In October 2015, the government of Chile started a constitution-making process, which allowed citizen participation. During the participatory phase, citizens gathered in local selfconvoked encounters to debate on four dimensions: constitutional values, rights, duties, and institutions. For each one of these four dimensions, participants collectively selected seven concepts from a list provided by the government or added new ones. For each concept, they wrote down a short argument explaining why this concept should be included in the new constitution. Although this process did not result in a new Constitution, the citizen consultation resulted in a valuable and unique source of information about people’s social and political preferences. The first objective of this work is related to the constitutional process itself and the citizen participation. The Chilean process exhibited two critical design weaknesses we analyze here. The first one is representativeness: the voluntary nature of the encounters increased participation biases, as those citizens who support the acting government were more likely to participate in the consultation. We study the determining factors of citizen participation in ELAs by setting up various regression models at the municipality-level. We found that engagement in politics and support for the government increases participation, which suggests that citizen involvement in the constitutional process may have been ideologically driven. The second weakness is the group deliberation quality. For a public deliberation to produce epistemic superiority, all the participants should have access to relevant and accurate information and evidence. Then, we analyze the written arguments for each selected concept, using structural topic modeling and natural language processing. We show that the emergent content can be ideologically differentiated, and that groups from municipalities with higher socioeconomic index, on average, produce higher-quality deliberation compared to groups from less developed municipalities. The second object of study comes from the data. The dataset gathered in the local participatory phase provides a rich source of information about people’s political preferences. To map the political ideology, we built co-occurrence networks where the nodes represent the constitutional concepts, and the links represent the association among them. Then, we aim to discover the structure of the ideology by examining the resulting networks, and identifying clusters - highly connected groups of concepts - inside them. The communities we found are consistent with the political conglomerates existing in Chile in 2016. Finally, using natural language processing techniques, we extracted psycho-linguistic features from the argument texts. These features are “internal factors”, for they respond to the intrinsic psychological, emotional, attitudinal or cognitive state of the subject, which affects their political ideology. Next, we set up a discrete choice model to study the effect of those features in cluster membership. We find that the progressive-left cluster shows a more propositive and non-agentic attitude when referring to values, as opposed to the traditional left. Regarding the dimension of rights, the right-wing cluster displays a more valorative attitude, suggesting that first-generation rights may also play the role of values. Throughout all chapters, and by the methods we use, this work attempts to contribute to the field of computational social science
