Of all the information our mind can process, 90% is taken in through the eyes.
But visual information isn't processed equally. Text, for example, is far harder to process than pictures. And when we read attentively, we listen poorly when someone is speaking to us. It's the reason why everyone hates keynotes with too much text: you're being asked to process difficult information — and way too much of it — simultaneously.
The juiciest tidbit?
Even when we see statistics visually, we often don't remember them unless they're shown in the context of a story. It's whyThe New York Times publishes articles with animated visualizations: we can more fully sense, imagine, and remember the story when pictures, metaphors, description, and narrative unite — and light up our sensory cortex.
So even if you're a top-notch analyst or first-rate data scientist with a PhD, there is far more to the art of visualization than meets the eye. Yes, you must combine data sources, analyze them, and create a compelling visualization. But you must also tell a meaningful story, one your audience cares about, can interact with, and take action from — all at the same time.
This may sound impossible, but trust us, it's not. So whether you use Qlik, Looker, Tableau, or Power BI for analysis, this article will help to orient you about which visualizations work best for telling a certain kind of story under a given set of circumstances.
But before we broach the visualization itself, let's first talk about how data can either help or hurt how we convey a visual message to an audience.
Blended Data Sources
Just as writers, to tell great stories, may research a topic intensively before typing so much as one word, so too do analysts gather data, structured and unstructured, before they arrive at a deeper understanding or observation of a certain phenomenon.
Collecting data from many sources has plagued writers and analysts for centuries. While it may be easy to interpret what each source means individually, interpreting what all the sources mean together — synthesis, in other words — is the real parlor trick.
In data analysis, the harder data sources are to combine, or "blend", the more effort and error tend to occur. Extracting data from a vast array of databases, on-premise and in the cloud, is far harder than if all that data were stored in a single place.
The existence of data silos is a major reason why Bedrock Data made Fusion, a MySQL database and data warehouse which stores all your customer data in a single place, in the cloud, for easy reporting. Fusion combines SaaS data from applications like HubSpot, Salesforce, NetSuite, and Zendesk, refreshing every 30 minutes so you can skip having to export data again via bulk extracts or custom API integrations. In fact, Fusion eliminates data prep altogether — no scripts, no code, no joins, no pivot tables.
And because we knew that analysts needed to spend time creating visualizations, we also built Fusion to refresh and backup data every 30 minutes. Once you connect Fusion's MySQL database to a BI tool, et voilà, you're all set for the exciting part: visualization.
Visualizations take all shapes and sizes in part owing to the diversity of metrics they can display. Overall, you can visualize metrics like:
- Counts, comparisons, quantities over time
- Proportions, Ratios, Percentages, & Rates
- Averages & Standard Deviations
- Recurring Patterns
- And more
Below we've compiled a few examples and tips for how to frame each metric in the context of telling a story.
Counts & Comparisons
How many contacts bought a certain product? Which companies bought the most or the least number of units of your product?
News flash: a mere subtotal is not a story. But since counts are great for measuring performance, you can focus on how those counts relate to your most important customer actions. For example, you can count and compare which customers installed a product feature and put this in the context of the number of customers who upgraded from your free trial to a paid account. In short, combine related counts when telling your stories. It may also help to revisit your OKRs. If one of your OKRs is to “generate n leads during Q4”, you can then count the number of leads generated after a holiday email campaign and tell a story by comparing this to another season or metric like the count of contacts converted into leads after you launched a new series of banner ads.
Proportions, Ratios, Percentages, Averages, & Rates
Sometimes you need more than just a simple count. For an average you need the total and the total number of inputs.
When calculating a ratio like Net Promoter Score (NPS) or rate like customer churn, perform a count for all your inputs first. To calculate NPS, for instance, you would first need the counts for the total number of respondents, promoters and detractors. Only then can you tally your NPS, or the (# Promoters — # Detractors) / (# Respondents) x 100.
