The Why Axis RSS Feed

Mind the Gap - An Economic Chart Remake

Tue, 09 Apr 2013 00:00:00 +0000

Guest author Jon Schwabish is an economist with the U.S. federal government and creator of policy-relevant data visualizations. His website, policyviz.com, contains details about his one-day workshops on visualizing and presenting data for people in public policy. You can reach him at jschwabish@gmail.com or by following him on Twitter @jschwabish.

Download the Excel file with the data tables and charts covered in this post.

This chart in a recent post by Catherine Rampell on the New York Times’ Economix blog struck me in need of a redo. Rampell’s post uses the chart to help illustration how there is a greater share of women in high-achieving jobs in the U.S. than in many other countries.

Rampell references an interactive version of the graph from the Organisation for Economic Co-operation and Development (OECD). The interactive visualization was made in Tableau and is essentially identical to the static version, but uses different colors (orange bars instead of red). The interactivity enables the user to choose any country and any year between 1990 and 2011. (Technically, the data are published by the International Labor Organization, but OECD has visualized the data).

I find the static graph difficult to decipher for five main reasons (and the interactive version suffers from many of the same shortcomings).

It’s difficult to compare the values for men and women because the data for men (blue diamonds) are so far from data for women (red columns)—there’s nothing that helps the eye connect the two.
The red columns for women take up a much larger proportion of the graph than do the blue diamonds for men, which results in overemphasizing the data for women. This may have been purposeful, but an active title—such as “Women’s Employment as Senior Managers Averaged 6% in 2008”—would have made that clear. I’m also not a fan of the color gradient in the columns—it emphasizes the base of the column instead of the top of the column, which is where the data are encoded.
The legend is completely disconnected from the graph and off to the right side. We typically start reading a graph from the top-left and then work our way down and into the chart. A better position for the legend would be just below the title or above the y-axis.
There are a lot of gridlines, they are a bit too heavy, and the percentage signs on all of the y-axis labels are probably redundant. (There are additional gridlines in the interactive version).
Why, oh, why do we continue to make column charts with vertically-oriented x-axis labels? (I bet a lot of people who just read that sentence looked up to the chart and rotated their heads 90 degrees to the left).

Okay, so let’s take a tour through some revisions of this chart by trying to address the issues just mentioned.

Just for comparison sake, here is my repeat of the same chart. I’m going to make a series of small changes, but keep the overall chart design the same.

Here are the changes I’ve made: different font (Cabin); different color scheme (orange and blue); fewer and lighter gridlines; top-left aligned legend; title with the “(Percent)” subheader.

So now let’s make what may be the easiest and most important change: rotating the chart 90 degrees. When the graph is rotated in this way, the metrics stay the same and how we perceive the data is unchanged, but the x-axis labels are now horizontally oriented, which are much easier to read.

Still, the orange bars overwhelm the blue diamonds and the two series are disconnected from one another. One way to address the visual imbalance of the two series is to create what I call a “stacked broken bar chart.”

Excel tip: This type of chart is seriously easy to make: it includes three stacked series, the middle series equals some number—in this case 20—minus the value of the base (far-left) series; that middle series is then given an empty fill. Unfortunately, when you create this sort of graph, the x-axis labels are 0, 10, 20, and 30, but that can be fixed by adding four scatterplot points and labeling each point with the series name (e.g., “0”, “10”, “0”, and “10”). I walk through the Excel process of making this type of chart in more detail in my “Making Excel Graphs Better” tutorial available on Slideshare.

I think this chart is a little better — it gives a more equal visual treatment of the two series, it’s easier to compare the values between men and women because the bars are aligned horizontally, and it’s easier to read the country names because they are also aligned horizontally.

But it’s still a little difficult to compare the values between men and women because the two series are treated separately. In the next figure, I link the two groups together using a dot plot graph and connect each pair with a thin, grey line.

Excel tip: In this tutorial, Jon Peltier details a step-by-step approach to creating a dot plot.

Notice again the integrated title, unit, and legend. I’ve not included data values directly on the chart because it clutters the graph, not to mention that the original chart did not include data values. (In another iteration of this chart I connected all of the points, but I didn't think that was really necessary).

Take note of one other important characteristic of this graph: the country labels are completely separate from the data values, which again makes the graph hard to read. I could draw another line from each orange circle to the y-axis. That would help, but that means I’m adding more non-data elements to the chart. Instead, I’m going to move the labels directly onto the chart:

Excel tip: You can move labels the easy way or the hard way. The hard way is to manually type all of these data labels one by one. The easy way is to write or steal a macro, or grab an add-in to do it for you — I like this one from The Spreadsheet Page. Note, however, that some of the labels will wrap onto two lines (in the chart dimensions I’m using, ‘United Kingdom’ and ‘Slovak Republic’). You can try resizing the graph, but you can’t directly manipulate data labels in Excel 2010—I hear that this is available in Excel 2013. To fix those two countries in my graph, I manually inserted two text boxes.

I’m satisfied with this last chart, but I’ll take it one step further. If you read the Rampell post, you’ll notice at the very end she discusses the ratio of women’s employment rates to men’s employment rates:

“In the United States, the ratio of the share of women who are in senior management positions to the share of men who are in senior management positions is about 0.85 (13.9/16.3), whereas for all O.E.C.D. countries it is about 0.59 (6.1/10.3). The country with the next highest ratio, behind the United States, is New Zealand, with a ratio of 0.76 (11.7/15.4).”

