Archive for the ‘Dashboards and Visualizations’ Category

A Preattentive Dashboard

Posted by: Jefferson on February 20th, 2012 Leave a Comment

A Preattentive Dashboard

The visual world is extraordinarily complex.  For example a quick scan of my desk reveals hand-written notes, dry erase markers and USB thumb-drive.  While I recognize these objects rapidly, I experience them at a basic visual perceptual level long before I can label or describe them.  This low level of perception is what is called preattentive processing, or visual processing that occurs without deliberate attention.  Preattentive processing can be used to create dashboards that easily communicate extraordinary amount of information per pixel and need very little effort to understand.

Characteristics such as shape, size, color, contrast, luminosity and motion are examples of features that are perceived at this low level of perception.  These factors are referred to as preattentive visual cues and help our brains categorize and filter our visual environment.  Simply put, preattentive features are the information we gain from a visual scene before we direct attention to salient features to extract deeper meaning.

Humans are very good at extracting meaning from complex visual environments.  However, this does not mean that we should be required to.  This is certainly the case when designing dashboards and data visualizations.  To keep things simple, CAN designs visualizations that focus on using preattentive imagery.  Preattentive imagery allows us to communicate complex information in a rapid and concise manner.  Our lives are complex enough.  We deserve simple dashboards.

CAN recently competed a project for one of our clients examining the accuracy of industry level forecasts for every Metropolitan Statistical Area (MSA) in the United States.  The report contained over 600 pages!  Six hundred page reports do not get read, and consequently are rarely of value.  We decide to go back to the white board.  A 600 page report contrasts with our goal of making complex information easy to understand and act upon.

We needed a way for our client to explore and understand the meaning of our  complex analysis.  The result of our research are meaningless if they are not implemented.  We started our design process by defining the business question our client needed to answer, “Which forecasts are inaccurate, and why?”  Our client needed to navigate forecast accuracies by geography, industry sector, and the duration for which the forecasts are accurate.  The dashboard we developed presents a 600 page report on one screen and can be fully navigated with three clicks.

Pre-Attentive Dashboard 1

 

Users explore the data by selecting areas on the map, concepts or MSAs individually or in group.  This action updates the State, MSAs, forecast accuracy durations and industry sections for the selected region.

Pre-Attentive Dashboard 2

For our client, forecasts with greater than 90% accuracy are deemed acceptable, and closer examination is need for forecasts with 75 to 80% accuracy.  We built these tolerances into our design.  Notice the positioning of the grey crosses in each pane.  The thin pink line shows 90% accuracy while the pink band shows 75-80% accuracy.  As users explore the dataset, this relationship allows them to quickly identify and focus on values which are below the desired range.  Glancing at the MSA window, it is clear that forecasts for Yuba City and Merced are suspect, and MSAs like Modesto should be examined more closely.

Let’s take a closeup look.

A Preattentive Dashboard 3

We’re looking at an overview of all Californian MSAs across several industry Concepts and at the Duration of Forecast.  It’s immediately clear that the accuracy for Concept #4 is ‘in the red’.  At this point, end users who are experts in the data can ask questions about what is going on in Concept #4, and discuss how this accuracy impacts future planning.

When elements on a page are judged to a similar standard, it is useful to maintain consistent visualization techniques.  For example, we kept the theme of the reference lines constant across the Concepts and Duration of Forecast window. This helps reduce the effort required to use the dashboard and frees up some cognitive bandwidth to focus on the meaning of the data.

To visualize the Duration of Forecast, we carried over the reference line theme used in other panes.  The purpose of this window is to let the user decide; for the region or categories they have selected, how accurate are the forecasts X quarters out.  All the user needs to do is watch for where the grey line crosses the pink lines.  This is a simple graphic.  Users know they can expect this combination of forecasts to be 90% accurate up to 14 quarters out, and after 18 quarters the usefulness of the forecasts dissolve.

This approach strikes true to CAN’s goal of helping businesses work smarter.  We turned a 600 page report into a single page that can be navigated with three clicks.  Rather than increase complexity, we just built simplicity.

Using Tableau Reference Lines to Explore Data

Posted by: Branden Collingsworth on February 1st, 2012 Leave a Comment

Using Tableau Reference Lines

At CAN, as needed we use the visualization software Tableau to create reports and dashboards for our clients.   Also, because Tableau is capable of handling large amounts of data very quickly, we’ve started using it to explore data visually during the data discovery stage of each project.  We use Tableau to check the quality of data, find outliers, and get a sense of the properties of a data set, such as dispersion, central tendency, clustering, etc., before we apply statistical analysis or build predictive models.  A Tableau feature, especially useful for exploring data, are Reference Lines.

This blog post explains a few ways that CAN uses Tableau to explore a data set.  For this example, we’ll look at this table of Metropolitan Statistical Areas of the United States of America (MSAs).  Within this table, we are interested in four variables for each MSA: Rank, 2010 population, 2000 population, the change in population between the decennial censuses. Tableau will geocode states to create maps, but to access that feature we need to separate the MSA name from the state.  Also, some MSAs are located within more than one state, like the Omaha-Council Bluffs, NE-IA MSA.  So, to access the maps feature, we can create a new variable called ‘primary state’ by manipulating the data.

