the power of categorization

There are still a few spots left in upcoming Dallas (5/5) and San Francisco (5/11) public workshops: details and registration can be found here.

I am writing this post on the heels of a lovely albeit short European trip. It included a few days in London, where I had the opportunity to conduct a day-long workshop and also present at Tucana Global's 2015 People Analytics conference. In our spare time, my husband and I ventured out to one of our favorite restaurants: Ottolenghi. As I was perusing the wine list, I was reminded of the importance of categorization (yes, apparently my data-brain is on even at dinnertime). Let's take a quick look at how categories help us make sense of things: both in life and in data visualization.

Here's a pic of the drink menu that inspired this post:

In the case of the drink list, categories ease our processing of the information. They appear on the left: aperitif, sparkling, rose, white, orange (!!), and red. Can you imagine how increasingly difficult the task of picking something to drink would be without this categorization to help us make sense of the list and understand where to focus our attention? There would also be a greater potential for misinterpretation - for example, without the categorization, I might have (incorrectly) assumed Dabouki to be a red wine. I certainly would have (again, incorrectly) believed Bianco Amphora to be a white wine. The processing of the information was made easier (and with less room for error) because I had a well-labeled construct to use as I interpreted the information.

Categories can be similarly useful when it comes to helping your audience interpret your data visualization. Let's look at one of the examples I discussed briefly at the People Analytics conference.

In the example below, data is plotted in a scatterplot across two dimensions. Imagine your organization collects information about its managers via an upward feedback survey, ultimately quantifying a manager's capabilities (as assessed by his or her team) with a single number. Your company also has a performance management process, through which everyone receives a performance rating. It might be useful to look at the upward measure (how employees feel about their boss) and the downward-looking measure (how the manager performs, as determined by their manager) together. This is shown below.

The vertical y-axis shows the manager rating. The horizontal x-axis shows performance rating. Each manager in your company is a point on the scatterplot.

We can add additional labels on each axis to help with the interpretation of the information. With this setup, the audience need not know that a higher % favorable on the Upward Feedback Survey indicates a better manager (in the same way that I didn't need to know that Dabouki is a white wine because of the categories on the drink menu).

We can take this a step further and add categories onto the x-y plane directly:

I'll admit that this final version does look a bit intimidating at first. For this reason, there can be value in starting with less and adding more, explaining what you're doing to your audience at each step so they can follow along with you, making the final visual feel less intimidating than it might otherwise. In my presentation, I started with a blank graph with only the axis labels and first described what I would plot (before showing any actual data; this can be a nice way to create anticipation among your audience as well). In the next view, I added the points to the previously blank graph. Then I emphasized the average. Following that, I drew the quadrants by adding vertical and horizontal lines based on the average. Then I drew attention to the points at the bottom left by making them red and adding the label Low/Low. Finally, I ended with the version shown above with all quadrants labeled and light shading at upper left and lower right.

In this final view, note also how the added labels on the graph make the data easier to talk about. With the quadrant titles, I can focus conversation on the cases where managers are scoring low from both the upward or downward perspectives (Low Perf/Low Mgr Score in red at the bottom left). Or there might be some interesting discussion in the cases where the signals don't align - Low Perf/High Mgr Score at the top left or the opposite on the bottom right.

Meta-lesson: categories (and more generally, descriptive and pithy labels) can help your audience interpret the data you show.

My other European destination this trip was Paris, where my husband and I enjoyed more amazing food, saw many sights, and perused many more drink menus. The overall trip was great, only too short. I hope to travel to Europe again this summer for a longer stay. If you are reading this and interested in discussing a potential workshop for your European team or organization, reach out to me at cole.nussbaumer@gmail.com.

I'll close with a couple pics of Parisian adventures with my favorite travel partner.

Eiffel Tower in the distance!

Musee D'Orsay

the great pie debate

You can't title a talk "Death to Pie Charts" and not expect to spur some debate on the topic. Sometimes being a little provocative can help generate interest and keep people's attention. It seemed to work last night at a talk I gave at the University of San Francisco as part of their Data Visualization Speaker Series.

We had an awesome turnout and I covered a condensed overview of the key lessons I teach in my workshops: understand the context, choose an appropriate visual display, identify and eliminate clutter, draw attention where you want your audience to focus, and tell a story. As part of the lesson on common visual displays, I noted one graph you won't see from me: the pie chart. We looked at an example to illustrate some of the challenges reading pie charts and discussed some alternative ways to visualize the data. Then we went on to cover the remaining lessons, followed by some lively Q&A.

The debate started with a simple question that went something like this: I've recently become interested in data visualization and I've been reading a lot about the field. Specifically on the use of pie charts, I've read some things that denounce them and others that say they have a place. Are you aware of any research comparing the takeaways that people get from pie charts compared to bar charts, for example?

