52 Jen Hay and Margaret Maclagan
Indeed, both the identity of the base and the F3 of the “real” /r/ were
signi¿ cant predictors of F3 of intrusive /r/. Strikingly, on top of these effects,
there is a signi¿ cant effect of social class. Holding real F3 and base identity
constant, people from higher social classes produce higher F3 (i.e., produce
the /r/ with less constriction).
The model statistics are shown in Tables 2.4 and 2.5. The effect of the
average F3 during Sarah is shown in Figure 2.6. As one would predict, the
greater the F3 is during Sarah, the greater it is during intrusive /r/. Includ-
ing this effect in the model, then, has a normalizing effect. The effect of the
identity of the base is shown in Figure 2.7. Some of the variation seen here
may relate to the effect of the preceding vowel, although we don’t expect radi-
cal differences in F3 amongst these vowels. The lower value for bra than ma
probably reÀ ects the fact that it also has an /r/ earlier in the word.
Figure 2.8 shows the effect of social class. When the base word and the
“real” /r/ F3 are held constant, individuals from lower social classes (and so
higher Elley-Irving scores), produce a lower F3 during the intrusive /r/. We
interpret this as reÀ ecting the fact that their intrusive /r/ is more /r/-like.
Table 2.4 Anova Table for Ordinary Least Squares Model Predicting F3 of /r/
Factor d.f. Partial SS MS F P
“real” /r/ F3 1 1483982 1483982 36.29 <.0001
Social cla ss 1 167535.2 167535.3 4.1 0.0 4 45
Base 4 1398549 349637.2 8.55 <.0001
REGRESSION 6 2594777 432462.8 10.58 <.0001
ERROR 178 7278811 40892.2Va lue
Intercept 1008.15
“real” /r/ F3 0.4454
Social class -15.2102
base=claw 111.3187
base=ma 204.0661
base=plough 83.1766
base=sofa -57.3682Table 2.5 Coef¿ cients for Ordinary Least Squares Model Predicting F3 of /r/