Let the Truth be Told

I was recently presented a line chart that looked very similar to the one below:

 

At first glance, I thought it was designed fairly well. After all, the designer wanted to show the scoring trends for these 5 countries over time. The line chart was seemingly a wise choice as they’re great for showing the shape, velocity and direction of data over time.

However, until I looked at the labels, I was thinking that highest score for 2012 Q3 (United States) was over 3x or 300% higher than the lowest score (England).When in actuality, the US score isn’t even 7% lower.

How could this be?

Hint: Look at y-axis.

Does it Start with Zero?

No, it starts with 3.9.

It’s doubtful that the designer did this to deceive his or her audience (although there are plenty of examples where deception is the ultimate goal). Most likely they did so to avoid what you see below:
 

So what can I do to avoid a situation like the one above where a y-axis with an origin of Zero lead to a visual that’s impossible to analyze?

The answer may seem a bit simple: Annotate! Let the truth be told. Don’t assume that your audience will figure out something like this so find a way to tell them.

Here’s one possible solution:

You want your annotation to be noticeable but avoid obstructing the view of the data itself.

Annotation is a feature that most tools will allow you to apply directly on your charts. If not, be creative and find an alternative solution to ensure your audience understands the story you are telling.

Cheers,

Kevin A. Taylor

If You Only Get One Chance...Make it Count!

Not too long ago,  had the pleasure of co-teaching a data visualization workshop at the Data Matters Workshop Series at The University of North Carolina at Chapel Hill with Rachael Brady (formerly of the Duke Visualization Center). Rachel delivered a great presentation covering some fairly advanced aspects of the visualization process. While I took in a ton, one quote really stuck out.

“You get one chance.”

In the world of data visualization and specifically dashboard design, you’ve got one chance and typically one chance only to win over your audience. Your work must come across a clear and professional. So attention to detail is paramount.

Every pixel on our screen, every drop of ink on our canvas should be applied with conscious intention. The same is true for every pixel we exclude. And every decision should be based on communicating the clear message we intend to deliver to answer our stakeholder’s questions.

So what does that mean for us as designers? It means understanding and applying visualization best practices. There’s too many to list in this blog but I’ll provide a good starter sample below. PLEASE comment with suggestions you feel strongly about as well.

• When we apply color we do so for a specific reason, not to make it prettier
• When we add legends, make sure the entire legend is visible
• When we add value labels, think about the formats and decimal places
• When we add titles we make them meaningful and dynamic if the scope can change
• When we want to compare values across multiple slices we choose the right chart type and not force your audience to compare a series of pie charts
• When we build in interactivity, the interactivity works exactly as we expect to which requires testing each option
• When we build in interactivity, it’s clear how the interactivity works
• When we build dashboards we consider the layout of the canvas and we don’t clutter it with unnecessary banners and logos, the filtering is intuitive in positioned appropriately, the most important visual is in the upper left and the need to scroll is minimized. White space can have a huge impact.

And most importantly, what we end up with should tell the story we intend to tell without the need for human intervention. My suggestion here is that we hand our drafts over to someone else to see if they are receiving the message we are attempting to deliver. And while they’re at it, ask them to challenge the visual design aspect as well.

These tips should help our dashboards go from good to great!

Cheers,

Kevin Taylor

Tableau Control Charts - Resetting Control Limits and Signals for Process Changes

1st off...Kudos to Ben Jones of Tableau for his work and his willingness to share his own knowledge on how to build process control charts using Tableau Desktop. This is something my 6 sigma leaders wanted badly and it's not an out-of-the-box option with Tableau...yet.

