Monthly Archives: April, 2014

An Introduction to Data Blending – Part 5 (Tableau’s Data Blending Architecture)

Readers:

In Part 4 of this series on data blending, we reviewed Tableau’s Data Blending Principles. We also reviewed an example of data blending in Jonathan Drummey’s Patient Falls Dashboard. [3]

Today, in Part 5 of this series, we will peel the onion a bit more and look at Tableau’s Data Blending Architecture.

Again, much of Parts 1 – 5 are based on a research paper written by Kristi Morton from The University of Washington (and others) [1].

You can learn more about Ms. Morton’s research as well as other resources used to create this blog post by referring to the References at the end of the blog post.

Best Regards,

Michael

Integrating Data in Tableau

In Part 5, we discuss in greater detail how data blending works. Then we discuss how a user builds visualizations using data blending using several large datasets involving airline statistics.

Data Blending Architecture

Part 5 - Figure 1

The data blending system, shown in Figure 1 above, takes as input the VizQL query workload generated by the user’s GUI actions and data source schemas, and automatically infers how to query the data sources remotely and combine their results on-the-fly. The system features a two-tier mediator-based architecture in which the VizQL query workload is analyzed and partitioned at runtime based on the corresponding data source fields being used. The primary mediator initiates this process by removing the visual encodings from the VizQL query workload to yield an abstract query. The abstract query is partitioned for further processing by the primary mediator and one or more secondary mediators. The primary mediator creates the mediated schema for the given query workload. It then federates the abstract queries to the primary data source as well as the secondary mediators and their respective data sources. The wrappers compile the abstract queries into concrete SQL or MDX queries and instantiate the semantic mappings between the data sources and the mediated schema for each query. The primary mediator joins all the result sets returned from all data sources to produce the mediated result set used by the rendering system. [1]

Part 5 - Figure 2

Post-aggregate Join

A visualization is organized by its discrete fields into pages, partitions, colors, etc., and like a GROUP BY clause in SQL, these grouping fields comprise the primary key of the visualization. In a blended visualization, the grouping fields from the primary data source become the primary key of the mediated schema. In Figure 2 above, these are shown as the dark-green fields in the primary data source, and the light green fields represent the aggregated data. Each secondary data source must contain at least one field that matches a visualization grouping field in order to blend into the mediated schema. The matching fields in a secondary data source comprise its join key, and fields appear in the GROUP BY clause issued by the secondary mediator wrappers. The aggregated data from the secondary data source, shown in light-purple, is then left-joined along its join key into the mediated result set.  Morton (et al) refer to this left-join of aggregated result sets as a post-aggregate join. [1]

Primary Key Cardinality

many mapping between the domain values of the primary key and those of the secondary join key, because the secondary join key is a subset of the primary key and contains only unique values in the aggregated secondary result set. Morton (et al) find that this approach is the most natural for augmenting a visualization with secondary data sources of uncertain value or quality, which is a common scenario for Tableau users.

Data blending supports many-to-one relationships between the primary and each secondary. This can occur when the secondary data source contains coarser-grained data than the mediated result set, as discussed in Part 3 of this series.

Since the join key in a secondary result set may match a subset of the blended result set primary key, portions of the secondary result set may be duplicated across repeated values in the mediated result set. This does not pose risk of double-counting measure values, becaused all aggregation is performed prior to the join. When a blended visualization uses multiple secondary data sources, each secondary join key may match any subset of the primary key. The primary mediator handles duplicating each secondary result set as needed to join with the mediated result set.

Finally, a secondary dimension which is not part of the join key (and thus not a grouping field in the secondary query) can still be used in the visualization. If it is functionally dependent on the join key, a secondary dimension can be used without affecting the result set cardinality. Tableau references this kind of non-grouping dimension using both MIN and MAX aggregations in the query issued to the secondary data source, which allows Tableau to determine if the dimension is functionally dependent on the join key. For each row in the secondary result set, if the two aggregated values are the same then the value is used as-is, reflecting the functional dependence on the grouping fields. If the aggregated values differ, Tableau represents the value using a special form of NULL called ManyValues. This is represented in the visualization as a ‘*’, but retains the behavior of NULL
when used in calculated fields or other computations. The visual feedback allows a user to distinguish this lack of data from the NULLs which occur due to missing or mismatched data.

Inferring Join Keys

Tableau uses very simple rules for automatically detecting candidate join keys:

  1. The secondary data source field name must match a field with the same name in the primary data source.
  2. The data types must match
  3. If they are date/time fields, they must represent the same granularity date bin in the date/time hierarchy, e.g. both are MONTH. A user can intervene to force a match either by providing field captions to rename fields within the Tableau data model, or by explicitly defining a link between fields using a simple user interface.

Part 5 - Figure 3

Another Simple Blending Example

A Tableau data blending scenario is shown in Figure 3 above, which includes multiple views that were composed in minutes by uniquely mashing up four different airline datasets, the largest of which include a 324 million row ticket pricing database and a 140 million row on-time performance database. A user starts by dragging fields from any dataset on to a blank visual canvas, iteratively building a VizQL statement which ultimately produces a visualization. In this example, the user first drags the VizQL fields, YEAR(Flight Date) and AVG(Airfare), from the pricing dataset onto the visual canvas.

Data blending occurs when the user adds fields from a separate dataset to an existing VizQL statement in order to augment their analysis. Tableau assigns the existing dataset to the primary mediator and uses secondary mediators to manage each subsequent dataset added to the VizQL. The mediated schema has a primary key composed of the grouping VizQL fields from the primary dataset (e.g. YEAR(Flight Date)); the remaining fields in the mediated schema are the aggregated VizQL fields from the primary dataset along with the VizQL fields from each secondary dataset.

Continuing our example, the user wishes to drag AVG(Total Cost per Gallon) from the fuel cost dataset to the visualization. The schema matching algorithm examines
the secondary dataset for one or more fields whose name exactly matches a field in the primary key of the mediated schema. While the proposed matches are often sufficient and acceptable, the user can specify an override. Since the fuel cost dataset has a field named Date, the user provides a caption of Flight Date to resolve the schema discrepancy. At this point the mediated schema is created and the VizQL workload is then federated to the wrappers for each dataset. Each wrapper compiles VizQL to SQL or MDX for the given workload, executes the query, and maps the result set into the intermediate form expected by the primary mediator.

The mapping is performed dynamically, since both the VizQL and the data model evolve during a user’s iterative analytical workflow. Finally, the primary mediator
performs a left-join of each secondary result set along the primary key of the mediated schema. In this example, the mediated result set is rendered to produce the visualization shown in Figure 3(a).

Evolved Blending Example

Figure 3(b) above shows further evolution of the analysis of airline datasets, and demonstrates several key points of data blending. First, the user adds a unique ID field named unique carrier from the primary dataset to the VizQL to visualize results for each airline ID over time. The mediated schema adapts by adding this field to its primary key, and the secondary mediator automatically queries the fuel cost dataset at this finer granularity since it too has a field named uniquecarrier. Next, the user decorates the visualization with descriptive airline names for each airline ID by dragging a field named Carrier Name from a lookup table.

This dataset is at a coarser granularity than the existing mediated schema, since it does not represent changes to the carrier name over time. Morton’s (et al) system automatically handles this challenge by allowing the left-join to use a subset of the mediated result set primary key, and replicating the carrier name across the mediated result set. Figure 4 below demonstrates this effect using a tabular view of a portion of the mediated result set, along with portions of the primary and secondary result sets.

The figure also demonstrates how the left-join preserves data for years which have no fuel cost records. Last, the user adds average airline delays from a 140 million row dataset which matches on Flight Date and uniquecarrier. This is a fast operation, since the wrapper performs mapping operations on the relatively small, aggregated result set produced by the remote database. Note that none of these additional analytical tasks required the user to intervene in data integration tasks, allowing their focus to remain on finding insight in the data.

