A separate analysis was conducted for each data set (Welsh-English and
Spanish-English) using the Goldvarb X program (Sankoff et al., 2005). This
program primarily uses logistic regression in order to model the factors that
affect variant use while simultaneously taking into account their interaction
(Tagliamonte, 2009). The results of the multivariate analysis reveal which of
the factors are significant and which are not, as well as the relative weight
of each factor. That is to say, the results show which factors favour or disfa-
vour the variant in question, and which factors are more significant than
others in predicting the variant chosen (i.e. the ML). Further analysis using
this method can also show the constraint hierarchy, which is the order of the
categories within a factor group. For example, for the variable language used
in primary school there are four possible responses: Spanish or Welsh, English,
Both, Other. The analysis could reveal the hierarchy of responses according
to the distribution of the ML.
In order to reveal community-wide trends, our study analysed the ques-
tionnaire responses from all of the participants for each corpus (85 Spanish-
English and 151 Welsh-English) using a χ^2 test of independence with the
statistical computation software R.
Results
MLF analysis
The MLF analysis of the simple bilingual clauses from both data sets
yielded the following results which are shown in Figure 6.1. For the Welsh-
English data, 100% of the clauses had Welsh as the ML. (For more discussion
of these results see Davies & Deuchar, 2010.) The Spanish-English data, how-
ever, provided more variation: 66% of the clauses had Spanish as the ML and
the remaining 34% were identified as having an English ML, reflecting the
switch patterns found by Sayahi (2002) for Spanish-English bilinguals in
New York.
Factors Influencing Code-Switching 127Welsh
100%Welsh-EnglishSpanish
66%English
34%Spanish-EnglishFigure 6.1 Matrix Language distribution for Welsh-English and Spanish-English fi nite
bilingual clauses