Many studies have demonstrated that improvements in infrastructure have been effective in reducing inequalities due to poverty. The Brazilian Government has invested a significant amount of resources to improve access to sanitation facilities in the municipalities in Brazil in the last decade. One of these programmes is the Bahia Azul programme of sanitation, which aimed to supply sanitation for all the population in the City of Salvador and the surrounding areas. In this programme, households have to pay the costs of the sewer connected to household excreta disposal to treatment plants. So far, models applied to sanitation studies were either misspecified, presenting serious bias, or did not demonstrate the causal relationship among variables. The objective of this study is to assess the demand for sanitation in Salvador, with focus on determinants of the choice for types of connections. Sanitation was assessed as a function of the objective variables (socioeconomic and demographic, alternative attributes) and non-observed variables, defined in this study as perception and attitude. The Hybrid Choice Model was the theoretical model used in this analysis. A questionnaire was administered to 721 households. The model was estimated using a sequential estimation, associating a latent model (MIMIC) to a mixed logit model. The analysis showed that the inclusion of latent variables in the model increased the magnitude and significance of the estimation of demand. Results indicated that the more educated' and wealthy household tended to choose a system of sanitation. The attributes of usefulness, suitableness, convenience, and healthy, not the latrine and connection themselves, were what the households really cared about. The results of my investigation supported the appropriateness of the Hybrid model for demand evaluation: latent variables incorporated to a discrete choice model improved the explanation of household behaviour, and filled the gap between behavioural theory and discrete choice models applied to sanitation.
