In large object-attribute matrices, irrelevant attributes often bias derived conceptual orderings. Attribute weights can be helpful to directly reduce the number of attributes. User-based attribute weighting methods allow the user to express attribute weights themselves in a structured form. A major drawback of user-based attribute weighting is that additional user judgments are required. Therefore, approximating attribute weights from existing object-attribute matrix data (or incidence-based attribute weighting) is an interesting alternative. This study investigates the usefulness of attribute derivations, which are popular measures of incidence-based attribute weighting, for approximating user-based attribute weights. For the analysis, we use an established general domain concept dataset. We observe a correlation between the first-order attribute derivation and user-based attribute weights. However, this correlation only applies to exemplar data with more than 100 concrete attributes and not to category data with significantly fewer abstract attributes. Our analysis includes two well-known methods for user-based attribute weighting: feature importance ratings and feature generation frequencies. Our results show that first-order attribute derivation is the preferred incidence-based attribute weight based on a restricted set of attributes. The application of different sampling methods indicates that these methods differ in terms of their suitability for categories. This limitations of attribute derivations needs to be considered by theories and algorithms for attribute weighting.