Now that we have data, we can connect it to Tableau.  If you are following along, you might find a problem at this point. The percent change or growth rate is listed as a dimension, is not a measure as we would expect.  This is due the fact that the formatting is off for some of the percentages in the change column: some of the numbers are entered as text.  This can show us very quickly that the data needs to be normalized before we can do the analysis.  With this data set we may have found the error in an excel spreadsheet or another program, but in a bigger data sets looking at all of the values, one by one, is a waste of time.  Tableau can help you assess the quality of the data very quickly, often faster than filtering and sorting in other programs.  Buy visualizing the data before running statistical tests or data mining tools, we can find these types of problems and get clear picture of the data in one step.

Here’s the file, MSA public data, for you to play with a primary state variable and all the data normalization issues resolved.

Tableau has many tools that are useful for exploring data. One tool that helps us explore data is Reference Lines.  You can add a Reference Line to a graph by right clicking on the axis of a measure variable.  The Add Reference Lines Dialogue box allows you to add Lines, Bands, and Distributions.  Lines can emphasize an important value or threshold.  Bands allow you to emphasize a range of values.  And Distributions allow you to emphasis more than one range of values, for example, values + or – one standard deviation, quartiles, and percentiles.  These are very flexible tools that can be used for many applications.  Sometimes we will set Distributions to show standard deviations so we can quickly visualize the dispersion of the data, whether it’s normally distributed, and if there are any statistical outliers that might need to be removed from the data set.

Adding Reference Line in Tableau 1

Tableau’s Reference Lines are a good example of how CAN emphasizes pre-attentive processing in our visualizations.  Lines, Bands and Distributions create enclosures that people understand intuitively: people do not need to think to understand that data above a line is different from data below the line. It’s intuitive. When the significance of the reference line or band is an important business metric like profitability, customer loyalty, or a budgeted value, our clients understand their performance against that metric immediately.

Adding Reference Line in Tableau 2

If you enjoyed this post about Tableau Reference Lines please visit CAN’s other posts on data visualization and dashboards.  Do you have any suggestions of data visualization topics that I should write about?

Dashboard Design: Bullet Graph vs. Bar Chart

Posted by: Grant Stanley on November 29th, 2011 Leave a Comment

We invest a lot of time and energy communicating our research, because unless we can effectively communicate our findings they are useless.  When goal is to communicate the most valuable information with the least most of ink that can be understood with the least amount of effort.  For your reference, our major influences are Deirdre McCloskey on writing, Stephen Few on dashboard design, and Edward Tufte on data visualization.

Recently, CAN conducted a customer satisfaction survey for the Georgia Regional transportation Authority.  In addition to developing, deploying and analyzing the customer survey, CAN went above and beyond to improve how GRTA reported the results of their annual survey.  In this post, I will explain why we used a modified bullet graph instead of a bar chart to answer the business question.

The purpose of the graph is to help answer the business question of how does GRTA compare to two competitors across 17 different metrics.  While GRTA needs to continually improve, for the purpose of  answering the business question the exact score was not important, but instead the difference between each competitor and compared to others how does GRTA score.  Comparing each company by metric was the main influence behind the design on CAN’s graph.

The Original Graph

 

 

The CAN Graph

- In the original graph, the bold vertical lines focus the viewer how each metric scored, by encouraging the eyes to go up and down.  In the CAN graph, the light gray horizontal lines encourage the eyes to travel left and right to compare each companies performance.  Also, we used light gray lines so that we did not dominate the graph with supporting data.

- In the original graph, there is no simple way to show the spread between the different competitors, besides comparing each line together.  However, it important to know how competitive each metric is when answering the business question.  When designing the CAN Graph, we darkened a length of the light gray horizontal lines to show the minimum and maximum score on the service quality index.  This

- In the original graph, using four different colors made it difficult to make a memorable distinction between each company, take up an unnecessary amount of space, and impossible for color blind (10% of males) to make distinctions.  Using different shades of gray CAN made it easy for everyone, including the colorblind, to distinguish between different companies.  In addition to adding an additional way to differentiate between companies, using different shapes allowed for better distinction when multiple companies score close to each other.

- In the original graph, the overall low graphical quality such as broken vertical lines, faded colors and pixilated font created an unnecessary distraction, and reduce the credibility of the results.  While this might seem petty, producing graphs that are crisp and well designed help develop trust with the audience.  In the CAN Graph, we produced the entire graph in black and white, so that the report can easily be reproduced on either a color or black and white printer.

Dashboard Design: Teaching Strategic & Analytical Thinking

Posted by: Jefferson on November 28th, 2011 Leave a Comment

At CAN, we exist to provide our clients with leading edge methodologies that are both effective and easy to use.  This requires that we constantly learn about new tools and techniques, and hone a fine edge on the ones we keep to provide to our clients.

Previously on our blog, we have discussed the application of dashboards and aspects of dashboard design that facilitate rapid perception by the human brain.  How about using dashboards as a way to teach users a way of thinking?  In this blog, we will discuss using dashboards to promote strategic thinking through guided analysis.

