Modeling partial knowledge in multiple-choice tests using elimination scoring