Abstract
The capability of superimposed random coding to support queries against large databases with sparsely populated search attributes is investigated with particular emphasis on the balance of false and correct drops, carefully distinguishing coding strategies that are often mixed up. A comparison with inverted files is made, showing that superposition is competitive for queries involving more than a very small number of terms if the correct random code generation strategy is used.
Acknowledgements
The author acknowledges many helpful suggestions provided by the referees.