This thesis consists of three essays that improve the general understanding of the public demand for safety programmes in the context of natural hazards. With the growing importance of this topic around the world, this study provides a practical and methodological contribution to the literature on Environmental Economics and Policy, especially local policy. In particular, this research examines people’s preferences and their willingness-to-pay for landslide mitigation programmes. The primary aim is to assess how the residents of and visitors to a mountain valley in the Alps value and trade off the multiple attributes of protection programmes for landslide risk reduction by applying Discrete Choice Modelling methodology. To address the current needs of local decision-makers, the investigation of the determinants of preference heterogeneity is the central theme of the research. The study is based on a panel choice dataset created from a Discrete Choice Experiment, based on full ranking, administered in person by the author to 250 respondents in the Boite Valley, Italy. The first essay examines the stability of preferences, investigating to what extent additional information has an impact on estimated values. Specifically, it studies whether respondents adjust their preferences based on scientific information provided on one specific attribute. A mixed logit model in willingness-to-pay space is implemented to account for preference heterogeneity. The findings suggest that respondents perceive the existing protection measures as insufficient. The provision of information affects only the attribute subject to additional information and the consideration of the current status of protection. Preferences for the other attributes remained stable. Preliminary evidence of spatial heterogeneity is also detected. The second essay addresses the issue of the stability of parameter estimates obtained through simulation using choice models with latent variables. Specifically, it analyses the stability of the coefficients to the number of simulation draws and the increasing number of latent variables. Three Random Parameter logit models with respectively one, two and three latent variables are fitted with six sets of increased numbers of draws. The landslide risk perceptions of respondents are modelled as latent sources of heterogeneity in the consideration of the riskiest scenario. Overall, the results show very stable estimates for the attributes’ coefficients but not for the latent variables. Thus, increasing the complexity by adding more latent factors into the model implies the necessity of additional draws in the simulation process to ensure empirical identification. The results also show how preferences are strongly related to the underlying perceptions of own mortality risk due to landslides and risk severity. The third essay explores multiple sources of preference heterogeneity, accounting for its spatial determinants. It emerges that the inclusion of more observables allows for a better segmentation of the policy based on respondents’ and municipalities’ characteristics. The findings show the importance of distinct spatial effects, such as geographical characteristics, spatial error components for road tracts and site-specific choice-sets, with relevant insights into the priority of intervention. In addition, residual unobserved heterogeneity is analysed at a higher hierarchical scale using spatial models at the municipality level. Overall, the empirical results of this thesis provide important policy implications for local decision-makers in charge of public safety, given the relevant information on the distributional effects of protection across different groups of beneficiaries.