Junk Charts

Visualization as an analysis tool

Visualizing data has many uses. We often explore how charts can be used to convey data insights and tell stories. We talk less on this blog about how slicing and dicing data helps us form impressions about the structure of the data sets we're analyzing.

I have been digging around some payroll employment data recently. (You can find the data at the Bureau of Labor Statistics website.) I thought the following two charts are quite instructive.

The first one surfaces one type of recurring patterns: there is a seasonal pattern running from January to December that repeats every year. I use a small-multiples setup, with each chartlet indiced by year.

The second chart shows a different kind of regularity: there is a cyclical pattern running from 2002 to 2012, no matter which month we're looking at. Again, we have a small-multiples setup, this time with each chartlet indiced by a month of year.

This second chart is a simple form of "seasonal adjustment". The data used in this plot are unadjusted. The chart shows that there is a larger cyclical pattern during the period of 2002-2012 that affects every month of the year.

I already hear grumbling about using a line chart when there is no continuity from one dot to the next. In this chart, in fact, time runs left to right, top to bottom, then starts again at the first chartlet, and so on. This is a profile chart. As the name suggests, we should be focused on the shape of the line. It doesn't have to have physical meaning; we are only looking for regularity.

***

Statisticians love to find this kind of regular patterns because they are easy to describe. Of course, most data are much messier.

There's life without animation

Just as we don't always need a map to do justice to geographic data, we don't always usually need animation to convey time-varying data.

Some examples of good visualizations of time-series information without moving parts include the "horse-race" charts on the Presidential election (link), and the NY Times plot of Olympic race times (link).

Reader Jeff Cole did some nice Web charts (link) that show NFL matches as horse races. The look of these charts is exceptionally clean and easy to understand. They tell us which matches were blow-outs, and which ones were close, and which ones were tales of two halves.

 

This Green Bay-New England match looked like a thriller, with five lead changes, and a final score gap of only 4 points. Green Bay scored pretty much regularly through the four periods while New England had two relatively long droughts in scoring.

I'm not sure if the Margin of Victory section is worthwhile. It seems redundant to me. The labeling can be better, showing that New England leads are above the zero reference line while Green Bay leads are below the line. I'd consider making this a cumulative amount of time in the lead up to a specific point of the match. That would give an extra piece of information that is difficult to grasp from the Game Scoring chart.

This was an exciting match for a different reason. Dallas was always ahead and eventually won by three points. But it was a shootout in which each team scored regularly. Dallas had a scoring drought for most of the seond half which allowed Washington to get even but then scored a field goal before the final whistle to come out on top. I didn't watch the game but this chart tells me all of that.

On the Game Scoring chart, I'd add vertical tickmarks to help read the intermediate scores. Also maybe put dots to highlight the crossover points, which are where lead changes occur.

NFL is a complex game that is difficult to fully capture in a simple chart like this. It would be nice to add some extra event indicators, such as interceptions, fumbles, and other turning points. That's easier said than done, especially when trying to automate the graph production.

The electoral map sans the map

Xan G. has a must-read post comparing different ways of showing the electoral map. See here.

The key learning is something I often point out on this blog: geographical data can have a greater impact when it is unshackled from the map.

Xan pointed to a series of ideas that are improvements upon the map.

Here's an attempt to portray the election night as a horse race. This borrows an idea from the sports world where a baseball game can be portrayed with such a chart.

I love this sort of presentation. Similar to a baseball game, someone can look at this chart after the fact and experience the ups and downs of an Obama/Romney supporter without actually being there.

Then Xan spoils some of the fun by transforming the above into the following chart, which portrays Obama's win as a rout. All the suspense is gone!

As Xan explains it, he took Nathan Silver's predictions of "sure wins" and plotted those first. Thus, Obama started the night at almost 200 while Romney started with about 170.

While indeed the fun is gone, this is a more accurate view of this just-concluded election. I was a spoilt sport myself that night as I kept telling my friends that the only reason why Romney seemed close at the start was that the Red States generally have smaller populations, and thus took less time to count their votes. In addition, the Red States also tend to favor Republican candidates by very large margins so that the winner could be called early without counting most of the votes.

***

I have other thoughts on the state of reporting on polls, which I'll cover in a later post.

Guest blog: Popcorn infographics

Note: This post is by Aleksey Nozdryn-Plotnicki, who blogs at ThinkDataVis.

