Bubble charts, ratios and proportionality

A recent article in the Wall Street Journal about a challenger to the dominant weedkiller, Roundup, contains a nice selection of graphics. (Dicamba is the up-and-comer.)

Wsj_roundup_img1


The change in usage of three brands of weedkillers is rendered as a small-multiples of choropleth maps. This graphic displays geographical and time changes simultaneously.

The staircase chart shows weeds have become resistant to Roundup over time. This is considered a weakness in the Roundup business.

***

In this post, my focus is on the chart at the bottom, which shows complaints about Dicamba by state in 2019. This is a bubble chart, with the bubbles sorted along the horizontal axis by the acreage of farmland by state.

Wsj_roundup_img2

Below left is a more standard version of such a chart, in which the bubbles are allowed to overlap. (I only included the bubbles that were labeled in the original chart).

Redo_roundupwsj0

The WSJ’s twist is to use the vertical spacing to avoid overlapping bubbles. The vertical axis serves a design perogative and does not encode data.  

I’m going to stick with the more traditional overlapping bubbles here – I’m getting to a different matter.

***

The question being addressed by this chart is: which states have the most serious Dicamba problem, as revealed by the frequency of complaints? The designer recognizes that the amount of farmland matters. One should expect the more acres, the more complaints.

Let's consider computing directly the number of complaints per million acres.

The resulting chart (shown below right) – while retaining the design – gives a wholly different feeling. Arkansas now owns the largest bubble even though it has the least acreage among the included states. The huge Illinois bubble is still large but is no longer a loner.

Redo_dicambacomplaints1

Now return to the original design for a moment (the chart on the left). In theory, this should work in the following manner: if complaints grow purely as a function of acreage, then the bubbles should grow proportionally from left to right. The trouble is that proportional areas are not as easily detected as proportional lengths.

The pair of charts below depict made-up data in which all states have 30 complaints for each million acres of farmland. It’s not intuitive that the bubbles on the left chart are growing proportionally.

Redo_dicambacomplaints2

Now if you look at the right chart, which shows the relative metric of complaints per million acres, it’s impossible not to notice that all bubbles are the same size.


Choosing between individuals and aggregates

Friend/reader Thomas B. alerted me to this paper that describes some of the key chart forms used by cancer researchers.

It strikes me that many of the "new" charts plot granular data at the individual level. This heatmap showing gene expressions show one column per patient:

Jnci_genemap

This so-called swimmer plot shows one bar per patient:

Jnci_swimlanes

This spider plot shows the progression of individual patients over time. Key events are marked with symbols.

Jnci_spaghetti

These chart forms are distinguished from other ones that plot aggregated statistics: statistical averages, medians, subgroup averages, and so on.

One obvious limitation of such charts is their lack of scalability. The number of patients, the variability of the metric, and the timing of trends all drive up the amount of messiness.

I am left wondering what Question is being addressed by these plots. If we are concerned about treatment of an individual patient, then showing each line by itself would be clearer. If we are interested in the average trends of patients, then a chart that plots the overall average, or subgroup averages would be more accurate. If the interpretation of the individual's trend requires comparing with similar patients, then showing that individual's line against the subgroup average would be preferred.

When shown these charts of individual lines, readers are tempted to play the statistician - without using appropriate tools! Readers draw aggregate conclusions, performing the aggregation in their heads.

The authors of the paper note: "Spider plots only provide good visual qualitative assessment but do not allow for formal statistical inference." I agree with the second part. The first part is a fallacy - if the visual qualitative assessment is good enough, then no formal inference is necessary! The same argument is often made when people say they don't need advanced analysis because their simple analysis is "directionally accurate". When is something "directionally inaccurate"? How would one know?

Reference: Chia, Gedye, et. al., "Current and Evolving Methods to Visualize Biological Data in Cancer Research", JNCI, 2016, 108(8). (link)

***

Meteoreologists, whom I featured in the previous post, also have their own spider-like chart for hurricanes. They call it a spaghetti map:

Dorian_spaghetti

Compare this to the "cone of uncertainty" map that was featured in the prior post:

AL052019_5day_cone_with_line_and_wind

These two charts build upon the same dataset. The cone map, as we discussed, shows the range of probable paths of the storm center, based on all simulations of all acceptable models for projection. The spaghetti map shows selected individual simulations. Each line is the most likely trajectory of the storm center as predicted by a single simulation from a single model.

