|
4.3. Multivariable
analysis: discriminant analysis
Descriptive analysis allows us to
identify certain tendencies regarding the relationship between language
of identity and language used with parents but it does not tell
us which of the variables studied provides the best explanation
for the figures. In this section the technique of discriminant analysis
is applied to the data in the study, focusing on those pupils who
use Spanish with both parents. The object of this analysis is to
determine which of the variables studied best explains the language
of identity of pupils who use Spanish with their parents. The pupils
were therefore divided into two groups (dependent variable: those
who identify only with Spanish and those who identify with both
Catalan and Spanish).
Canonical correlation determines
the predictive capacity of the independent variables (oral competence,
social networks, attitudes, confidence and language used with siblings)
included in the analysis. The value for canonical correlation ranges
from 0 to 1. Values of around 0.7 are considered to indicate a good
model for distinguishing between the two groups.
Discriminant analysis of the data
for the Aragonese border area shows that the five sociolinguistic
variables permit quite good discrimination for the two groups of
language of identity (canonical correlation higher than 0.7). A
large number of respondents could not be included in this first
analysis, as many of the pupils left one or more questions unanswered
and discriminant analysis excludes them completely, even if they
have answered all the other questions. Only 22 individuals could
be considered, those who had answered all the questions. To improve
the analysis a neutral value was assigned to missing cases (they
totalled 59). The value assigned was the average obtained from the
other pupils who spoke Spanish with their parents and who had given
a valid response. In this second analysis the value for canonical
correlation was lower (0.548), as had been expected, since the missing
values were replaced by values which neutralised the initial differences
between cases.
The matrix of correlations between
discriminant variables and functions shows that of the five independent
variables two are fundamental in determining the function which
gives the best discrimination between those Spanish speakers who
identify with Spanish and those whio identify with both Spanish
and Catalan: the presence of Catalan in the social network of friends
(0.67) and knowledge of spoken Catalan (0.642). The predictive value
of the other three variables - confidence in Catalan (0.443), attitude
to Catalan (0.427) and language spoken with siblings (0.416) –
is noticeably lower. These three variables can be excluded from
the model with hardly any reduction in prediction of the dependent
variable.
Table 13. Matrix of correlation
between discriminant variables and canonical discriminant functions
for the Aragonese border area.
Independent variables |
Value of the canonical
discriminant functions |
Presence of Catalan in the social network of friends |
0,670 |
Knowledge of Catalan: speaking |
0,642 |
Confidence in Catalan |
0,443 |
Attitude to Catalan |
0,427 |
Presence of Catalan in interaction with siblings |
0,416 |
Discriminant analysis of the data from Mataró shows that
the power of prediction of the five independent variables is not
so great as in the case of the Aragonese border area. In the analysis
without substitution of missing values (141 valid cases) a canonical
correlation of 0.332 is obtained. If the missing values are replaced
by the average for those who speak Spanish with their parents (324
cases), an even lower canonical correlation is obtained (0.258).
Data from the correlation structure matrix for discriminant variables
and functions shows, as in the case of the Aragonese border area,
that it is the presence of Catalan in social networks that best
discriminates between the two groups of pupils who speak Spanish
with their parents (those who identify with Spanish and those who
identify with both Catalan and Spanish). Moderate correlation is
given by confidence and attitude to Catalan. However, oral competence
and the use of Catalan with siblings show relatively little correlation
with the discriminant function and the model is not weakened if
they are excluded.
Table 14. Matrix of correlation between discriminant
variables and canonical discriminant functions for Mataró.
Independent variables |
Value of the canonical
discriminant function |
Presence of Catalan in the social network
of friends |
0,648 |
Confidence and anxiety when using Catalan |
0,644 |
Attitude to Catalan |
0,528 |
Knowledge of Catalan: speaking |
0,485 |
Presence of Catalan in interaction with siblings |
0,154 |
5.
Discussion and conclusions
The results of this study show that in the population
of sixth-year primary pupils in the two areas studied, bilingual
identification (Catalan and Spanish) is much more common than in
the adult population. It also shows that the language which these
pre-adolescents use with their parents does not necessarily coincide
with the language with which they identify. Discriminant analysis
of the population who use Spanish with their parents has revealed
that it is the presence of Catalan in social networks which leads
some pupils to include Catalan as one of their languages of identity,
both in Mataró and in the Aragonese border area. It was also
observed that in the Aragonese border area, as well as social networks,
oral competence in Catalan discriminates the two groups, while in
Mataró discrimination depends on confidence and the attitude
to Catalan.
Two main differences can be observed between the
populations of Mataró and the Aragonese border area. Firstly,
in Mataró a greater number of Spanish speakers report that
they identify with both Catalan and Spanish. In Mataró this
category accounts for 40% of all those who speak Spanish with their
parents (Table 4), while in the Aragonese border area the figure
is just over a quarter (Table 5). In addition to this major difference
we also find two distinct situations regarding oral competence in
Catalan. While in Catalonia self-assessment of competence is very
high in the two groups studied (Spanish as language of identity
vs. both Catalan and Spanish), with very little difference between
them, in the Aragonese border area the figure is high only for those
who identify with both languages. These low levels of competence
for many Spanish speakers in the Aragonese border area may be a
factor impeding greater identification with both languages. However,
as shown by data from Mataró, once high levels of oral competence
are generalised in all groups, this factor does not normally enable
us to discriminate between the two groups studied and the differences
are best explained by the social network and other variables (confidence
and attitude).
The second difference between the two populations studied is precisely
the type of variable that discriminates between the two groups.
While the first variable (social networks) is the same, those which
follow are not. In the Aragonese border area it is competence which
largely explains identification with a single language (Spanish
in the majority of cases) or two. The fact that in the Aragonese
border area competence in Catalan is not so widespread may make
it a differentiating factor for some speakers, affecting whether
they identify with the language or not. In Mataró, on the
other hand, we find that the majority of respondents report that
they have a high level of competence in Catalan (as a result of
the education system). This makes it easy for a larger number of
young people to identify with Catalan. It is, however, possible
that competence in both languages may reach a maximum level, after
which this factor will no longer favour identification with the
two languages, and will therefore cease to be a discriminating factor.
Other factors, such as confidence and attitude, are then likely
to be those which allow us to discriminate between the two groups.
The data thus shows that language competence may be a factor which
explains identification with a language but that, once competence
is guaranteed, there are other factors (less linguistic and more
social) which help to explain the language identification of Spanish
speakers.
|