Part 5 - Figure 4

Filtering

Tableau provides several options for filtering data. Data may be filtered based on aggregate conditions, such as excluding including airlines having a low total count of flights. A user can filter aggregate data from the primary and secondary data sources in this fashion, which results in rows being removed from the mediated result set. In contrast, row level filters are only allowed for the primary data source. To improve performance of queries sent to the secondary data sources, Tableau will filter the join keys to exclude values which are not present in the domain of the primary data source result set, since these values would be discarded by the left-join.

Data Cleaning Capabilities

As mentioned in the Inferring Join Keys section above, Tableau supports user intervention in resolving field names when schema matching fails. And once the schemas match and data is blended, the visualization can help provide feedback regarding the validity of the underlying data values and domains. If there are any data inconsistencies, users can provide aliases for a field’s data values which will override the original values in any query results involving that field. The primary mediator performs a left-join using the aliases of the data values, allowing users to blend data despite discrepancies from data entry errors and spelling variations. Tableau provides a simple user interface for editing field aliases. Calculated fields are another aspect of Tableau’s data model which support data cleaning. Calculated fields support arbitrary transformations of original data values into new data values, such as trimming whitespace from a string or constructing a date from an epoch-based integer timestamp.

As with database fields, calculated fields can be used as primary keys or join keys.

Finally, Tableau allows users to organize a field’s related data values into groups. These ad-hoc groups can be used for entity resolution, such as binding multiple variations of business names to a canonical form. Ad-hoc groups also allow constructing coarser-grained structures, such as grouping states into regions. Data blending supports joins between two ad-hoc groups, as well as joins between an ad-hoc group and a string field.

 Next: Data Blending Using MicroStrategy

———————————————————————————-

References:

[1] Kristi Morton, Ross Bunker, Jock Mackinlay, Robert Morton, and Chris Stolte, Dynamic Workload Driven Data Integration in Tableau, University of Washington and Tableau Software, Seattle, Washington, March 2012, http://homes.cs.washington.edu/~kmorton/modi221-mortonA.pdf.

[2] Hans Rosling, Wealth & Health of Nations, Gapminder.org, http://www.gapminder.org/world/.

[3] Jonathan Drummey, Tableau Data Blending, Sparse Data, Multiple Levels of Granularity, and Improvements in Version 8, Drawing with Numbers, March 11, 2013, http://drawingwithnumbers.artisart.org/tableau-data-blending-sparse-data-multiple-levels-of-granularity-and-improvements-in-version-8/.

 

The Lean UX Manifesto – Principle Driven Design

Lean UX Manifesto

Who Wrote The Manifesto?

This was a joint effort between Anthony Viviano, Ajay Revels and Ha Phan.

Anthony and Ajay work at a large financial institution and are trying to apply Lean UX within their enterprise. Ha works at a startup called Porch.

Why a Manifesto?

So, why a manifesto? Anthony, Ajay and Ha were inspired by the Agile Manifesto. Anthony stated that it is simple and to the point. It’s not a list of rules, but a value statement that can be used to guide you through a project or an organizational change. It’s tempting to lay down rules. As if to say, “this list of methods are required to practice Lean UX. Check these boxes in your process and you can brand this a Lean UX project.” We don’t like rules. We prefer principles that drive the methods needed.

Lean UX applies well to uncertainty, but not everything is uncertain. You may know your customer, so you can breeze through customer development. Or, you may already have a design, so a design studio is not needed.In addition to their anti-rules stance, there’s another reason why a manifesto makes sense. Anthony heard a few practitioners say that only a startup can apply this process in its purest form. While that might be true, enterprise entrepreneurs (a.k.a. intraprenuers) shouldn’t be excluded from this great thinking. We can take advantage of it by doing what we can to customize it to our unique culture and structure.

Anthony, Ajay and Ha hope you allow these values to guide you through your Lean UX journey.

The Lean UX Manifesto

Anthony, Ajay and Ha are developing a way to create digital experiences that are valued by our end users. Through this work, we hold in high regard the following:

  • Early customer validation over releasing products with unknown end-user value
  • Collaborative design over designing on an island
  • Solving user problems over designing the next “cool” feature
  • Measuring KPIs over undefined success metrics
  • Applying appropriate tools over following a rigid plan
  • Nimble design over heavy wireframes, comps or specs

As stated in the Agile Manifesto, “While there is value in the items on the right, we value the items on the left more.”

How The Lean UX Manifesto Works

Let’s take each of these in turn and see how we can follow the principles of lean UX.

Early Customer Validation Over Releasing Products With Unknown User Value

What if you worked at a company where usability testing just wasn’t done? Unfortunately, this is the sad state in which many of our fellow UX practitioners find themselves. How, then, do they follow the principles of lean UX?

With usability testing, we seek customer validation or early failure. Customer validation may be sought through other means as well. For example, does your company gather feedback from users? If that feedback is circulated, are you on the list of people who receive it?

Here are other sources of learning about customer needs:

  • Customer service representatives
    Their focus is on helping customers overcome experience issues. Try to speak to them regularly. They are likely documenting their calls, so see whether you can create some system for tagging experience issues that you can follow up on.
  • Sales representatives
    This is another group that is focused on the customer. They will understand what customer problems are out there to be solved. They’ll also know which features are important and which innovations customers want.
  • Website search data
    This is an invaluable source of customer desires. Search data can uncover website navigation problems and features or problems that customers are looking for.

Salespeople and customer service reps can be great sources of customer needs.
Salespeople and customer service representatives are great sources of learning about customer needs. (Image: Renato Ganoza)

Collaborative Design Over Designing on an Island

Design should not be a solo exercise. Being a design team of one is no excuse. Anthony uses the design studio process and adopt the role of facilitator. Gather team members who own a piece of the project, and host a design studio workshop. Include at least the following people (adjusting to suit your unique organization):

  • Domain owner
    Your subject matter expert
  • Requirements Owner
    A business analyst or the person who gathers and writes the requirements
  • Data provider
    A data analyst on hand who is familiar with the analytics and can pull the info you need
  • Technology owner
    A developer, someone who understands the technology constraints and design patterns
  • Product or business owner
    A product manager or the person who owns this piece of business
  • Designer
    The UX or visual designer or person who owns the design and can facilitate the design studio
  • Researcher
    The usability analyst or UX researcher or person who owns customer development and persona creation

Solving User Problems Over Designing the Next Cool Feature

When you’re handed a requirements document, a thought-out solution, a feature, a brief or whatever artifact you receive to inform your work, begin by asking, “What problem are we trying to solve?” Ideally, you should clearly understand the customer’s problem. Design is problem-solving, so if you don’t know the problem, you can’t design a solution. If you do this enough, then the stakeholders will understand that you’re more than just a wireframe jockey. You’re a professional problem-solver with a system for creating solutions that make sense.

Measuring KPIs Over Undefined Success Metrics

You can’t measure success if you aren’t… er, measuring. Avoid vanity metrics. Anthony loves Dave McClure’s pirate metrics:

  • Acquisitions
    Users come to the website from various channels.
  • Activation
    Users enjoy their first visit (a “happy” user experience).
  • Retention
    Users come back, visiting multiple times.
  • Referral
    Users like the product enough to refer others.
  • Revenue
    Users conduct some monetization behavior.

Applying Appropriate Tools Over Following a Rigid Plan

Lean UX should be a flexible process. As Anthony started to develop the process steps for one cycle, he found himself overwhelmed with the number of tools being recommended. Anthony’s advice, similar to what he had said when creating a minimum viable product, is to apply the minimum tools required to get you to “pivot” or “persevere.”

Here are a few tools that Anthony found useful (not an exhaustive list):

  • provisional personas, right sized for the effort;
  • persona map (which we learned from Menlo Innovations);
  • assumptions, with the riskiest identified;
  • design studio;
  • paper prototyping in early stages;
  • digital prototyping (HTML preferred) in later stages;
  • guerilla design assessment (a better name for usability testing);
  • co-location wherever possible.

The design studio method is popular for collaborative design
The design studio method is popular for collaborative design. (Image: visualpun.ch)

Everything else should be applied as it makes sense. For example, if more customer development is needed, then take the time to interview as a team and to internalize customer needs. The lean startup world has no shortage of tools. Use only the ones that make sense to your project and that get you to a validated solution faster.

Nimble Design Over Heavy Wireframes, Comps or Specs

The goal is to release a product. Once it’s released, users won’t interact with the wireframes or requirements document as part of the product. They will interact with the product itself. So, try to spend less time on your design artifacts.

How can you lighten your wireframes?

  • Lighter annotations and more presentation
    Anthony found that if I take the time to present my unfinished wireframes to stakeholders, He would get valuable feedback sooner and save time.
  • Co-design
    If developers, quality assurance testers and business analysts are involved in the design, then they will share ownership and internalize the annotations. When this happens, you can pass off sketches as wireframes because team members will already understand the interactions.
  • Paper prototypes
    These serve a dual purpose. They get you to design validation (i.e. usability testing) sooner, but they also demonstrate the interactions. No need to write detailed wire annotations if the user can see the interactions firsthand.

It’s All About Principle-Driven Design

This all boils down to something that I call principle-driven design. As stated, some lean UX is better than none, so applying these principles as best you can will get you to customer-validated, early-failure solutions more quickly. Rules are for practitioners who don’t really know the value of this process, while principles demand wisdom and maturity.

By allowing principles to drive you, you’ll find that you’re more nimble, reasonable and collaborative. Really, you’ll be overall better at getting to solutions. This will please your stakeholders and team members from other disciplines (development, visual design, business, etc.). To quote the late Stephen Covey:

“There are three constants in life: change, choice and principles.”

——————————————————

Sources:

[1] Anthony Viviano, Ajay Revels and Ha Phan, The Lean UX Manifesto, http://www.leanuxmanifesto.com/.

[2] Anthony Viviano, The Lean UX Manifesto: Principle-Driven Design, Smashing magazine, January 8, 2014, http://www.smashingmagazine.com/2014/01/08/lean-ux-manifesto-principle-driven-design/.

 

Tips & Tricks #10: How to Remove the Underlining from Hyperlinks Created on an Attribute or Metric

This is another one of those little tricks that can save you from pulling your hair out trying to figure it out.

By default, hyperlinks in a Report Services document are underlined. The underlining can not be removed by changing the format of the attribute or metric.

Tip 10-1

There is an Enhancement Request open with MicroStrategy on this, but MicroStrategy does currently provide a workaround.

Workaround to Remove the Underlining is to Clear the Default Link

Step 1: Right click the object that is linked and choose Edit Links.

Step 2: Click the Clear Default button while the link is highlighted.

Tip 10-2

 

Step 3: The underlining should now be removed as shown below.

Tip 10-3

CAVEAT: The only limitation to this workaround is that in order to now get to the link, the object has to be right clicked on. The user can not directly click on the object and be taken directly to the new page (see screenshot below).

Tip 10-4

Charles Apple: Two Recent Infographic Fails You Ought to Know About

Readers:

Charles AppleI have been a big fan of Charles Apple’s work for a long time. I have blogged about him and his work in the past (see “Charles Apple” in my Categories on the right or do a search for “Charles Apple” on my blog).

Charles Apple (photo, right) is a longtime news artist, graphics reporter, designer, editor and blogger. The former graphics director of the Virginian-Pilot and the Des Moines Register, he spent five years as an international consultant and instructor. Currently, he’s Focus page editor of the Orange County Register.

I always like to reshare articles and blogs about what NOT to do in regards to data visualization and infographics. This morning, Mr. Apple posted a blog entry titled “Two recent infographic fails you ought to know about.” Charles has always shown a keen eye for detail and accuracy. He is also very reflective of his own work as today’s blog entry shows.

I hope you enjoy Mr. Apple’s thoughts as much as I do.

Have a great Good Friday and Happy Easter.

Best Regards,

Michael

Source: Charles Apple, Two recent infographic fails you ought to know about, http://www.charlesapple.com, April 18, 2014, http://www.charlesapple.com/2014/04/two-recent-infographic-fails-you-ought-to-know-about/.

Two recent infographic fails you ought to know about

A couple of charting debacles popped up this week of which you might want to take note.

POSITIVE VS. NEGATIVE SPACE

First, Reuters moved this fever chart showing the number of gun deaths in Florida going up after the state enacted its “stand your ground” law in 2005.

Just one little problem: The artist — for some unknown reason — elected to build the chart upside down from the usual way a fever chart is drawn.1404GunDeaths01Meaning the chart appears to show the number of gun deaths going down… if you focus on the white territory and consider the red to be the background of the chart.

After a lively discussion on a number of forums — most notably at Business Insider — a reader volunteered to flip the chart right-side around for clarity’s sake.1404GunDeaths02Is that better? Most folks seem to think it is.1404GunDeaths03Three important rules about infographics that I’m making up right here:

Rule 1: A graphic must be clear. If it’s not clear, then it’s not doing its job and should probably be put out of its misery.

Rule 2: It’s OK for a graphic to offer the reader a longer, more complicated view that requires more time spent observing a piece. But that’s not typically the job of a freakin’ one-column graphic.

Rule 3: Occasionally, it’s OK to flip a graphic upside down. But you’d better have a damned good reason for doing it. Other than, y’know, “I thought it’d look cool.”

This graphic fails all three: It’s not immediately clear — at least to many readers — and it’s a small graphic. So it has no business getting fancy. If the artist had a reason for turning it upside down, that reason eludes me.

Read more about the debate over this piece at…

UPSIDE DOWN YOU’RE TURNING ME

Full disclosure: I feel a little guilty criticizing this piece because I myself did something funky last week: I turned a map upside down:Unnamed_CCI_EPS

That ran in the middle of a page about John Steinbeck‘s the Grapes of Wrath. The intent was to show the route the fictional Joad family took in the book from the dust bowl of Oklahoma to what they hoped would be a better life here in Southern California.But vI really wanted to get those two pictures in there, which needed to read from left to right. I wanted those to sit atop my map showing the journey. I tried mapping it the usual way, but it was difficult to get the reader to stop — and then read this one segment of my page from right to left — and then resume reading the rest of the page from left to right.This would take quite a bit more vertical space and some very careful use of labels. And I was plum out of vertical space.So I elected to flop the map upside down. My logic: This time, it was more important to follow the narrative — to feel the twists and turns in the Joads’ journey — than to take in the geographical details of the trip. If the upside-down map was vetoed, Plan B would have been to kill the map and run the list of cities in a timeline-like format. There was just one problem with that: I already had a timeline on the page, just above the map:

Unnamed_CCI_EPS

We debated this and decided I was right to flip the map — This time. I can’t imagine too many times we’d ever want to run a map with the north arrow pointing down.And, y’know, perhaps we did the wrong thing. Another editor might have made a different choice.But the point is: We made a conscious decision here to let the map support the narrative. I don’t know what point Reuters was making with its upside-down fever chart. Whatever it was, it’s not apparent to me.It’s OK to make unusual choices. Just make sure your data is clear, your story is clear and readers don’t walk way from your piece puzzled as hell.

WHEN IS A MAP NOT A MAP?

This seems like a good time to present the other infographics debacle this week: This one is by NBC News.1404DemographicsOh, dear. I was just talking about using a map when the map wasn’t the most important element.What we have here is another fever chart, but this one has been pasted inside a map of the U.S. This has a number of effects that harm the greater good we do by presenting the data in the first place:Fever charts (and pie charts and bar charts and most other charts, for that matter) are all about showing proportions. If the proportions get screwed up — by, say, varying the widths of your bars or by covering up part of the chart — then the reader can’t make the visual comparisons you’re asking her to make.And that’s the case here: We see territory marked as “Asian” in the upper left of the chart and also at the upper right. But where is that set of data in 2010? I’m guessing it’s there, but it’s hidden outside the area of the map.

Rule 4: If you’re going to hide important parts of your chart, then your chart is no good. And, yes, it should be put out of its misery.

The data is displayed over a map. What is the artist trying to tell us? Where white people live in the U.S.? That Hispanics only live near Canada and Asians in Washington State and New England?No, the map is merely a decorative element. It has nothing at all to do with the data.

Rule 5: If you don’t need an element to tell your story, then eliminate it. Or I will.

Rule 6: If your decorative element gets in the way of your story, then not only do I demand you eliminate it, I also insist you come over here so I can smack you upside your head.

Rule 7: Don’t use a map if you’re not telling a story that includes some type of data that needs geographical context.

Oh, and don’t forget this last one:

Rule 8: Don’t tilt a map or turn it upside down. Not unless you have a good reason.

Go here to read more about the perils of rotating maps.

Fast Company: Is Flat Design Already Passé?

Source: John Brownlee, Is Flat Design Already Passé?, Fast Company, Co.DESIGN, April 11, 2014, http://www.fastcodesign.com/3028944/is-flat-design-already-passe.

Skeuomorph CalendarBlog Note: A skeuomorph is a derivative object that retains ornamental design cues from structures that were necessary in the original. Examples include pottery embellished with imitation rivets reminiscent of similar pots made of metal and a software calendar that imitates the appearance of binding on a paper desk calendar (see image to the right).

Over the last few years, we’ve seen an upheaval in the way computer interfaces are designed. Skeuomorphism is out, and flat is in. But have we gone too far? Perhaps we’ve taken the skeuomorphic death hunts as far as they can go, and its high time we usher in a new era of post-flat digital design.

John Brownlee

John Brownlee

Ever since the original Macintosh redefined the way we interact with computers by creating a virtual desktop, computer interfaces have largely been skeuomorphic by mimicking the look of real-world objects. In the beginning, computer displays were limited in resolution and color, so the goal was to make computer interfaces look as realistic as possible. And since most computer users weren’t experienced yet, skeuomorphism became a valuable tool to help people understand digital interfaces.

But skeuomorphism didn’t make sense once photo-realistic computer displays became ubiquitous. Computers have no problem tricking us into thinking that we’re looking at something real so we don’t need to use tacky design tricks like fake stitching or Corinthian leather to fool us into thinking our displays are better than they are. In addition, most people have grown up in a world where digital interfaces are common. UI elements don’t have to look like real-world objects anymore for people to understand them.

This is why Jony Ive took over the design of Apple’s iOS and OS X operating systems and began a relentless purge of the numerous skeuomorphic design elements that his predecessor, skeuomaniac Scott Forstall, created. To quote Fast Company’s own John Pavlus, “skeuomorphism is a solution to a problem that iOS no longer has,” and that’s true of other operating systems and apps too. Google, Microsoft, Twitter, Facebook, Dropbox, even Samsung, they’re all embracing flat design, throwing out the textures and gradients that once defined their products, in favor of solid hues and typography-driven design.

This is, without a doubt, a good thing. Skeuomorphism led to some exceedingly one-dimensional designs, such as iOS 6’s execrable billiard-style Game Center design. But in an excellent post, Collective Ray designer Wells Riley argues that things are going too far.

Flat design is essentially as far away from skeuomorphism as you can get. Compare iOS 7’s bold colors, unadorned icons, transparent overlays, and typography-based design to its immediate precessor, iOS 6. Where once every app on iOS had fake reflections, quasi-realistic textures, drop shadows, and pseudo-3-D elements, iOS 7 uses pure colors, no gradients, no shadows, and embraces the 2-D nature of a modern smartphone display. But while flat design has made iOS 7 look remarkably consistent, it has also removed a lot of personality from the operating system. It’s like the Gropius house, when the old iOS 6 was a circus funhouse. Maybe it needs to get a little bit of that sense of madness back.

Here’s how Riley defines elements of a post-flat interface:

• Hierarchy defined using size and composition along with color.

• Affordant buttons, forms, and interactive elements.

• Skeuomorphs to represent 1:1 analogs to real-life objects (the curl of an e-book page, for example) in the name of user delight.

• Strong emphasis on content, not ornamentation.

• Beautiful, readable typography.

 

Riley’s argument is that flat design has allowed digital designers to brush the slate clean in terms of how they approach their work, but it has also hindered a sense of wonder and whimsy. Software should still strongly emphasizes content, color, and typography over ornamentation, but why is, say, the curl of a page when you’re reading an e-book such a crime, when it so clearly gives users delight?

“Without strict visual requirements associated with flat design, post-flat offers designers tons of variety to explore new aesthetics—informed by the best qualities of skeuomorphic and flat design.” Riley writes. “Dust off your drop shadows and gradients, and introduce them to your flat color buttons and icons. Do your absolute best work without feeling restricted to a single aesthetic. Bring variety, creativity, and delight back to your interfaces.”

Maybe Riley has a point. Why should mad ol’ Scott Forstall be allowed to ruin skeuomorphism for everyone? With the lightest of brush strokes, skeuomorphism can be used to bring back a sense of personality and joy to our apps. For those of us growing listless in the wake of countless nearly identical “flat” app designs, he makes a good point. It is time for the pendulum towards flat and away from skeuomorphism to swing back, if only a little bit.

An Introduction to Data Blending – Part 4 (Data Blending Design Principles)

Readers:

In Part 3 of this series on data blending, we examining the benefits of blending data. We also reviewed an example of data blending that illustrated the possible outcomes of an election for the District 2 Supervisor of San Francisco.

Today, in Part 4 of this series, we will discuss data blending design principles and show another illustrative example of data blending using Tableau.

Again, much of Parts 1, 2, 3 and 4 are based on a research paper written by Kristi Morton from The University of Washington (and others) [1].

You can learn more about Ms. Morton’s research as well as other resources used to create this blog post by referring to the References at the end of the blog post.

Best Regards,

Michael

Data Blending Design Principles

In Part 3, we describe the primary design principles upon which Tableau’s data blending feature was based. These principles were influenced by the application needs of Tableau’s end-user. In particular, we designed the blending system to be able to integrate datasets on-the-fly, be responsive to change, and driven by the visualization. Additionally, we assumed that the user may not know exactly what she is looking for initially, and needs a flexible, interactive system that can handle exploratory visual analysis.

Push Computation to Data and Minimize Data Movement

Tableau’s approach to data visualization allows users to leverage the power of a fast database system. Tableau’s VizQL algebra is a declarative language for succinctly describing visual representations of data and analytics operations on the data. Tableau compiles the VizQL declarative formalism representing a visual specification into SQL or MDX and pushes this computation close to the data, where the fast database system handles computationally intensive aggregation and filtering operations. In response, the database provides a relatively small result set for Tableau to render. This is an important factor in Tableau’s choice of post-aggregate data integration across disparate data sources – since the integrated result sets must represent a cognitively manageable amount of information, the data integration process operates on small amounts of aggregated, filtered data from each data source. This approach avoids the costly migration effort to collocate massive data sets in a single warehouse, and continues to leverage fast databases for performing expensive queries close to the data.

Automate as Much as Possible, but Keep User in Loop

Tableau’s primary focus has been on ease of use since most of Tableau’s end-users are not database experts, but range from a variety of domains and disciplines: business analysts, journalists, scientists, students, etc. This lead them to take a simple, pay-as-you-go integration approach in which the user invests minimal upfront effort or time to receive the benefits of the system. For example, the data blending system does not require the user to specify schemas for their data sets, rather the system tries to infer this information as well as how to apply schema matching techniques to blend them for a given visualization. Furthermore, the system provides a simple drag-and-drop interface for the user to specify the fields for a visualization, and if there are fields from multiple data sources in play at the same time, the blending system infers how to join them to satisfy the needs of the visualization.

In the case that something goes wrong, for example, if the schema matching could not succeed, the blending system provides a simple interface for specifying data source relationships and how blending should proceed. Additionally, the system provides several techniques for managing the impact of dirty data on blending, which we discuss in more in Part 5 of this series.