The problem is that each predictive model type has its own historical accuracy (known as "skill"), and so the lines embody different levels of importance. Further, it's not immediately clear if all possible lines are drawn so any reader making conclusions of, say, the envelope containing x percent of these lines is likely to be fooled. Eyeballing the "cone" that contains x percent of the lines is not trivial either. We tend to naturally drift toward aggregate statistical conclusions without the benefit of appropriate tools.

Plots of individuals should be used to address the specific problem of assessing individuals.


As Dorian confounds meteorologists, we keep our minds clear on hurricane graphics, and discover correlation as our friend

As Hurricane Dorian threatens the southeastern coast of the U.S., forecasters are fretting about the lack of consensus among various predictive models used to predict the storm’s trajectory. The uncertainty of these models, as reflected in graphical displays, has been a controversial issue in the visualization community for some time.

Let’s start by reviewing a visual design that has captured meteorologists in recent years, something known as the cone map.

Charley_oldconemap

If asked to explain this map, most of us trace a line through the middle of the cone understood to be the center of the storm, the “cone” as the areas near the storm center that are affected, and the warmer colors (red, orange) as indicating higher levels of impact. [Note: We will  design for this type of map circa 2000s.]

The above interpretation is complete, and feasible. Nevertheless, the data used to make the map are forward-looking, not historical. It is still possible to stick to the same interpretation by substituting historical measurement of impact with its projection. As such, the “warmer” regions are projected to suffer worse damage from the storm than the “cooler” regions (yellow).

After I replace the text that was removed from the map (see below), you may notice the color legend, which discloses that the colors on the map encode probabilities, not storm intensity. The text further explains that the chart shows the most probable path of the center of the storm – while the coloring shows the probability that the storm center will reach specific areas.

Charley_oldconemap

***

When reading a data graphic, we rarely first look for text about how to read the chart. In the case of the cone map, those who didn’t seek out the instructions may form one of these misunderstandings:

  1. For someone living in the yellow-shaded areas, the map does not say that the impact of the storm is projected to be lighter; it’s that the center of the storm has a lower chance of passing right through. If, however, the storm does pay a visit, the intensity of the winds will reach hurricane grade.
  2. For someone living outside the cone, the map does not say that the storm will definitely bypass you; it’s that the chance of a direct hit is below the threshold needed to show up on the cone map. Thee threshold is set to attain 66% accurate. The actual paths of storms are expected to stay inside the cone two out of three times.

Adding to the confusion, other designers have produced cone maps in which color is encoding projections of wind speeds. Here is the one for Dorian.

AL052019_wind_probs_64_F120

This map displays essentially what we thought the first cone map was showing.

One way to differentiate the two maps is to roll time forward, and imagine what the maps should look like after the storm has passed through. In the wind-speed map (shown below right), we will see a cone of damage, with warmer colors indicating regions that experienced stronger winds.

Projectedactualwinds_irma

In the storm-center map (below right), we should see a single curve, showing the exact trajectory of the center of the storm. In other words, the cone of uncertainty dissipates over time, just like the storm itself.

Projectedactualstormcenter_irma

 

After scientists learned that readers were misinterpreting the cone maps, they started to issue warnings, and also re-designed the cone map. The cone map now comes with a black-box health warning right up top. Also, in the storm-center cone map, color is no longer used. The National Hurricane Center even made a youtube pointing out the dos and donts of using the cone map.

AL052019_5day_cone_with_line_and_wind

***

The conclusion drawn from misreading the cone map isn’t as devastating as it’s made out to be. This is because the two issues are correlated. Since wind speeds are likely to be stronger nearer to the center of the storm, if one lives in a region that has a low chance of being a direct hit, then that region is also likely to experience lower average wind speeds than those nearer to the projected center of the storm’s path.

Alberto Cairo has written often about these maps, and in his upcoming book, How Charts Lie, there is a nice section addressing his work with colleagues at the University of Miami on improving public understanding of these hurricane graphics. I highly recommended Cairo’s book here.

P.S. [9/5/2019] Alberto also put out a post about the hurricane cone map.