I think the ratios are important and really help tell the story—the relative share of women who are senior managers across countries is revealing in its own way, but comparing those shares to another group (e.g., men) within each country helps take certain country-specific employment characteristics into account.

There are two ways I try to visualize these ratios. First, I add a bar chart to the right side of the dot plot graph—in this way, the focus of the visual is still on the levels of men and women, but the ratio information is also included. (Alternatively, if you think the bars are distracting, you could just include the numbers of the ratio and not the bar chart).

Of course, another option is to take a much simpler and direct approach: visualize the ratios.

This final graph is clearly a departure from the initial graph, but that’s because this graph has a different focus. Here, countries take on different ranks: for example, Italy ranks 12th in the share of women employed as senior managers (9.1%), but the ratio between women and men (75%) puts Italy behind only the United States (85%) and New Zealand (76%).

To sum up, I’ve redesigned a static graphic (in Excel) from a column-line chart combination that was very easy to make to a dot plot (a bar-scatterplot combination), which was only slightly harder to make. The goal was to improve the static graphic such that the series for women and the series for men were given more even visual weights and to make it easier for the reader to compare those two groups. The last couple of graphics incorporated the ratio between women and men because Rampell discusses that metric in the last part of her post. Whether you think that information should be included or not largely depends on whether you think it helps to visually support Rampell’s argument.

Download the Excel file with data and charts.

The Washington Post Plots Quality and Quantity of Life

Thu, 21 Mar 2013 00:00:00 +0000

In December the Washington Post published a visualization which attempts to answer the question asked directly in its title - "How long will we live - and how well?" This interactive scatterplot was created in four days to support this article about the release of the Global Burden of Disease Study.

I talked to Emily Chow from the Post Graphics team (which also includes Bonnie Berkowitz and Todd Lindeman). Emily told me the team's goal with this visualization was:

to illustrate how healthy life expectancy has changed since 1990.

Every news organization should have a good scatter plot in their tool belt. The Washington Post makes good use of this one to show a broad distribution of worldwide data. The addition of some faster filtering and more intuitive point selection would help make it even more of a workhorse for future visualizations. This particular incarnation animates to reflect the choices made by the user (selecting Men / Women and 1990 / 2010).

The angle that we found interesting was the fact that life expectancy is generally longer than it was 20 years ago, but that doesn't mean that we are actually living healthier lives… The percentage of years that are considered healthy has actually decreased overall.

Emily, Bonnie and Todd decided to show this by plotting life expectancy against the percentage of those years lived in good heath. Although this is a slight abstraction of raw healthy years, the real question to me is whether or not this animated scatter plot effectively demonstrates that a smaller percentage of healthy years are lived over time.

Because the chart simultaneously plots the multitude of points on two axes and then animates them over time, it ends up being extremely difficult for me to see the overall trend represented with this type of chart.

The team did explore several other chart types in their process, some of which had very promising aspects. My favorite was this set of parallel coordinates which quickly shows some dramatic outliers that are difficult to pick out in the animated scatter plot. The team encoded the years of unhealthy life as the thickness of the bars in this example. They had also planned to taper the thickness to show direction from 1990 to 2010 in later iterations.

In the end they decided to moved away from this version.

The chart bordered on not being able to fit on one screen, which would affect comprehensibility, and I personally found the amount of overlap in Africa to be overwhelming, to the point in which the patterns are harder to draw out.

They also tried a mirrored bar graph that looks a bit like a population pyramid. It seems they were playing with bar thicknesses here as well, but in the end:

The bar chart didn't go far. The graphic was just too deep. It's hard to compare Afghanistan with Zambia and was fairly one dimensional. It's easier to understand the number of years lost to disability, but the strengths pretty much stop there.

--
Scatterplots were on the table as an option from the beginning of the process. The original studies the team was provided with contained a number of them. We saw something similar happen when the WIDE visualization extended precedents set in the original tableau charts created by UNESCO.

Emily and the team tried a number of iterations using the scatter plot form:

If we overlay the years much like the Institute of Health Metrics and Evaluation did, the scatterplot tells more of a story but also gets hairy and cluttered in the middle where most countries are positioned. We also considered building out smaller multiples laid out in a grid, but the points became too small to interact with.

She also drew inspiration from an earlier Washington Post graphic published by Wilson Andrews giving a global look at body mass index and diabetes. Here the labels help the chart tremendously and hovering over any point reveals its trail over time, cleaning up any hairball that might result from displaying them all.

However, in the case of the Life Expectancy chart, I don't think trails and labels would solve the ultimate problem of showing the smaller percentage of healthy years in 2010.

To remedy this I would suggest the team revisit the parallel coordinates option. Here I've come up with a sketch that addresses the concerns that led the team away from this type originally (space and visual noise).

The chart (a pair of slopegraphs) separates life expectancy from healthy years and displays the trends independently. Basic filtering and hover states make comparison much easier. Showing all countries would get quite complicated but I believe it would still show the overall trends and outliers. Showing all data points simultaneously is not necessarily a visualization requirement. The slopegraph also works particularly well with this small data set that spans only 2 points in time.

While the scatter plot displays all the data efficiently it doesn't communicate that ultimate insight that the team discovered. This is part of the reason it's important to explore less common chart types. In this case the team did explore other options but ruled them out, some because they were less effective but others because of space and filtering concerns. It's important to note that these types of design challenges are easier to overcome than delivering your message through a blurry lens.

Thank you Emily Chow and The Washington Post Graphics team for sharing their process with everyone. I hope they continue to develop their interactive scatterplot and perhaps add a slopegraph to their toolbelt for showing simple trends between two dates.

WIDE Shows the Gaps in Education Equality Around the Globe.

Thu, 07 Mar 2013 00:00:00 +0000

After Interactive Things (IXT) published their visualization, World Inequality Database on Education (WIDE) I decided to sit down with one of the chief architects, Christian Siegrist, to get a look at the design process for this piece.

Christian told me that WIDE visualizes important data in disparity, deprivation and marginalization in education worldwide. The creation of this visualization stems from the need to make the underlying education data more accessible to researchers, teachers, journalists and policy makers. IXT defined these groups as the visualization’s target audience.

Structure

As all creators do, Christian and the IXT team started by exploring the raw materials they were working with. For them, the data exploration process yielded the shape, structure and granularity of the data through a series of interim visualizations.

Created with Tableau, this overarching visualization attempts to show every intersection between the attributes in the dataset. In doing so it reveals patterns repeated across countries, making comparison from a bird’s eye view easy.

The structure of these attributes and the results of the initial visualization led the team to create more tableau charts that draw heavily on some of UNESCO’s earlier visualization techniques. The goal of each is to show a more fine grained “spread” that went beyond the broad averages, down into its detailed segments and demographic groups.

From Christian:

“You don’t always need to re-invent the wheel. You do have to vet the process and the form to make sure it makes sense.”

Functionality

To determine the functional capabilities any final visualization would have, IXT workshopped the central questions at hand. Together, in the same room, the team wrote down the main questions a user from the target audience might have.

They ended up grouping these questions into different uses for the tool, i.e. find, explore and ask. They also arranged the questions on a continuum from general to specific to help organize their thinking.

Once the team had their questions refined and organized, they were free to envision the types of visualizations that might best answer these questions.

The part of process that’s more or less invisible is the team’s discussion that lead them away from complex parallel sets and radial charts towards sketches in the bottom right that resembled UNESCO’s existing charts.

These initial sketches generated a long trail of mockups, prototypes and high fidelity designs by the IXT team. These designs try to illustrate how some visual forms could look and work with the WIDE source data. They explore how the data can be viewed across countries or across WIDE indicators and helped the team tease out how these different cross cuts to the data would be used in the final product.

Visual Design

Besides the original Tableau visualizations, IXT shared that they drew from several visualizations as visual and functional precedents. Chief among them is Jan Willem Tulp’s Ghost Counties for its treatment of links and nodes between data points.

The structure of the data attributes revealed several tiers of information, each with its own set of functionality. This concept fit nicely with the pattern developed by 37Signals in their recent redesign of Basecamp and WIDE draws heavily on this layered page metaphor.

Final Product

The final platform is the culmination of IXT’s design thinking distilled into an interface built for exploration. The front page of the site helps explain the information architecture that goes from countries to demographics to individual indicators. There are also many ways on this page to dive into specific stories without starting from scratch. There’s the newly added Popular Views in the righthand column that gives you a sense of what other users are interested in.

There’s also an interesting slideshow further down that presents short stories centered around 6 goals of the Education For All initiative and how they're supported by the data being presented. This is a great way to communicate a sense of purpose and entice users to explore some of the most compelling stories presented in the platform.

Exploring the data by one of the WIDE indicators reveals a sorted list of countries with the full disparity between all of their demographic groups laid out on one line. The can be ordered by the average amount or total range of disparity, giving you an interesting ranking that begs further exploration.

While exploring a country + an indicator you’re able to see these demographic groups broken out much better. They represent the same line seen in the overall indicator view but paint a much better picture of disparities within countries. Often the greatest disparities are seen in gender and wealth, though some countries have very different patterns.

Countries can also be explored as top level elements so that you can compare the disparities for all of WIDE’s indicators at the same time. This view becomes particularly powerful when you filter down the results and can see how a single demographic group fairs across the various indicators.

At its most complex WIDE can display disparities cross cut by multiple demographic groups. This allows you to see the difference between the poorest female rural group and the richest male urban group for any of WIDE’s indicators.

Because the user has near ultimate control of the way the data is filtered and cross cut it’s crucial that the complex exploratory system is explained. At the top of each page IXT included a help graphic that illustrates how to read the individual lines from most deprived to most privileged. The layered page metaphor also goes a long way to explaining the current view quickly.

However, as we saw in Google's Consumer Barometer, labels in sentence structure could go along way to make this complex visualization more understandable. The easiest way to accomplish this would be in the tooltips for any group. Instead of saying:

“Gender: Male, Urban/Rural: Urban, Wealth: Poor, 44%”

it instead could read:

“44% of Nigeria's Urban Poor Males have Less than 4 year of schooling”

The relationship between the myriad of dimensions could also be explained even more clearly. As it currently stands, the three main representations of the data exist on separate pages. To some extent each is a more detailed breakdown of the last so I wonder if there's a way to actually transition between them. Clicking on a country could animate down a list of the demographic groups and clicking on a group would break it out further by the remaining demographic types.

Social Extensions

Since WIDE was first launched, IXT has added a ‘Popular’ section showing individual views within the platform that are viewed more often. What's interesting here is that WIDE is essentially publishing its analytics of platform use as a feature and the feature works well. It’s a more social way to deliver interesting stories to the forefront and it lines up nicely with the ability to share or export any page within the report. These social extensions will become more common as larger visual reports like these become easier to produce.