***

On my way to Crete recently, I was flipping through the in-flight magazine when I stumbled upon this treat. This full-page piece was about Claire Cock-Starkey’s upcoming (at the time) book, Seeing the Bigger Picture.

The book sells itself as “Global Infographics” and the article says it is “swapping dry words for colourful illustrated visuals”. The baby and the iPhone are pure decoration, but there are also some information graphics here at the top and the bottom which bear a closer look.

Above we have what at first looks innovative, but is actually a disguised bar chart. That’s fine, but:

• Bars have been arched, challenging our ability to compare them
• Outer bars actually have further to go as the radius and therefore circumference increases. So while Japan has the lowest percentage, its bar appears to be equally as long as that of Norway, the largest. In fact, since the values are sorted, for the most part all bars are the same length and size.
• The legend is far larger than the chart itself, and is what really delivers the information at all. Using that space for a larger chart and labelling the bars directly (like in a usual bar chart) might be better.
• There is no axis with any ticks or labels
• The chart has too many categorical colours, so knowing what any colour represents requires looking it up in the legend where the raw data is anyway.
• Why this circular shape? I suspect it was a clock-face for time, but the decoration, presumably informing our sense of “leisure activity” has removed the clock hands, so the metaphor is weak.
• Why does the Norway bar go only 90 degrees around? This seems equivalent to not properly scaling the Y-axis on a bar chart and leaving copious empty space above. Maybe this is meant to indicate that even the most leisurely Norwegians only have time for gardening, being a kite, and drinking at a table.
• Consolation points, however, for taking the time to clearly state what leisure time was defined as in this data.

At first this looks more like a traditional bar chart, until you realise that:

• Larger data is at the top and smaller at the bottom, so the data is tied to the blue lines on the left, rather than the visually-weighty bars on the right. Or maybe the height of the pyramid is meant to be tied to age at marriage?
• Bars are artificially grouped and forced to be the same length, i.e. Sweden 34.3 and Germany 33.7. This leads to a “lie factor”.
• In any event the data is so loosely encoded that it can hardly be considered encoded at all. The lines and the data are both sorted.
• It has a non-zero baseline at roughly 20 or so, a “sin” in bar charts, though you could argue for a non-zero baseline of around 18 for marriage since you would never expect to see values below that

Ultimately, what I think we have here belongs in a genre of its own, perhaps “popcorn infographics”.  At the time of writing the one review on amazon.co.uk reads “Bought this for my 14 yr old - absolutely loves it and showed friends who were also suitably impressed. Thank you” which says a lot,  and not all negative. Perhaps there is room for popcorn infographics in this world or perhaps it’s just junk.

***

Aleksey Nozdryn-Plotnicki an analyst/consultant and data visualisation blogger at ThinkDataVis.com. He is @alekseynp on Twitter.

 

Purists stay clear; many rules to be broken

This chart, from Internet Retailer (March 2012, p. 26), is okay, at least they didn't use pie charts. But it could have been much more effective.

To make it better, we have to break all the rules:

• Use lines instead of columns. The following reproduces the right side of the chart above which deals with shopping behavior by income groups. One of the issues with this chart is that the gray partition between the gender section and the income section is too incognito.

• Use rounded percentages instead of two decimal places. If we need to see two decimal places in order to tell two categories apart, then the difference is surely insignificant.

• Round up group boundaries. Notice that I committed the sin of imprecision by not specifying whether $60,000 belongs to $30-60K or $60-100K. Is that level of precision necessary? Is it worth the ugliness of printing a number like $59,999? If there is information in whether $60,000 belongs to $30-60K or $60-100K, what does that say about your analysis?

• Dare to throw away information. Look at this final version of the above charts:

I have collapsed the five income groups into two. That's because the bottom three income groups are more or less the same in terms of online/offline shopping behavior, and the top two groups are more or less the same. Bear in mind that any analysis of this type has a margin of error so differences of a few percentage points are not worth representing. It is possible that even the 10 percent or so differences between the two remaining income groups are not meaningful--we won't know unless we know the margin of error of their methodology.

(Note that you have to consult the Census to get the relative proportions of each income group in order to average the data shown in the original chart.)

In any case, the key message of the article becomes a lot clearer than in the original chart.

• Use black and white instead of color. Having simplified the presentation--but retaining all the important information--we no longer need colors.