 

 

 


Water stress served two ways

Via Alberto Cairo (whose new book How Charts Lie can be pre-ordered!), I found the Water Stress data visualization by the Washington Post. (link)

The main interest here is how they visualized the different levels of water stress across the U.S. Water stress is some metric defined by the Water Resources Institute that, to my mind, measures the demand versus supply of water. The higher the water stress, the higher the risk of experiencing droughts.

There are two ways in which the water stress data are shown: the first is a map, and the second is a bubble plot.

Wp_waterstress

This project provides a great setting to compare and contrast these chart forms.

How Data are Coded

In a map, the data are usually coded as colors. Sometimes, additional details can be coded as shades, or moire patterns within the colors. But the map form locks down a number of useful dimensions - including x and y location, size and shape. The outline map reserves all these dimensions, rendering them unavailable to encode data.

By contrast, the bubble plot admits a good number of dimensions. The key ones are the x- and y- location. Then, you can also encode data in the size of the dots, the shape, and the color of the dots.

In our map example, the colors encode the water stress level, and a moire pattern encodes "arid areas". For the scatter plot, x = daily water use, y = water stress level, grouped by magnitude, color = water stress level, size = population. (Shape is constant.)

Spatial Correlation

The map is far superior in displaying spatial correlation. It's visually obvious that the southwestern states experience higher stress levels.

This spatial knowledge is relinquished when using a bubble plot. The designer relies on the knowledge of the U.S. map in the head of the readers. It is possible to code this into one of the available dimensions, e.g. one could make x = U.S. regions, but another variable is sacrificed.

Non-contiguous Spatial Patterns

When spatial patterns are contiguous, the map functions well. Sometimes, spatial patterns are disjoint. In that case, the bubble plot, which de-emphasizes the physcial locations, can be superior. In our example, the vertical axis divides the states into five groups based on their water stress levels. Try figuring out which states are "medium to high" water stress from the map, and you'll see the difference.

Finer Geographies

The map handles finer geographical units like counties and precincts better. It's completely natural.

In the bubble plot, shifting to finer units causes the number of dots to explode. This clutters up the chart. Besides, while most (we hope) Americans know the 50 states, most of us can't recite counties or precincts. Thus, the designer can't rely on knowledge in our heads. It would be impossible to learn spatial patterns from such a chart.

***

The key, as always, is to nail down your message, then select the right chart form.

 

 


Where are the Democratic donors?

I like Alberto's discussion of the attractive maps about donors to Democratic presidential candidates, produced by the New York Times (direct link).

Here is the headline map:

Nyt_demdonormaps

The message is clear: Bernie Sanders is the only candidate with nation-wide appeal. The breadth of his coverage is breath-taking. (I agree with Alberto's critique about the lack of a color scale. It's impossible to know if the counts are trivial or not.)

Bernie's coverage is so broad that his numbers overwhelm those of all other candidates except in their home bases (e.g. O'Rourke in Texas).

A remedy to this is to look at the data after removing Bernie's numbers.

Nyt_demdonormap_2

 

This pair of maps reminds me of the Sri Lanka religions map that I revisualized in this post.

Redo_srilankareligiondistricts_v2

The first two maps divide the districts into those in which one religion dominates and those in which multiple religions share the limelight. The third map then shows the second-rank religion in the mixed-religions districts.

The second map in the NYT's donor map series plots the second-rank candidate in all the precincts that Bernie Sanders lead. It's like the designer pulled off the top layer (blue: Bernie) to reveal what's underneath.

Because all of Bernie's data are removed, O'Rourke is still dominating Texas, Buttigieg in Indiana, etc. An alternative is to pull off the top layer in those pockets as well. Then, it's likely to see Bernie showing up in those areas.

The other startling observation is how small Joe Biden's presence is on these maps. This is likely because Biden relies primarily on big donors.

See here for the entire series of donor maps. See here for past discussion of New York Times's graphics.


What is a bad chart?

In the recent issue of Madolyn Smith’s Conversations with Data newsletter hosted by DataJournalism.com, she discusses “bad charts,” featuring submissions from several dataviz bloggers, including myself.

What is a “bad chart”? Based on this collection of curated "bad charts", it is not easy to nail down “bad-ness”. The common theme is the mismatch between the message intended by the designer and the message received by the reader, a classic error of communication. How such mismatch arises depends on the specific example. I am able to divide the “bad charts” into two groups: charts that are misinterpreted, and charts that are misleading.

