340 Paul Foulkes, Gerard Docherty, Ghada Khattab, and Malcah Yaeger-Dror
(“boy” or “girl”), and the independent variables were amplitude, f0, rate,
voice quality and variant (plain versus laryngealized for the run on medial
responses, plain versus pre-aspirated for pre-pausal responses). Post hoc anal-
ysis was performed using chi square tests.
- Results
Since the listening test yielded a binary outcome we report all results arbi-
trarily in terms of “girl” responses.
5.1 Overall responses
Although we were not concerned with how accurate listeners were in iden-
tifying the sex of the speaker, it is worth reporting that the proportion of
correct responses was very similar for the three listener groups (Table 14.3).
Moreover, the ¿ gures approached chance level at 50%, which was perhaps to
be expected. The proportion of “girl” responses, however, was slightly higher
than “boy” responses for all three groups (recall also that more of the tokens
did in fact come from girls’ speech). Neither the correct responses nor the
“girl” responses differed signi¿ cantly across the two British listener groups.
However, the American group gave signi¿ cantly fewer correct responses than
the non-local UK group (chi sq = 5.992, df = 1, p < .025).
Table 14.3 Overall Distribution of Results
listener group correct responses (%) “girl” responses (%) N
Ty n e s i d e r s 4 8 .7 52 .0 1, 3 4 0
non-local UK 49.4 53.3 2,343
Americans 46.5 51.4 7,6485.2 Logistic regression analysis
The results of the exploratory logistic regression analyses are summarized
in Table 14.4. Unsurprisingly the results were complex, as is to be expected
when we consider the number of factors included in the analysis, and the fact
that the stimuli were both small in number and relatively uncontrolled. All
factors were returned as signi¿ cant in one or more of the runs for at least one
of the listener groups.