46 Jen Hay and Margaret Maclagan
We ¿ rst investigated the factors affecting the likelihood that an /r/ would
be produced at all. We then turned our attention to the cases where /r/ was
produced, in order to assess whether there was any meaningful variation in
the F3 values.
- Results
3.1 Likelihood of Intrusive /r/
First, we ¿ t a logistic regression model investigating the likelihood of use of
intrusive /r/. Independent factors considered were the speaker gender, age,
and social class, as well as the base word, and the identity of the af¿ x. All
factors but age and gender proved signi¿ cant predictors of /r/-insertion. Gen-
der was signi¿ cant to .06, and so we have left it in the model. This allows us
both to consider the direction of this near-signi¿ cant effect and to consider
the nature of the remaining effects while holding this potential gender effect
constant. Recall, however, that there are only four males in the data set, so
any observed effect of gender is tentative at best. In addition to the effect
of the af¿ x, the model also includes a signi¿ cant interaction between social
class and the identity of the base. The model details are given in Tables 2.2
and 2.3. For factors involved in interactions, the signi¿ cance level in Table
2.2 incorporates both the main effects and the relevant interactions (i.e., Fac-
tor + Higher Order Factors).
Figure 2.2 Utterance of the word clawify produced by Speaker 2, with an indication
of where the F3 reading was taken.