Visualizing the impossible

Note [July 6, 2022]: Typepad's image loader is broken yet again. There is no way for me to fix the images right now. They are not showing despite being loaded properly yesterday. I also cannot load new images. Apologies!

Note 2: Manually worked around the automated image loader.

Note 3: Thanks Glenn for letting me about the image loading problem. It turns out the comment approval function is also broken, so I am not able to approve the comment.

***

A twitter user sent me this chart:

It's, hmm, mystifying. It performs magic, as I explain below.

What's the purpose of the gridlines and axis labels? Even if there is a rationale for printing those numbers, they make it harder, not easier, for readers to understand the chart!

I think the following chart shows the main message of this poll result. Democrats are much more likely to think of immigration as a positive compared to Republicans, with Independents situated in between.

***

The axis title gives a hint as to what the chart designer was aiming for with the unconventional axis. It reads "Overall Percentage for All Participants". It appears that the total length of the stacked bar is the weighted aggregate response rate. Roughly 17% of Americans thought this development to be "very positive" which include 8% of Republicans, 27% of Democrats and 12% of Independents. Since the three segments are not equal in size, 17% is a weighted average of the three proportions.

Within each of the three political affiliations, the data labels add to 100%. These numbers therefore are unweighted response rates for each segment. (If weighted, they should add up to the proportion of each segment.)

This sets up an impossible math problem. The three segments within each bar then represent the sum of three proportions, each unweighted within its segment. Adding these unweighted proportions does not yield the desired weighted average response rate. To get the weighted average response rate, we need to sum the weighted segment response rates instead.

This impossible math problem somehow got resolved visually. We can see that each bar segment faithfully represent the unweighted response rates shown in the respective data labels. Summing them would not yield the aggregate response rates as shown on the axis title. The difference is not a simple multiplicative constant because each segment must be weighted by a different multiplier. So, your guess is as good as mine: what is the magic that makes the impossible possible?

[P.S. Another way to see this inconsistency. The sum of all the data labels is 300% because the proportions of each segment add up to 100%. At the same time, the axis title implies that the sum of the lengths of all five bars should be 100%. So, the chart asserts that 300% = 100%.]

***

This poll question is a perfect classroom fodder to discuss how wording of poll questions affects responses (something called "response bias"). Look at the following variants of the same questions. Are we likely to get answers consistent with the above question?

As you know, the demographic makeup of America is changing and becoming more diverse, while the U.S. Census estimates that white people will still be the largest race in approximately 25 years. Generally speaking, do you find these changes to be very positive, somewhat positive, somewhat negative or very negative?

***

As you know, the demographic makeup of America is changing and becoming more diverse, with the U.S. Census estimating that black people will still be a minority in approximately 25 years. Generally speaking, do you find these changes to be very positive, somewhat positive, somewhat negative or very negative?

***

As you know, the demographic makeup of America is changing and becoming more diverse, with the U.S. Census estimating that Hispanic, black, Asian and other non-white people together will be a majority in approximately 25 years. Generally speaking, do you find these changes to be very positive, somewhat positive, somewhat negative or very negative?

What is also amusing is that in the world described by the pollster in 25 years, every race will qualify as a "minority". There will be no longer majority since no race will constitute at least 50% of the U.S. population. So at that time, the word "minority" will  have lost meaning.

Funnel is just for fun

This is part 2 of a review of a recent video released by NASA. Part 1 is here.

The NASA video that starts with the spiral chart showing changes in average global temperature takes a long time (about 1 minute) to run through 14 decades of data, and for those who are patient, the chart then undergoes a dramatic transformation.

With a sleight of hand, the chart went from a set of circles to a funnel. Here is a look:

What happens is the reintroduction of a time dimension. Imagine pushing the center of the spiral down into the screen to create a third dimension.

Our question as always is - what does this chart tell readers?

***

The chart seems to say that the variability of temperature has increased over time (based on the width of the funnel). The red/blue color says the temperature is getting hotter especially in the last 20-40 years.

When the reader looks beneath the surface, the chart starts to lose sense.

The width of the funnel is really a diameter of the spiral chart in the given year. But, if you recall, the diameter of the spiral (polar) chart isn't the same between any pairs of months.

