Bad-Viz: The Silent Killer of Data Science Careers – The Shocking Truth About How Poor Data Visualisation “Hurts”.

1.0       Introduction

In the age of big data, the ability to glean valuable insights from complex datasets is no longer a competitive edge – it’s a fundamental requirement for organisational success (Manyika et al., 2013). Data scientists stand at the forefront of this endeavour, wielding sophisticated algorithms and analytical techniques to unearth hidden patterns and trends. However, the true power of data lies not just in its analysis, but in its effective communication. Data visualisation acts as the crucial bridge, translating complex data points into actionable insights readily digestible by stakeholders (Few, 2009). Despite its immense importance, many data scientists struggle with creating impactful visualisations, inadvertently falling victim to the silent killer of data insights: poor data visualisation, or “Bad-Viz”. This article delves into the multifaceted world of Bad-Viz, exploring its various forms, the underlying factors that contribute to its prevalence, and its demonstrably negative impact on decision-making processes. Through the inclusion of concrete examples, case studies, and real-world scenarios, this article aims to equip data scientists with the knowledge and tools necessary to overcome Bad-Viz and effectively communicate their data-driven insights.

1.1       Demystifying Bad-Viz: A Multifaceted Problem

Bad data visualisation, or “Bad-Viz”, manifests through a spectrum of issues that impede the clarity and efficacy of data presentations. These shortcomings, ranging from cluttered layouts to deceptive portrayals, pose a significant threat to effective data communication ([Kirk, 2016]). Some prevalent manifestations of Bad-Viz include:

1. Information Overload: Visualisations crammed with unnecessary elements, such as excessive gridlines, labels, or data points, overwhelm viewers and obscure essential insights. This clutter hinders comprehension and impairs the ability to glean meaningful information, as documented in studies on the psychological effects of visual complexity ([Cleveland & McGill, 1983]).

2. Misleading Scales and Data Manipulation: Manipulating axes scales or omitting crucial data points to exaggerate trends or downplay fluctuations is a form of deception. Such practices lead to skewed interpretations and can ultimately result in misguided decisions, as highlighted in research on the ethics of data visualisation ([Cairo, 2019]).

3. Inappropriate Chart Selection: Choosing the wrong chart type can obfuscate patterns and relationships within the data. This inappropriate selection hinders comprehension and prevents viewers from extracting clear insights. Choosing the right chart type is crucial for effective data storytelling ([Few, 2009]).

4. Lack of Contextual Information: Omitting adequate context or explanations for the presented data leaves viewers without a clear understanding of its relevance or implications, potentially leading to misinterpretations. Providing context is essential for audience engagement and informed decision-making ([Yoon et al., 2012]).

5. Inconsistent Design Elements: Inconsistent use of colours, fonts, and styles across multiple visualisations can confuse viewers and detract from the overall coherence of the presentation. A lack of unified visual language hinders accurate data interpretation and informed decision-making, as emphasised in research on the principles of visual design ([Tufte, 1990]).

6. Overcomplicated Visualisations: Trying to convey an excessive amount of information in a single visualisation can overwhelm viewers and dilute the key message. While visually impressive, overly complex visualisations often fail to communicate insights effectively, leading to confusion and misinterpretation. Simplicity and focus are key principles for creating impactful data visualisations ([Ware, 2013]).

7. Neglecting Accessibility: Failing to consider accessibility requirements, such as colour-blindness or visual impairments, excludes a significant portion of the audience and limits the effectiveness of the visualisation. By designing with accessibility in mind, data scientists ensure that all stakeholders can access and understand the presented information, adhering to the principles of inclusive design ([Universal Design Project]).

2.0       Illustrating the Impact of Bad-Viz: Real-World Examples

To solidify the detrimental effects of Bad-Viz, let’s delve into real-world scenarios highlighting its potential pitfalls:

1. The Case of Misleading Scaling:

Imagine a company’s quarterly revenue growth is visualised using a line chart. However, the y-axis starts from a value significantly higher than zero, thereby exaggerating the growth trajectory and masking minor fluctuations. As a result, stakeholders may erroneously perceive the company’s performance as more robust than it is, potentially leading to unwarranted optimism or complacency.

Bad Viz Example: Using a line chart where the y-axis starts from a value significantly higher than zero, exaggerating the growth trajectory.

Reason why this is bad: Starting the y-axis from a value significantly higher than zero exaggerates the differences between data points, making minor fluctuations appear more significant than they are. This can lead to misinterpretation of the data and erroneous conclusions about the company’s performance.

Appropriate Visualisation Type: Line chart with a proper y-axis scale starting from zero.

