How do you avoid common pitfalls when telling stories with data?

This is something I think that gets overlooked the most, especially when first starting out. When you're telling a story your core audience is the people who will view and use this more than anyone. You should cater heavily towards your intended audience. I made this mistake a lot when I started as a Data Analyst. I tried to make things more generic so anyone would understand it. I really needed to know exactly who was going to be using it and create it for those end users.

One of the most powerful things you can do before any development is to ask this question to your stakeholders: What business questions aren't you asking today of your business? That really gets them to think about their business in ways that might lead to better and deeper insights being revealed, leading to optimal decision making and increasing the chances that your business outcomes will actually mean something in terms of ROI. You can then wireframe and start the iterative process of selecting the right viz options. Think 80% on the questions and 20% on your viz impact. Don't do it the other way round.

Understanding your audience, especially the key decision-makers who drive actionable outcomes, is crucial in my experience. Discovering their motivations for listening to your presentation allows you to tailor your visualizations specifically to address their questions and interests effectively.

One harsh reality of being a data analyst is the realization that business stakeholders are not at all impressed by the sophistication of your analysis or advanced data visualization skills you are showcasing. They are rather interested in understanding how your work could help them solve their problems. Thus, if you say, are presenting to the leadership team, instead of opting into building an interactive dashboard, you might be better off showing a line chart of business performance clearly summarizing recent trends in consumer demand and company's ability to meet them. However, if you are working with stakeholders who pride themselves on making data-driven decisions, a self-serve dashboard solution might be just what the doctor ordered.

When telling data-driven stories, it's always essential to consider your audience's knowledge level, ensure a clear and accurate message, and use high-quality, contextualized data. I try to choose appropriate visualizations, provide context, and avoid bias. I also focus on keeping the narrative simple, engaging with relatable examples, and encouraging interaction. Iterate, seek feedback, and maintain relevance to your audience's goals. Ethical considerations and avoiding common pitfalls, like overinterpretation and cherry-picking, round out a comprehensive approach to effective data storytelling.

I have found there needs to be a balance here between: best practice, industry standard, and end-user preference. The most common example out there I feel is "DEATH TO PIE CHARTS". Causing developers to avoid at all costs. However, this ignores the fact, (if done correctly!) some users prefer them and more importantly more people know how to read them then perhaps the more robust chart you may prefer. So, you need to know the data enough as well as the question the visual needs to answer while also knowing the data literacy and aesthetic preferences of your audience so that they interpret the data correctly without additional effort. If you have to explain how to read the chart repeatedly, it's probably not the best one to use.

Similarly, it's important to consider the representation of the y-axis in a chosen chart. Alterations, such as truncating the y-axis, can distort data interpretation. While adjusting the range can emphasize differences, extreme changes can exaggerate and confuse, potentially conveying a false impression. Inverting the y-axis can also intentionally or unintentionally lead to confusion, like the misleading 'Gun-death in Florida' chart.

Another consideration when working with charts is their scale: - There are likely no scenarios that necessitate using a 3-D chart, which more often than not distorts the data it aims to visualize. - It might be tempting to use custom images to make your graph more visually appealing, yet likely you won't be able to make your comparison in scale, thus misrepresenting your facts. - Lastly, don't take a page from some of the popular media's books. Unless it's absolutely warranted, attempt to start your vertical axis at the 0 point to avoid mischaracterization of the findings on your chart.

I believe that selecting appropriate visualizations is crucial in data storytelling. The chosen visual should be clear, relevant, and match data types. Avoid complexity and misleading charts that overwhelm or distort information. Keep it simple, provide context, proper labeling and avoid misrepresentation and confusion. Choose the right chart type and avoid selective data or stacked charts that hinder accurate interpretation. By adhering to these principles, visualizations enhance understanding and convey insights accurately.

I advocate for adhering to one simple rule. Don’t make your audience “think”. If your audience can’t digest what your visualization means within a few seconds, then your visualization is too complicated. Remove distractions that do not provide any value to the audience. Be ruthless about removing chart junk. One huge pitfall, especially for beginners, is the use of too many colors in one visualization. Color should serve a purpose, be accessible (e.g. color-blind friendly palettes), and not distract from your visualization. For example, if you want to emphasize and highlight differences, use a contrasting color so the audience can quickly ingest “what is different?”

Consistently prioritize the ease of reading, interpretation, and decision-making for the audience. For this, the principle of 'less is more' applies perfectly, as well as giving a journalistic approach as if you were reporting news.

I usually follow some rules on these : 1. Start with a tripwire in the first 3 seconds to capture the audience's attention by focusing on their area of interest and removing plots not serving the purpose. 2. Avoid overloading annotations elements to maintain clarity and simplicity. 3. Limit legends to a maximum of 10-15 colors, emphasizing important information and breaking down complex data. 4. Connect all visualizations through a coherent story to engage the audience and provide context. 5. Keep plots dynamic by incorporating features like sorting, filtering changes to maintain interactivity.