In the particular rendering of this video, the width of the funnel is the diameter linking the April and October values.

Remember the polar gridlines behind the spiral:

Notice the hole in the middle. This hole has arbitrary diameter. It can be as big or as small as the designer makes it. Thus, the width of the funnel is as big or as small as the designer wants it. But the first thing that caught our attention is the width of the funnel.

***

The entire section between -1 and + 1 is, in fact, meaningless. In the following chart, I removed the core of the funnel, adding back the -1 degree line. Doing so exposes an incompatibility between the spiral and funnel views. The middle of the polar grid is negative infinity, a black hole.

For a moment, the two sides of the funnel look like they are mirror images. That's not correct, either. Each width of the funnel represents a year, and the extreme values represent April and October values. The line between those two values does not signify anything real.

Let's take a pair of values to see what I mean.

I selected two values for October 2021 and October 1899 such that the first value appears as a line double the length of the second. The underlying values are +0.99C and -0.04C, roughly speaking, +1 and 0, so the first value is definitely not twice the size of the second.

The funnel chart can be interpreted, in an obtuse way, as a pair of dot plots. As shown below, if we take dot plots for Aprils and Octobers of every year, turn the chart around, and then connect the corresponding dots, we arrive at the funnel chart.

***

This NASA effort illustrates a central problem in visual communications: attention (what Andrew Gelman calls "grabbiness") and information integrity. On the one hand, what's the point of an accurate chart when no one is paying attention? On the other hand, what's the point of a grabby chart when anyone who pays attention gets the wrong information? It's not easy to find that happy medium.

Dots, lines, and 2D histograms

Daniel Z. tweeted about my post from last week. In particular, he took a deeper look at the chart of energy demand that put all hourly data onto the same plot, originally published at the StackOverflow blog:

I noted that this is not a great chart particularly since what catches our eyes are not the key features of the underlying data. Daniel made a clearly better chart:

This is a dot plot, rather than a line chart. The dots are painted in light gray, pushed to the background, because readers should be looking at the orange line. (I'm not sure what is going on with the horizontal scale as I could not get the peaks to line up on the two charts.)

What is this orange line? It's supposed to prove the point that the apparent dark band seen in the line chart does not represent the most frequently occurring values, as one might presume.

Looking closer, we see that the gray dots do not show all the hourly data but binned values.

We see vertical columns of dots, each representing a bin of values. The size of the dots represents the frequency of values of each bin. The orange line connects the bins with the highest number of values.

Daniel commented that

"The visual aggregation doesn't in fact map to the most frequently occurring values. That is because the ink of almost vertical lines fills in all the space between start and end."

Xan Gregg investigated further, and made a gif to show this effect better. Here is a screenshot of it (see this tweet):

The top chart is a true dot plot so that the darker areas are denser as the dots overlap. The bottom chart is the line chart that has the see-saw pattern. As Xan noted, the values shown are strangely very well behaved (aggregated? modeled?) - with each day, it appears that the values sweep up and down consistently.  This means the values are somewhat evenly spaced on the underlying trendline, so I think this dataset is not the best one to illustrate Daniel's excellent point.

It's usually not a good idea to connect lots of dots with a single line.

 

[P.S. 3/21/2022: Daniel clarified what the orange line shows: "In the posted chart, the orange line encodes the daily demand average (the mean of the daily distribution), rounded, for displaying purposes, to the closed bin. Bin size = 1000. Orange could have encode the daily median as well."]

 

There's more to the composite rating chart

In my previous post, I sketched a set of charts to illustrate composite ratings of maps platforms (e.g. Google Maps, TomTom). Here is the sketch again:

For those readers who are interested in understanding these ratings beyond the obvious, this set of charts has more to offer.

Take a look first at the two charts on the left hand side.

Compare the patterns of dots between the two charts. You should note that the Maps Data ratings (blue dots) are less variable than the Platform ratings (green dots).

For Maps Data, the range is from 30 to 85 (out of 110) but the majority of the dots line up around 50.

For Platform, the range is 20 to 70 (out of 90) and the dots are quite spread out within this range.

This means competitiveness based on Platform is more differentiating among these brands than is Maps Data.

In the previous post, I already noted that the other key insight is that the Maps Data values hang quite closely to the overall average ratings while the Platform values are much less correlated.

