If you’ve read this blog before, or heard me speak, then you know that designing data visualization makeovers is one of my favorite activities of all time.

I love redesigning pie charts, in particular.

The vast majority of data visualization trainings just advise people to stop using pie charts… without teaching people what to do instead.

Every year, I work with dozens of organizations, and every single organization still has pie charts sprinkled throughout their reports, slideshows, dashboards, and infographics.

Hearing that you’re supposed to avoid 3D exploding pie charts with a hundred tiny slices is beginner-level stuff. I need to train you on useful alternatives–that’s the advanced-level stuff.

I created this before/after pie chart makeover a million years ago but forced myself to wait until March 14 to share it. Phew! It’s been a long wait.

Before: Two Pie Charts

I recently worked with a grantmaking organization. They awarded grants to support various research projects. For anonymity, I’ve changed the names of the research projects to A, B, C, etc.

The grantmaking organization simply wanted to look for patterns in their funding over time.

Their “before” version looked like this:

Two Slices Only

A major guideline for pie charts is that they’re easiest to read with only two slices.

At the very least, we’d need to collapse the seven slices into just two slices. For example, you may choose to focus on research topic A with a dark-light contrast.

Dark-light contrast is helpful, but it’s not enough. We have to keep editing.

Don’t Make Viewers Zig-Zag Their Eyes Across Different Graphs

Another guideline—for all charts, not just pies—is that you don’t want to make viewers make comparisons across multiple charts.

If you want them to make comparisons, then put those things next to each other, not several inches apart from each other.

Here’s how my eyes have to zig-zag back and forth to read the original:

The collapsed, two-slice version still requires zig-zagging eye movements:

Let’s make the comparisons faster and easier.

The grantmaking team and I put our heads together. We came up with a few alternatives.

After: Stacked Columns

Being researchy types, the first alternative we came up with was stacked columns. Stacked columns are just rectangular versions of pie charts.

There’s nothing inherently wrong with stacked columns. But, yikes! This alternative felt way too busy. There are seven segments in each column, which is too many…

… unless…

…. you use dark-light contrast to focus your viewer’s attention one just one segment at a time.

After: Slope Graphs

Since we’re comparing patterns over time, how about a slope graph?

A slope graph is a fancy name for a line graph that has exactly two points in time.

Spaghetti Slope

Here’s what we tried:

Yikes! We had good intentions, but we accidentally created a spaghetti slope graph.

When this happens, don’t fret. It’s not your fault. It just means your particular dataset had percentages that criss-crossed and overlapped too much.

Spaghetti Slope with Highlighting

There are a couple ways to detangle spaghetti graphs.

One option is to guide viewers’ eyes to just one thing at a time with dark-light contrast:

Which line(s) will you highlight?

Use your best professional judgment.

Think about your unique audience. You might highlight something that increased, like B. Or you might highlight something that decreased, like D. Or, you might highlight something that remained steady over time—perhaps that thing was supposed to go up or down, but didn’t, and you’ve got an interesting story to discuss.

Small Multiples Slope

Another way to fix a spaghetti graph is with a small multiples layout.

Small multiples means multiple small charts.

You could produce seven mini charts, one for each of the seven research topics.

But better yet, let’s group them into categories that will give our viewers more insights into the patterns—the fact that some research topics received more funding while other topics received less funding.

I color-coded by category (one color for increases, another color for decreases).

Finally, I added icons to boost memorability.

This one’s my favorite. Swoon.

After: Dot Plot with Arrows

I intentionally sorted the categories into those that increased and those that decreased. It’s kind of like a small multiples dot plot.

I also included arrows, instead of just regular ol’ circles or dots, to reinforce the direction of the changes.

I color-coded by category (increases in one color, decreases in another color).

Finally, I added icons to boost memorability.

Join the Conversation

Team, you know the drill! Comment and let me know which alternative is your favorite and why. I’m personally drawn to the small multiples slope graph.

Bonus! Download the Materials

Purchase the spreadsheet that contains these graphs.

Purchase the Materials

The post Stop Making People Read Across Multiple Pie Charts (& Here’s What to Do Instead) appeared first on Depict Data Studio.

We were brainstorming different ways to display the number of new HIV infections in each subgroup. The organization wanted to compare HIV infections among men who have sex with men (MSM), African Americans, Hispanics/Latinos, and injection drug users (IDUs) between 2006 and 2010.

As always, there are several correct ways to display these infection numbers.

Option A: A Traditional Line Graph

This is the traditional way to display the data, a classic line chart.

We’ve got all the years displayed: 2006 through 2010, and all four subgroups of people are displayed within the same chart.

But this isn’t the only way to display that dataset.

Option B: A Slope Graph

Another option is a slope chart. A slope chart is basically a line chart that only shows two points in time.

You might only have two points to display, like pre and post data. Maybe that’s the only time you collected data.

Or, you might meet with your teammates and do some brainstorming and decide that your viewers don’t need to be distracted by all the peaks and valleys in the dataset. You might decide that your viewers really need to focus on the big picture. In general, did things go up, or did things go down?

I’m usually a big fan of slope charts, but I don’t think a slope chart works best for this particular dataset. The lines almost look flat – like barely anything happened during that timespan. And maybe that’s the story you’re trying to tell – that HIV infection rates were relatively stable during this time period. But if you’re trying to emphasize that there were huge increases or decreases, then this chart isn’t for you in this particular circumstance.

Option C: A Small Multiples Line Graph

A small multiples line chart helps us disaggregate the four different subgroups.

This style would be useful if you wanted to add subtitles or annotations above each graph. For example, you might type “MSM” and then below, “HIV infections rose slightly in 2007, decreased in 2008 and 2009, and then increased again in 2010.”

Here’s a twist on the small multiples line chart. It combines a technique called color highlighting with the small multiples design.

I use this approach when I want my viewers to see the contextual information,  like the number of new HIV infections for men who have sex with men (the blue line on the first chart) compared to HIV infection rates for the other groups.

Option D: A Small Multiples Area Graph

When you’ve only got one line per graph, another option is to transform your line chart into an area chart. An area chart shades in the area underneath each graph.

This is a really popular design because people say it’s easier for our brains to see that big, filled-in, dark shape than to spot the skinny line all by itself. The skinny line on the line chart is competing with the axes and the grid lines for our brain’s attention. But there’s no way you can look at this graph and not spot those big blue chunks right off the bat.

Option E: A Small Multiples Slope Graph

A fifth option for this dataset is a small multiples slope graph.
Remember, slope graphs have two points in time. So we’ve just got 2006 and 2010 HIV infections here.

The takeaway message is that there are at least a dozen correct ways to visualize each of your datasets. It’s your job to sit down and sketch all the options on paper, and to weigh the pros and cons of each layout, and then to choose which layout corresponds to the message that you’re going for.

Which style do you prefer for this particular dataset?
 
 
 
 

The post One Dataset, Five Line Charts appeared first on Depict Data Studio.

My vote for worst chart? The clustered bar chart.

I see clustered bar charts everywhere. E-V-E-R-Y-W-H-E-R-E. On pages 1, 2, 3, 4, 5, 6, 7, 8, 9, and 10 of reports. In slides 1, 2, 3, 4, 5, 6, 7, 8, 9, and 10 of presentations. In every handout, dashboard, and infographic.

Clustered bar charts aren’t inherently bad, but their overuse is killing me. What’s worse than a text-heavy report with zero graphics? A report that only contains the same chart type over and over and over, regardless of whether that chart is really the best tool for the job.

When a colleague was looking to overhaul her clustered bar chart, I jumped at the opportunity.

Let’s explore alternatives to this overused chart. Don’t forget to vote for your favorite alternative at the end of the post.

Before: An Unformatted Clustered Bar Chart

Here’s the chart style I see most often among researchers, analysts, and communications specialists: the unformatted clustered bar chart.
This particular graph is courtesy of Saint Wall Street.

Remake #1: A Formatted Clustered Bar Chart

At the very least, let’s format this chart. Regardless of which software program you’re using, make sure to question every single default setting. Default settings are fine for the privacy of your own computer – when you’re just exploring datasets or drafting several different options. But before charts get shared with coworkers or clients, you’ll need to adjust nearly every setting to ensure that your chart is easy to understand.

Crucial edits include:

• I removed the 3d, bringing the graph into the 21st century and making it easier to read.
• I flipped the order of the bars. Now, pre-intervention data is listed first and post-intervention data is listed second.
• I reduced chart clutter. I removed the border, the grid lines, and the tick marks, none of which served a purpose in our before chart. This streamlined design shocks folks at first, but trust me, sleep on it and you’ll awaken a completely new analyst. Once you get into the habit of reducing chart clutter you’ll never go back to your old ways.
• I used the color highlighting strategy. Previously, viewers’ eyes were equally drawn to the orange and the blue. Now, pre-intervention data is a light color and post-intervention data (what we want viewers to focus on) is a deeper, saturated color. I want viewers to know exactly which set of bars is most important.
• I swapped the generic title (“Fathers’ Increased Commitment to Family”) for a descriptive title (“Fathers showed an increased commitment to families”). Notice how the title is the largest font in the chart, a technique called hierarchical font sizing.

Remake #2: A Side-by-Side Bar Chart

In theory, this small multiples layout should be easy to read. I purposefully added light gray shading to indicate part-to-whole relationships.
In practice, viewers’ eyes have to zig-zag back and forth between the pre and post data. For instance, they’d have to read the 68%, and then the 75%,  and then try to compare the lengths of those bars to each other. Since the bars are beside each other – not on top of each other – comparing the two lengths takes too much precious mental energy. For that reason, the side-by-side bar chart isn’t my favorite option.

Remake #3: A Slope Chart

Slope charts are line charts that only display two points in time. They’re an excellent option for pre-intervention and post-intervention datasets like this. The upward-sloping trend jumps off the screen and into our brains.

Remake #4: A Panel Chart

A small multiples version of the slope chart, this remake allows viewers to examine each of the metrics separately.

Remake #5: A Panel Chart with Contextual Shading

I use this style when I really want viewers to compare the lines to each other. I highlight one line at a time and gray-out the others.

Remake #6: A Dot Plot

Dot plots are another stellar choice for displaying two points in time. I use dot plots for pre/post data, for Grant Year 1/Grant Year 2 data, for Fall/Spring data, and many other timeframes. View additional examples here.

Share your perspective in the comments section: Which option do you prefer? Which option(s) do you already use in your own work, and which ones will you try now that you’ve seen them in action? 

The post Alternatives to the Clustered Bar Chart appeared first on Depict Data Studio.

Most of these charts from my prior work were ineffective. Of course they were. Data visualization is a new skill for most researchers and evaluators.

One of my old charts was so horrific that I just had to share it with you:

What’s so bad? you ask. I see these all the time, you say. Can’t they just tilt their head to read the words, you argue.

What’s Wrong with my Previous Chart

Each of these mistakes–from the before I knew better period of my career, never to be revisited again–kept viewers from understanding the information:

• Generic and centered title (Should have a “so what?” and be left-justified.)
• No subtitle or annotations (So the viewer can’t skim the information.)
• Border (Should be removed.)
• Full grid lines (Should be lightened or removed altogether.)
• Legend (Should have direct labeling instead of the legend to avoid those back-and-forth eye movements.)
• Speaking of direct labeling… why did I label the axes and the bars? Overkill.
• Diagonal text (Should swap this vertical bar chart for a horizontal bar chart, leaving more room for the labels.)
• Default color scheme (Should use the client’s brand colors).
• An action color is there, kind of. Not really. It’s used incorrectly. Our brains are drawn to darker colors. Why was I drawing attention to the “possible items correct” section with the dark blue?
• Most importantly, why did I display the raw scores vs. the percentages? Can you imagine how much mental energy it must’ve taken my viewers to figure out what 2.9 vs. 2.6 vs. 6 means?

Before/After Makeovers

The first phase in learning about data visualization is usually critiquing charts. What, exactly, is wrong with your work or someone else’s? How can you learn from those mistakes?

The second phase is articulating specific ideas for how you’d make charts better.

The third phase is actually remaking your charts.

Some people stay in the first phase forever. My goal is to move more people into the third phase: actually improving your work.

