Two experiments in one: How accounting for context matters for welfare estimates