***

Another informative observation can be found in the bottom row of charts.

The yellow dots (Developer Ecosystem) are mostly to the right of the overall ratings, meaning most of these brands were given scores on Developer Ecosystem that are higher than their average scores.

That is not the case with the green dots (Platform). For this sub-rating, most of the brands score lower than they do in the overall rating.

***

None of these insights are readily learned from the stacked column chart. A key skill in data visualization is whether one can pile on insights without overloading the chart.

 

 

Visualizing composite ratings

A twitter reader submitted the following chart from Autoevolution (link):

This is not a successful chart for the simple reason that readers want to look away from it. It's too busy. There is so much going on that one doesn't know where to look.

The underlying dataset is quite common in the marketing world. Through surveys, people are asked to rate some product along a number of dimensions (here, seven). Each dimension has a weight, and combined, the weighted sum becomes a composite ranking (shown here in gray).

Nothing in the chart stands out as particularly offensive even though the overall effect is repelling. Adding the overall rating on top of each column is not the best idea as it distorts the perception of the column heights. But with all these ingredients, the food comes out bland.

***

The key is editing. Find the stories you want to tell, and then deconstruct the chart to showcase them.

I start with a simple way to show the composite ranking, without any fuss:

[Since these are mockups, I have copied all of the data, just the top 11 items.]

Then, I want to know if individual products have particular strengths or weaknesses along specific dimensions. In a ranking like this, one should expect that some component ratings correlate highly with the overall rating while other components deviate from the overall average.

An example of correlated ratings is the Customers dimension.

The general pattern of the red dots clings closely to that of the gray bars. The gray bars are the overall composite ratings (re-scaled to the rating range for the Customers dimension). This dimension does not tell us more than what we know from the composite rating.

By contrast, the Developers Ecosystem dimension provides additional information.

Esri, AzureMaps and Mapbox performed much better on this dimension than on the average dimension. 

***

The following construction puts everything together in one package:

Ridings, polls, elections, O Canada

Stephen Taylor reached out to me about his work to visualize Canadian elections data. I took a look. I appreciate the labor of love behind this project.

He led with a streamgraph, which presents a quick overview of relative party strengths over time.

I am no Canadian election expert, and I did a bare minimum of research in writing this blog. From this chart, I learn that:

• the Canadians have an irregular election schedule
• Canada has a two party plus breadcrumbs system
• The two dominant parties are Liberals and Conservatives. The Liberals currently hold just less than half of the seats. The Conservatives have more than half of the seats not held by Liberals
• The Conservative party (maybe) rebranded as "progressive conservative" for several decades. The Reform/Alliance party was (maybe) a splinter movement within the Conservatives as well.
• Since the "width" of the entire stream increased over time, I'm guessing the number of seats has expanded

That's quite a bit of information obtained at a glance. This shows the power of data visualization. Notice Stephen didn't even have to include a "how to read this" box.

The streamgraph form has its limitations.

The feature that makes it more attractive than an area chart is its middle anchoring, resulting in a form of symmetry. The same feature produces erroneous intuition - the red patch draws out a declining trend; the reader must fight the urge to interpret the lines and focus on the areas.

The breadcrumbs are well hidden. The legend below discloses that the Green Party holds 3 seats currently. The party has never held enough seats to appear on the streamgraph though.

The bars showing proportions in the legend is a very nice touch. (The numbers appear messed up - I have to ask Stephen whether the seats shown are current values, or some kind of historical average.) I am a big fan of informative legends.

***

The next featured chart is a dot plot of polling results since 2020.

One can see a three-tier system: the two main parties, then the NDP (yellow) is the clear majority of the minority, and finally you have a host of parties that don't poll over 10%.

It looks like the polls are favoring the Conservatives over the Liberals in this election but it may be an election-day toss-up.

The purple dots represent "PPC" which is a party not found elsewhere on the page.

This chart is clear as crystal because of the structure of the underlying data. It just amazes me that the polls are so highly correlated. For example, across all these polls, the NDP has never once polled better than either the Liberals or the Conservatives, and in addition, it has never polled worse than any of the small parties.

What I'd like to see is a chart that merges the two datasets, addressing the question of how well these polls predicted the actual election outcomes.

