Data visualizations and information dashboards are useful but complex tools. They must be fully understood to draw proper insights and to avoid misleading conclusions. However, several elements and factors are involved in this domain, which makes it difficult to learn. In previous works, we proposed a meta-model to capture the primitive elements that compose visualizations and dashboards. This meta-model has served as a framework for conducting data visualization research, but also to develop a graphical tool for generating data visualizations and dashboards. This tool (namely MetaViz) enables users to create data visualizations through fine-grained components based on the entities represented in the meta-model. The main goal of the system is to provide a learning experience in which users can freely add and configure elements to understand how they influence the final display. This work describes work in progress to validate the pedagogical value of MetaViz in terms of the understanding of data visualization concepts.
Conveying information effectively has become a challenge over the years; the increase of data in terms of quantity and complexity has hampered their analysis and communication to different audiences. However, along with the increase in quantity and complexity, there is also an increase in relevance of data within informed decision-making processes [1]. Using a data-driven philosophy to make decisions allows better communication, measurability, accountability and more objective approaches to tackle new problems [2].
Visual tools such as data visualizations and information dashboards are being widely used to address the extraction of knowledge from raw data [3]. Although these mechanisms provide means to convey data easier [4-7], audiences must understand the employed visual metaphors [8]. Many factors are involved at this point: potential biases [9], beliefs [10-12], and even -purposely or not- misleading designs [13, 14].
Regarding the latter, several platforms have been developed to facilitate the process of designing and implementing data visualizations. Systems like Tableau (https://www.tableau.com/), Microsoft Excel (https://www.microsoft.com/microsoft-365/excel), Power BI (https://powerbi.microsoft.com/), etc., provide graphical interfaces that allow users with no experience in programming to create data visualizations and even assist them in the design process to choose the best encodings. However, it is crucial to understand and account for every element involved in understanding data visualizations to deliver effective and clear displays of information [5, 6, 15, 16].
In this context, we propose MetaViz, a new system designed and implemented through a modeldriven approach. A dashboard meta-model was developed in previous works [17-21] to provide a decomposition of the dashboards and data visualizations domain and to fuel the model-driven development.
The meta-model represents the most primitive elements of data visualizations (tangible and conceptual entities) and how they influence one another. In this sense, MetaViz takes advantage of this artifact and provides a graphical interface to generate data visualizations and dashboards through the composition of basic shapes and configurations. The aims of this research are two-fold: 1) to generate data visualizations through a usable interface and 2) to provide a learning experience in which users can learn how different configurations and visual elements influence how we see data and our reached insights.
This paper describes a work-in-progress for a pilot study to measure the understanding of the elements involved in the design of data visualizations and information dashboards. To do so, we designed a questionnaire focused on determining the level of awareness of users of Tableau and MetaViz in terms of the visual metaphors and configurations they employed to create a data visualization.
The rest of this work is organized as follows. Section 2 outlines the dashboard meta-model and the MetaViz platform. Section 3 details the study procedure and the developed questionnaire. Finally, section 4 discusses the impressions of the pilot study.
The foundation of this work is a meta-model which has been subject to several improvements and modifications through domain engineering [22, 23] to capture the most relevant elements and factors in the domain of data visualizations and information dashboards. The current version of the dashboard meta-model is shown in Fig 1. The user [21], the layout and visual components [20] are represented along with concepts related to data, such as its domain and context [17].
Starting from this meta-model and following the model-driven development [24, 25] and software product lines [26, 27] approaches, we implemented a graphical meta-model instantiator, i.e., a platform that allows users to create models following the meta-model abstract entities. MetaViz also enables the generation of functional, real-world visualizations following the instantiated models’ specifications. MetaViz’s interface is shown in Fig 2. The platform is available at https://metaviz.grial.eu/.
MetaViz is focused on developing factual, conceptual, and procedural visualization knowledge [28]. The study focuses on the learning experience that the MetaViz system offers. MetaViz’s strengths are flexibility and the fine-grained configuration of every element. In this sense, we want to test if the system’s interface and architecture improve attention during the design process of data visualizations.
While powerful tools like Tableau offer assisted and straightforward implementation of data visualizations, MetaViz forces the user to be aware of the elements and underlying mappings they define when creating a visualization.
For this matter, we designed a procedure and a questionnaire that aims at measuring two cognitive dimensions following Bloom’s taxonomy of educational objectives [28-30]: • Remember - Identification of the elements that compose a data visualization at first sight, i.e., to test if the student can recognize and recall basic elements of data visualizations. • Understand – Understandability of the data visualization design process, i.e., to test if the student is aware of the dimensions and elements involved in the display.
The study procedure consists of two similar parts. The first part involves a widely used data visualization tool, in this case, Tableau. Users are asked to download a test dataset and create a scatter plot that shows data values from two numerical variables and one categorical variable with Tableau. When finished, users close Tableau and answer the following questions about the scatter plot they just created: 1. Which variable was represented in the X-axis? 2. What was the maximum value of the Y-axis scale? 3. How many visual encodings were employed in your visualization? 4. How many scales were involved in your visualization?
This set of questions are focused on testing if users were aware of the design process of their own data visualizations and if they remembered basic features of their charts. Textual and graphical indications about the meaning of the data visualization terms involved in the questions are provided to avoid confusion. Explanations of the meaning of “scale” and “visual channel/encoding” are shown in Fig. 3 and Fig. 4, respectively.
The remaining questions in this part test the ability of users to identify the elements that compose an already implemented data visualization. Different screenshots of data visualizations created in Tableau are displayed, following the next questions: 1. How many variables are involved in the (screenshot’s) visualization? 2. How many scales are involved in the (screenshot’s) visualization? 3. How many visual channels or encodings are involved in the (screenshot’s) visualization? Figure 5 (a) shows an example of a question from this block. Once finished, the user is asked to create an additional scatter plot with MetaViz with the same dataset. A tutorial is also provided due to the novelty of this system. When the task has been completed, the previous questionnaire is presented in the context of the MetaViz system, that is, questions 1 to 4 referring to the scatter plot created by the user in MetaViz and questions 5 to 7 with screenshots of data visualizations created in this same system, as shown in Fig. 5 (b).
This research presents a work-in-progress to measure the understanding of the elements involved in data visualizations and information dashboards. The study is set to validate the learning dimension of a novel system focused on the creation of data visualizations and dashboards through fine-grained components. This system (MetaViz) is based on a dashboard meta-model developed through domain engineering that captures different concepts and relationships from the data visualization domain.
We developed a study procedure and a questionnaire to determine whether users can understand and recall concepts related to data visualization with MetaViz. The questionnaire aims at measuring two cognitive dimensions –understanding and remembering– [28-30] in the context of education. In this case, MetaViz is compared to a commercial tool (Tableau), but other visual analytics tools can be employed to conduct this research.
The study will be carried out with a sample of students with data visualization skills. Future research will involve replicating this study with other samples, including lay users and people with knowledge from other domains, to compare the outcomes.
This research work has been supported by the Spanish Ministry of Education and Vocational
Training under an FPU fellowship (FPU17/03276). This research was partially funded by the Spanish
Government Ministry of Economy and Competitiveness through the DEFINES project gr