2. Cluttered Visuals Obscuring Insights:

Consider a marketing dashboard that displays customer demographics using a pie chart with numerous small segments representing various age groups. The cluttered visualisation makes it difficult for marketers to identify dominant age demographics and tailor their strategies accordingly. Consequently, marketing efforts may lack precision and fail to resonate with target audiences effectively.

Bad Viz Example: Using a cluttered pie chart with numerous small segments representing various age groups, making it difficult to interpret.

Reason why this is bad? A pie chart with numerous small segments representing various age groups can be cluttered and difficult to interpret. The clutter obscures the insights that the visualisation is intended to convey, making it challenging for marketers to identify dominant age demographics and tailor their strategies effectively.

Appropriate Visualisation Type: Stacked bar chart or grouped bar chart.

3. Inappropriate Visualisation Type:

Suppose a dataset showcasing monthly website traffic attempts to visualize trends using a pie chart instead of a line chart. Pie charts are ill-suited for displaying trends over time, as they do not convey the sequential nature of the data. Consequently, the visualisation fails to effectively communicate fluctuations in website traffic, hindering the identification of peak periods and potential opportunities for optimisation.

Bad Viz Example: Using a pie chart to visualise monthly website traffic, which fails to effectively communicate fluctuations over time.

Reason: Pie charts are not suitable for displaying trends over time, as they do not convey the sequential nature of the data. Using a pie chart to visualise monthly website traffic fails to effectively communicate fluctuations over time, hindering the identification of peak periods and potential optimisation opportunities.

Appropriate Visualisation Type: Line chart or bar chart.

4. Lack of Consistency:

A dashboard displaying key performance indicators (KPIs) for various departments within an organization uses different colour schemes and chart types for each department, creating inconsistency and confusion. Without a consistent visual language, stakeholders may find it challenging to compare performance across departments or identify trends and patterns.

Bad Viz Example: Using different colour schemes and chart types for each department within a dashboard, creating inconsistency and confusion.

Reason why this is bad viz: Using different colour schemes and chart types for each department within a dashboard creates inconsistency and confusion. Without a consistent visual language, stakeholders may find it challenging to compare performance across departments or identify trends and patterns effectively.

Appropriate Visualisation Type: Consistent chart types (e.g., bar charts, line charts) and colour schemes throughout the dashboard.

5. Over-complication:

A scatter plot attempting to visualise the relationship between multiple variables includes too many data points and overlapping labels, making it difficult to discern any meaningful patterns or correlations. Despite the complexity of the visualization, viewers are left with little actionable insight, rendering it ineffective for decision-making purposes.

Bad Viz Example: Including too many data points and overlapping labels in a scatter plot, making it difficult to discern meaningful patterns or correlations.

Reason: Including too many data points and overlapping labels in a scatter plot makes it difficult to discern any meaningful patterns or correlations. Despite the complexity of the visualisation, viewers are left with little actionable insight, rendering it ineffective for decision-making purposes.

Appropriate Visualisation Type: Simplified scatter plot with fewer data points or using other visualisation techniques like heatmaps or regression analysis.

6. Ignoring Accessibility:

A data visualisation platform relies heavily on colour to convey information, with critical insights represented solely through variations in colour intensity. However, this approach excludes colourblind individuals from fully understanding the data, limiting their ability to engage with the visualisation and make meaningful contributions to discussions or decision-making processes.

Reason: Relying solely on variations in color to convey information excludes color-blind individuals from fully understanding the data. Ignoring accessibility requirements limits the ability of a significant portion of the audience to engage with the visualisation and make meaningful contributions to discussions or decision-making processes.

Appropriate viz:

2.1       Impact on Decision-Making:

The ramifications of Bad-Viz extend far beyond aesthetics, profoundly affecting decision-making processes:

1. Informed Decision-Making: Effective data visualisation is paramount for extracting meaningful insights and facilitating informed decision-making. However, Bad-Viz obscures insights, leading decision-makers to base judgments on incomplete or inaccurate information, thereby jeopardising the organisation’s success.

2. Misallocation of Resources: Misinterpreted data resulting from Bad-Viz can lead to misallocation of resources, such as investing in underperforming strategies or overlooking potentially lucrative opportunities. Without clear and accurate visualisations, organisations risk squandering valuable resources and failing to capitalize on growth opportunities.

3. Loss of Trust: Repeated encounters with Bad-Viz erode trust in data-driven approaches, leading stakeholders to question the reliability of analytics and undermining organisational confidence. Moreover, instances of misleading or inaccurate visualisations can damage the reputation of data scientists and undermine their credibility within the organisation.

