280 Erik R. Thomas, Norman J. Lass, and Jeannine Carpenter
analysis. Logistic regression was applied to the results plotted in Figure 12.1.
The main effects are shown in Table 12.4. All four independent variables
proved signi¿ cant for the African American listeners. For the North Caro-
lina Eu ropean A mer icans, presence/absence of diag nostic vowels and nat u ral /
swapped prosody showed signi¿ cant effects, but sex and ethnicity of speaker
did not. For the West Virginia listeners, only natural/swapped prosody proved
signi¿ cant, though sex of speaker nearly did.
It is clear that different groups of listeners base their identi¿ cations on
different sets of cues. For African Americans, the fact that they identify
males more accurately than females suggests that they associate features
of African American English most strongly with African American male
identity, or perhaps features of European American English most strongly
with European American male identity. The failure of the West Virginians to
show any signi¿ cant effect for presence/absence of diagnostic vowels prob-
ably has to do with the fact that the speakers were from North Carolina.
Figure 12.3 Interaction between sex of speaker and natural/swapped prosody in
Experiment B.