Statistical Prediction of Programming Errors.

Abstract
This report presents and discusses the results obtained for statistical predictions of programming errors using multiple linear regression analysis. Programming errors were predicted from linear combinations of program characteristics and programmer variables. Each of the program characteristics variables were considered to be measures of the program's complexity and structure. Two distinct data samples comprising 783 programs with approximately 297,000 source instructions written for command and control software applications were analyzed. Background data on both samples is provided which includes discussions related to each sample's software development environment, testing conditions, predictor variables, definition of programming errors, and general data characteristics. Results are presented which give the prediction equations obtained and a discussion of the predictability of errors and error rate in each sample. Conclusion of the study and recommendations for further research are also provided. (Author)

This publication has 0 references indexed in Scilit: