**2. Methods**

To display the structure of secessionist vs. unionist preferences among Catalonian citizenry we started by using the official electoral data from the last regional election (21 December 2017). The geographical distribution of the pro-secession vote was examined by adding the results of the three secessionist political parties: JxCat, ERC and CUP. We obtained a map (Figure 1) with 947 current Catalan municipalities coloured in distinctive tones: In red when the sum of the secessionist vote was greater than 50% of the municipal census and in pale pink, otherwise. Secondly, we obtained a map with the 947 current Catalan municipalities coloured in five green hues according to percentages of unionism votes on the electoral census of each municipality (Figure 2). These results are also summarized in Table 1.

**Figure 1.** Geographic distribution of secessionism at regional elections 21 December 2017. The percentages by municipalities are computed over their corresponding electoral census. Secessionism is majoritarian (in red) at 76% of the municipalities, representing 78% of the surface of the whole region, but only 20% of the electoral census live in this area, while the remaining 80% live in the rest of Catalonia (in rose pale). Secessionism is concentrated at inland counties mainly, whereas unionism predominates on overpopulated coastal areas and in some Pyrenean and peripheral counties as well.

**Figure 2.** Geographic distribution of non-secessionist vote at regional elections 21 December 2017. The percentages by municipalities are computed over all their corresponding electoral census. The different shades of green correspond to the different percentages of unionist forces at each municipality. Observe the pattern complementary to that observed in Figure 3.

**Figure 3.** Catalonian municipalities sorted by the value of an index measuring "unionism" degree. This is an index based on a factor analysis over electoral data in the period 2008–2012. For each abscissa point, we show the accumulated electoral census to the left of that point, from 0% to 100% in green. We show also a smoothed average (in red) of the corresponding secessionism at each municipality. Note that the highest values appear at the smallest municipalities, while they progressively fall at the largest municipalities in the Barcelona metropolitan area. The Barcelona municipality had intermediate behavior.


**Table 1.** Secessionists and non-secessionists in different geographic areas.

We built then a way to visualize the distribution of the pro-secession vote across municipalities of the region. They were ordered according the ranks of an index obtained considering municipal, regional and general election results in the period 2008–2012 and a factorial analysis. Such an index can be interpreted as a measure of "unionism". Then we plotted simultaneously the cumulative electoral census, from 0% to 100% and a smooth moving average of percentages of secession support at each municipality (Figure 3). The shape of both plots shows the relationship between municipality sizes and support for secession. Further details could be supplied by the authors to any interested reader.

We then proceeded to analyze the evolution of several measures obtained by consecutive CEO barometers (the official survey agency of the Regional Government), during the period 2006–2019, from representative samples of Catalonian citizenry through personal interviews. Sample sizes for each survey oscillated between 2500 and 1500 citizens, with the exception of autumn 2017, which had only 1338 citizens. This longitudinal analysis was based on responses of 88,538 individuals through a regular series of 45 official surveys, in the period 2006–2019. Our variables came from particular questions on these CEO barometers. Specifically, our main economic variable was "household net incomes" for the whole population and its interaction with the main family/mother language segments (Catalan vs. Spanish). We estimated medians of "household net incomes" of segments defined by the second one. We also used a binary variable "household net incomes ≥3000 €/month", with two possible results "Yes" or "No", discarding the relatively few answers "do not know" or "do not answer" ("DK/NA"). We also used the qualitative variables "economic resistance limit in case of economic breakdown" (based on questions asked in two surveys in 2016 and 2017), and "perception of own economy last year", both in combination with support for secession in a (hypothetical) referendum of self-determination (as is specifically asked in these CEO barometers, but only from 2015 onwards).

## *Longitudinal Variations on National Identity Feelings Depending on "Household Net Incomes" and Linguistic Segments*

We focused our analyses on the evolution of national identity (sense of belonging feelings) because these CEO barometers maintained a specific question on this issue, without changes, along the whole period. It had six options: "only Catalan", "more Catalan than Spanish", "as Catalan as Spanish", "more Spanish than Catalan", "only Spanish", and DK/NA. Previous work had established substantial covariation between national identity feelings and preferences in favour of or against secession, despite addressing different features (Oller et al. 2019a, 2019b). We considered the variable family/mother language and the dichotomous variable net income plus. First, we plotted the evolution of the national identity "only Catalan" on the four groups obtained crossing the main two segments of family/mother language (Catalan vs. Spanish) with the dichotomous variable net income plus (with also two levels, according to whether the household net incomes were greater than, equal to, or less than 3000 €/month). The graphics allowed comparisons of different evolution for these groups. Secondly, we obtained the same graphs for the national identity "as Catalan as Spanish" on the same groups, and finally we did the same with the national identity "only Spanish".

We additionally analysed this data, just as an approximation, through the perspective of Analysis of Covariance (ANCOVA). For the sake of simplicity, we started with two ordinary univariate analyses and continued with a combination of them in a trivariate version of MANCOVA that we shall describe hereafter. At the start, we considered two factors for both analyses the levels of which were the possible values of the qualitative variable family/mother language (with two levels corresponding to the main linguistic segments, Catalan and Spanish) and the dichotomous variable net income plus (according to whether the household net income was greater than, equal to, or less than 3000 €/month), using time (from 2006) as a covariate and considering as a dependent variable the percentages of those who self-identified as "only Catalan", for the first analysis, "as Catalan as Spanish", for the second one, and "only Spanish" for the third one. Since the interactions between factors were highly significant, to facilitate the analysis we built a new factor, group, with four levels, combining the two levels of family/mother language factor and both levels of the factor net income plus. We used this factor to replace the previously mentioned original binary factors at the final analysis.

Notice that there were two sources of randomness: One corresponding to the sample procedure and another corresponding to the political and communicative events throughout the period: We used a linear model as a simple way to deal with both sources of variability, just as an approximation, introducing time (in years, from 2006) to try to capture potential trends and checking the global adequacy of this approach examining the standard output supplied by the function lm of R package statistics. We supply also a graphic plot illustrating the dependence of each percentage ("only Catalan" and "as Catalan as Spanish"), with respect to the time covariate, in each one of the four levels determined by both factors.

As a final remark, in most of the plots, we marked relevant historical events that might have been crucial to understanding the evolution of the variables along the period. These events were: The date when a new home rule was approved (New Statute 2006); the resolution of the Spanish High Court (Tribunal Constitutional-TC) that sanctioned 14 articles (over 223) as contrary to the Spanish constitution and restricted the preamble and another 27 articles (June 2010); the peak protests of the social 15M movement (15M Peak Protests, June 2011); the regional elections of 25 November 2012 (25N); the illegal consultation about independence of 9 November 2014 (9N); the regional elections of 27 September 2015, (27S); the illegal referendum about secession, 1 October 2017 (1 Oct) and the regional elections 21 December 2017 (21D).

We were fully aware of the limits and restrictions we adopted in our analyses of an obviously multi-causal phenomenon. We limited ourselves to studying significant stochastic dependencies between variables, a strictly statistical work (descriptive plus correlational), although it is true that, in this context, high stochastic associations might sugges<sup>t</sup> plausible explanations, of at least part of the mechanisms that shaped the observed trends.