My response went something like the following. This is a difficult space to study. Many of the studies that come out demonstrating one thing are opposed via counter-studies that show the opposite. My personal dislike of pie charts is more anecdotal - when I see them used in a business setting, inevitably they fail. 

I didn't talk about this last night, but upon further reflection, as I think back through the many pie charts I've encountered over time (hundreds, at least), I can think of only two cases where I tolerated them:

1. At Google when we first started sharing diversity stats on the workforce internally - the team wanted to show the general breakdown of men vs. women (for example) but didn't want to communicate the specific numbers. In this case, the fact that our eyes don't do a great job of accurately measuring two-dimensional space worked in their favor. So in a way, they were taking advantage of one of the pie chart's biggest disadvantages.
2. More recently, I encountered this data visualization highlighted in Best American Infographics 2013 - ten years of art history. Each pie represents an individual painting with the five most prominent colors shown proportionally. You can see the shift in color usage over time. Art via pies. I actually really like this!

Personally, I don't use pie charts because when I pause and think about what I want to show, I've always found a way that seemed to get the information across better than the pie chart.

That said, intelligent people will disagree with me and point out use cases for the pie. I welcome this diversity of perspective! Last night, after giving my viewpoint, I opened the question up to the audience. Santiago Ortiz (Moebio Labs) was in attendance and offered some great perspective. I'll paraphrase the viewpoint he shared: There are studies, and usually bar charts win in terms of people remembering the numbers. But it's difficult to research the Gestalt feeling of a "percent of whole" where pie charts are actually effective. So is the story about the specific numbers, or the relative amounts, as a percent of the whole? If it's the latter, then pie charts can work. (I'll note also that this is a similar point to one raised by Robert Kosara as part of his highly valued feedback on my forthcoming book).

Still, I'm standing firm. I won't use pies.

Does that mean you shouldn't use pies? Not necessarily.

First and foremost, always think about what you want your audience to be able to do with the data you are showing. Choose a visual that will make this easy. I often recommend the following. If you find yourself reaching for a pie, pause and ask yourself why. If you can answer that question, you've probably put enough thought into it to make it work. I should point out that this is something you should do for any type of visual you are using. Making yourself articulate why the chosen visual works for your needs is one way to help ensure that it actually does.

We didn't solve the great pie debate last night and we won't solve it here. People stand on different sides of the fence and I actually think that is ok. When it comes to data visualization, rarely is there an absolute right or wrong. You should constantly be applying your critical thinking skills. Don't do something blindly because of a statement you read or hear. Think about your audience, what point you are trying to make, and how you can do that in an effective way. If unsure, create your visual and seek feedback.

Big thanks to the event organizers and sponsors for last night's event: Scott, Sha, Alark, Sophie, Chris, all of the student volunteers, and everyone else who helped. Thanks also to those who participated in Q&A and everyone who showed up to the talk. I had a great time and I hope you did, too!

the biggest bang for your buck

After trekking through some surprise springtime snow, I had a great public workshop in Chicago this afternoon (want to join in the fun? see here for upcoming sessions, including workshops in London, Dallas, and San Francisco). Discussion and Q&A are some of my favorite components of the workshops, because we can tackle specific challenges that folks are facing. There are always great questions and today was no exception. There was one super practical question that stuck with me that I thought I'd share more widely here.

You've likely heard of the 80-20 rule. Basically, in business it's the idea that you can put in 20% of the effort and get 80% of the result (and avoid the remaining 80% of work that only yields an additional 20% of result). The question was: "how can we apply the 80-20 rule to what we've learned today?" In other words, out of all of the meaty content we've covered, where should you start when it comes to having the greatest impact? Or, as I'll paraphrase it - where should you focus your energy to get the biggest bang for your data visualization buck?

My answer? There are two easy things you can start doing today to have greater impact when it comes to communicating with data:

First: always tell a story. Think about what you want your audience to get out of every graph you show and STATE IT IN WORDS. Doing this simple step goes an amazingly long way when it comes to helping make the data you show make sense to your audience. When you put the takeaway into words, your audience knows what they are meant to look for in the visual. We spend the hands-on portion of the workshop looking at a number of real-world example graphs. All made by well-intending people. And the question that comes up again and again and again is: what point are they trying to make? Don't make your audience work to figure this out - state it for them!

Second: use color sparingly and strategically. Rethink how you use color - don't use it to make your graph colorful. When used sparingly, color is your single biggest tool for drawing your audience's attention to where you want them to pay it. I often start by making every single component of my visual light grey, pushing it all to the background - the data, the axes, the titles. This forces me to think about where I want to draw attention and use color intentionally and with purpose to emphasize those pieces of the visual.

Pair these two things - state your story in words and use color strategically to highlight where you want your audience to look - and you'll have gone a long way down the path of communicating effectively with data. Bonus: you don't even need crazy technical skills to do either of these things.

Thanks, Bill, for the thought-provoking question!