Another Example: Patient Falls Dashboard [3]

NOTE: The following example is from Jonathan Drummey via the Drawing with Numbers blog site. The example uses Tableau v7, but at the end of the instructions on how he creates this dashboard in Tableau v7, Mr. Drummey includes instructions how the steps became more simplied in Tableau v8. I have included a reference to this blog post on his site in the reference section of my blog entry. The “I”, “me” voice you read in this example is that of Mr. Drummey.

As part of improving patient safety, we track all patient falls in our healthcare system, and the number of patient days – the total of the number of days of inpatient stays at the hospital. Every month report we report to the state our “fall rate,” a metric of the number of falls with injury for certain units in the hospital per 1000 patient days, i.e. days that patients are at the hospital. Our annualized target is to have less than 0.7 falls with injury per 1000 patient days.

A goal for our internal dashboard is to show the last 13 months of fall rates as a line chart, with the most recent fall events as a bar chart, in a combined chart, along with a separate text table showing some details of each fall event. Here’s the desired chart, with mocked-up data:

 

combo bars and lines

On the surface, blending this data seems really straightforward. We generate a falls rate very month for every reporting unit, so use that as the primary, then blend in the falls as they happen. However, this has the following issues:

  • Sparse Data – As I’m writing this, it’s March 7th. We usually don’t get the denominator of the patient days for the prior month (February) for a few more days yet, so there won’t be any February row of measure data to use as the primary to get the February fall events to show on the dashboard. In addition, there still wouldn’t be any March data to get the March fall events. Sometimes when working with blend, the solution is to flip our choices for the primary and secondary datasource. However, that doesn’t work either because a unit might go for months or years without a patient fall, so there wouldn’t be any fall events to blend in the measure data.
  • Falls With and Without Injury – In the bar chart, we don’t just want to show the number of patient falls, we want to break down the falls by whether or not they were falls with injury – the numerator for the fall rate metric – and all other falls. The goal of displaying that data is to help the user keep in mind that as important as it is to reduce the number of falls with injury, we also need to keep the overall number of falls down as well. No fall = no chance of fall with injury.
  • Unit Level of Detail – Because the blend needs to work at the per-unit level of detail as well as across all reporting units, that means (in version 7 at least) that the Unit needs to be in the view for the blend to work. But we want to display a single falls rate no matter how many units are selected.

Sparse Data

To deal with issue of sparse data, there are a few possible solutions:

  • Change the combined line and bar chart into separate charts. This would perhaps be the easiest, though it would require some messing about with filters, hidden reference lines, and continuous date axes to ensure that the two charts had similar axis ranges no matter what. However, that would miss out on the key capability of the combined chart to directly see how a fall contributes to the fall rate. In addition, there would be no reason to write this blog post. :)
  • Perform padding in the data source, either via a query/view or Custom SQL. In an earlier version of this project I’d built this, and maintaining a bunch of queries with Cartesian joins isn’t my favorite cup of tea.
  • Building a scaffold data source with all combinations of the month and unit and using the scaffold as the primary data source. While possible, this introduces maintenance issues when there’s a need for additional fields at a finer level of detail. For example, the falls measure actually has three separate fall rates – monthly, quarterly, and annual. These are generated as separate rows in our measures data and the particular duration is indicated by the Period field. So the scaffold source would have to include the Period field to get the data, but then that could be too much detail for the blended fall event data, and make for more complexity in the calculations to make sure the aggregations worked properly.
  • Do a tiny bit of padding in the query, then do the rest in Tableau via Show Missing Values aka domain padding. As I’d noted in an earlier post on blending, domain padding occurs before data is blended so we can pad out the measure data through the current date and then include all the falls. This is the technique I chose, for the reason that padding one row to the data is trivial and turning on Show Missing Values is a couple of mouse clicks. Here’s how I did that:

In my case, the primary data source is a Microsoft Access query that gets the falls measure results from a table that also holds results for hundreds of other metrics that we track. I created a second query with the same number of columns that returns Null for every field except the Measure Date, which has a value of 1/1/1900. Then a third query UNION’s those two queries together, and that’s what is used as the data source in Tableau.

Then, in Tableau, I added a calculated field called Date with the following formula:

//used for padding out display to today
IF [Measure Date] == #1/1/1900# THEN 
    TODAY() 
ELSE 
    [Measure Date] 
END

The measure results data contains a row per measure, reporting unit, and the period. These are pre-calculated because the data is used in a variety of different outputs. Since in this dashboard we are combining the results across units, we can’t just use the rate, we need to go back to the original numerator and denominator. So, I also created a new field for the Calculated Rate:

SUM([Numerator])/SUM([Denominator])

Now it’s possible to start building the line chart view:

  1. Put the Month(Date) – the full month/year version as a discrete – on Columns, Calculated Rate on Rows, Period on the Color Shelf. This only shows the data that exists in the data source, including the empty value for the current month (March in this case):

 

Screen Shot 2013-03-09 at 1.11.25 PM

 

  1. Turn on Show Missing Values for Month(Date) to start domain padding. Now we can see the additional column(s) for the month(s) – February in this case between January to the current month that Tableau has added in:

 

Screen Shot 2013-03-09 at 1.14.19 PM

 

With a continuous (green pill) date, this particular set-up won’t work in version 8. Tableau’s domain padding is not triggered when the last value of the measure is Null. I’m hoping this is just an issue with the beta, I’ll revise this section with an update once I find out what’s going on.

Even though the measure data only has end of month dates, instead of using Exact Date for the month I used Month(Date) because of two combined factors: One is that the default import of most date fields from MS Jet sources turns them into DateTime fields, the second is that Show Missing Values won’t work on an Exact Date for a DateTime field, you have to assign an aggregation to a DateTime (even Second will work). This is because domain padding at this level can create an immense number of new rows and cause Tableau to run out of memory, so Tableau keeps the option off unless you want it. Also note that you can turn on Show Missing Values for an Exact Date for a Date Field.

  1. Now for some cleanup steps: for the purposes of this dashboard, filter Period to remove Monthly (we do quarterly reporting), but leave in Null because that’s needed for the domain padding.
  2. Right-click Null on the Color Legend and Hide it. Again, we don’t exclude this because this would cause the extra row for the domain padding to fail.
  3. Set up a relative date filter on the Date field for the last 13 months. This filter works just fine with the domain padding.

Filtering on Unit

Here’s a complicating factor: If we add a filter on Unit, there’s a Null listed here:

 

Screen Shot 2013-03-09 at 1.18.31 PM

I’d just want to see the list of units. But if we filter that Null out, then we lose the domain padding, the last date is now January 2013:

 

Screen Shot 2013-03-09 at 1.18.58 PM

 

One solution here would be to alter the padding to add a padding row for every unit, instead of just one unit. Since Tableau doesn’t let us just hide elements in a filter, and we actually have more reporting units in our data than we are displaying on the dashboards, I chose to use a parameter filter because there are more reporting units in our production data than we are displaying on the dashboards, yet the all-unit rate needs to include all of the data. Setting this up included a parameter with All and each of the units, and a calculated field called “Chosen Unit Filter” with the following formula, that is set to Filter on False:

[Choose Unit] == "All" OR [Choose Unit] == [Unit]

Falls With and Without Injury

In a fantasy world, to create the desired stacked bars I’d be able to drag the Number of Records from the secondary datasource, i.e. the number of fall events, drag an Injury indicator onto the Color Shelf, and be done. However, that runs into the issue of having a finer level of detail in the secondary than in the primary, which I’ll walk through solutions for in the next section. In this case, since there are only two different numbers, the easy way is to generate two separate measures, then use Measure Names/Measure Values to create the stacked bars – Measure Values on Rows, and Measure Names on the Color Shelf. Here’s the basic calculation for Falls with Injury:

SUM(IF [Injury] != "None" THEN 1 ELSE 0 END)

We’re using a row-level calculated field to generate the measure, and a slightly different calc for Falls w/out Injury.