So I’ll lead by example. Here’s how I’d re-do this chart today. I designed five different remakes.

Makeover #1: A Horizontal Clustered Bar Chart

First and foremost, I changed the chart type from a vertical column chart to a horizontal bar chart. Horizontal bar charts are great when the data labels are pretty long, like in this example. Your goal is to avoid diagonal text at all costs. It’s harder to read, so viewers get distracted, bored, or generally turned off and stop trying to decode your messy chart. I’m not a fan of vertical text either. (Not sure how to transform your vertical bar chart into a horizontal bar chart? I’ve got a tutorial.)

Next, I transformed the raw scores (“2.6”) into percentages (“43%”).

Finally, I addressed all the Low Hanging Fruit formatting issues. For example, I added a 10-word title and a 1-sentence caption; I deleted the border, grid lines, and tick marks; I placed the percentages on the bars so I could delete the axis; and I placed the years directly on the bars so I could delete the legend.

Why the black and white? Just for fun, I want to show you that you can still emphasize patterns without using bright, showy colors. In real life, would I match the action color to my client’s RGB codes? Absolutely.

What do you think of remake #1?

Makeover #2: A Clustered Stacked Bar Chart

I’ve experimented plenty with real-life clients to see whether they prefer regular bar charts or stacked bar charts. Here’s the typical response:

• Me: Here’s the first chart. [Regular bar chart.] What’s the message here?
• Client: Oh wow! Our participants are doing so well! 46%! That’s high! … Right? Or is that number low? Out of what? Out of 100%, right? Hmm…
• Me: You’re on the right track. Here’s the second chart. [Stacked bar chart.] What’s the message in this one?
• Client: Oh darn, we’ve got a long way to go. 46 out of 100%?! I need to speak with our program director about this. In fact, our whole team better see this. We need to figure out what we’re doing wrong before it’s too late!

A dozen conversations later, and the result is still the same: Nearly all my clients gain deeper insights about the findings through stacked bar charts instead of regular bar charts. What’s the response in your projects?

Makeover #3: A Small Multiples Bar Chart

The first two remakes are better than the original. That being said… clustered bar charts are my least favorite chart in the history of the world. They’re so cluttered. And worse, the comparisons are lost. With so many bars smushed together, it’s nearly impossible to see at-a-glance patterns between the two series of data.

In this example, I created a side-by-side bar chart so viewers could more easily see 1) the 2009 pattern on its own, 2) the 2010 pattern on its own, and 3) the difference between the two. Still in the bar chart family, but different patterns pop out, don’t you think?

I also used the action color (dark gray) to emphasize the Social and Ethical scores, and I added an annotation (the call-out box on the chart) to make my viewer’s comprehension even easier.

Want to make your own side-by-side bar chart? I’ve got a tutorial.

Makeover #4: A Dot Plot

Dot plots are often the superior chart. I use them to compare two points in time (like this example); two distinct groups (Program A and Program B); or, when I triangulate data, two distinct viewpoints (students’ perspectives vs. teachers’ perspectives).

But they’re not always superior. This dot plot doesn’t work. It’s too cluttered, isn’t it? Its more confusing than helpful. It’s because students improved on some areas, declined on other areas, and didn’t change at all on other areas.

The annotation isn’t helping, either; instead of adding clarity, it adds clutter.

For the rare viewer who’s willing to spend 60+ seconds interpreting the chart, it’s great because it shows more nuanced patterns than the other charts. But my guess is that the majority of viewers will lose interest after a few seconds because they can’t immediately grasp what it means.

Makeover #5: A Slope Graph

A slope chart is basically a line chart for two or three points in time.

This chart type is also effective at showing rankings (i.e., it’s easy to see how the skills are ordered on the left-hand side). 

For me, this chart is the winner. It’s easiest to understand at-a-glance. It ranks each of the skills areas. It shows differences over time. There’s a lot of cool stuff going on for the viewer to explore. It’s supplemented with a non-intimidating title, subtitle, and annotation. Yet, the information takes up very little ink and space.

Your Turn

Which remake would suit your audience best?

The post Avoiding Diagonal Text in Your Charts appeared first on Depict Data Studio.