Experimental Analysis of Textual and Graphical Representations for Software Architecture Design

Abstract

Software architecture design documentation should communicate design decisions effectively. However, little is known about the way recipients respond to the different types of media used in documentation. We therefore conducted a controlled experiment to study whether visual or textual artifacts are more effective in communicating architecture software design decisions to software developers. Our participant group consisted of 47 participants from both industry and academia. Our results show that neither diagrams nor textual descriptions proved to be significantly more efficient in terms of communicating software architecture design decisions. Remarkably, participants who predominantly used text, scored significantly better; overall and with respect to topology related questions. Furthermore, surprisingly, diagrams were not able to alleviate the difficulties participants with a native language other than English had in extracting information from the documentation. In combination, these findings at the very least question the role of diagrams in software architecture documentation.

An empirical study on the efficiency of graphical vs. textual representations in requirements comprehension

“…Recently, Heijstek [2] et al reported findings pointing to the fact that neither textual nor graphical representations were significantly effective for understanding a software architecture. They also reported that the more experienced subjects, mostly, preferred textual representations.…”

“…Heijstek et al [2] adopted this measure to investigate whether the difference between their subjects' native language and English can impact their efficiency and accuracy while understanding a model. In this paper, we are also interested to investigate if the distance of our subjects' native language from English can impact our findings.…”

“…Graphical representations (e.g., UML diagrams) are effective tools to (1) promote a quick understanding of data, (2) facilitate data processing and data comparison, and (3) enhance the communication process between end-users and developers.…”

“…Moreover, graphical representations do not heavily dependent on natural languages, thus they could mitigate the language barrier problem [2]. Conventional wisdom assumes that graphical representations carry information more effectively to nontechnical people than textual one [1].…”

“…Based on all these reasons, it can be expected that developers prefer to use graphical representations in contrast with structured textual ones to understand the software under study. Yet, to the best of our knowledge, only a handful of studies investigated the effectiveness of graphical vs. textual representations or the developers' preferences (i.e., textual vs. graphical representations) in program understanding tasks [2], [3], [4].…”