Survey reports of zero expenditure result from either genuine non-consumption, or purchases undertaken too infrequently to observe during a survey, with hidden consumption from stocks. Infrequent purchase models rely on untested hypotheses to distinguish these types of zeros. We test such models with data from an unusual survey where food stocks are measured at the start and end of the survey reference period. Parameter estimates using these direct measures of hidden consumption out of stocks are compared with estimates from infrequent purchase models that attempt to recover this hidden consumption. The results suggest considerable bias when using the infrequent purchase models.