Here's the links to a few of the blogs Ben has written on the topic:

http://dataremixed.com/2011/09/tom-brady-and-control-charts-with-tableau

http://dataremixed.com/2011/10/how-to-make-control-charts-with-tableau

https://public.tableau.com/s/blog/2013/how-to-make-control-charts-tableau

(ALSO CHECK OUT BEN'S BOOK: Communicating Data With Tableau

Notice that there is an outlier on the right portion of this viz that falls within the LCL. This raised a red flag (or dot...haha). After a bunch of comparing some rather lengthy calculations, we came to the conclusion that the table calculations needed to be "computed using Pane Across" rather than the default "Window Across:"

This step will ensure that your LCL, UCL & Signals calculations will reset at each pane as shown below for this example...no more red alert:

In all honesty, I never could get Ben's dashboard to break. As I look closer, Ben was most likely able to accomplish the same by using "Compute at Date". Not sure which will work for you, but I wanted to be put this out there in case you run into something similar.

Cheers!

Kevin Taylor

It's Time to Leave the 90's

How are you running your meetings?

How are your leaders running their meetings?

Unfortunately, for many, your meetings are probably run the same they were for the past couple of decades…yes decades.

Although it was available under other names earlier, Powerpoint was officially launched in 1990. I won’t go as far as to say that Powerpoint has had the same impact on the world as say Lotus 1-2-3, but it has certainly dominated the corporate meeting room!

Having grown up in the corporate world during the 1990’s I have been bombarded with “slide deck” after “slide deck”. While some have been far better than others, the vast majority of these presentations have failed to do one thing: They do not allow the audience to dive deeper.

Traditionally, a presenter builds slides with static images and text. Hopefully, these slides are authored in a way that attempts to tell a story about the content.

That’s great…BUT…what if your audience wants more detail about the content you are discussing?

Consider this scenario. At year-end, your entire leadership team travels to it’s annual business review. The most important slide shows that Customer Satisfaction is down 30% year –over-year. As expected, this creates a stir.

Now…if all the presenter can show is that the CSAT score is down 30%, this leaves a LOT of questions. Arguments will be made but answers will not be given. Questions will go unanswered and left as an “action item” for someone to later share with the team.

This is an opportunity lost!

What if the data were presented “live” or “interactively”? What if, with a single click, the presenter could drill down into Region? And then spot a single region where CSAT was down? And then drill into that and see where a particular pillar was accountable?

Then you could have a real conversation. One that not only answers questions, but one that generates new questions along the way.

The tools are available now! Advances in visualization technology have changed the landscape. They’ve actually been for a while. This type of “Minority Reports” style of  meeting is changing how we communicate in business. It’s high time that we move our meetings in this direction.

Cheers,

Kevin Taylor

Visualization @ Home

Not sure how I feel about this, but once again I am providing testimony to how much of a geek I am. Although, sometimes being a geek pays off in big ways...

Take a look at the "highlight table", or as many like to mistakenly call it, the "heatmap".

                                                                                                                                                                   

I dug this "goodie" up this morning. A few years back now, my wife and I were in the market for a new home. We had our favorites but we still had a list of 11...and unfortunately, could only afford one.

I’ve fudged the numbers in all columns for the sake of not wanting to disclose information regarding our personal finances, but the point is not lost…This nifty lil’ visual helped me land our best option, House E (all green), which happened to be my #1 choice.

This is no small feat. Let me key in something here. Without being able to SHOW my wife how the houses stacked up, here is what our analysis may have looked like:

Anyone care to wager which house we would have moved into?

3 years later…couldn’t possibly be happier with our data-driven decision. Coincidence? I like to think not.

Cheers,

Kevin Taylor

Dashboard Layout Re-Thought

Having studied data visualization and dashboard design for many years, I had been programmed to regard screen space in a very particular way.

You may have seen something similar to this picture before:

According to this, the upper left part of your screen (or dashboard) should always contain your most important data, or what you hope to emphasize.

Others describe the layout in terms of following the shape of the letter “Z”. The upper left is the most important, followed by the upper right, then the lower left and then the lower right accordingly.

With this in mind, I have always pushed my filters out to the right of my visualizations. This made sense thinking strictly about the approaches above.

Then someone challenged me on this, saying “Go out on the web and find me a site that has the filtering or navigation on the right side of the screen.”

Although there are plenty of exceptions, the majority of websites have these controls on the left side of the screen. Almost as if to say, “Here, Right Here, This is where you tell me what you want to see!!!”