 

Charts that are misinterpreted

The Causes of Death entry, submitted by Alberto Cairo, is a “well-designed” chart that requires “reading the story where it is inserted and the numerous caveats.” So readers may misinterpret the chart if they do not also partake the story at Our World in Data which runs over 1,500 words not including the appendix.

Ourworldindata_causesofdeath

The map of Canada, submitted by Highsoft, highlights in green the provinces where the majority of residents are members of the First Nations. The “bad” is that readers may incorrectly “infer that a sizable part of the Canadian population is First Nations.”

Highsoft_CanadaFirstNations

In these two examples, the graphic is considered adequate and yet the reader fails to glean the message intended by the designer.

 

Charts that are misleading

Two fellow bloggers, Cole Knaflic and Jon Schwabish, offer the advice to start bars at zero (here's my take on this rule). The “bad” is the distortion introduced when encoding the data into the visual elements.

The Color-blindness pictogram, submitted by Severino Ribecca, commits a similar faux pas. To compare the rates among men and women, the pictograms should use the same baseline.

Colourblindness_pictogram

In these examples, readers who correctly read the charts nonetheless leave with the wrong message. (We assume the designer does not intend to distort the data.) The readers misinterpret the data without misinterpreting the graphics.

 

Using the Trifecta Checkup

In the Trifecta Checkup framework, these problems are second-level problems, represented by the green arrows linking up the three corners. (Click here to learn more about using the Trifecta Checkup.)

Trifectacheckup_img

The visual design of the Causes of Death chart is not under question, and the intended message of the author is clearly articulated in the text. Our concern is that the reader must go outside the graphic to learn the full message. This suggests a problem related to the syncing between the visual design and the message (the QV edge).

By contrast, in the Color Blindness graphic, the data are not under question, nor is the use of pictograms. Our concern is how the data got turned into figurines. This suggests a problem related to the syncing between the data and the visual (the DV edge).

***

When you complain about a misleading chart, or a chart being misinterpreted, what do you really mean? Is it a visual design problem? a data problem? Or is it a syncing problem between two components?


Morphing small multiples to investigate Sri Lanka's religions

Earlier this month, the bombs in Sri Lanka led to some data graphics in the media, educating us on the religious tensions within the island nation. I like this effort by Reuters using small multiples to show which religions are represented in which districts of Sri Lanka (lifted from their twitter feed):

Reuters_srilanka_religiondistricts

The key to reading this map is the top legend. From there, you'll notice that many of the color blocks, especially for Muslims and Catholics are well short of 50 percent. The absence of the darkest tints of green and blue conveys important information. Looking at the blue map by itself misleads - Catholics are in the minority in every district except one. In this setup, readers are expected to compare between maps, and between map and legend.

The overall distribution at the bottom of the chart is a nice piece of context.

***

The above design isolates each religion in its own chart, and displays the spatial spheres of influence. I played around with using different ways of paneling the small multiples.

In the following graphic, the panels represent the level of dominance within each district. The first panel shows the districts in which the top religion is practiced by at least 70 percent of the population (if religions were evenly distributed across all districts, we expect 70 percent of each to be Buddhists.) The second panel shows the religions that account for 40 to 70 percent of the district's residents. By this definition, no district can appear on both the left and middle maps. This division is effective at showing districts with one dominant religion, and those that are "mixed".

In the middle panel, the displayed religion represents the top religion in a mixed district. The last panel shows the second religion in each mixed district, and these religions typically take up between 25 and 40 percent of the residents.

Redo_srilankareligiondistricts_v2

The chart shows that other than Buddhists, Hinduism is the only religion that dominates specific districts, concentrated at the northern end of the island. The districts along the east and west coasts and the "neck" are mixed with the top religion accounting for 40 to 70 percent of the residents. By assimilating the second and the third panels, the reader sees the top and the second religions in each of these mixed districts.

***

This example shows why in the Trifecta Checkup, the Visual is a separate corner from the Question and the Data. Both maps utilize the same visual design, in terms of forms and colors and so on, but they deliver different expereinces to readers by answering different questions, and cutting the data differently.

 

P.S. [5/7/2019] Corrected spelling of Hindu.


