Beyond individual attitudes: Modelling deliberation on shared rooftop photovoltaics in Czech and Polish housing cooperatives

This preprint investigates how residents of housing cooperatives form and revise their attitudes towards shared rooftop photovoltaic (PV) investments during collective decision-making. While research on PV adoption has largely focused on individual homeowners, investments in multi-family buildings require collective agreement among residents with different financial circumstances, environmental values, levels of institutional trust, and opportunities to participate. The study introduces a large language model (LLM)-assisted agent-based modelling framework informed by survey data collected in Czechia and Poland within the ENBLOC project. Using two illustrative housing cooperative scenarios with 30 semi-synthetic resident agents each. The framework simulates successive stages of decision-making, from initial attitudes towards PV through institutional assessments and project notification to residents' responses during cooperative members' meetings. The analysis combines quantitative attitude scoring, qualitative analysis of generated reasoning, and reliability assessments across repeated simulation runs. The findings illustrate how initially favourable attitudes towards PV can become more conditional when residents confront specific investment costs, financial risks, and collective obligations. The simulations highlight the importance of repair-reserve financing, environmental commitments, institutional trust, leadership signals, and participation opportunities. Opposing arguments produced stronger downward shifts in support than supporting arguments produced upward shifts in the reported scenarios. The framework offers an exploratory approach to investigating the social and institutional dimensions of collective energy investments and generating hypotheses for further research. Its findings should not be interpreted as predictions of actual cooperative decisions: the scenarios are illustrative, formal voting was not simulated, and generated reasoning requires validation against observed human behaviour. Preprint status: This manuscript has been submitted for consideration for publication in Energy, Sustainability and Society. It has not yet undergone peer review and may be revised during the editorial process.