[via Visual Economics] A Department of Labor visualization showing where Americans are spending their money (annual numbers). The data is very interesting. The visualization is nice and colorful. Unfortunately, you need to work pretty hard to extract the information. The display mechanism is effectively a pie chart. I bet a different display would show off the data much better.

If anyone wants to take a stab displaying this data more effectively, I would be happy to post your results here.

Over the weekend, Simple Complexity changed its design. Please forgive some of the dust as we clean up a few problems with some formatting. If you find anything that is messed up or you just want to give us your opinion, it would be very appreciated.

People have a lot of trouble visulizing big numbers, particularly, the relationship between one big number and another big number.

There is a YouTube video that is getting a lot of attention from the Republicans. It does a good job demonstrating the cognitive challenge people experience. Take a look it is about a minute and a half.

Obama Budget Cuts Visualization

I love the way this video illustrates that the comparison of “$100 million to $3.5 trillion” generally gets reduced by us humans to “a big pile of money to a bigger pile of money”. When the video lays out how small 100 million is to 3.5 trillion, at least for me, I was shocked (even though I am well aware that a trillion is significantly larger than a million).

Here is another good visualization along the same lines:

Obama Health Reform and Wait Times Visualization (In Lego!)

While I may not agree with the conclusions that narrator is making, he has created exceptionally effective visualizations.

[via Future Perfect]

Although this is a little off topic, I did find this funny, and it does relate to data.

Future Perfect’s caption read: “Redundancy, quality, over-compensation…”

Via Mashables, below are several Twitter visualizations collected by Stan Schroeder.

[View with PicLens]

Twittervision maps tweets from around the world.

TwitterThoughts graph Twitter trends based on a variety of factors.

Twittearth is a 3D model of tweets from around the world.

Twitt3D lets you follow your tweets in 3D.

Twistori takes tweets containing certain words and displays them as a stream.

TweetWheel no longer exists, but was an interesting visualization.

SXSW Twitter Visualizer pulls and graphs recent tweets containing certain activities.

Twitter StreamGraphs uses stream graphs to display words and tags for a keyword.

Social Collider tracks conversations based on users or topics showing the flow of conversation.

Monitter is a real-time visualization of keywords.

Goodnight Sweet Tweets parses Twitter for tweets containing the strings “zzz” or “zzzz” and then animates them with some vector-based graphics.

Excellent article from Stacey Higginbotham at Gigaom:

As folks increasingly store and access information online, the data centers powering cloud services need to be managed more like a single computing entity rather than a bunch of servers, according to a Google white paper (Google calls it a mini-book) released today.

Check out the article.

A few weeks ago I evaluated a randomly selected dashboard using Stephen Few’s 13 Common Mistakes in Dashboard Design. Today, I have picked another Dashboard, this one from VisualCalc, a web-based analysis software provider.

Purpose of Dashboard:

The Swine Flu Dashboard tracks the daily progression of the H1N1 Flu (Swine Flu) virus on both a country-by-country and U.S. state-by-state basis. This public dashboard tracks both laboratory-confirmed cases and deaths resulting from the virus. The Swine Flu Dashboard contains four graphical indicators, each focused on a different aspect of the virus:

• Total laboratory-confirmed cases and deaths on a country-by-country basis.
• Total laboratory-confirmed and probable cases and deaths on a state-by-state basis within the U.S.
• Cumulative global laboratory-confirmed cases on a day-by-day basis, broken out by country.
• Overall composition of the different types of flu’s tested within the United States.

Description via VisualCalc Press Release

click to expand

Evaluation Areas:

Display Honors the Boundary of a Single Screen

Pass

Something powerful happens when things are seen together. Seeing everything at once is an advantage. The Analyze button does take you to a screen where you can dive deeper into the data. This is well done.

Supplying Adequate Context for the Data

Partial Fail

Measures of what’s currently going on rarely do well in isolation. It is more valuable to include comparison, whether the result is good or bad, how good or bad, are we on track, are we doing better than in the past, or better than the forecast. In the bottom right graph, there is a comparison between last week and year-to-date providing some context. That said, most of the data is provided without context. The lack of labeling makes the context particularly hard to determine.

Displaying Appropriate Detail or Precision

Pass

It appears that the appropriate precision is being used.

Choosing an Appropriate Measure

Partial Pass

For a measure to be meaningful, we must know what is being measured. A measure is deficient if it isn’t the one that most clearly and efficiently communicates the meaning that the dashboard viewer should discern. It appears that the measures are appropriately picked. Again, clearer labeling would clarify this some.

Choosing an Appropriate Display Media

Fail

I think in several cases the wrong display media was chosen.