Say it thrice: a nice example of layering and story-telling

I enjoyed the New York Times's data viz showing how actively the Democratic candidates were criss-crossing the nation in the month of March (link).

It is a great example of layering the presentation, starting with an eye-catching map at the most aggregate level. The designers looped through the same dataset three times.

Nyt_candidatemap_1

This compact display packs quite a lot. We can easily identify which were the most popular states; and which candidate visited which states the most.

I noticed how they handled the legend. There is no explicit legend. The candidate names are spread around the map. The size legend is also missing, replaced by a short sentence explaining that size encodes the number of cities visited within the state. For a chart like this, having a precise size legend isn't that useful.

The next section presents the same data in a small-multiples layout. The heads are replaced by dots.

Nyt_candidatemap_2

This allows more precise comparison of one candidate to another, and one location to another.

This display has one shortcoming. If you compare the left two maps above, those for Amy Klobuchar and Beto O'Rourke, it looks like they have visited roughly similar number of cities when in fact Beto went to 42 compared to 25. Reducing the size of the dots might work.

Then, in the third visualization of the same data, the time dimension is emphasized. Lines are used to animate the daily movements of the candidates, one by one.

Nyt_candidatemap_3

Click here to see the animation.

When repetition is done right, it doesn't feel like repetition.

 


Quick example of layering

The New York Times uses layering to place the Alabama tornadoes in context. (link)

Today's wide availability of detailed data allows designers to create dense data graphics like this:

Nyt_alabamatornado_3

The graphic shows the starting and ending locations and trajectory of each tornado, as well as the wind speeds (shown in color).

Too much data slows down our understanding of the visual message. The remedy is to subtract. Here is a second graphic that focuses only on the strongest tornadoes (graded 4 or 5 on a 5-point scale):

Nyt_alabamatornado_2

Another goal of the data visualization is to place in context the tornado that hit Beauregard:

Nyt_alabamatornado_1

The area around Beauregard is not typically visited by strong tornadoes. Also, the tornadoes were strong but there have been stronger ones.

***

The designer unfolds the story in three stages. There are no knobs and sliders and arrows, and that's a beauty. It's usually not a good idea to make readers find the story themselves.


Check out the Lifespan of News project

Alberto Cairo introduces another one of his collaborations with Google, visualizing Google search data. We previously looked at other projects here.

The latest project, designed by Schema, Axios, and Google News Initiative, tracks the trending of popular news stories over time and space, and it's a great example of making sense of a huge pile of data.

The design team produced a sequence of graphics to illustrate the data. The top news stories are grouped by category, such as Politics & Elections, Violence & War, and Environment & Science, each given a distinct color maintained throughout the project.

The first chart is an area chart that looks at individual stories, and tracks the volume over time.

Lifespannews_areachart

To read this chart, you have to notice that the vertical axis measuring volume is a log scale, meaning that each tick mark up represents a 10-fold increase. Log scale is frequently used to draw far-away data closer to the middle, making it possible to see both ends of a wide distribution on the same chart. The log transformation introduces distortion deliberately. The smaller data look disproportionately large because of it.

The time scrolls automatically so that you feel a rise and fall of various news stories. It's a great way to experience the news cycle in the past year. The overlapping areas show competing news stories that shared the limelight at that point in time.

Just bear in mind that you have to mentally reverse the distortion introduced by the log scale.

***

In the second part of the project, they tackle regional patterns. Now you see a map with proportional symbols. The top story in each locality is highlighted with the color of the topic. As time flows by, the sizes of the bubbles expand and contract.

Lifespannews_bubblemap

Sometimes, the entire nation was consumed by the same story, e.g. certain obituaries. At other times, people in different regions focused on different topics.

***

In the last part of the project, they describe general shapes of the popularity curves. Most stories have one peak although certain stories like U.S. government shutdown will have multiple peaks. There is also variation in terms of how fast a story rises to the peak and how quickly it fades away.

The most interesting aspect of the project can be learned from the footnote. The data are not direct hits to the Google News stories but searches on Google. For each story, one (or more) unique search terms are matched, and only those stories are counted. A "control" is established, which is an excellent idea. The control gives meaning to those counts. The control used here is the number of searches for the generic term "Google News." Presumably this is a relatively stable number that is a proxy for general search activity. Thus, the "volume" metric is really a relative measure against this control.