A Reader in Sociophonetics

Phonetic Detail in the Perception of Ethnic Varieties of US English 303

we see that for groups 3 and 5 the multivariate analysis accounts for more
than an additional 15% of the variation, whereas for group 1 the multivari-
ate analysis accounts for an additional 10% and, for group 4, an additional
7% of variation. Group 2 has the slightest difference between the two values
(2%) largely because one measure, percent glottal pulsing, carries the most
statistical weight for that group. The strongest difference occurs in the group
3 perception data where the multivariate model accounts for 25% more varia-
tion than the strongest univariate measure. Strength of the multivariate analy-
sis is that the ¿ ve non-perception study groups in Table 13.5 are signi¿ cant,
while some of the measures’ univariate F values—change in F1 for groups
1 and 3, change in F0 for groups 3 and 4, speci¿ cally—are above the usual
0.05 signi¿ cance level. On Table 13.6, the accounted-for difference across
groups is mirrored by the total-sample standardized canonical coef¿ cients
(TSSCC). This value is suggestive of the discriminatory power between the
acoustic predictors with respect to some underlying factor pertaining to the
voicing contrast. The TSSCC value indicates that when the tokens are stan-
dardized, the percent glottal pulsing receives the highest coef¿ cient for all
Wisconsin English groups (1.036, 1.030, 1.170, and 0.864, respectively) and
the perception data (2.026), and the second highest coef¿ cient for the control
group (0.639). When vowel duration is present (i.e., excluding group 2 and
including the group 3 perception data), it has the second strongest TSSCC
value except for the control group where it has the strongest value. Figure
13.2 presents the changing difference between vowel duration and the percent
glottal pulsing. First, it is worth pointing out the shared pattern across all of

Table 13.3 Results of Stepwise Discriminant Analysis Using a Forward Method

for Four Test Groups of WI English Speakers and One Control Group

(continued)

Group, Variable Partial r^2 F p > F

Ave r age

squared

canonical

correlation

6: Controls (df=4, 93; N voiced=52; voiceless=46)

pulsing duration 0.233 29.1 0.0001 0.233*

V:C ratio 0.144 16.0 0.0001 0.343*

vowel duration 0.035 3.5 0.0664 0.366*

% glottal pulsing 0.052 5.1 0.0261 0.399*

*p<0.05