Robert Renehan, "The Michigan Alcidamas-Papyrus: A Problem in Methodology," Harvard Studies in Classical Philology
75 (1971) 85-105 (at 93):
Three fundamental principles may be stated here:
(i) A word which survives only in postclassical authors, if it is of rare occurrence even there and if there is an obvious reason why it would not be of common occurrence at any period, should not be used as evidence for dating.
(ii) In general, a word which occurs often, and exclusively in post-classical authors is significant. A word which occurs chiefly in post-classical authors is less good evidence but, under certain conditions, may be used with some probability. A word which is adequately attested for the classical period, no matter how frequent its occurrences later, should not be used as evidence for a late date. All that is here set down is subject to the obvious principle that
(iii) In employing statistics a certain minimal sampling is necessary before probable results of any validity can be obtained.
Id., p. 100:
[W]e should remind ourselves that the greater bulk of classical Greek has perished and that therefore our knowledge of classical Greek diction is imperfect. The use of linguistic evidence to date a document is a valid procedure, but it has its limitations which should be recognized.
More useful reminders (id., pp. 89-90):
I remind the reader that it is the exception and due to special circumstances (say, the coinage of a new philosophical or medical term or of a new poetic compound) when the oldest extant example of a word is in fact the oldest actual occurrence of that word. I further remind the reader that it is the exception when the oldest actual occurrence of a word is a written occurrence at all and not rather an oral one.