A study of errors, error-proneness, and error diagnosis in Cobol
This paper provides data on Cobol error frequency for correction of errors in student-oriented compilers, improvement of teaching, and changes in programming language. Cobol was studied because of economic importance, widespread usage, possible error-inducing design, and lack of research. The types of errors were identified in a pilot study; then, using the 132 error types found, 1,777 errors were classified in 1,400 runs of 73 Cobol students. Error density was high: 20 percent of the types contained 80 percent of the total frequency, which implies high potential effectiveness for software-based correction of Cobol. Surprisingly, only four high-frequency errors were error-prone, which implies minimal error inducing design. 80 percent of Cobol misspellings were classifiable in the four error categories of previous researchers, which implies that Cobol misspellings are correctable by existent algorithms. Reserved word usage was not error-prone, which implies minimal interference with usage of reserved words. Over 80 percent of error diagnosis was found to be inaccurate. Such feedback is not optimal for users, particularly for the learning user of Cobol.