Insights

Thanks to Christian’s insights into the IXT process we are able to get a better understanding of the conditions that went into building the impressively complex WIDE report. Publishing their process is old hat to Christian and the IXT team though. They have a whole series of similar ‘inside’ posts on their news site, datavisualization.ch. It’s an exercise in self-discipline to write these postmortems for your own projects but IXT believes in the process because it helps them grow as designers. With just a little effort along the way data visualization practitioners can give us a glimpse of their design thinking. And for the visualizations that don’t have detailed postmortems, The Why Axis is at your service.

The New York Times and Tesla Motors Engage in Chart Warfare

Sat, 16 Feb 2013 00:00:00 +0000

Tesla, the car maker named after the famed electrical engineer, has been attempting to revive the dream of the electric car with its powerful Model S. The company’s latest PR nightmare comes after NYT writer John Broder published an account of his test drive experience. The piece includes a graphic depicting the trip, annotating recharges and the eventual breakdown along the way.

Graphics produced by The New York Times come with certain expectations of quality and this annotated map fails to measure up to them. It’s an extremely unclear way to communicate information about distance traveled and expected range and ends up confusing the Time’s readership more than enlightening them.

Elon Musk, Tesla’s founder, has been firing back, most recently publishing the logged data from the test drive in a series of charts.

These charts haven’t silenced the critics quite the way Elon hoped they would. Instead the’ve added fuel to the fire after being further analyzed and compared to Broder’s story.

But all the back and forth on small details clouds the big picture. Each side ignores some of the larger points which I’ve tried to illuminate in this graphic.

First, there is definitely a significant drop in expected range when the car is parked overnight in Groton. This is visible in Elon’s charts and in Broder’s report when he needs to make the trip to Norwich to refuel.

Second, the Tesla always reached its destination when the range indicated it could do so. The only instance in which the car fell short of its destination was when the range indicated it could only make it 50% of the way. Even then it ran for an additional 20 miles while the range read 0.

Beyond that, it’s almost impossible to tell with any detail whether the range indicator is over-optimistic in communicating to the driver. There remain discrepancies in the speed and conditions of the car as it traveled which probably won’t be solved by warring charts and graphs. However, it could be said that having 4 numbers representing your range in miles has the potential to confuse. (image from Fresh Dialogues)

No matter how you spin it, this is a lot of bad press for Tesla motors and, in this case, the bad press may not be better than no press at all. The electric car has already had its share of trouble and with a leader like Elon Musk at the helm, Tesla should be the one to do it.

Background Reading

Original NYT Article

Musk's Rebuttal and data

The Atlantic Chart Analysis

CNN's Tesla Test Drive (spoiler, they made it)

Breaking Excel Defaults – A Government Chart Remake

Tue, 29 Jan 2013 00:00:00 +0000

Guest author Jon Schwabish – an economist with the U.S. federal government and creator of policy-relevant data visualizations. You can reach him at jschwabish@gmail.com or by following him on Twitter @jschwabish.

To download the excel file for this post (data tables and charts) just click here.

I spend a lot of time looking at different visualizations produced by government agencies, universities, think tanks, and other policy shops. Unfortunately, many of those graphics fail to convey a clear, visual story. I can generally group those shortcomings into two groups:

Visualizations that fail to follow best data visualization practices such as avoiding exploding pie charts or 3D column charts; and
Visualizations that use appropriate types of visualizations, but use software defaults such that the graphic does not tell the whole story, is not clear, or is not visually appealing.

In that vein, a recent graph by the Bureau of Labor Statistics (BLS) struck me, not because it did not use the appropriate chart type, but because it used default Excel design and colors. More importantly, the graph does not convey the story BLS was trying to tell.

The purpose of this post is to walk you through my thought and design process as I revise this simple chart, a set of revisions that I hope will result in a graph that does a better job telling the story. I use Excel throughout the redesign process of this graph because, as you’ll see, it’s clearly the tool BLS used. Further, I want to demonstrate that by avoiding default settings, analysts can make high quality graphics using a relatively basic tool like Excel. My revisions could, however, be made in R, Tableau, or other graphics tools. I suspect, however, that it would be more difficult in packages such as Stata or SAS, which are used by many government analysts.

I don’t claim to have come up with the “correct” answer, but my hope is that more discussion of this sort will encourage government agencies to make better visualizations and thus tell their stories and share their data in better ways.

Job Openings in November 2012

In early January, as part of its “The Editor’s Desk” (TED) publication series, the BLS published this bar chart of job openings by industry. There are two tabs on the screen shot; one tab has the chart and the other tab has the chart data in a simple table.

Let me start with the text that sits below the actual chart. In case it’s not clear in the screen shot, here it is:

From November 2011 to November 2012, job openings increased most in retail trade (144,000, within the trade, transportation and utilities industry) and health care and social assistance (91,000, within the education and health services industry).
Government job openings increased the least, by 6,000.
These data are from the Job Openings and Labor Turnover Survey. Data for the most recent month are preliminary and subject to revision. For additional information, see Job Openings and Labor Turnover — November 2012” (HTML) (PDF), news release USDL-13-0015. More charts featuring data on job openings, hires, and employment separations can be found in Job Openings and Labor Turnover Survey Highlights: November 2012 (PDF).

Notice how BLS discusses the change in job openings between November 2011 and November 2012. Although the October 2012 values are presumably important because they illustrate the month-over-month change, they are not mentioned in the text. Does that mean BLS does not find those values important? If not, why are they included in the chart at all? And if the October 2012 value is not important and the change between November 2011 and November 2012 is the central story, why not just show that change? For the moment, let those questions marinate a bit as we look more closely at the chart.