But just as a count isn't a story, you should then seek to answer questions around your NPS score. Look to see whether it rose or fell over time. If it went up, how rapidly did it rise? If it fell, why?As quantities are anything but static, pan out from the single ratio and see how multiple ratios change over a longer period of time. Then choose an increment — second, minute, hour, day, week, month, quarter, year — that fits your use case. This might be looking at the count of support ticket requests over the year of 2018.
Variation from the Norm
In a dataset, there are outliers and exceptions to the rule. Measuring variation allows you to answer "What is normal anyway? How far from normal do values get?"
Visualize population variability or margin of error (error of the mean).
Luckily, variation is always interesting. And as a probability measure of variation or dispersion, standard deviation shows the extent of a dataset’s distribution's stretching or squeezing. (The lower the standard deviation, the closer the data points are to the mean; the higher, the wider range of values. The standard deviation of an observation variable is the square root of its variance.) Your next line of business is to figure out what caused such variation, why is there so little or so much variation from the mean?
While line charts are the old faithful of charting quantities over time and variation, more complex datasets may require you to categorize their quantities (e.g. demographics of election poll results). If this is the case, the less common visualizations (e.g. bubble chart or tree map) may help you tell a more compelling story if there are several different snapshots or animations. In a recent feature article about the current global trade war, for example, The New York Times animated a chronological series of bubble charts to visualize the mounting volley of tariffs between the U.S. and foreign nations. This visualization technique was far more engaging than showing a simple line chart with tariffs increasing by country. By dating and sloshing bubble charts, journalists evinced a story of power and of quid pro quo geopolitics.
When history repeats itself, you may want to highlight these patterns through a time series analysis of the phenomenon that lets you forecast future values.
Seasonal commodity prices for electricity or gas.
You can show recurrence in a lot of ways: markers, overlay, time frames, layers, aggregates, etc, which save you from plotting each individual point and help people see the pattern more quickly. Explore these options as you seek to extrapolate identified pattern and predict future events.
Now that we have a basic grasp of metrics to visualize, remember that before you even attempt to tell a story, in general the best visualizations:
- Have a title & label variables
- Avoid clutter
- Use color — in moderation
- Educate, inform, surprise, delight, or even disturb
- Convince the user to take action or make a decision
- Let the user interact or collaborate somehow
Now let's dive into which charts and plots work best for certain situations.
Bar chart / Histogram
Show categorical data with rectangular bars with heights or lengths proportional to the values that they represent, shown vertically, horizontally, or both.
Bar charts don't so much tell a story as summarize a story's point visually. Often they convey the lede of a news article, a thesis, or conclusion. In the bar chart below, our eye knows the gist: "Of those who use Twitter, almost all use it as a news source, especially for business and news purposes."
Histograms are a visual representation of the distribution of a dataset. Their shape allows the audience to easily see where a relatively large or small amount of the data is situated. Overlaying histograms is thus a very powerful storytelling technique, especially when used with a line graph, to elucidate the relative distribution of disparate variables (x, y, z).
Show trends over time with the slope between two points of change. In the visual below, we have the average tickets per account per month. In mid summer, there's a slight dip. But for the most part, we see little variability year round. Such a point is very hard to convey with the scatter plot below.
A circular statistical graphic divided into slices to illustrate numerical proportion, or composition. While pie charts are simple, like bar charts and line graphs they get the point across more directly. Avoid cluttering your pie charts with excess categories.
Used to display steps in a process or identify bottle necks in a workflow, funnel charts don't necessarily have to look like a funnel. They can also show steps that bring about a certain end. For example, in the funnel report below are various stages (Product Search, View Product Page, Add to Shopping Cart, Enter Payment Info, Complete Purchase) of the buyer's journey. Funnel charts are therefore an excellent visualization technique for calculating overall conversion rate as well as where and why certain consumers segments bounce during their purchase.
One of the best if-then visualizations out there, flowcharts mark the logic of choice as it relates to a literal consequence. For telling stories, they’re a great option for showing the outcome of many codependent decisions. Although they can get complicated in a business context, they’re a great go-to for ironic, wry humor when you’re in the mood to amuse.
