurring within the field of investigation. A
small number may suffice in treating of points in which men tend to bear
a strong res