Try it for yourself. Facebook.com, CNN.com, Gmail.com and even most internal corporate sites have the controls placed strategically on the left side of the screen, consuming at least a portion of the “Emphasized” quadrant.

So while approaches like the ones I had read over and over again are still very valuable in terms of where to place each visualization, User Experience should never be overlooked. For this reason, I have started designing most of my dashboards with the filters on the left.

Cheers, 

Kevin Taylor

Never Say Never! Except When...

Sorry “Beiber Nation! This is not a blog about a pop song. Your Welcome to the rest of the world.

A mentor of mine once warned me against using absolute terms like never and always.  In the world of visualization there are a lot of rules and principles, but rarely are there laws that are written in stone, never to be broken…Although some of the purists might wish to argue this point.

While there are some rules that really should be avoided in almost any situation, there are still typically exceptions. The main 2 arguments I tend to hear are :

1. The customer asked for it.
2. My boss asked for it.

Both of these are very valid arguments in the real world. For the former, I’d suggest designing as they have requested but also show them a better alternative. For the latter, good luck!

With that said, I want to raise one design rule/principle/guideline that should never, ever, ever, ever be broken…ever. Do NOT sort your data alphabetically when your intention is to show a trend over time.

The issue here may be obvious, but let’s look at how this can blind one’s analysis:

Looking at the chart below it’s quite difficult to determine what direction the data is trending…at best, it will take some time.

Not so hard in the chart below is it?

So if your boss demands the 1st chart, you might consider a different role.

I saw this recently in a dashboard and I do not believe it was by design…but it was there. So the lesson here is to check your work and when you’re done checking, check it again and they have someone else check it. Otherwise your credibility could be challenged.

In one of my 1^st blogs titled “Do You See What I See” , I suggest that as designers we should have others validate our visualizations. This situation might have been avoided had the designer heeded this advice.

I’d like to know if you all have any “Never-Break” rules? Please leave a comment.

Cheers,

Kevin Taylor

If Data is the New Bacon, How Can We Prepare?

It’s no secret that more and more and more data is becoming available. When I got started working with data in 1999, Terabyte was very rarely heard word. Today we’re speaking an entirely different language: Exabytes and Zettabytes and, before too long, Yottabytes (May the force be with you).

What there’s not necessarily a lot of…people with the skills and “know-how” manage the new landscape or to transform the data from data to information to insight.

"There is a shortage of big data experts," said Michael Rappa, director of advanced analytics and distinguished professor at North Carolina State University. "I don't see the gap narrowing. Universities aren't producing enough. We have 80 grads per year" with master's degrees in analytics. "We could be producing 800 per year and still not meet demand. With each class, the demand goes up."

The Advanced Analytics program referred to above is gaining a lot of attention from the corporate world and thus, admissions has become extremely competitive. However, there are alternatives.

One such alternative is the 1^st Associates Degree Program in Business Analytics. Offered at Wake Technical Community College, the program is now in its 2nd class which includes over 70 students, most of which already possess Bachelor and Master degrees in various fields.

In addition to the degree program, the school also offers 2 different  certifications (Business  Intelligence Certificate & Business Analyst Certificate). To meet the demands of the modern world, classes are offered both in-person as well as online.

If you or one of your colleagues is interested in the program, please don’t hesitate to ask questions and be sure to check out the program website below:

 http://www.waketech.edu/programs-courses/credit/business-analytics/program-information

Cheers, 

Kevin A. Taylor

Pies for Binary Comparisons? Not so fast Skippy

If you study data visualization enough, you’re probably aware there is a large constituent of folks that believe pie charts suck. However, most have at least some level of acceptance for their use. ***although as I found out on Twitter last week, plenty of hardliners do exist***

Some experts say a pie chart should never exceed 5 slices, some say 3 and others say no more than 2. Personally, I’d say no more than 2…with a possible exception for 3 slices if you must.

Yet even if you only have only 2 slices, pie charts can still have their shortcomings.

