Input errors were analyzed in a data base of a Japanese academic journal, Japan Analyst (Bunseki Kagaku). The data base contained, among other data, the text ef abstracts of all the articles. Frequencies of words in the abstracts were counted and all the words with one occurrence in a year's data were listed. The list was scanned for possible misspellings due to the original text and due to the coding and card-punching operations. The error occurrences were classified according to their symptoms: 1. missing character (e.g. ABSTRCT, 53% of the total), 2. wrong character (e.g. AVSTRUCT, 27%), 3. excessive character (e.g. ABSTTRACT, 15%), 4. vocabulary error due mostly to inaccurate knowledge of English (e.g. MESSFLASK, 5%). The origin of each error was traced and classified into errors in the original publication and errors caused by the coding and the punching operations. Measures for reducing such errors were discussed.
View full abstract