***

The project goes very deep as Stephen provides charts for individual "ridings" (perhaps similar to U.S. precincts).

Here we see population pyramids for Vancouver Center, versus British Columbia (Province), versus Canada.

This riding has a large surplus of younger people in their twenties and thirties. Be careful about the changing scales though. The relative difference in proportions are more drastic than visually displayed because the maximum values (5%) on the Province and Canada charts are half that on the Riding chart (10%). Imagine squashing the Province and Canada charts to half their widths.

Analyses of income and rent/own status are also provided.

This part of the dashboard exhibits a problem common in most dashboards - they present each dimension of the data separately and miss out on the more interesting stuff: the correlation between dimensions. Do people in their twenties and thirties favor specific parties? Do richer people vote for certain parties?

***

The riding-level maps are the least polished part of the site. This is where I'm looking for a "how to read it" box.

It took me a while to realize that the colors represent the parties. If I haven't come in from the front page, I'd have been totally lost.

Next, I got confused by the use of the word "poll". Clicking on any of the subdivisions bring up details of an actual race, with party colors, candidates and a donut chart showing proportions. The title gives a "poll id" and the name of the riding in parentheses. Since the poll id changes as I mouse over different subdivisions, I'm wondering whether a "poll" is the term for a subdivision of a riding. A quick wiki search indicates otherwise.

My best guess is the subdivisions are indicated by the numbers.

Back to the donut charts, I prefer a different sorting of the candidates. For this chart, the two most logical orderings are (a) order by overall popularity of the parties, fixed for all ridings and (b) order by popularity of the candidate, variable for each riding.

The map shown above gives the winner in each subdivision. This type of visualization dumps a lot of information. Stephen tackles this issue by offering a small multiples view of each party. Here is the Liberals in Vancouver.

Again, we encounter ambiguity about the color scheme. Liberals have been associated with a red color but we are faced with abundant yellow. After clicking on the other parties, you get the idea that he has switched to a divergent continuous color scale (red - yellow - green). Is red or green the higher value? (The answer is red.)

I'd suggest using a gray scale for these charts. The hardest decision is going to be the encoding between values and shading. Should each gray scale be different for each riding and each party?

If I were to take a guess, Stephen must have spent weeks if not months creating these maps (depending on whether he's full-time or part-time). What he has published here is a great start. Fine-tuning the issues I've mentioned may take more weeks or months more.

****

Stephen is brave and smart to send this project for review. For one thing, he's got some free consulting. More importantly, we should always send work around for feedback; other readers can tell us where our blind spots are.

To read more, start with this post by Stephen in which he introduces his project.

Working hard at clarity

As I am preparing another blog post about the pandemic, I came across the following data graphic, recently produced by the CDC for a vaccine advisory board meeting:

This is not an example of effective visual communications.

***

For one thing, readers are directed to scour the footnotes to figure out what's going on. If we ignore those for the moment, we see clusters of bubbles that have remained pretty stable from December 2020 to August 2021. The data concern some measure of Americans' intent to take the COVID-19 vaccine. That much we know.

There may have been a bit of an upward trend between January and May, although if you were shown the clusters for December, February and April, you'd think the trend's been pretty flat. 

***

But those colors? What could they represent? You'd surely have to fish this one out of the footnotes. Specifically, this obtuse sentence: "Surveys with multiple time points are shown with the same color bubble for each time point." I had to read it several times. I think it simply means "Color represents the pollster." 

Then it adds: "Surveys with only one time point are shown in gray." which simply means "All pollsters who have only one entry in the dataset are grouped together and shown in gray."

Another problem with this chart is over-plotting. Look at the July cluster. It's impossible to tell how many polls were conducted in July because the circles pile on top of one another. 

***

The appearance of the flat trend is a result of two unfortunate decisions made by the designer. If I retained the chart form, I'd have produced something that looks like this:

The first design choice is to expand the vertical axis to range from 0% to 100%. This effectively squeezes all the bubbles into a small range.

The second design choice is to enlarge the bubbles causing copious amount of overlapping. 

In particular, this decision blows up the Pew poll (big pink bubble) that contained 10 times the sample size of most of the other polls. The Pew outcome actually came in at 70% but the top of the pink bubble extends to over 80%. Because of this, the outlier poll of December 2020 - which surprisingly printed the highest number of all polls in the entire time window - no longer looks special. 