4. Delayed Decision-Making: Inaccurate or misleading visualisations can lead to prolonged decision-making processes as stakeholders grapple with conflicting or unclear information. Delays in decision-making can have significant consequences, especially in fast-paced industries where agility and responsiveness are paramount.

5. Poor Communication: Bad-Viz can hinder effective communication between data scientists and decision-makers, leading to misunderstandings and misalignment of objectives. Without clear and concise visualisations, data scientists may struggle to convey complex insights, while decision-makers may misinterpret or overlook critical information.

6. Risk of Costly Errors: Relying on flawed visualisations to inform strategic decisions increases the risk of costly errors and missed opportunities. Whether it’s allocating resources, entering new markets, or launching products, decisions based on inaccurate or incomplete data can have far-reaching implications for the organisation’s bottom line.

7. Stifled Innovation: In environments where Bad-Viz is prevalent, innovation may suffer as decision-makers hesitate to take risks or explore new opportunities. Without confidence in the accuracy and reliability of data visualisations, organisations may opt for conservative approaches, stifling creativity and limiting potential breakthroughs.

3.0       Addressing Bad-Viz:

Mitigating the prevalence of Bad-Viz requires a concerted effort and a multi-faceted approach, including:

1. Education and Training: Providing data scientists with comprehensive training in data visualization principles and best practices to enhance their ability to create clear and effective visuals. Training programs should emphasize the importance of simplicity, clarity, and accuracy in visualization design, equipping data scientists with the skills needed to communicate complex information effectively.

2. Collaborative Review Processes: Implementing peer review mechanisms to evaluate visualizations for clarity, accuracy, and relevance before dissemination to stakeholders. By soliciting feedback from colleagues with diverse perspectives, data scientists can identify potential flaws or biases in their visualizations and make necessary revisions to ensure their accuracy and effectiveness.

3. Leveraging Automation and Tools: Harnessing the power of advanced visualization tools and automation techniques to streamline the creation of visually appealing and informative graphics while minimizing human error. These tools can help data scientists generate interactive, dynamic visualizations that engage viewers and facilitate deeper exploration of the data.

Standardised Templates: Implementing standardised templates and design guidelines for data visualisations can promote consistency and coherence across different projects and teams. By adhering to established design principles, data scientists can ensure that their visualisations are clear, intuitive, and easily understandable by stakeholders.

5. User-Centered Design: Adopting a user-centered approach to data visualisation design involves actively soliciting feedback from end-users and incorporating their preferences and needs into the design process. By understanding the perspectives and priorities of stakeholders, data scientists can create visualisations that resonate with their audience and facilitate decision-making.

6. Continuous Improvement: Embracing a culture of continuous improvement encourages data scientists to iteratively refine and enhance their visualisation skills. By seeking out opportunities for learning and professional development, data scientists can stay abreast of emerging trends, technologies, and best practices in data visualisation.

7. Stakeholder Engagement: Engaging stakeholders throughout the visualization process fosters buy-in and ownership of the data-driven insights generated. By involving stakeholders from the outset, data scientists can ensure that visualizations are aligned with organisational goals and priorities, increasing their relevance and impact. Additionally, ongoing communication and collaboration with stakeholders enable data scientists to address feedback and refine visualisations to better meet their needs.

Conclusion:

In conclusion, Bad-Viz represents a significant challenge for data scientists and organisations alike, undermining the credibility of insights and impeding informed decision-making. By understanding the causes and consequences of Bad-Viz and adopting proactive measures to address it, data scientists can elevate their craft, empowering stakeholders with clear, actionable insights for driving organisational success. Ultimately, effective data visualisation is not just about making data look good; it’s about making data meaningful and actionable, and that begins with avoiding the pitfalls of Bad-Viz.

References:

Cairo, A. (2019). The ethical case for good data visualization. https://medium.com/@peter.haferl/the-ethical-responsibilities-of-data-visualization-4d12b7c9640d

Cleveland, W. S., & McGill, R. (1983). Graphical perception: Theory, experimentation, and application to the development of graphical methods. Cambridge University Press.

Few, S. (2009). *https://www.edwardtufte.com/

Kirk, A. D. (2016). Data visualisation: A practical introduction (2nd ed.). SAGE Publications Ltd.

Tufte, E. R. (1990). The visual display of quantitative information. Graphics Press.

Universal Design Project. (n.d.). https://cee.ucdavis.edu/universal-design-learning-institute

Ware, C. (2013). Information visualisation: Perception for design. Morgan Kaufmann Publishers.

Yoon, S., Lee, S., & Wohn, D. Y. (2012). The role of context in information visualisation for decision making. Information Visualisation, 11(1), 42-54.