• Swine Flu Totals by Country (Top Left) – This would be better as a horizontal bar chart (particularly a bullet chart), where the countries are at the left and the deaths are either shaded areas within the horizontal bar or as a separate chart. The multiple axis seem to me to make the graph very hard to interpret. I frequently forgot which graph went with which axis. There is no need to mix line graph and bar graph together in this case. I think it was chosen to visually differentiate the data.
• Swine Flu Totals by U.S. State (Top Right) – The best display would probably have been a map with some color coding to show comparative incidence. Regional patterns could be identified. The way it is, a table of numbers would be more effective than this. This graph is just impossible to discern. Death data does not turn up because it is so low, and without labels the bars also don’t provide information.
• Cumulative Swine Flu Cases (Bottom Left) – A horizontal bar chart based on % of total would have been better. A table would have been a good choice too.
• Composition of U.S. Flu Cases (Bottom Right) – I think this has the right media choice.

Use of Effectively Designed Display Media

Fail

I think this is the weakest area of this dashboard. In general, the data is either unlabeled or poorly labeled. Examples include:

• Swine Flu Totals by Country (Top Left) – I had trouble determining which data points belonged to which axis. After scrolling over the data points, something I did accidentally at first, I was able to figure out that green was deaths and the red bar graph was confirmed cases. It might have been useful to sort the results largest to smallest. It would have allowed that information to be obtained instantly. I have to assume that the data here is cases to date, but a label providing this information (including a current-to date) would have helped.
• Swine Flu Totals by U.S. State (Top Right) – This graph is totally ineffective. There are no labels as to which line represents which state. The user must scroll over the line to see the label. When I did that, I found that the number of deaths are plotted with a red bar graph (that is what causes the space between the bars). The scale of the deaths compared to the cases is just so small it doesn’t turn up. This data absolutely should be sorted from largest to smallest. In my opinion, this graph provides no insight.
• Cumulative Swine Flu Cases (Bottom Left) – Besides the wrong type of graph being used here, I’d recommend a horizontal bar chart with the countries along the x-axis. Once again, the x-axis is missing any date labels. The two end points and a couple of interim points need to be labelled. With the display media chosen, it would have been more effective to reduce the number of countries to US (light-blue), Mexico (brown), Canada (purple), Australia (green) and other. There are a number of countries with data so small it can’t be seen. Again a horizontal bar chart would have avoided most of these problems.
• Composition of U.S. Flu Cases (Bottom Right) – Again the labeling is poor here. I had trouble figuring out what composition meant. I guessed that was the case. I don’t know a lot about flu. The bars are labeled (if you scroll over them): Blue – Other Type A, Green – Swine, and Red – Other Type B. I think for this dashboard comparing Swine versus all other flu would have been better. I would have put the two other types at least next to each other as opposed to Swine flu in the middle. The real problem is how the data is labeled when you roll over it. was that this showed the percentage of swine flu versus general flu outbreaks. After finding out the labels for the bars, I found that I could not understand what the label meant. Looking at the graph, you see that the green bar is significantly higher than the year to date average on the right. What does the (225%) mean? I don’t understand the average incidence of 33% as looking at the green bar to the right, you see that it is around 16%. The huge difference between 75% of the reported flu up from 16% seems massive. This is where some text really is necessary. I am left thinking the data is wrong.

Encoding Quantitative Data Accurately

Mostly Fail

The scaling of the top two graphs is quite poor. The scale of deaths to the number of cases is so small, putting them on the same graph (right) or using two axes (left) makes interpretation of the pattern hard to discern. I had spent some time with this visualization pondering how Mexican deaths could be higher than the incidence. The use had 27 deaths compared to almost 14,000 cases, having the marking about 1/3 up the US graph is misleading. In Mexico, there were 106 deaths compared to 5,600 cases. Mexico had almost 1/3 of the incidence, but almost 5 times the deaths. This can be represented better by either having two side-by-side bar graphs or by using a bullet graph, where the portion of the incidence is shaded in the bar. The deaths are such a small percentage of the total occurance, more attention needed to be paid to representing this clearly. On the right, the death data does not even turn up.

Effective Arrangement of the Data

Partial Pass

There is a lot of meaty information here, and the graphs seem broken up even into reasonable analysis categories. I think if other problems were fixed, it would be easier to see that the data is arranged well.

Avoiding Meaningless Variety

Pass

I don’t not get the feeling that the graphing methods were haphazardly chosen. I suspect that tool limitations restricted chosing the right graph medium for the data.

Important Data is Highlighted

Fail

When you look at a dashboard, your eyes should immediately be drawn to the information that is most important. In this design, everything is visually prominant, and consequently nothing stands out.

Avoiding the Use of Useless Decoration

Pass

There is not a lot of distracting and useless decoration. There is a lot of screen real estate that is not used for conveying information.

Appropriate Use of Color

Partial Fail

While the coloring is not distracting as it is in many displays, the color also has absolutely no meaning within the dashboard. I wanted to group all the green, red and blue data in some relative way. I went through the graphs trying to see if all of the green related to the same thing. Better media choices would have alleviated this confusion.

Designing an Attractive Visual Display

Pass

I think the display is clean and attractive.

