En el Gràfic 3 es pot comprovar com els fills de parelles catalanoparlants (primera columna) utilitzen el català amb els seus amics, mentre que els fills de parelles castellanoparlants (tercera columna) utilitzen el castellà. Tot i això, hi ha un percentatge d’alumnes força visible en els dos grups que no només parla la llengua dels seus pares, si no que també utilitza l’altra llengua.

Graph 3. Use of Catalan with friends from school according to the language spoken between the parents when the pupil was small

Note: The box graph shows the dispersion of the values of the dependent variable. Each box consists of a central horizontal line indicating where the median value is situated (2nd quartile, at 50%), two red areas (a higher and a lower one) which, in their limits indicate where the value of the 1st and 2nd quartiles are (at 25% and 75% of cases respectively) and, finally, a vertical line (whisker) on each side indicating the distance of 95% of cases. The external points indicate extreme cases.

The multi-variant analysis we present below attempts to answer the question of which factors are most related to the maintenance or change of family language in order to talk to friends? Below we present, firstly, the main elements interacting with linguistic usages with friends, to place the results in a general context. Afterwards, we deal with the pupils according to their initial language in order to go into depth on the different language usages within the Catalan-speaking group and, then, the different usages within the Spanish-speaking group. This will help us to understand the most important circumstances defining usages tending towards the use of the parents' language or the incorporation of others.

3.2. Multivariable analysis of the language spoken with friends

In models with very few variables, a high capacity for predicting linguistic usages with friends (from school) has been obtained. The language spoken with brothers and sisters is the best situated variable and, with this one alone, an explanation of 49.9% of the variation in language usages with friends is obtained. The logic is simple, as most pupils who speak Catalan with their brothers and sisters also use only Catalan with their friends (51.8%). Moreover, 71% of those speaking Spanish with all their brothers and sisters also speak only Spanish with their friends. With this important correspondence between one variable and another, a very robust model is formed from the very first variable.

But, in order to form a complete model with the variables needed in order to predict usage in the best possible way, we select a total of as many as 13 variables. See them here Table 6. With these, the model is adjusted until 64.8% of the existing variance is explained. The variables can clearly be grouped into three sets. The first set, as we saw above, is made up of the socio-linguistic variables of family origin (presence of Catalan in interactions with brothers and sisters and parents), establishing the most important nucleus in order to explain the presence of Catalan in interactions with friends. Secondly, the school variables in informal uses and socio-linguistic representations take over: more informal linguistic uses with teachers (outside class) and the linguistic reference group ("I'd like to belong to the group of those who speak...") reinforce the role of family usages. And finally, in third and last place, the analysis is completed with a total of nine additional and miscellaneous variables which obtain much lower correlation indices. These include linguistic competences, language usages with the teacher in class, linguistic representations, consumption of communications media in Catalan or family geographical origin. As we see, these are very diverse variables which help to complete a good approximation of language usages with friends.

Table 6. Linear regression model. Dependent variable: presence of Catalan in language spoken with friends from school