This study explores the impact of vocabulary scheme arrangement on the quality of author-generated metadata, specifically specificity and frequency of vocabulary terms chosen from schemes to describe websites. By evaluating vocabulary assigned using hierarchical and flat schemes, and by comparing these evaluations, this study seeks to isolate the arrangement of the scheme used from other variables, such as skill level and intentions of metadata generators, which have been the focus of previous research into the viability of author-generated metadata. This study suggests a relationship between term specificity and scheme arrangement, and possible relationships between term frequency and scheme arrangement, and submits that it is therefore possible that non-professional status, lack of skills, or intentions to misrepresent web page content via metadata are not the sole contributing factors to quality of author-generated metadata. New methods for researching metadata quality are tested and their validity discussed.