One of our clients approached CAN with the following predicament.  Their enterprise operates nationwide with several districts responsible for operations within their unique geographic region.  Every year, the strategic planning division would produce a thick binder reviewing each districts market forecasts, opportunities, and past performance.  The intent was to assist the non-technical managers and business development of each district to think about trends in the market and industry to get more sales.  Although very well produced and full of useful information, these binders acted mostly as a reference and did little to encourage analysis by the end-user.

Our solution was to use the same information used to build the binders and create views using Tableau.  At first, these views replicated the familiar visualizations found in binders with an added level of interaction.  Then, we started to add new data sources into the existing information.  We connected industry forecasts, census data, economic indicators, past performance and connected all this functionality to a dashboard where the end user is able to bring in these factors at their command.  Populating the dashboard with the raw materials required for analysis, is the first stage.

The second stage is defining the business questions that the users need to answer to run their business.  We interviewed the executives on the strategic planning team and in several of the district offices to define what the most important business questions they needed to answer to run their business.  Instead of providing managers of each district with binders that pushed facts and figures at them, we created a work book of questions that needed to be answered and how the answers could be applied to running their district.

The third stage is doing most, not all, of the users’ work for them.  What I mean by this is producing dashboards that are 90% completed for the types of questions the user will want to answer.  Our goal is to support the user in asking questions and getting answers, not simply handing them the answers or making them build their own dashboards.  So, we build pre-made views for them.  For example, one aspect of our client’s business functions was closely related to population growth.  We produced a dashboard that integrated population growth figures for the past several years with our client’s historical sales figures and billable hours.  The district manager, interested in staffing requirements, can population changes across the region with his current staffing and identify where adjustments and hiring are likely to take place.

In designing guided analysis, the bottom line is producing dashboards that solve the business question that users need to answer.  This requires that the designers understand the purpose of each dashboard, how it will be used, and what the user intends to get out of it.  If your goal is to achieve data-driven decisions from non-technical managers, you must design so that the user is on the right track with the controls, but ultimately require their interaction and thinking to reach the outcome.

Dashboard Design: Design for Parallel Processing

Posted by: Grant Stanley on September 12th, 2011 2 Comments

The value of dashboards and visualizations are that they allow users to shift from serial to parallel processing.  When reading a block of text you can only process the information serially by starting at the top left of the text and finishing the bottom right. Dashboards and data visualizations allow you to absorb information in parallel making it easier to absorb information quickly, identify relationships and trends.

However, the lack of serial processing requires that dashboards be effectively designed so that information can be absorbed as easily as possible.  This requires that dashboard be designed for pre-attentive processing or for “the unconscious accumulation of information from the environment” (Wikipedia).  Pre-attentive processing is specifically designed for parallel processing.  Pre-attentive processing allowed our ancestors to continually scan the horizon to identify opportunities and threats.  If well designed, a dashboard is modern-day equivalent of the horizon of the savanna, a data rich experience where it is easy to absorb the most important information, identify relationships and spot trends.

The basic principle of designing a pre-attentive dashboard that enables parallel processing is to keep element natural.  Replace bright bold colors with neutral and natural hues, and pie charts, gauges and traffic lights with hue, intensity, location, orientation, line length, line width, size, shape, added marks, enclosure, and motion.

Three Types of Dashboards

Posted by: Grant Stanley on September 9th, 2011 2 Comments

A dashboard is a single display that in a glance provides essential information for a specific objective.  Since you are limited to a single display capable of being monitored at a glance, the first step of dashboard design is to select the purpose of your dashboard.  This provide you a filter to make sure that your dashboard effectively accomplishes its intended purpose.  Will it be strategic, analytical or operational?  Answering this question will keep your dashboard from falling victim to trying to be everything to everyone.

Strategic dashboards provide managers and executives at all levels of the organization the information they need understand the health of the organization and help identify potential opportunities for expansion and improvement.  Strategic dashboards do not provide all the detailed information needed to make complex decisions, but instead help executives identify opportunities for further analysis.  A strategic dashboard should be simple and contain aggregate metrics the represent the over all health of the organization.  Typically there is no need for interactive features and the data should be updated no more than monthly.

Analytical dashboards provide users with the data they need to understand trends and why certain things are happening by making comparisons across time and multiple variables.  Analytical dashboards often contain more information per square inch than both strategic and operational dashboards.  Since understanding is the goal analytical dashboards can be more complex than strategic or operations dashboards.  Also, while analytical dashboards should facilitate interactions with the data, including viewing the data in increasing detail, it is important to maintain the ability to compare data across time and multiple variables.  If you lose the ability to compare data then an analytical dashboards is no longer able to accomplish the goal of allowing users to understand trends and why things are happening.

Operational dashboards are used to monitor real time operations and alert the users to deviations for the norm.  This often means that operational dashboards need to be updated frequently if not in real time, contain less information than analytical or strategic dashboards, and make it nearly impossible to avoid or misunderstand an alert when something deviates from the acceptable standards.  Operational dashboards should provide users with specific alerts and provide them with exactly what information they need to quickly get operations back to normal.