Masking data: a solution to social desirability bias in paired comparison experiments