While speaking with an industry expert some time ago, I shared a dashboard I had built. I had mistakenly thought hat using a series of pie charts would be a good choice to show the ratio between Red Badge and Blue Badge workers across 4 different regions. Needless to say, my colleague blasted this.

Here’s the reasoning. And it’s not that the pie charts don’t show the ratio well for each region…it’s the extra bit of insight that the pie charts can’t deliver…at least not very well.

Here’s what I originally had (I have omitted all labeling and axes as they’re not pertinent to this discussion & Red = Red Badge and Blue = Blue Badge in all visualizations):

While it’s easy to see the split of Red to Blue for any given region, it’s impossible to compare the regions. OK, may you be able look back and forth to see the slice..but which has more?

We could use size to show that right?

Sort of. But it’s been proven that we suck at comparing 2-D surfaces & Angles. What we do excel at is comparing length.

Using a bar graph, it’s easy to see  not only the ratio of red to blue, but also the total number for each region and how they compare to one another. And it takes up considerably less space! Very glad that someone pointed this out to me and hope you’ll find this valuable too!

I wrote about this one as I know that stacked bars have their shortcomings as well and I'm really hoping you will be vocal about what works even better!

Cheers, 

Kevin Taylor

Sometimes You Just Gotta Give 'Em What They Want

I’m probably guilty of this as much, if not more than anyone else:

Not letting go of the Principles of Data Visualization.

I say this because I have studied these principles inside and out. I have a tendency to read the words of Stephen Few, Edward Tufte and several other experts as though they are written in stone.

This is something I have to consciously recognize and then quickly get over. Even Stephen Few will tell you, the best dashboard is the one that is used.

The fact of the matter is, if we are building a dashboard, we are probably building it for someone else. That someone probably has less knowledge of data viz design and often has no desire to understand data viz design. But in the end, we have to deliver something that our client will adopt, use, understand and ultimately take action on. If we don’t accomplish these things, our dashboard will not be a success, no matter how good of job we did applying the principles!

So, if a client or manager or director wants to see all good/bad data scenarios to be marked with green/yellow/red, then that’s what we deliver. Do we know that statistically, approximately 10% of the population may not be able to decipher between green and red? Yes! Then why do we do it? Because it’s not a show-stopper and if it helps to increase adoption, that’s a win.

With this said, you still don’t want to create “3-D, Spinning, Flaming Pie Charts”. There are certain things like chart selection that you will want to save your battles for.

And when you do get asked to build that “3-D, Spinning, Flaming Pie Chart”, if your tool will actually build it, go for it. But be sure to provide a better solution right beside it. Explain the difference and hope that your client SEES the better way.

I chose to write about this in response to a healthy debate I had based on the following tweet I posted:

While I stand by my comment and believe that there are much better ways to display data than through a donut chart...I am developing a dashboard now for an executive that will include a series of donut charts with a % Value in the center.

#BendButDon’tBreak

Cheers! 

Kevin Taylor

Analyzing at the Speed of Color

One of our goals when designing visualizations should be to help accelerate the analysis process for the analyst. There are many ways to do this, one of which includes being conscious about color resonance…or how colors might resonate with our audience.

  

So how can color choice speed up analysis? Let’s look at a few examples.

In this first example, rank Sprite, Coke and Sunkist in order by sales:

Obviously it can be done, but how efficiently? I think most would agree that this next graph, in which we’ve changed only the color, requires much less time and effort to analyze:

This is because the colors in the second chart in most cases will resonate with the audience. “Thanks” to to heavy marketing of these brands, we associate red with Coke, Green with Sprite and Orange with Sunkist. (Granted, labeling the bars would have helped quite a bit as well)

We could provide seemingly endless examples, but I’ll leave you with this last one that came up when I was at Cisco. In the 2 charts below, I challenge you to determine which is a larger, the ‘% of blue badge’ or ‘% of red badge’ employees?

 

      

The second one should be much easier to analyze. Now this may seem silly and petty, but keep in mind that dashboards often have multiple charts so the time lost can easily multiply.