Start breaking some rules today, and you'll make better charts.

Fifteen points, add confusion

Bloomberg Markets (November) has this chart showing national debt as a percentage of GDP for selected countries.

If you count the points, there are only 15 on the chart, three annual numbers for each of five countries.

The theme appears to be redundancy. Countries are identified by both their names and their flags, one placed vertically and one placed horizontally. The 2011 debt-to-GDP ratios are displayed twice, once as the rightmost points on those lines, and once as data labels on the far right. The vertical axis plus gridlines too provide information that is quite unnecessary when there are labels on the right.

The five dots soak up much of our attention but there is no data in them. In particular, the Greek and U.S. dots fall in between two years where the straight lines are just convenience.

If all the lines were given the same color, then it would be straightforward to highlight the U.S. data by giving it a different color. The switch to an area chart for U.S. data is both ugly and distorting. The distortion is due to the vertical axis not starting at zero -- this is acceptable for line charts but certainly not for area charts.

***

I find that the data themselves present a challenge for interpretation. Based on my reading of economic news, Greece, Spain and Italy are current poster children for countries with debt problems while Japan has had a debt problem but is more associated with a stagnation problem more recently.

The issue is the comparison of each of these countries to the U.S. brings almost zero insight. Italy, for example, probably has some kind of legislative control that puts a lid on debt at 120 percent of GDP. Japan, while not currently in the conversation about countries in danger of default, has much worse debt-to-GDP ratio than Greece, which is in serious trouble. By this measure, Spain is nowhere near as indebted as the U.S. and yet it is in much trouble.

In the following version, I plotted the relative change in the metric with 2009 set to 100.

When simple is too simple

Thanks to reader Don M, I came across this fascinating chart published in the New York Times Review recently (link). The main article, about gender segregation in job categories, is found here.

This is one of those charts that require a reader's guide.

The chart shows the proportion of women in each job category in year 1980 and in year 2010 (and nothing in between). The jobs are divided into three large chunks: the top chunk (shaded) consists of jobs in which women account for more than 70 percent of the total; the middle chunk (white background) are those jobs with 30 to 70 percent women; the bottom chunk (also shaded) are jobs with more than 70 percent men.

The designer then uses the red, green and gray colors (apologies to the color-blind folks) to group the jobs into three clusters. This is usually a great idea except that it is poorly executed here. Don is very annoyed with this because these colors lead the readers to the wrong conclusion, and I agree.

***

The color scheme is unnecessarily convoluted. Here is an alternative I prefer:

• if the change is 5 percent or less, color as gray no matter where the line is. (It is insane to color the line for housekeepers "red" for going from 87 to 89 percent in 30 years). 
• if the change is over 5 percent in the female direction, color it red to indicate the occupation is becoming more female. (There would be many red lines, such as for managers in education, HR staff, social workers, architects, etc.)
• if the change is over 5 percent in the male direction, color it blue to indicate the occupation is becoming more male (There would be only one blue line, and that is for welfare service aides.)

This would mean the lines for dentists and architects would be labelled progress. So too with most of the jobs that were predominantly male in 1980. In fact, there really isn't any occupation that went backwards--all those red lines in the bottom shaded chunk indicate shifts of only 1 to 4 percent, over 30 years!

This conclusion usurps the premise of the column in which the author claims that the conventional wisdom is wrong.

***

The other precaution in reading this chart is to realize that each occupation is put on equal footing in this chart even though some job categories employ a lot more people than others. Also confounded with this data is the differential growth/decline in job categories over the 30-year period. Further, the proportion of women entering the labor force must be accounted for.

This is a case in which less is less. The structure of the problem is complex, and it requires a more sophisticated approach.

Expanding circles of error

Reader James H. spotted this offensive pie chart in Forbes (link).

This chart tells us that emerging markets will be responsible for the greatest growth in medical spending up to 2016.

It is hard to find this message in the chart. The gray sector for Japan in 2006 reads 10%, the exact same number as the gray sector in 2016, which appears several times as large. In a pie chart, it is hard enough to compare the sectoral areas within a pie, let alone sectors of different-sized pies.

James noticed that the pie areas are incorrect. The 2016 pie should be roughly double the area of the 2006 pie. This is not the case. It seems like the radius of the 2016 pie is at least three times larger than that of the 2006 pie.

