arXiv subjects

Carlo Romano Marcello Alessandro Santagiustina

Publications and source records attributed to Carlo Romano Marcello Alessandro Santagiustina.

3 recordsLinked to original sources

arXiv2026

Reproducibility is not construct validity: LLM measurement of institutionally situated communication

High annotation reproducibility does not necessarily imply that an LLM-inferred measure captures the construct it is intended to measure. We test this distinction using a dataset from the European Commission's AI Act consultation, linking structured survey responses to free-text consultation submissions from the same stakeholders. LLM annotations of consultation submissions are highly reproducible (intraclass correlations > 0.99), yet show limited convergence with survey-reported measures of the nominal construct they were intended to approximate. Divergence between survey-and LLM-inferred text-based measures varies systematically across stakeholder groups: business associations express greater concern about AI risks in text-based consultations than in survey responses ({g} = +1.0), whereas public authorities and several nonbusiness groups show smaller or negative divergences. Divergences between scores suggest positive spatial autocorrelation across European countries (Moran's I = 0.347, p = 0.036), indicating that stakeholders from neighboring countries tend toward more similar text-based stances towards AI safety concerns. Despite divergence, survey-reported concerns remain strongly associated with support for explainability across all divergence levels. These results demonstrate that LLM annotation reproducibility can coexist with poor construct correspondence and motivate validation procedures that distinguish reproducibility, construct validity, and communication context variation when LLMs are used as measurement instruments.

cs.AI↗

arXiv2026

Expressing One's Identity Online: Left-Right and cross EU-country variation in self-representation in social media

We examine how social media users from eight European Union (EU) member states express their socio-political identities, focusing on users' online self-presentation and group identity cues conveyed through bios. Our goal is to explore commonalities and differences in topics discussed in social media profiles, across Left-and Right-wing user groups, within and across EU countries. Through a novel approach we map how identity-related discourse varies by country and political orientation, revealing how group identity is expressed within the EU. We find that topics related to democracy, national way of life, and decentralization emerge as particularly divisive, showing considerable variation both within and between EU countries. A subset of topics, which includes education, environmentalism, sustainability, equality, freedom & human rights, and traditional morality, among others, clearly differentiate Left-from Right-leaning user groups. These partisan topics are relevant as they could be leveraged for mobilizing ideological groups and highlight Left-Right identitarian differences at the EU level. Finally, we show that our Left-Right identity similarity metrics reflect aspects of real-world political fragmentation, which are closely aligned to the perceptions of political conflict intensity by country, as measured by the 2022 PEW survey.

cs.SI↗

arXiv2020

Visualizing and comparing distributions with half-disk density strips

We propose a user-friendly graphical tool, the half-disk density strip (HDDS), for visualizing and comparing probability density functions. The HDDS exploits color shading for representing a distribution in an intuitive way. In univariate settings, the half-disk density strip allows to immediately discern the key characteristics of a density, such as symmetry, dispersion, and multi-modality. In the multivariate settings, we define HDDS tables to generalize the concept of contingency tables. It is an array of half-disk density strips, which compactly displays the univariate marginal and conditional densities of a variable of interest, together with the joint and marginal densities of the conditioning variables. Moreover, HDDSs are by construction well suited to easily compare pairs of densities. To highlight the concrete benefits of the proposed methods, we show how to use HDDSs for analyzing income distribution and life-satisfaction, conditionally on continuous and categorical controls, from survey data. The code for implementing HDDS methods is made available through a dedicated R package.

stat.ME↗