On the impact of size to the understanding of UML diagrams (Introduction…
On the impact of size to the understanding of UML diagrams
Visual languages are superior to textual languages in that they support human perceptual and thought processes, also true for UML.
No compelling research results to support it.
No general research of layout of UML diagram.
The paper's working hypothesis is that modeler performance correlates negatively with diagram size.
Well-known rules of good layout
Four levels of design principles governing the layout of UML : Aesthetic criteria for the layout of UML diagrams and its effects on model understanding
principles of graphical design and visualization.
layout principles applying to all structures that can be considered a graph in the mathematical sense.
layout principles that apply mostly only to notations like those found in UML.
the level of pragmatics supports for underlining the purpose of a diagram in order to better address the audience.
Diagram layout factors
There is no single layout algorithm and set of calibration factors across visual notations.
If a diagram layout does not (significantly) violate any of the rules on the first two levels but (more or less) adopts the rules described in the latter two levels then it's a good layout.
Defining the size of UML diagrams
Metric 1: Number of diagram elements (NODE)
count every element of a diagram that has discernible visual features
Regarding bended or multi-legged lines, count each segment individually, including integral parts and adornments such as arrow heads, attached names, or stereotypes required for element disambiguation.
count textual labels as additional individual elements if and only if they are not a necessary part of an element.
Metric 2: Weighted number of diagram elements (WNODE)
Decomposition of complex graphemes
Relative weights of graphical primitives
Metric 3: Adjusted number of diagram elements (ANODE)
three disjoint populations of students at different levels of expertise
students were given a set of sheets where each sheet contained a UML diagram, ten questions about the model visualized by the diagram, and questions to assess the difficulty and clarity of the diagram, personal preference, and subjective assessment of layout quality
created by students as part of their coursework in a requirements engineering course
the model of the team achieving the highest grade was selected
very similar levels of modeler capability and model quality may be assumed
the models are realistic wrt
Correlations between diagram size and modeler performance
there is a large correlation between increasing diagram size and decreasing mean scores.
also see a positive correlation between diagram size and layout quality
might see here a weakness of our experimental design
participants intuitively understand layout quality relative to the diagram size
Expertise level effect the correlations
Participants perceived larger diagrams as more difficult to understanding than smaller diagrams
Threats to validity
Internal validity : reduced the danger of introducing bias through the experimental procedure by using several alternative measurements for each variable, thus also corroborating our observations and reducing the implications of poor readings, outliers, and noise.
External validity : applied objective and rational criteria to the selection
Conclusion validity : used nonparametric tests. most of the inferences we present are equipped with high or very high levels of statistic significance and large effect sizes, using Cohn’s thresholds for the effect size levels for want of any better guideline.
Construct validity : developed three different plausible metrics of increasing complexity and have opportunistically adopted the simplest of these metrics.
Implementation and validation
Implemented the DiagramMetrics tool as part of the MagicWand toolset that integrates with the MagicDraw UML CASE tool, one of the leading commercial tools for UML-based modeling.
used them in a modeling class for undergraduate students
collected the students’ experience as part of the general course feedback sessions.
Summary : the results suggest that the number of diagram elements is a useful metric for diagram size.
Findings : high layout quality is particularly helpful for large diagrams
Relevance : it is not just worthwhile to provide evidence also for the seemingly obvious, but in fact indispensable to seek this evidence
Validity : the findings have a high level of validity