Unit Level of Detail

When we want to blend in Tableau at a finer level of detail and aggregate to a higher level, historically there have been three options:

  • Don’t use blending at all, instead use a query to perform the “blend” outside of Tableau. In the case that there are totally different data sources, this can be more difficult but not impossible by using one of the systems or a different system to create a federated data source, for example by adding your Oracle table as an ODBC connection to your Excel data, then making the query on that. In this case, we don’t have to do that.
  • Use Tableau’s Primary Groups feature “push” the detail from the secondary into the primary data source. This is a really helpful feature, the one drawback is that it’s not dynamic so any time there are new groupings in the secondary it would have to be re-run. Personally, I prefer automating as much as possible so I tend not to use this technique.
  • Set up the view with the needed dimensions in the view – on the Level of Detail Shelf, for example – and then use table calculations to do the aggregation. This is how I’ve typically built this kind of view.

Tableau version 8 adds a fourth option:

  • Tell Tableau what fields to blend on, then bring in your measures from the secondary.

I’ll walk through the table calculation technique, which works the same in version 7 and version 8, and then how to take advantage of v8′s new feature.

Using Table Calculations to Aggregate Blended Data

In order to blend the the falls data at the hospital unit level to make sure that we’re only showing falls for the selected unit(s), the Unit has to be in the view (on the Rows, Columns, or Pages Shelves, or on the Marks Card). Since we don’t actually need to display the Unit, the Level of Detail Shelf is where we’ll put that dimension. However, just adding that to the view leads to a bar for each unit, for example for April 2012 one unit had one fall with injury and another had two, and two units each had two falls without injury.

 

Screen Shot 2013-03-09 at 1.30.27 PM

 

To control things like tooltips (along with performance in some cases), it’s a lot easier to have a single bar for each month/measure. To do that, we turn to a table calculation, here’s the Falls w/Injury for v7 Blend calculated field, set up in the secondary data source:

IF FIRST()==0 THEN
	TOTAL([Falls w/Injury])
END

This table calculation has a Compute Using of Unit, so it partitions on the Month of Date. The IF FIRST()==0 part ensures that there is only one mark per partition. I’m using the TOTAL() aggregation here because it’s easier to set up and maintain. The alternative is to use WINDOW_SUM(), but in Tableau prior to version 7 there are some performance issues, so the calc would be:

IF FIRST()==0 THEN
	WINDOW_SUM(SUM(Falls w/Injury]), 0, IIF(FIRST()==0,LAST(),0))
END

The ,0 IIF(FIRST()==0,LAST(),0 part is necessary in version 7 to optimize performance, you can get rid of that in version 8.

You can also do a table calculation in the primary that accesses fields in the secondary, however TOTAL() can’t be used across blended data sources, so you’d have to use the WINDOW_SUM version.

With a second table calculation for the Falls w/out Injury, now the view can be built, starting with the line chart from above:

  1. Add Measure Names (from the Primary) to Filters Shelf, filter it for a couple of random measures.
  2. Put Measure Values on the Rows Shelf.
  3. Click on the Measure Values pill on Rows to set the Mark Type to Bar.
  4. Drag Measure Names onto the Color Shelf (for the Measure Values marks).
  5. Drag Unit onto the Level of Detail Shelf (for the Measure Values marks).
  6. Switch to the Secondary to put the two Falls for v7 Blend calcs onto the Measure Values Shelf.
  7. Set their Compute Usings to Unit.
  8. Remove the 2 measures chosen in step 1.
  9. Clean up the view – turn on dual axes, move the secondary axis marks to the back, change the axis tick marks to integers, set axis titles, etc.

This is pretty cool, we’re using domain padding to fill in for non-existent data and then having a blend happening at one level of detail while aggregating to another, just for the second axis. Here’s the v7 workbook on Tableau Public:

Patient Falls Dashboard - Click on Image to go to Tableau Public

Patient Falls Dashboard – Click on image above to go to Tableau Public

Tableau Version 8 Blending – Faster, Easier, Better

For version 8, Tableau made it possible to blend data without requiring the linking fields in the view. Here’s how I build the above v7 view in v8:

  1. Add Measure Names (from the Primary) to Filters Shelf, filter it for a couple of random measures.
  2. Put Measure Values on the Rows Shelf.
  3. Click on the Measure Values pill on Rows to set the Mark Type to Bar.
  4. Drag Measure Names onto the Color Shelf (for the Measure Values marks).
  5. Switch to the Secondary and click the chain link icon next to Unit to turn on blending on Unit.
  6. Drag the Falls w/Injury and Falls w/out Injury calcs onto the Measure Values Shelf.
  7. Remove the 2 measures chosen in step 1.
  8. Clean up the view – turn on dual axes, move the secondary axis marks to the back, change the axis tick marks to integers, set axis titles, etc.

The results will be the same as v7.

Next: Tableau’s Data Blending Architecture

———————————————————————————-

References:

[1] Kristi Morton, Ross Bunker, Jock Mackinlay, Robert Morton, and Chris Stolte, Dynamic Workload Driven Data Integration in Tableau, University of Washington and Tableau Software, Seattle, Washington, March 2012, http://homes.cs.washington.edu/~kmorton/modi221-mortonA.pdf.

[2] Hans Rosling, Wealth & Health of Nations, Gapminder.org, http://www.gapminder.org/world/.

[3] Jonathan Drummey, Tableau Data Blending, Sparse Data, Multiple Levels of Granularity, and Improvements in Version 8, Drawing with Numbers, March 11, 2013, http://drawingwithnumbers.artisart.org/tableau-data-blending-sparse-data-multiple-levels-of-granularity-and-improvements-in-version-8/.

 

Tips & Tricks #9: How Do Changes on the Source Report (Dataset) Get Reflected in MicroStrategy Report Services 9.x Documents

In MicroStrategy Report Services Documents, document datasets and their original source reports  (such as a grid report being used as a dataset) are not completely connected to each other. Depending on the changes made on the source report, it can be reflected differently on the document. Basically, the change can be divided into two types.

Type 1 – Formatting Changes

Formatting changes, for example, changing autostyles, thresholds, subtotals.

If a user chooses the option Add to Section without Formatting, the grid/graph showing on the document will not use the report’s stored formatting. Any formatting changes on the source report will not be reflected on the document.

Tip 9-1If a user chooses the option Add to Section with Formatting, the grid will be added with the current format of the report.  However, any formatting changes made to the source report AFTER the dataset has been included in the document will NOT be reflected on the document.

For example, a user disabled the subtotal (see screenshot below) for the original report after the report has been included as a dataset in a document, the document will still show the subtotals.

Tip 9-2

To force the document to recognize the report’s formatting changes, the user needs to delete the grid/graph from document section and add it again using the With Formatting option. By doing this, the latest formatting properties of the grid/graph on the source report are retrieved.

Type 2 – Adding/Removing Objects and Modifying Report Filters

Another type of changes made on the report involving adding/removing objects and modifying report filters.

Unlike the formatting change, this type of change does carry over from the source report to the document datasets.

For example, if users add/remove/modify report filters (see screenshot below), the change will be reflected on the data when running the document.

Tip 9-3

If an object is removed from the source report (see screenshot below), the user can see the change in the document’s Dataset Objects window. After the user runs the document, the object will be removed from grid/graph on the document.

Tip 9-4

If an object is added to the report (see screenshot below), the change will show in the document’s dataset objects window. However, the object will not be automatically added to the grid/graph. The user has to manually add the object to the grid/graph or add the dataset to the section again to make it show up.

Tip 9-5

NOTE:   As of MicroStrategy 9.0, a new feature was introduced where a user can add a dataset report to a Report Services document as a shortcut by selecting the Add to Section As a shortcut option, as shown below:

Tip 9-6

If the grid/graph is added to the document using this option, then the document would be updated automatically if ANY type of change is made on the source report.

