Rhythm Types and the Speech of Working-Class Youth in a Banlieue of Paris 107
instance read overall faster that Ramey and Mousa from the AF group (Table
4.2). The U Mann-Whitney non-parametric t-test for vocalic intervals as
dependent variable and ethnicity as grouping variable also showed signi¿ cant
differences (U=165432.00, z = 4.249, p<0.001, two-tailed). The EF group’s
mean duration (778 ms) was signi¿ cantly higher than the AF group’s (582
ms). As for consonantal interval length, there were also signi¿ cant differ-
ences between the two groups (U=210862.00, z = -3.175, p<0.001, two-tailed),
with the EF group’s mean consonantal durations (718 ms) signi¿ cantly higher
than the AF group’s (650 ms).
3.2 Rhythm type and inter-speaker variation
Regression analyses were carried out to explore what information about the
speakers and their speech led to their grouping in two distinct categories. In
the ¿ rst stepwise analysis, the dichotomous outcome variable, ethnicity, was
tested against ¿ ve rhythmic predictor variables: the three rhythm type indices
%V, ǻV, a n d ǻC, speakers’ articulatory rate, and the total number of segments
per phrase. In a second series of analyses, two external predictor variables,
age (as grade in school) and performance in school (moyenne générale), were
added (see Table 4.1). These variables have been introduced, since speakers’
degree of pro¿ ciency and/or choice of a more or less careful reading style
could be expected to have an effect on the hyper- vs. hypo-articulation of
vowels and consonants. Grade in school (approximate age) was expected to
inversely correlate with reading pro¿ ciency and the ability to reproduce a
careful reading style learned in school. Overall performance in school, based
on the of¿ cial score moyenne générale put out by the school each year, could
also show positive correlation with reading style, as better students could be
expected to read more carefully and thus have a lesser tendency to reduce or
elide vowels. In the third stepwise regression analysis, the speaker as an addi-
tional source of variations was added to the model.
Three backward stepwise regression analyses were performed. The
¿ rst was carried out with ¿ ve rhythmic predictor variables, the second and
third were performed with grade in school (approximate age) and moyenne
générale, respectively, added to the model. The fourth analysis combined
rhythmic predictors with grade in school and moyenne générale, while the
¿ fth analysis added speaker to all the other predictor variables. The initial
step in each analysis consisted in building a model with all available predic-
tor variables included. After this initial step, several runs were carried out to
test whether any of the predictors could be removed from the model without