(although many, myself included, might argue it’s still a pie so it still “sucks”…at least it’s binary)

Color can be a great mark type for categorizing your data. Just keep in mind that all color should be used with purpose. So when you have a chance to use colors that will resonate with your audience, consider doing so.

Cheers! 

Kevin A. Taylor

Dueling Views on Dual Axis Charts

Maybe it’s the recent birth of my twin girls that’s got me thinking in 2’s. Whatever it is, I can’t seem to get my head off the effectiveness of dual-axis charts.

For a long time, I used dual axes and thought nothing of it. After all, they enable us to show 2 measures on the same graph and we’re seemingly always looking for more real estate to work with.

However, I’ve found that there are fewer situations that truly benefit from leveraging this technique than I might have originally believed. Now I’m not gonna  go to an extreme here and say “Never Use Dual Axes”, I’ll leave the absolutism to the hard asses in our field.

While a case can be made for (although I’m not completely sold) dual axes when you have want to compare 2 measures of different units (i.e. Sales $ and Units Sold), I would suggest that we savoid using 2 scales for measures of the same unit (i.e. Sales $ and Shipping Cost $).

I’ll use a simple plot chart to illustrate the potential shortcoming here. Let’s say we want to show Sales $ and Shipping Cost $ for each of our Customer Segments. It might look something like this:

The issue with the chart above is the Sales $ are exponentially larger than the Shipping Costs $ leaving the latter in a state where the differences can’t be visualized. A common tendency here would be to create 2 axes, one for Sales and a 2^nd for Shipping Costs as shown below:

So we’ve resolved our problem of not being able to see the pattern for Shipping Cost $. However, in doing so we’ve created an illusion as our minds will instinctively attempt to compare the magnitude of difference.

The best solution here might be to use a separate chart for each measure…i.e. Small Multiples. Or, if your data is granular enough, a scatter plot might be more revealing depending on the question you seek to answer. Just a couple of alternatives, happy to hear yours

Cheers,

  

Kevin Taylor

10 Easy Ways to Improve Your Tableau Workbook Performance

I don’t typically write “tool-specific” blog posts. However, tomorrow I’m presenting at a local Tableau User Group Meeting so I figured I’d kill 2 birds with 1 stone. If you are not a Tableau user, hopefully you can find some value in the concepts.

Part of my presentation pertains to optimizing your Tableau workbooks. The following 10 suggestions are some of my favorites for reducing workbook size and enhancing the overall performance. Most of these take only a few seconds to complete and will pay huge dividends.

Every little bit counts!

Please feel free to add your own suggestions in the comments! I know there are many more Jedi tricks. (& if you know of a way to do both #4 AND #5 on the same workbook…do tell!)

1. Use Extracts when Possible
• Tableau’s Columnar Data Engine is the way to go if you don’t require real-time data
2. Optimize Extracts
• Adds your calculated fields to your extract instead of calculating on your desktop
3. Filter Data at Data Source Level
• If you have 10 years of data and you’re only reporting the last 5, filter out what you don’t need before you extract.
4. Aggregate for Visible Dimensions (& “Roll Dates To”)
• Aggregates your data based on your visualizations (i.e. you have data to Engineer level but we only need to report at Director Level)
5. When using dates, specify how far down date hierarchy to view.
• No need to aggregate to the minute or second if you need to display at a weekly view
6. Incremental Extracts
• Appends only new data rather than regenerating all data
7. Hide All Unused Fields
• Best used when ready for production
8. Use Actions instead of quick filters
• Actions Filter the Visualization rather querying your data source
9. If you must use filters, use context filters
• Context Filters create a temp table with the results of your filter so that all subsequent filters query only the reduced result set
10. String Calculations perform worse than Numeric Calculations.
• Parameters are a great place to leverage this advice
11. Do not attempt to Boil the Ocean
• Use workbook case specific datasets rather than 1-size-fits-all solutions, this is not like building a Universe in BO if you’ve worked with that tool.

Cheers,

Kevin Taylor