Mode-switch protocols: how a seemingly small design difference can affect attrition rates and attrition bias