Partisan Support for the Welfare State in Québec: A Supervised Text Analysis of Parliamentary Debates (1995–2025)

Quebec stands out within Canadian for its unique welfare state for having developed a distinctive welfare state that incorporates social-democratic elements into an otherwise liberal model. Despite changes in government, these policies have largely been maintained and, in some cases, expanded (Hallé-Rochon et al., 2024). While institutional feedback effects (Béland et al. 2023) and the popularity of these policies (Bélanger and Chassé, 2021; Déry et al. 2024), help explain their enduring nature in the province, little research has investigated how elected officials in Quebec articulate welfare policy preferences in legislative debates. This study investigates how partisan actors discuss social welfare and early childhood policies in Quebec’s National Assembly from 1995 to 2025. Using the Agora+ database of legislative texts, we apply a combination of dictionary-based classification (Topic Lexicoder) and transformer-based models (BERT), trained on annotations generated through a large language model. This hybrid approach aims to assess the validity and efficiency of sequential models for classifying political discourse in large corpora. A human-annotated gold standard is currently in development and will serve as a benchmark for evaluating classification performance. Our findings assess whether legislative discourse aligns with ideological expectations or reveals alternative patterns of welfare state politics. The study contributes to both welfare state research and methodological innovation in text-as-data analysis, showcasing a new application of machine learning to Canadian legislative data.