An Introduction to Data Blending – Part 3 (Benefits of Blending Data)

Readers:

In Part 2 of this series on data blending, we delved deeper into understanding what data blending is. We also examined how data blending is used in Hans Rosling’s well-known Gapminder application.

Today, in Part 3 of this series, we will dig even deeper by examining the benefits of blending data.

Again, much of Parts 1, 2 and 3 are based on a research paper written by Kristi Morton from The University of Washington (and others) [1].

You can learn more about Ms. Morton’s research as well as other resources used to create this blog post by referring to the References at the end of the blog post.

Best Regards,

Michael

Benefits of Blending Data

In this section, we will examine the advantages of using the data blending feature for integrating datasets. Additionally, we will review another illustrative example of data blending using Tableau.

Integrating Data Using Tableau

In Ms. Morton’s research, Tableau was equipped with two ways of integrating data. First, in the case where the data sets are collocated (or can be collocated), Tableau formulates a query that joins them to produce a visualization. However, in the case where the data sets are not collocated (or cannot be collocated), Tableau federates queries to each data source, and creates a dynamic, blended view that consists of the joined result sets of the queries. For the purpose of exploratory visual analytics, Ms. Morton (et al) found that data blending is a complementary technology to the standard collocated approach with the following benefits:

  • Resolves many data granularity problems
  • Resolves collocation problems
  • Adapts to needs of exploratory visual analytics

Figure 1 - Company Tables

Image: Kristi Morton, Ross Bunker, Jock Mackinlay, Robert Morton, and Chris Stolte, Dynamic Workload Driven Data Integration in Tableau. [1]

Resolving Data Granularity Problems

Often times a user wants to combine data that may not be at the same granularity (i.e. they have different primary keys). For example, let’s say that an employee at company A wants to compare the yearly growth of sales to a competitor company B. The dataset for company B (see Figure 1 above) contains a detailed quarterly growth of sales for B (quarter, year is the primary key), while company A’s dataset only includes the yearly sales (year is the primary key). If the employee simply joins these two datasets on yearly earnings, then each row from A will be duplicated for each quarter in B for a given year resulting in an inaccurate overestimate of A’s yearly earnings.

This duplication problem can be avoided if for example, company B’s sales dataset were first aggregated to the level of year, then joined with company A’s dataset. In this case, data blending detects that the data sets are at different granularities by examining their primary keys and notes that in order to join them, the common field is year. In order to join them on year, an aggregation query is issued to company B’s dataset, which returns the sales aggregated up to the yearly level as shown in Figure 1. This result is blended with company A’s dataset to produce the desired visualization of yearly sales for companies A and B.

The blending feature does all of this on-the-fly without user-intervention.

Resolves Collocation Problems

As mentioned in Part 1, managed repository is expensive and untenable. In other cases, the data repository may have rigid structure, as with cubes, to ensure performance, support security or protect data quality. Furthermore, it is often unclear if it is worth the effort of integrating an external data set that has uncertain value. The user may not know until she has started exploring the data if it has enough value to justify spending the time to integrate and load it into her repository.

Thus, one of the paramount benefits of data blending is that it allows the user to quickly start exploring their data, and as they explore the integration happens automatically as a natural part of the analysis cycle.

An interesting final benefit of the blending approach is that it enables users to seamlessly integrate across different types of data (which usually exist in separate repositories) such as relational, cubes, text files, spreadsheets, etc.

Adapts to Needs of Exploratory Visual Analytics

A key benefit of data blending is its flexibility; it gives the user the freedom to view their blended data at different granularities and control how data is integrated on-the-fly. The blended views are dynamically created as the user is visually exploring the datasets. For example, the user can drill-down, roll-up, pivot, or filter any blended view as needed during her exploratory analysis. This feature is useful for data exploration and what-if analysis.

Another Illustrative Example of Data Blending

Figure 2 (below) illustrates the possible outcomes of an election for District 2 Supervisor of San Francisco. With this type of visualization, the user can select different election styles and see how their choice affects the outcome of the election.

What’s interesting from a blending standpoint is that this is an example of a many-to-one relationship between the primary and secondary datasets. This means that the fields being left-joined in by the secondary data sources match multiple rows from the primary dataset and results in these values being duplicated. Thus any subsequent aggregation operations would reflect this duplicate data, resulting in overestimates. The blending feature, however, prevents this scenario from occurring by performing all aggregation prior to duplicating data during the left-join.

Figure 2 - San Francisco Election

 Image: Kristi Morton, Ross Bunker, Jock Mackinlay, Robert Morton, and Chris Stolte, Dynamic Workload Driven Data Integration in Tableau. [1]

Next: Data Blending Design Principles

——————————————————————————————————–

References:

[1] Kristi Morton, Ross Bunker, Jock Mackinlay, Robert Morton, and Chris Stolte, Dynamic Workload Driven Data Integration in Tableau, University of Washington and Tableau Software, Seattle, Washington, March 2012, http://homes.cs.washington.edu/~kmorton/modi221-mortonA.pdf.

[2] Hans Rosling, Wealth & Health of Nations, Gapminder.org, http://www.gapminder.org/world/.

An Introduction to Data Blending – Part 2 (Hans Rosling, Gapminder and Data Blending)

Readers:

In Part 1 of this series on data blending, we began to explore the concepts of data blending as well as the life-cycle of visual analysis.

Today, in Part 2 of this series, we will dig deeper into how data blending works.

Again, much of Parts 1, 2 and 3 are based on a research paper written by Kristi Morton from The University of Washington (and others) [1].

You can learn more about Ms. Morton’s research as well as other resources used to create this blog post by referring to the References at the end of the blog post.

Best Regards,

Michael

Data Blending Overview

Data Blending allows an end-user to dynamically combine and visualize data from multiple heterogeneous sources without any upfront integration effort. [1] A user authors a visualization starting with a single data source – known as the primary – which establishes the context for subsequent blending operations in that visualization. Data blending begins when the user drags in fields from a different data source, known as a secondary data source. Blending happens automatically, and only requires user intervention to resolve conflicts. Thus the user can continue modifying the visualization, including bringing in additional secondary data sources, drilling down to finer-grained details, etc., without disrupting their analytical flow. The novelty of this approach is that the entire architecture supporting the task of integration is created at runtime and adapts to the evolving queries in typical analytical workflows.

A Simple Illustrative Example

In this section we will discuss a scenario in which three unique data sources (see left half of Figure 1 below for sample tables) are blended together to create the visualization shown in Figure 2 below. This is a simple, yet compelling mashup of three unique measures that tells an interesting story about the complexities of global infant mortality rates in the year 2000.

Figure 1

 

Image: Kristi Morton, Ross Bunker, Jock Mackinlay, Robert Morton, and Chris Stolte, Dynamic Workload Driven Data Integration in Tableau. [1]

In this example, the user wants to understand if there is a connection between infant mortality rates, GDP, and population. She has three distinct spreadsheets with the following characteristics: the first data source contains information about the infant mortality rates per 1000 live births for each country, the second contains information about each country’s total population, and the third source contains country-level GDP. For this analysis task, the user drags the fields, “Country or Area” and “Infant mortality rate per 1000 live births”, from her first data source onto the blank visual canvas. Since these fields were the first ones selected by the user, then the data source associated with these fields becomes the primary data source.

This action produces a visualization showing the relative infant mortality rates for each country. But the user wants to understand if there is a correlation between GDP and infant mortality, so she then drags the “GDP per capita in US dollars” field onto the current visual canvas from Data Table A. The step to join the GDP measure from this separate data source happens automatically: the blending system detects the common join key (ı.e. “Country or Area”) and combines the GDP data with the infant mortality data for each country. Finally, to complete her analysis task, she adds the “Population” measure from Data Table B, to the visual canvas, which produces the visualization in Figure 2 below associated with the blended data table in Figure 1.