Okay, onto the chart. The chart itself is quite simple: it’s a horizontal bar chart of job openings (in thousands) for 7 different industries in 3 different months. Relative to a lot of visualizations I’m sure you have seen, here is what I liked:

Horizontal orientation. This orientation means that industry labels are not rotated or vertical or in some other alignment that makes it difficult to read as some might do in a vertical column chart.
Values measured in thousands. This results in the (appropriate) omission of a lot of zeros in the axis labels.
Legend order matches the bars. For example, November 2012 is at the top of both the legend and the set of bars.
Title is left-aligned. The title stands out from the rest of the chart and is descriptive. This is a simple chart and all of the pertinent information is contained in that simple, descriptive title.
Lightened vertical gridlines. This helps emphasize the data (although I would also lighten the x-axis line).
Source. The source is clear and placed at the bottom of the chart.

Now, here are things I think can be improved:

Text. As mentioned, the content of the chart does not match the text.
Vertical axis tick marks. Bar/column charts generally do not need tick marks because the column labels serve to separate the groups; in this case, the tick marks are just unnecessary lines in the chart (so-called chartjunk).
Legend. They legend forces the reader to scan back and forth between the chart content and the legend; a better strategy is to weave the two together (although this strategy doesn’t always work as I’ll demonstrate in the redesigns below).
Order of the industries. The order of the industries seems completely arbitrary; they are not sorted alphabetically, which is usually also unnecessary but is at least a common strategy. Bar and column charts that use categorical data (as opposed to time series data) are generally best served by sorting the data in some way.
Order of the bars. It seems to me that November 2012 should be the last/bottom bar since it is the most recent value. If this was a vertical column chart, November 2012 would appear on the right side; in the horizontal layout, putting the most recent value at the top seems counterintuitive.
Spacing. Another possible issue is that the three bars are not spaced in a way that accurately reflects the gap in time; that is, the first data point is November 2011 should be much further from the other two bars. This doesn’t bother me as much in the original chart because the point is to show the year-over-year and month-over-month comparisons and not a complete time series. If BLS wants to show the monthly time series, they can easily do so.
Defaults. They use default Excel colors and the default font (Calibri). You don’t need to be a graphic designer to make better color choices than the defaults Excel provides. (Note also that many people with color blindness often cannot distinguish between red and green).

Redesigning the Chart

When I first started thinking about redesigning this chart, I started simple: So what if BLS didn’t mention October 2012 in the text? There’s a whole report that accompanies the graphic and this graph is supposed to act as the teaser. Fine. So, for this exercise, I started by fixing the basic things in my above list: drop the y-axis tick marks, move the legend onto the chart, sort the industries (descending by the November 2012 value), and change the colors and font (lately, I’ve been into the sans serif Cabin font, which is available for free here).

Two quick notes about this approach:

You don’t need fancy color schemes to make basic charts—I used a basic blue color scheme here and made the darkest color the most recent value (which is also now on the bottom). The consistent color scheme makes the chart more cohesive and the darkest color helps the reader focus on the most recent value, which I think is the most important.
Excel note: For those who are interested, the date labels are inserted to the chart by using three different scatterplot points (an approach similar to that often used by Jon Peltier at www.peltiertech.com).

I like this chart. Simple. Clear. Follows good data visualization practices. And uses all the data provided by BLS (naturally, I would probably revise the text to accompany this chart as well). But I don’t love the way the time series is shown vertically. I am accustomed to seeing time series data plotted horizontally. Some of the long industry labels are potentially a problem, but we can let them run on 2 or 3 lines if necessary. So let’s rotate the figure.

Ah, feels more intuitive. Notice a key change here: I’ve kept the legend and placed it at the top-left of the chart. Because the time series runs from left to right for each category in this approach, if I keep the legend in the same order and lay it out horizontally, it’s more intuitive than having a vertically stacked legend as in the original chart. In addition, moving the labels directly onto this chart ends up looking cluttered, so the legend works better. I’ve also integrated what is now the y-axis label “(Thousands of jobs)” into the title; this avoids having rotated text along the y-axis. The descending sorting (based on the most recent November 2012 data point) and blue color scheme continues to give the chart a more intuitive look and feel. I’ve also added the data labels, which could have been done in the original bar chart as well. Because the data are a quick click away, including them in the chart may not be necessary.

Okay, so this is good. But I want a bit more. The columns are really using a lot of space to show only three data points for each industry. Roughly speaking, the blue columns account for about half of the total area in the graph. That’s not necessarily a bad thing (think of Tufte’s data-ink ratio), but I’m distracted by all the vertical bars and think the reader would be helped by having a clearer visual cue to the top of the bars. That said, let’s give the chart some air and change it to a line chart (well, 7 separate line charts).

That’s a little better; more breathing room. I’ve directly labeled three of the dots here with the dates, but an alternative would be to color-code the points as in the column chart and add a legend at the top. Once that’s done and because there’s so much less ink than in the column chart, I can now label the data values and it shouldn’t be too distracting.

Lovely. Lighter still, plus it includes the data and the color scheme is softer than the Excel red-green-blue default. (I could also color-code the data labels, but am not going to do so here.) But, I still want more. The original BLS text emphasized the change from November 2011 to November 2012, so let’s add that as a set of columns at the bottom of the chart. I’ll also move the legend closer to the top-left of the graph since that’s where most people will start reading.

Excel notes: For those that are interested, just a few points about the Excel work involved here.

I first plotted the three values for each industry as scatterplots and then added the difference series as a column chart.
The legend is also three separate scatterplots with the series name used as the data labels
The “Difference: Nov2011-Nov2012” label is based off a single point; another point is plotted at the elbow of the connector line—the two orange lines are vertical and horizontal error bars.

