Dealing with missing data in a multi-question depression scale: a comparison of imputation methods