This not only speeds up understanding but also encourages the audience to interact with the data, unencumbered by unnecessary distractions.

Presentation order matters. Consider “Sales dropped 50% last quarter. This was due to a change in manufacturer” versus “The recent change in manufacturer last quarter led to a 50% drop in sales”. Starting with the negative insight first can (understandably) lead to a lot of confusion and frustration. But providing context before the negative insight softens the ensuing negative reactions. When structuring your presentation, be aware of how the order in which you present data affects the audience’s perception of it. Include context to help the audience understand why something has happened and why it’s important to take action on.

While storytelling with data can be a powerful tool, one must tread cautiously to avoid falling into the trap of assuming correlation implies causation. A common pitfall is misinterpreting trends as causal relationships. Just because two variables appear to move in tandem does not necessarily mean one causes the other. It is essential to exercise caution and rigorously analyze the data to establish a genuine cause-and-effect relationship. To avoid such pitfalls, focus on data literacy and statistical analysis. Ensure you understand the data's context, limitations, and potential biases.

Answering the "So What" question is arguably the primary goal of any data story. Instead of simply restating the obvious, always try to provide additional context on the narrative as told by your data and understanding of the business. Compare and contrast two summaries, both of which are true: * Year over year revenue for #2 pencils rose 4.5%. vs. * Volume of units for #2 pencils dropped 5% ... as the 10% price increase effective this month lowered product demand. Another example could be: * Month over month sales jumped 15% in the Midwest region. vs. * Month over month sales increase of 15% in the Midwest region was below 25% forecast due to Seasonality ... Out of stock issues in the Chicago distribution center are being addressed.

Provide context for your data by explaining the background, methodology, and any limitations. Also, try to give examples that align closely with the situation. Doing so will paint a clear picture into the end user's mind, in this case, the audience. This helps your audience grasp the insights and interpret the visualization accurately.

Getting feedback on my visualizations has been very helpful. When you're working on something for a long time, it's easy to overlook certain details. Getting an outside perspective can spot missed aspects or potential misinterpretations of what you are visualizing. A great way to do this is to ask people what they take away from the visualization before you explain it to them. Their responses can provide valuable information as you further develop your visualization.

Testing and refining the data visualization is as important as building the visualization. Sometimes, when we don't check them properly, the visualizations might not work as they should. Interactive visualizations for example allow the users to have freedom to explore the data while using the links you provided. These links can be broken when you're exploring on different interfaces or with different devices or browsers. This can be really annoying for users and make their overall experience not so great. Therefore, before handing over visualizations to users, proper testing is critical.

Another crucial thing to consider is avoiding the pitfall of data misinterpretation in data storytelling. It's essential to thoroughly understand the data you're working with, including its limitations, biases, and context with respect to the audience, and the narrative we are trying to convey. Neglecting this aspect can lead to cherry-picking data, resulting in misleading correlations, or overlooking critical factors that might significantly influence the story's outcome.

A great place to learn is by watching television presenters. Wether its Bloomberg, BBC, or Sky News their editors know how to tell a good data-led story, within a incredibly short time window. They will, --- Usually start by reading the question out load? For e.g. How will today's rate increase affect the sale of new properties? --- Then bring up visualisations support the narrative. The presenter knows the answer already. But they will not spill the beans just yet. Instead they want the audience to interpret the data in a way that leads up to the answer. --- To add more authenticity, they will thrown in a few short interviews/blurbs to further drive the narrative. --- And then finally reveal the answer.

Sometimes in dashboards, the aggregated data that you show may not reconcile with, what the data says overall. It’s very much essential to show the numbers that are same throughout your dashboard. It’s easy to fall into this category because of various reasons: 1. The top level aggregator is different from the record level aggregator. 2. Duplicates in one or more dimensional tables could also cause different values. 3. Null values not correctly handled. 4. Filters not properly applied. 5. Simple human mistake of writing different expressions in different places. It is crucial to test your data and check them before presenting, in each chart, table, text box, KPI box everything that you have on your dashboard.

The very first thing to consider is "what is the message?". To be honest, that drives *everything* and also helps in avoiding the other pitfalls. Different audiences need different messages. The chart type needs to fit the message. Anything that does not contribute to the message = clutter. Same with adding titles and annotations. You'd be surprised to find how often people are actually not sure what the main message of their data visualization is or should be. And once you have that message, don't be afraid to test your (draft) visualization on people. If they are getting something different from the visualization than what you intended, rethink and adjust (either the viz, or the message!).

Telling impactful data stories requires clarity, relevance, and precision. Start by defining your purpose and target audience. Choose data that directly supports your message, avoiding clutter. Contextualize the data's origin, methodology, and limitations. Simplify visuals for easy comprehension, using appropriate charts and colors. Ensure data accuracy and transparency to build trust. Address biases, differentiating between correlation and causation. Tailor your story's complexity to your audience's understanding. Craft a well-structured narrative with a beginning, middle, and end. Iterate based on feedback, and adhere to ethical standards. Ultimately, your data story should inform, engage, and drive meaningful insights.