***

As usual, a line chart brings out the trend more clearly:

 

***

The projected numbers should be clearly labelled as such. "2016" should read "2016P". I'm not sure if the 2011 number was projected also - depends on when the data source was published.

The worst thing about this chart is it's completely misleading. It fails to recognize that there are many billions of people in emerging markets and "rest of the world" while U.S, Europe and Japan combined have just over one billion people. Thus, all this chart is really saying is that population growth in the next several years will mostly occur in emerging markets. One can substitute medical spending with any kind of mass market spending and have essentially the same picture.

Below are a rough estimate of the per-capita medical spending by region using population sizes in 2011. For emerging markets, I have substitued BRIC i.e. Brazil, Russia, India and China, which underestimates the population and thus overestimates the per-capita spend. These parts of the world spend a fraction of what industrialized countries are spending. So what's the story?

 

 

Gelman remakes a grouped bar chart

Like Gelman, I also cannot stand grouped bar charts. In this recent post, he tries making the chart several ways.

A general lesson learned here is that there really is no one-size-fits-all charts. Multiple line charts often work well as a replacement for grouped bar charts but in this instance, Andrew didn't like the lines being too crowded together. If the lines (i.e. data) look different, then the line chart might suffice. The point is: just make the chart, look at it, and decide if it's good enough.

I left a comment suggesting that in his last chart, it might be more telling if the data are turned into indices so that every line starts from the same level, and then we observe the percent changes over time.

Mountain, molehill

Reader Jordan G. wasn't impressed by an attempt to visualize medal counts by country and sport in the Olympics over 112 years by Christian Gross at Visualizing.org (link).

The author chose to use the metaphor of "mountains" to portray the cumulative medals earned by each country. Each country is treated as a unit in a small-multiples-style presentation (see right). The bars represent different sports, and they are arranged as if arranging lanes in a swimming contest, with the largest haul in the middle, and second largest on the left, third largest on the right, etc.

This exercise highlights two important considerations from the designer's perspective.

The first is scaling. You'll notice that the first page (for Athens, 1896; excerpted below) is essentially unreadable. This is because the designer uses the same scale for every single page, and because he is plotting the cumulative number of medals over time. These two decisions mean that the initial pages would have much lower values than the latter pages. 

It also means that on other pages, the extreme values walk off the edge of the chart area. (I think the reason is that if the scale has been tailored to present these extreme values, then pretty much every chart that doesn't contain extreme values would become unreadable.)

The choice of making countries units (discussed further below) makes for some awkwardness in latter years as the medals became more spread out among more countries. In the first Olympics, only 10 countries won any medals but in 2008, 127 different countries won at least one medal, among which 50 countries or so had never won more than 20 medals in all sports combined. This skewed distribution causes the designer to break one of the cardinal rules of small multiples, which is that the design of each unit must be the same, with only the data varying. Here, the top countries have their data plotted on a different scale from that of the other countries, as we can see from the different sized squares. When you mouse over a particular bar for a particular country, that sport is now colored red and corresponding bars are highlighted for every other country -- the problem is that the scales are not the same so the lengths of the bars give us misleading comparisons.

***

The second consideration is pagination. The data set has three dimensions: country, sport and time. In this presentation, the designer places sport within country within time. Put differently, the time categories are placed furthest apart - in fact, the reader must load a different page to see the evolution from one Olympics to another. The sport categories are placed closest to each other, in the same chart unit and so it requires the least effort for the reader to compare the number of medals won by the US in athletics (say) compared to gymnastics.

This goes back to the top corner of our Trifecta Checkup. What is the most important question the designer is trying to address? If it is the evolution over time, then the time dimension should not be placed furthest apart. If it is comparison across sports, then the sport dimension should be placed innermost.

For me the country dimension is the least important because everyone knows the US typically wins the most medals, and the top 5-10 countries are quite stable. Within a sport, I might wonder if certain countries are dominant in certain periods, and if certain countries started developing particular sports from a certain time period onwards. In this case, I'd place country and time within sport. 

The following gives an idea of an alternative way of visualizing this data:

Apologies for not completing the dataset. Both charts are missing countries as well as years of history. But you can see where I'm going with this. There would be one chart per sport.

In gymnastics, we see that the US and China are latecomers, Russia has been the superpower until recently while Japan and Germany have stagnated.