Conclusion

There is much that could be fixed in this dashboard if our goal is the communication of information from which people could make decisions.

Please let me know your comments.

Photo Credit: PauliePaul on Flickr

• There are no off-the-shelf plug and play solutions that can fully address the needs of all end user(s).
• There are no “one-size-fits-all” solutions!
• The process of bringing all affected interests together to collaborate toward a shared goal seems impossible.

These are all challenges on any data warehouse development project or for that matter on any project I have been involved with. After close to 12 years helping organizations’ of all shapes and sizes develop and implement a data warehouse, I have found one issue or flaw that keeps the project from getting full organizational buy-in: inadequate end user(s) or functional user(s) input during the planning and development stages of the project.  We always wait until the last phase of the project “reporting” to allow them to tell us what the reports should look like and what data they currently have.  Why not ask them “what data do you need, but don’t have?” or “how can we make your job easier through data and reporting?” My thought is simple we should take a “bottoms-up” rather than a “tops-down” approach for Data Warehouse development and implementation. 

One of the best resources for insuring the success of any project is the people using what you are building. So why wait until the end – let’s have them drive the project. When we get to the implementation stage, we will have users banging on the door to get in versus us forcing it done their throats. There’s probably a group of people already in your organization who will volunteer to help design the system and evaluate it. Think of these people as Power User(s).

Finding the best power user(s) requires a little effort. The following list should help you identify, inspire, and make best use of the Power User(s) in your organization:

1. Why do I need Power User(s)?

Power User(s) work side-by-side with regular users – the same people you’re trying to reach with your data warehouse project. If your Power User(s) are excited about the data warehouse, you can be sure they’ll let their colleagues know and that excitement can be contagious. Power User(s) are also vital to testing. They can help you identify potential problems and areas of improvement you may not have found on your own.

2. How do you identify these Power User(s)?

You will need the managers of the functional area in the organization to suggest candidates. Do not, however, let managers pick the Power User(s) for you. Their idea of what may make a good Power User may not match yours or the projects.

3. Qualities of a good Power User

• Likes to try new things and able to provide honest, useful feedback
• Excited about volunteering time to make the project a success, and has the time and support of their management team to add this to their schedule
• Able to get up in front of their peers and tell them that data warehouse will make their lives easier
• Very organized, not only in their own jobs, but also as they work with others
• Clear communication

Power User(s) should represent a good cross section of the organizations workforce. I suggest avoiding employees that have been at the organization less than a year, while they can offer you outside experience, they could slow down your efforts by providing too much basic information about the organization.

Power User(s) do not need to have advanced computer skills already. In fact, having advanced computer skills may be a disadvantage. Advanced users may not help you identify problems less experienced users would see.  Technical skills can be taught; the ability to speak for and to others cannot.

4. Management on Power User(s)?

Even if you don’t use managers as a Power User, try to keep them involved in the process. The more involved they are, the more excited they will be about the deployment, and the more they will work to make sure it’s a success. Ask them to nominate Power User(s) and to check in regularly with the users. This helps managers see the success of the project. It will also help motivate the Power User(s), since they will see their boss is invested in the project.

5. How many do I need?

Ideally, you want one at least Power User per functional area. Of course, the number of users needed varies by organization. In a very large organization, you may want a couple of Power Users per functional area.

Amit Agarwal wrote an informative post a few weeks ago regarding Web 3.0. In his post, he links to 4 presentations that do a great job explaining what Web 3.0 is. I have pulled a few of the better slides below.

Make sure you check out Amit Agarwal for the full presentations.

I spent some time looking at Daytum the other day. On the Daytum website, it states:

Daytum can help you count and communicate the most important statistics in your life.

The idea for Daytum came out of Nicholas Feltron’s annual reports, where he displays statistics about his personal life.

Nicholas was asked on, interaction design, why is it important that people have access to visualization and tracking tools?

One of the most obvious discoveries about my Annual Reports has been that people would love this kind of documentation of their own lives. The barriers to this are two-fold: the data-logging of daily habits is difficult and time-consuming; and communicating the data is a challenge. I believe that if we can reduce the burden of recording the data and even add a social stimulus, then remove all obstacles to designing and sharing the results, this type of personal reporting may gain some popularity.

As someone who works with organizations trying to analyze what is happening within their businesses, what is true at the individual level is also true at the organization. Obtaining and then displaying the right measures is quite challenging.

I wanted to see how people were using Daytum, so I went to Flickr and searched on Daytum. Here are some of the results that I found. I’d love to hear your observations.

[View with PicLens]

From emeffer

From Philipp_Roth

From Allison McCarthy

From miss_yasmina

From Pengdo-oing

From ni[do]

From ni[do]

From ni[do]

From ni[do]

From BRADSHAWK

From _heather_r_

From ravina sniper

From ni[do]

From ni[do]

From theskinimin

From kooop

From *mynameisjosh

From bee combs

From joy-z