One other note: Instead of sorting by the value in November 2012, I could have sorted on the change, but the dots then look sort of odd because they don’t descend in the same way.

I mentioned earlier that in the original bar chart the spacing between columns does not accurately reflect the gap between months. This is probably even more of an issue in the line chart, but one that can be easily fixed. I’ve also made the difference columns a bit wider so the graph looks more balanced. (For labeling purposes, I centered the orange columns, but they could be lined up below the November 2012 data points.)

Excel note: For those who are interested, creating this graph simply required moving the data into a monthly layout. To make the orange “Difference” columns a bit fatter than in the previous graph, I added three bars (with no gap width) instead of one column.

To sum up, I moved from the original horizontal BLS chart that probably took about 3 minutes to make, to a blue-colored vertical bar chart that probably took about 6 minutes to make, to a more complicated chart that probably took about 15 minutes to make (not including a few minutes to get the data in the proper format). But I think this graph has more information, has a better color scheme and layout, and (hopefully) tells the story in a more cohesive way.

Is it worth it? Should BLS pursue such a graphic approach? I’m not sure they need to go all the way to create new graphic approaches for each TED chart they release, but at the very least, I hope they start to think a bit harder about the basics: color, font, and good data visualization practice.

One final set of questions. Does any of this matter? So what if one little chart on the BLS website uses default Excel colors and fonts? The story is (mainly) still there, isn’t it? I think that if BLS could improve this chart by using good data visualization practices and strategies, their readers might be better able to understand their data, their analysis, and their stories. In general, if government agencies can improve how they tell visual stories with data—be it about job openings, unemployment rates, farming subsidies, poverty rates, border security, or what have you—then perhaps our understanding of the nation’s public policy challenges will improve as well.

What do you think?

Update: You can now download the excel file (data tables and charts) used in this post.

Posts on Process – November 2012

Tue, 13 Nov 2012 00:00:00 +0000

From Frank Chimero’s The Shape of Design:

“We can get closer to the wisdom of other people by having them explain their decisions – not just in How they were executed, but Why they were made… Asking Why unlocks a new form of beauty by making choices observable so they can be discussed and considered.”

I’m constantly trying to improve the content on The Why Axis and I’m realizing the best way to do that is to get more voices involved. Not everyone can commit to being a regular contributor (though I’m always open to it) so what I’ve decided to do is go out and interview the designers behind the visualizations to get answers to both the “why” and “how” questions that come up during every design process.

My hope is that this kind of post will be extremely useful for anyone who makes or uses modern data visualization. I’ve been seeing some evidence of this in some other blog posts published recently. The following are posts that go behind the scenes and expose the process of creating complex and beautiful visualizations. They reveal practical insights which I hope to be able to bring you from future articles published here.

National Infographic – Charting Weather Disasters

Juan Velasco is the Art Director at National Geographic and his new blog is called National Infographic. In his recent post he talks about “Charting Weather Disasters” an infographic published in the print edition of the magazine and then redesigned for the iPad edition. He walks us through the iterative design process showing early versions and explaining design decisions. The finished piece has a unique look and effectively tells the story of mounting damages incurred by frequent sever weather events.

Mapbox Blog – How we built USA Today’s Election Night Maps

For the geospatially minded, Mapbox has written up an excellent explanation of their technical process for creating a maps application for USA Today. The article doesn’t talk much about the visual design but there is lots of beautiful efficiency, stability and scalability to be found here. There's nothing like building great visualizations on a rock solid platform.

Source – The NYT’s Visual Election Outcome Explorer

Shan Carter and Mike Bostock do a write up of their NYT published 512 Paths to the White House, my favorite visualization of the year. The article has enough technical detail for coders and enough practical insight for any visualization enthusiast. Plus there’s a Easter egg of sorts revealed if you read the article.

The Ask

Are you a practitioner who would like to help the visualization community by talking about your process? Or is there a visualization you’re dying to know the secrets behind? Either way, don’t hesitate to get in touch with @thewhyaxis on twitter or through the contact page.

Oh The Political Possibilities: 512 Paths to the White House

Mon, 05 Nov 2012 00:00:00 +0000

With the outcomes of presidential ballot battles in 80% of U.S. states seemingly decided, all attention is focused on the remaining battlegrounds. Candidates are spending vast amounts of money and time to sway just a few supposedly undecided voters in just a few states.

In such an uncertain climate The New York Times has published one of their final visualizations of the presidential race and I think :it’s their finest.

512 Paths to the White House maps all the possibilities for each candidate to make their way to the oval office. Built on top of a mountain of political analysis and number crunching, this graphic simplifies it all and presents the information in a format that anyone can understand.

As you hover over graphic the interactions are very intuitive and the sentence structure that’s revealed as you cascade down the tree makes perfect sense. The buttons across the top make the visualization more accessible for touch users and help you narrow down the possibilities in case you miss the fact that double clicking isolates your current selection.

The graphic clearly illustrates just how important Florida and Ohio are in this election and helps explain why candidates spend and outsized amount of time there. I’m not sure what happens to this tree if there’s a major upset in one of the seemingly “decided” states but the visualization includes Nevada and North Carolina in with states much tougher to call, perhaps to hedge against this situation.

To me, this is the perfect visualization to have up on election night as the results are rolling in. I’ll be on the edge of my seat, feverishly clicking this decision tree along with news coverage on Tuesday night. The NYT graphics department has had a stellar political season of data driven journalism and delightful visualizations. Congratulations to them for creating a visualizations that’s both addictively fun to play with and vitally important.

