A comparison of models for clustered binary outcomes: analysis of a designed immunology experiment