Bibliotheca Arnamagnæana - 01.06.1999, Page 348
328
Editor’s postscript
4. The statistical method in Chapter VII
An earlier version of the chapter on the expletive particle was published
in Norwegian in a Festschrift for Finn Hødnebø (Fidjestøl 1989). This
led to a critical review by Hans Olav Egede Larssen (1991), directed
more against the use of Charles Muller’s handbook (1981) than against
the author. Professor Fidjestøl responded to the criticism (1991) and lat-
er supplemented his statistical analysis with a partial rank correlation, as
can be seen from the chapter in the present book. Professor Fidjestøl
reports on the debate in the chapter (p. 216, n. 10), and there is reason to
believe he would have pursued the methodological discussion before
completing work on the book. In view of this, it was felt appropriate to
ask for a new appraisal by a statistician. Associate Professor Håkon K.
Gjessing at the University of Oslo has gone through the whole chapter
and made the following observations:
The general idea of comparing the relative frequency of the particle
with the rank of age seems valuable from a statistical point of view.
To use Spearman’s rank correlation is quite reasonable since it yields
a comprehensive measure of the correspondence between two rank-
ings. As Fidjestøl points out, it is important to attach more weight to
the longer of the poems, since in those the relative frequency will be
Jess influenced by random disturbances.
However, the use of “reduced deviations” (deviations normed by
the theoretical standard deviation) is not to be recommended in this
context. Firstly, the computation of the reduced deviation will alter
the ranking of the relative frequencies, thus possibly obscuring their
relation to the (unaltered) sequence of age ranks. Secondly, since the
short poems retain equal weights to all other poems in the computa-
tion of Spearman’s rank correlation, the net effect of the “reduction”
is not to reduce their influence but rather to disturb the resulting cor-
relation coefficient in an unpredictable manner.
A natural way to deal with the problem is simply to compute the
correlation coefficient directly from the ranks obtained from the rel-
ative frequencies, without the “reduction”. In the computation, each
poem should then be weighted according to its length. This will
achieve the desired effect, and it will also make it less important to re-
move the shortest poems, since their weights will be correspondingly