writings. This examination will show
us how he worked: whether he was capable of abstraction, reasoning,
generalisation, and what were the mistakes he was in the habit of
making. In order to determine the value of the data, we must criticise
each statement separately; we must imagine the conditions under which
the author observed, and ask ourselves whether he was able to procure
the necessary data for his statement. This is an indispensable
precaution in dealing with large totals in statistics and descriptions
of popular usages; for it is possible that the author may have obtained
the total he gives by a process of conjectural valuation (this is the
ordinary practice in stating the number of combatants or killed in a
battle), or by combining subsidiary totals, all of which were not
accurate; it is possible that he may have extended to a whole people, a
whole country, a whole period, that which was true only of a small group
known to him.[155]

VI. These two first series of questions bearing on the good faith and
the accuracy of the statements in the document are based on the
supposition that the author has observed the fact himself. This is a
feature common to all reports of observations in the established
sciences. But in history there is so great a dearth of direct
_observations_, of even moderate value, that we are obliged to turn to
account documents which every other science would reject.[156] Take any
narrative at random, even if it be the work of a contemporary, it will
be found that the facts observed by the author are never more than a
part of the whole number. In nearly every document the majority of the
statements do not come from the author at first hand, but are
reproductions of the statements of others. Even where a general relates
a battle in which he commanded, he does not communicate his own
observations, but those of his officers; his narrative is in a large
measure a "second-hand document."[157]

In order to criticise a second-hand statement it is no longer enough to
examine the conditions under which the author of the document worked:
this author is, in such a case, a mere agent of transmission; the true
author is the person who supplied him the information. The critic,
therefore, must change his ground, and ask whether the informant
observed and reported correctly; and if he too had the information from
some one else (the commonest case), the chase must be pursued from one
intermediary to anothe