480 JOURNAL OF LAW AND POLICY
Examining CB’s messages in a similar way, there was no
indication for removal of any messages, and this left the original
set of 203 messages intact.
D. Identifying Features with Discriminative PowerThe selected undisputed texts and features were further
examined to determine whether there were features that
discriminated consistently to some degree between the two
writers in their known texts. Features were only retained in the
analysis if one author used them in at least twice as many
messages as the other—that is to say where one author used a
feature at a rate of more than sixty-six percent of its total
occurrence. This left a set of just eighteen features that are
tabulated in Table 1.
This frequency table based on the known messages of AB
and CB clearly indicates, in a general sense, that CB and AB
write text messages using distinctive styles, and for some
features the degree of distinctiveness is absolute. AB never, in
this data, writes “with” as “wiv” and CB never writes “had” as
“ad.” Other features are features of predominately one author;
CB tends to use commas, and AB does so very rarely. This
contrasts, for example, with AB’s tendency to use “t” for “the,”
which is only rarely used by CB. For each author, lists can thus
be created of features characteristic of their text messaging style;
ten features for AB and eight features for CB. In contrast to
Coulthard’s method in the Jenny Nicholl inquiry, these feature
lists are not necessarily reciprocal alternates. For two words,
“don’t” and “with,” each writer has as a feature a preferred
spelling variant of that word, but for the other features this is
not the case.
For neither author can these lists be considered identifying in
an absolute sense. The features contained in these lists are not
linguistic “fingerprints” identifying individuals against a
population. Rather, they demonstrate a relative consistency of
habit and a pairwise distinctiveness which thus can be used to
stylistically discriminate between messages of the suspect and
the victim in this case. As we do not have good knowledge of
the distribution of texting features across the population of all