Line charts where the x-axis wraps around 360 degrees and with one y-axis for each x-value. The result is similar to a spider web or a radar screen. Using R and the RadarChart function, you can easily paste your own data over the library's sample dataset.
Similar to a scatter plot, grid charts use the dimension values (strings) on the x-axis, y-Axis, and use an expression to visualize the data. As we wrote in this post, example dimensions could include a contact persona or lead source. Or they could include buyer preferences, survey results, poll responses, or any other subcategory whose circles will denote size. In the first example, pie charts divide the categories selected for the grid chart.
To avoid information overload, the grid chart below doesn't use pie charts in the grid chart. It simply shows the size of purchase decisions by stage and relative size, telling a story of what buyers tend to think pre-sale.
Present data using variable width bars. They can display up to three levels of data in a two-dimensional chart. Great for market analysis, they tell a story over time by making proportions self-evident to the observer. The example below was made in Qlik.
For biographical purposes, timeline charts can’t be beat for how they delineate the highlights of history so viewers can see how proximate causes informed future events. Best of all, timeline charts area easy to make. You don’t need a fancy business intelligence tool necessary. Excel can do the trick just fine.
The stats 101 failsafe for causation and correlation, scatterplots cluster points. Use them with or without a line graph to show a trend, aggregation, spread of extremes, or, as with the scatterplot below, to give visually striking evidence for comparing two variables and proving a thesis statement (e.g. where income is higher, lifespans are longer).
A method for displaying hierarchical data using nested figures, usually rectangles, treemaps render each “branch” of the tree a rectangle tiled with smaller rectangles representing sub-branches. Every rectangle has an area proportional to a specified dimension category (e.g. Urban vs. Rural and Mix vs. Youth). The treemap below was made in Power BI.
Treemaps are a great option for telling stories when you want to display a lot of hierarchical data but not overwhelm your audience with too many values in a bar chart. Just be careful. Using size to encode the data can mislead the eye. For precise quantitative comparisons, the bar chart is a safe bet. For categories, especially regional ones (e.g. an electorate or population size), treemaps tell a compelling story far quicker than a bar chart.
Besides being really fun to say and picture eating, donut multiples are a hollowed out version of pie charts. Here is how Looker renders them in their dataset of clothes worn by day of the week:
If you can remember this oldie but goodie from grade school, box-and-whisker plots display the five-number summary (minimum, first quartile, median, third quartile, and maximum) of a dataset. Box and whisker plots are ideal for comparing distributions because the center, spread, and overall range are immediately apparent, even if you have a large dataset with a wide range of outliers.
A classic, gantt charts are bar charts which can illustrate a schedule with overlapping chronologies. They’re so handy at depicting milestones, in fact, that apps like Asana use them for helping teams visualize their work week ahead and the relative distribution of time allotted per task.
Bubble chart - sometimes depicted as a force-directed graph which form a collection of bubbles that move around based on data, bubble charts show data in three dimensions. When put on maps, bubble charts are known as “cartograms”. Proportional symbol maps are great for showing quantitative data for individual locations. Below we’ve plotted the earthquakes around the world and sized them by magnitude.
Keep in mind that there are dozens more visualizations from bullet charts for progress to heat maps for positional data. Each tells its own kind of story. And every BI tool uses them differently.
Some might tell you to be an expert in one tool like Qlik, Tableau, or Power BI. Others might advise being familiar with all three so you can adapt to what your employer has purchased. While it's not Bedrock's part to review the strengths and weaknesses of each tool, you should remember that each tool offers a different suite of visualizations, data modeling capabilities, and maintenance costs. Go beyond how tocreate custom visualizations. Explore how to setup, save, and share dashboards. And learn how to connect your database.
Whatever you decide, remember that analysis is more than finding insights valuable to a business. Those insights must convince people to do something, to take some action or make a decision.
In this article, for example, the goal was primarily to inform you about metrics and visualization types. But it was also to convince you that:
- Unified data lays the foundation for great data visualization.
- You should try Fusion for free by signing up today.