The Effects of Identifier Characteristics on Software Product Quality

Abstract

Background: Software projects consist of many identifiers, and how those identifiers should be named to improve software product quality is an opinionated subject. Some studies have tried to find the relation between identifiers characteristics and software product quality, but these studies have been conducted on small sets of software projects (N 12). Consequently, a need exists of large scale studies that investigate how identifier characteristics effect and can be used as predictors of software product quality.

Aim: This study evaluates if source file averages of the identifier characteristics length, number of containing words, number of containing digits, and casing consistency can predict the software product quality of source files. Method: With identifier and software product quality data found in 60,315 source files from 1,000 open-source Java software repositories, linear regression models are fitted on the data. The models can then be evaluated in terms of accuracy and coefficient importance to find if identifier characteristics could be used as predictors of software product quality.

Results: Bayesian linear regression models with identifier characteristics and the size of source files as independent variables had R-squared accuracies ranging from 0.008-0.545 of predicting the software product quality of source files. But, none of the models’ identifier characteristics were accepted as important for the change of the prediction based on the ROPE + HDI decision rule with a [-0.1,0.1] ROPE interval.

Conclusions: The identifier characteristics investigated in the study could not be used to predict the study’s measurement of software product quality. Further work is needed in investigating the same identifier characteristics relation to other measures of software product quality and with source files written in other programming languages than Java.