On Efficient and Statistical Quality Estimation for Data Annotation