We non-parametrically test a general collective consumption model with public consumption and externalities inside the household. We further propose a novel approach to model special cases of the general collective model. These special cases include alternative restrictions on the 'sharing rule' that applies to each household, and which defines the distribution of the household budget over the household members. A limiting case is the unitary model. Our application uses data from the Russia Longitudinal Monitoring Survey (RLMS); the panel structure of this data set allows non-parametric testing of the behavioral models without relying on preference homogeneity assumptions across similar individuals. This application includes test results but also a power analysis for different specifications of the collective consumption model. Our main findings are that the most general collective model, together with a large class of special but still fairly general cases, cannot be rejected by the data, while other, restricted, versions of the general model, including the unitary alternative, are rejected. Since these tests are entirely non-parametric, this provides strong evidence in favor of models focusing on intra-household decision-making.