Google Consumer Barometer Promises Insight Through Visualization

Thu, 04 Oct 2012 00:00:00 +0000

The Consumer Barometer is a beautiful and immersive visualization presented by Google and produced by CLEVER°FRANKE. Experience it for yourself.

The three main sections of the barometer (browse, graph, and data map) are all variations on a theme of visualizing the wealth of data collected by Google and others about how consumers research and make purchases. “Browse” guides you through a selected set of stories that have been pulled from the data. “Graph” lets you selectively compare and contrast all the metrics collected in the barometer. “Data Map” gives you a sense of the organizational structure of these metrics.

The Consumer Barometer does a lot of things right. The interactions are helpful and plentiful and the whole thing is visually engaging. What’s more, it is clear that restraint and careful visual reduction have been applied to the entire visualization. You’re never overwhelmed with options or controls. What it comes down to is a very well defined information architecture.

Browse

Let’s dive into Browse, probably the most complex section of the tool. The four “stories” you’re presented with are top level trends that the collective team has identified and designed specifically for. Within each story you have the ability to select a region (bubble) and/or industry (hexagon) to see more relevant data. The interactions on these elements are so responsive and entertaining that it’s not immediately clear that you can click on these to explore the data further.

As you browse, I think the strangest thing about this section is the lack of consistency between the “stories” as illustrated by the image at right.

No two have the same flow through the different types of visualizations and not all lead to an “insight” screen in the end. It makes navigation more confusing and the metaphor weaker. I’m sure these choices were dictated by the available data but there’s no obvious reason why ”compare purchase behaviors across countries” can’t start with a country selector like the other three stories.

The other issue here is the navigational labeling. CLEVER°FRANKE and the team have done so much work to make everything semantically correct and readable (“People who are [age range] with [income] purchase [online only]“) but I still have to hover to see this information in most cases. This becomes important when you consider the ability to socially share any page of this visualization with a direct link. Someone landing on a page deep in the tool will have no immediate indicator that says “you’re looking at a visualization for technology purchases in Ireland.” They see the selected story and generic breadcrumbs in the upper left but there’s no true “title for what you’re seeing.” The breadcrumbs could get you half way there if they updated with your selection of country and industry. Those breadcrumbs are pretty important too considering the browser’s back button doesn’t really work within the tool.

The different types of visualization within these screens are designed quite nicely. I find myself drawn more toward the hexagons despite there being no clear reason to stray from the simple circles (except, perhaps, to avoid the bubble plague). I think this attraction is due to the fact that the hexagons display more data more immediately and are therefore more useful to me. I can scan them quickly and see relative values of areas I might want to explore. In the bubble cartogram these values are hidden in the hover state and the filled circle areas are harder to compare.

On of my favorite bits is the care taken in the final “insight” screens. Firstly, the tool automatically compares your selected country to another relevant one and lets you adjust or remove the comparison. What’s more, every element in this screen is interactive in a simple way. Hovers tie the sections of visualization to their labels, making scanning and comprehension even faster. Hovering in one area highlights another element and you’re forced to draw connections between them. These details make the difference in explaining the radar bubble charts which would otherwise fall apart.

Data Map

This section is a little more straightforward and uses a novel navigational construct to show the relationship between the different ideas visualized by the Consumer Barometer. It also serves as an link to the Graph section, allowing you to explore and select the metrics you want to compare.

Graph

The graphing section is built much more for the self directed user who wants to query specific sections of the data collected. This robust yet simple graphing tool lays out all your possible options and combinations and delivers them in straightforward bar graphs that can be downloaded and shared.

CLEVER°FRANKE and the team have certainly raised the bar with this comprehensive visualization that does a great job of communicating the complex.They mention on their project page that:

“The sphere consists of three different environments for users to explore the immense data set, each environment build specifically for different target audience groups.”

Yet I can’t help feeling that these three separate sections are all parts of a whole that longs to be unified into a more powerful visualization. Is it too much to expect a single visualization to guide users through stories, show them the structure of the data and allow them to make their own specific queries and explorations? Has the complex research in the Consumer Barometer been simplified too much? Does the tool give its users enough credit in the level of sophistication it presents them with? The Why Axis and CLEVER°FRANKE would love to hear your answers in the comments.

FF Chartwell Comes to the Web

Wed, 26 Sep 2012 00:00:00 +0000

I’ve created a number of example charts using FF Chartwell here.

FontFont has released Travis Kochel’s Chartwell as a web font. If you weren’t familiar with the font before, it employs a clever use of the OpenType system and has been lauded as “disruptive” and “a graphing tool for the rest of us” by blogs across the web.

FF Chartwell has been available for purchase for some time but its recent release in web font form exponentially increases its potential uses and audiences. While relatively straightforward to work with as a desktop font, manipulating numerical values and the design of the chart is even easier with the web font form. You just need to be able to work with simple HTML and CSS.

For a much longer explanation take a look at Travis Kockel's post about this font.

Instead of replacing number sets with glyphs using OpenType, FF Chartwell for the web abstracts the idea of a “font” again and renders the charts using a series of javascript libraries. FF Chartwell includes seven different stylistic sets for different chart types, some more useful than others.

The vertical bars in FF Chartwell web are straightforward and useful, though it would be nice to be able to condense the space between bars to form more of a histogram. I tried to achieve this effect by layering multiple charts in my FF Chartwell demo page. To create a bar chat with negative values I created a second graph and rotate it so it aligned correctly.

