| While the cost of computer hardware has declined over the past decade, the cost of computer software has skyrocketed over this same time frame. Much of the cost associated with computer software has involved the maintenance of programs: program debugging and modification. A primary assumption of the present research was that computer programs which are more readable are also easier to debug and modify, and therefore easier to maintain. Therefore, the goal of this study was to identify a set of program characteristics which would best predict program reliability.; Structured FORTRAN programs produced by 68 novice programmers at the University of Illinois were rated for overall readability by four raters. The average readability rating constituted the criterion to be predicted. Multiple regression equations were generated employing both machine-scored and human-judged predictors. Machine-scored predictors included Halstead's metrics of software science, the cyclomatic number of McCabe, measures of program comments, and others. Human-judged predictors included ratings of the quality of identifiers, indentation, comments, data structures, control structures, and general stylistic features.; Three machine-scored predictors, the cyclomatic number, the number of program comments, and the index of variable concentration, produced a multiple -R of .82 accounting for two-thirds of the variance of readability. Adding the human-judged predictor measuring the quality of indentation significantly increased the multiple -R to over .90, accounting for over four-fifths of the variance in readability.; The four-predictor model chosen fit well with hypothesized aspects of program understanding. The cyclomatic number and the index of variable concentration were deemed "heuristic inhibiting" factors and appeared to be critical for the first stage of comprehension (interpretation). The number of comments, and the quality rating of indentation, were deemed "heuristic facilitating" factors and appeared to be important for the second stage of program comprehension (learning). |
