Recent advances in the collective model literature suggest ways to estimate the complete allocation of resources within households, using assignable goods and assuming adult preference similarity across demographic groups (or across spouses). While it makes welfare analysis at the individual level possible, the predictive power of the model is unknown. We propose the first validation of this approach, exploiting a unique dataset from Bangladesh in which the detailed expenditure on private goods by each family member is collected. Individualized expenditure allows us to test the identifying assumptions and to derive 'observed' resource sharing within families, which can be compared to the resource allocation predicted by the model. Sharing between parents and children is well predicted on average while the model detects key aspects like the extent of pro-boy discrimination. Results overall depend on the identifying good: clothing provides the best fit compared to other goods as it best validates the preference-similarity assumption. The model leads to accurate measures of child and adult poverty, indicating the size and direction of the mistakes made when using the traditional approach based on per adult equivalent expenditure (i.e. ignoring within-household inequality). This assessment of existing approaches to measure individual inequality and poverty is crucial for both academic and policy circles and militates in favor of a systematic use of collective models for welfare analyses.