FF Chartwell Lines, though perhaps more accurately labeled as area charts, are also rendered simply. I’m not sure why you’d want to individually color each segment but the font gives you that ability. The Chartwell User Manualalso shows stacked area diagrams, though you need to layer and offset the values manually as I’ve attempted to do in the demo.

Pie charts, at their best, can only hold a few segments before they become unwieldy, but in FF Chartwell Pies you have the option to add up to 100 segments. The donut chart variation makes perception even more difficult and especially requires restraint with the number of segments used. Originally FF Chartwell Pies would create a new chart if its cumulative values exceeded 100 but on the web it seems to just overlap additional segments on top of old ones. Limiting the pie to 100 forced you to chart segments in terms of a percent and avoided any confusion with how to use a pie chart. The overlapping issue in the web font is particularly dangerous because it’s so hard to notice once it’s happened.

Chartwell’s horizontal bars add a margin of error to the way people perceive the comparative lengths. It employs a sort of isometric or dimensional diamond design that seems to represent the need to make bar charts “pop” in a lot of infographics. Despite this trendy shape I actually think it does a good job when used to show parts of a whole (or a simple stacked bar chart, depending on how you look at it).

FF Chartwell Rose, like Florence Nightingale’s original visualization, is really best suited for visualizing cyclical data. Unfortunately it gets used a lot in infographics as a cool variation on the pie chart and I fear the visual appeal of this chart type will see it abused further in FF Chartwell.

The radar style of Chartwell also has lots of potential for misuse and is often hard to read no matter how simply it is rendered. Comparison is only really possible when radars are layered on top of each other as I’ve done in the demo. The number and visibility of axis in the radar are also important to be aware of. They're the only aid to comprehension in this graph type.

FF Chartwell Rings seems like the chart type with the least utility. This representation of data is no where near as easy to read as a simple bar chart. There are very few instances where this design enhances the dat. Because of that there’s a very small likelihood that this chart font will be used for more than decoration.

There’s no doubt that FF Chartwell breaks down a lot of the barriers for someone to create their first chart but I’d argue that their first chart probably shouldn’t use Chartwell Rose or Radar. Travis mentions that Rose, Radar and Rings are a more experimental set for Chartwell but they’re also very eye catching. Like a yo-yo that lights up and makes noise but breaks if you use it the wrong way.

I’m all for the democratization of chart creation and design but I think part of that process has to be education. What I’d propose is that a field guide or instructional manual come with the purchased fonts that explains not just how to use the chart types but why you’d use them.

Travis clearly knows a lot about graphing and visualization. All of his examples use appropriate data for the chart type chosen and it takes a passionate person to put together a tool like this. He used cyclical data for FF Chartwell Rose and even colored the first segment to indicate a starting position. But the importance of this is never communicated to the font purchaser. I’ve used the demo page to start to put together a little guide for using FF Chartwell. Leave a comment and help me improve it for the Chartwell users of the world.

The Map of Life, A Field Guide For The Future

Wed, 12 Sep 2012 00:00:00 +0000

The Map of Life plans to plot every species on the planet when it is complete. Talk about big data. For now it’s in a beta version and visualizes ranges and observations for 25,000 species compiled from a variety of data sources.

Eventually there will be a set of Wikipedia-like functionality added to the map including the addition of user-generated data sets and flagging / editing distribution data. The map will also be able to provide more sophisticated filters, calculations and displays of patters within data and make it all available for download and analysis.

In its current state the map does a lot of things well. Google maps is a familiar platform and the surrounding interface provides you with enough options to visualize almost exactly what you’re looking for. The search is quite powerful though the focus on scientific names is somewhat intimidating to non-experts like myself.

For the map I think a lot more cartographic finess could be applied to the way things are rendering. Perhaps a future iteration using Mapbox and tilemill could reduce issues of contrast and overlap. My personal favorite amongst the map options along the bottom is the pure terrain map with the political boundaries toggled off. There’s something to be said for the pure power of Earth’s geography without the complications of borders. I think part of the goal with a project like this, as we saw with Google’s Endangered Languages map, is to make people care, whether they’re experts or not.

Something else that might aid in both emotional impact and scientific progress would be the inclusion of photos for species listed in the search results. It would sure help me with choosing which species of sloth to map! And I think it would be of use to scientists as well. It also wouldn’t hurt if I could share specific URLs with friends and colleagues, pointing them to a specific version of the map I’ve created, telling a short story about a specific species.

When it comes to the interface I think there are a lot of simplifications to be made that would improve the situation. Many of the menu items surrounding the map (on almost all sides) could be removed in a streamlined iteration. For the search and results I’d rather have a dedicated sidebar than a hovering modal box I have to worry about closing to see a full map. In general the interface just gets in the way of the map experience. Each interaction seems to call up a new window over the map. In future iterations I hope the interface gets out of the user’s way as much as possible so they can be immersed in an enjoyable cartographic experience.

With sponsors and partners including Yale, NASA, The National Science Foundation and more I am hopeful to see the project continue to develop. I do worry for the complexity of the experience at present and how fast progress can be made with the amount of coordination that must need to occur between agencies. As much as possible I think the group should open source parts of the project. Adopt a model that allows people to generate different map styles, edit the data and renderings, and generally enrich the community and experience of what could potentially be a truly significant digital map of the world’s biodiversity.

Edit: Visuallity, run by @Saleiva, has received funding to develop CartoDB in the context of mapping biodiversity. CartoDB is the foundation for The Map of Life and any infrastructure improvements will directly benefit it.