"Data was collected from primary and bibliographic sources (see details in Table 1). Analysis was developed using R statistical language [49] and Gephi software [50] (details specified below). Script and data are included in the supplementary material to promote a fully reproducible study."

"Once the adjacency matrices were constructed, the second phase, which covers the network construction, was developed with Gephi [50]. In this case, because this study does not consider the direction of the interaction, we have built an undirected network for every time-bin.

As part of analysing the obtained networks, we calculated several structural measures to determine the diachronic evolution. Fig 3 presents the calculated metrics for describing the evolution of the social network. By calculating these metrics, we intend to decode the evolution of the network at different scales, namely, macroscale (structure), which includes those metrics that allow us to know the general structure of the network, and microscale (node level), which consists of those metrics that reveal the role played by the different nodes in the network of which it is a part."

"The intention is to test whether our observed networks have the properties of a small-world network. Thus, following Buchanan’s [23] approach, the clustering coefficient from the original networks should be higher than the random values; however, in the case of APL, the comparison is more complex because even though small-world have a low APL, it is not enough to compare original APL values with random APL values since random ones typically have shorter APL than real networks [23]. Therefore, the original network needs to have a pretty low APL to be lower than the average random APL. Likewise, we could not discard small-world in cases when the original APL were higher than random."

"To avoid the previous situation, we follow Buchanan et al., [23], who incorporate in the small-world test protocol some additional calculations, including the coefficient σ from Watts and Strogatz [70]. The small-world coefficient σ is the ratio of the clustering coefficient of the original network and an equivalent random network divided by the ratio of the original APL to the APL of the equivalent random network. Here, we previously calculated such measures for the 1000 iterated random network, so we use the mean for both the clustering coefficient and the APL random of all this created network as the “equivalent” random measure. Except for the small-world test protocol, developed with R studio, we ran the different study steps with Gephi. Code for this and other steps (temporal samples construction) are available in the supplementary."