 

Figure 2

Image: Kristi Morton, Ross Bunker, Jock Mackinlay, Robert Morton, and Chris Stolte, Dynamic Workload Driven Data Integration in Tableau. [1]

Hans Rosling, Gapminder and Data Blending

The Gapminder World interactive graph below shows how long people live and how the number of children a woman has is affected by how much money they earn using different data sources.

Gapminder World for Windows

Image: Hans Rosling’s Wealth and Health of Nations (Gapminder.org) [2]

Hans RoslingIn the screenshot above, the y-axis shows us Children per women (total fertility) . The x-axis shows us Income per person (GDP/capita, PPP$ inflation-adjusted). The series data points (the bubbles) show us population for each country. If you were to click the Play button, you would see as an interactive “slide show” how countries have developed since 1800.

This demonstrates the flexibility of the data blending feature, namely that users can dynamically change their blended views by pivoting on different data sources and measures to blend in their visualizations.

In the screenshot below, Mr. Rosling explains how to use the interactive Gapminder World application.

Also, Mr. Rosling has provided Gapminder World Offline, which you can use to show animated statistics from your own laptop! It can be run on Windows, Mac and Linux. Here is a link to the download installation page on the Gapminder.org site.

And here is a link to the PDF for the Gapminder World Guide show above.

Gapminder World Guide

Image: Hans Rosling’s Gapminder World Guide (PDF) [2]

Next: Benefits of Blending Data

——————————————————————————————————–

References:

[1] Kristi Morton, Ross Bunker, Jock Mackinlay, Robert Morton, and Chris Stolte, Dynamic Workload Driven Data Integration in Tableau, University of Washington and Tableau Software, Seattle, Washington, March 2012, http://homes.cs.washington.edu/~kmorton/modi221-mortonA.pdf.

[2] Hans Rosling, Wealth & Health of Nations, Gapminder.org, http://www.gapminder.org/world/.

An Introduction to Data Blending – Part 1 (Introduction, Visual Analysis Life-cycle)

Readers:

Today I am beginning a multi-part series on data blending.

  • Parts 1, 2 and 3 will be an introduction and overview of what data blending is.
  • Part 4 will review an illustrative example of how to do data blending in Tableau.
  • Part 5 will review an illustrative example of how to do data blending in MicroStrategy.

I may also include a Part 6, but I have to see how my research on this topic continues to progress over the next week.

Much of Parts 1, 2 and 3 are based on a research paper written by Kristi Morton from The University of Washington (and others) [1].

Please review the source references, at the end of each blog post in this series, to be directed to the source material for additional information.

I hope you find this series helpful for your data visualization needs.

Best Regards,

Michael

Introduction

Tableau and MicroStrategy’s new Analytics Platform are commercial business intelligence (BI) software tools that support interactive, visual analysis of data. [1]

Using a Web-based visual interface to data and a focus on usability, these tools enable a wide audience of business partners (IT’s end-users) to gain insight into their datasets. The user experience is a fluid process of interaction in which exploring and visualizing data takes just a few simple drag-and-drop operations (no programming skills or DB experience is required). In this context of exploratory, ad-hoc visual analysis, we will explore a feature originally introduced in Tableau in 2006, and in MicroStrategy’s new Analytics Platform v9.4.1 late last year (2013).

We will examine how we can integrate large, heterogeneous data sources. This feature is called data blending, which gives users the ability to create data visualization mashups from structured, heterogeneous data sources dynamically without any upfront integration effort. Users can author visualizations that automatically integrate data from a variety of sources, including data warehouses, data marts, text files, spreadsheets, and data cubes. Because data blending is workload driven, we are able to bypass many of the pain points and uncertainty in creating mediated schemas and schema-mappings in current pay-as-you-go integration systems.

The Cycle of Visual Analysis

Unlike databases, our human brains have limited capacity for managing and making sense of large collections of data. In database terms, the feat of gaining insight in big data is often accomplished by issuing aggregation and filter queries (producing subsets of data).

However, this approach can be time-consuming. The user is forced to complete the following tasks.

  1. Figure out what queries to write.
  2. Write the queries.
  3. Wait for the results to be returned back in textual format. And, then finally,
  4. Read through these textual summaries (often containing thousands of rows) to search for interesting patterns or anomalies.

Tools like Tableau and MicroStrategy help bridge this gap by providing a visual interface to the data. This approach removes the burden of having to write queries. The user can ask their questions through visual drag-and-drop operations (again, no queries or programming experience required). Additionally, answers are displayed visually, where patterns and outliers can quickly be identified.

Visualizations leverage the powerful human visual system to help us effectively digest large amounts of information and disseminate it quicker.

Cycle of Visual Analysis

Image: Kristi Morton, Ross Bunker, Jock Mackinlay, Robert Morton, and Chris Stolte, Dynamic Workload Driven Data Integration in Tableau. [1]

Figure 1, above, illustrates how visualization is a key component in turning information into knowledge and knowledge into wisdom.

Ms. Morton discusses the process as follows,

The process starts with some task or question that a knowledge worker (shown at the center) seeks to gain understanding. In the first stage, the user forages for data that may contain relevant information for their analysis task. Next, they search for a visual structure that is appropriate for the data and instantiate that structure. At this point, the user interacts with the resulting visualization (e.g. drill down to details or roll up to summarize) to develop further insight.

Once the necessary insight is obtained, the user can then make an informed decision and take action. This cycle is centered around and driven by the user and requires that the visualization system be flexible enough to support user feedback and allow alternative paths based on the needs of the user’s exploratory tasks. Most visualization tools, however, treat this cycle as a single, directed pipeline, and offer limited interaction with the user. Moreover, users often want to ask their analytical questions over multiple data sources. However, the task of setting up data for integration is orthogonal to the analysis task at hand, requiring a context switch that interrupts the natural flow of the analysis cycle. We extend the visual analysis cycle with a new feature called data blending that allows the user to seamlessly combine and visualize data from multiple different data sources on-the-fly. Our blending system issues live queries to each data source to extract the minimum information necessary to accomplish the visual analysis task.

Often, the visual level of detail is at a coarser level than the data sets. Aggregation queries, therefore, are issued to each data source before the results are copied over and joined in Tableau’s local in-memory view. We refer to this type of join as a post-aggregate join and find it a natural fit for exploratory analysis, as less data is moved from the sources for each analytical task, resulting in a more responsive system.

Finally, Tableau’s data blending feature automatically infers how to integrate the datasets on-the-fly, involving the user only in resolving conflicts. This system also addresses a few other key data integration challenges, including combining datasets with mismatched domains or different levels of detail and dirty or missing data values. One interesting property of blending data in the context of a visualization is that the user can immediately observe any anomalies or problems through the resulting visualization.

These aforementioned design decisions were grounded in the needs of Tableau’s typical BI user base. Thanks to the availability of a wide-variety of rich public datasets from sites like data.gov, many f Tableau’s users integrate data from external sources such as the Web or corporate data such as internally-curated Excel spreadsheets into their enterprise data warehouses to do predictive, what-if analysis.

However, the task of integrating external data sources into their enterprise systems is complicated. First, such repositories are under strict management by IT departments, and often IT does not have the bandwidth to incorporate and maintain each additional data source. Second, users often have restricted permissions and cannot add external data sources themselves. Such users cannot integrate their external and enterprise sources without having them collocated.

An alternative approach is to move the data sets to a data repository that the user has access to, but moving large data is expensive and often untenable. We therefore architected data blending with the following principles in mind: 1) move as little data as possible, 2) push the computations to the data, and 3) automate the integration challenges as much as possible, involving the user only in resolving conflicts.

Next: Data Blending Overview

——————————————————————————————————–

References:

[1] Kristi Morton, Ross Bunker, Jock Mackinlay, Robert Morton, and Chris Stolte, Dynamic Workload Driven Data Integration in Tableau, University of Washington and Tableau Software, Seattle, Washington, March 2012, http://homes.cs.washington.edu/~kmorton/modi221-mortonA.pdf.