allowed for a study of the consistency of the indexing of this file. The results showed a bipolar distribution: indexing matched completely almost half of the time and did not match at all almost half of the time. The indexing policies of ISA require one mainheading and one or two subheadings per document. This restriction in the number of terms and the fact that ISA has a very small vocabulary from which to draw these terms may be the reason for this bipolar distribution. The indexing consistency was highest for the descriptors, drawn from a small controlled vocabulary, and lowest for identifiers, drawn from natural language or the controlled vocabulary. The data suggested that as the number of terms assigned per article increased indexing consistency decreased.
Concern with indexing consistency
is not new. Given the vast complexity of natural languages and the many connotations of the words in those languages, absolute precision in indexing is probably impossible, even with controlled vocabularies.
Studies have been done to measure how consistently a given indexer or group of indexers will index a set of documents (intraindexer consistency) and how consistently a group of indexers will index the same document or set of documents (interindexer consistency).
Most research on indexing consistency has been experimental, not examining operational systems. Funk et al. (1983), however, studied the phenomenon in MEDLINE and found results similar to those of the experimental studies. Further exploration of operational systems, therefore, could be valuable.
In this research a subject-specific or subject-restricted document file, Information Science Abstracts (hereafter referred to as ISA) was searched for documents entered and indexed twice. Like Funk's, this research focused on an operational system rather than an experimental situation. Several questions were examined: