FACTORS AFFECTING CODING ERRORS
Abstract
This Memorandum describes experiments which sought to identify the factors that contribute to coding errors. The experiments used code-stimulus materials: numeric codes, consisting only of numbers; alpha codes, consisting only of letters; alpha-numeric codes of mixed letters and numbers; and mnemonic codes (natural abbreviations, such as 'OVH' for 'overheated'). Only three-character codes were used. Major findings were: (1) Coding errors are proportional to the alpha content. (2) Errors committed with mixed codes reveal a position effect. (3) Perceptual set can increase or reduce error rates. (4) There was evidence of interaction (in the statistical sense) among the first, second, and third variables. (5) Unless perceptual set is specifically controlled most errors in recording mixed codes consist of 'substitutions of opposites'. (6) Over half the erroneous substitutions appeared to be non- random. (7) Nonrandom substitutions tended to be unidirectional. (8) Most coding errors (75 to 95 per cent) result from having a single digit in error. (9) The use of mnemonic codes did not reduce coding errors. (10) Letter-pattern familiarity affects coding errors. (11) Coding errors can be substantially reduced by providing keypunchers with a list of codes. (12) The amount of usable information that is retrievable from coded information depends on three factors: (a) the error rate; (b) the number of codes used; and (c) the number of codes possible with the format in question.
Document Details
- Document Type
- Technical Report
- Publication Date
- Apr 01, 1965
- Accession Number
- AD0614415
Entities
People
- Anders Sweetland
- Sidney Owsowtiz
Organizations
- RAND Corporation