***

Now, let's see what else we can do to enhance this chart. 

I don't like how bubble size is used to encode the sample size. It creates a weird sensation for anyone who's familiar with sampling errors, and confidence regions. The Pew poll with 10 times the sample size is the most reliable poll of them all. Reliability means the error bars around the Pew poll outcome is the smallest of them all. I tend to think of the area around a point estimate as showing the sampling error so the Pew poll would be a dot, showing the high precision of that estimate. 

But that won't work because larger bubbles catch more of the reader's attention. So, in the following version, all dots have the same size. I encode reliability in the opacity of the color. The darker dots are polls that are more reliable, that have larger sample sizes.

Two of the pollsters have more frequent polling than others. In this next version, I highlighted those two, which reveals the trend better.

 

 

 

Reading this chart won't take as long as withdrawing troops from Afghanistan

Art sent me the following Economist chart, noting how hard it is to understand. I took a look, and agreed. It's an example of a visual representation that takes more time to comprehend than the underlying data.

The chart presents responses to 3 questions on a survey. For each question, the choices are Approve, Disapprove, and "Neither" (just picking a word since I haven't seen the actual survey question). The overall approval/disapproval rates are presented, and then broken into two subgroups (Democrats and Republicans).

The first hurdle is reading the scale. Because the section from 75% to 100% has been removed, we are left with labels 0, 25, 50, 75, which do not say percentages unless we've consumed the title and subtitle. The Economist style guide places the units of data in the subtitle instead of on
the axis itself.

Our attention is drawn to the thick lines, which represent the differences between approval and disapproval rates. These differences are signed: it matters whether the proportion approving is higher or lower than the proportion disapproving. This means the data are encoded in the order of the dots plus the length of the line segment between them.

The two bottom rows of the Afghanistan question demonstrates this mental challenge. Our brains have to process the following visual cues:

1) the two lines are about the same lengths

2) the Republican dots are shifted to the right by a little

3) the colors of the dots are flipped

What do they all mean?

A chart runs in trouble when you need a paragraph to explain how to read it.

It's sometimes alright to make complicated data visualization that illustrates complicated concepts. What justifies it is the payoff. I wrote about the concept of return on effort in data visualization here.

The payoff for this chart escaped me. Take the Democratic response to troop withdrawal. About 3/4 of Democrats approve while 15% disapprove. The thick line says 60% more Democrats approve than disapprove.

***

Here, I show the full axis, and add a 50% reference line

Small edits but they help visualize "half of", "three quarters of".

***

Next, I switch to the more conventional stacked bars.

This format reveals some of the hidden data on the chart - the proportion answering neither approve/disapprove, and neither yes/no.

On the stacked bars visual, the proportions are counted from both ends while in the dot plot above, the proportions are measured from the left end only.

***

Read all my posts about Economist charts here

 

Two commendable student projects, showing different standards of beauty

A few weeks ago, I did a guest lecture for Ray Vella's dataviz class at NYU, and discussed a particularly hairy dataset that he assigns to students.

I'm happy to see the work of the students, and there are two pieces in particular that show promise.

The following dot plot by Christina Barretto shows the disparities between the richest and poorest nations increasing between 2000 and 2015.

The underlying dataset has the average GDP per capita for the richest and the poor regions in each of nine countries, for two years (2000 and 2015). With each year, the data are indiced to the national average income (100). In the U.K., the gap increased from around 800 to 1,100 in the 15 years. It's evidence that the richer regions are getting richer, and the poorer regions are getting poorer.

(For those into interpreting data, you should notice that I didn't say the rich getting richer. During the lecture, I explain how to interpret regional averages.)

Christina's chart reflects the tidy, minimalist style advocated by Tufte. The countries are sorted by the 2000-to-2015 difference, with Britain showing up as an extreme outlier.

***

The next chart by Adrienne Umali is more infographic than Tufte.

It's great story-telling. The top graphic explains the underlying data. It shows the four numbers and how the gap between the richest and poorest regions is computed. Then, it summarizes these four numbers into a single metric, "gap increase". She chooses to measure the change as a ratio while Christina's chart uses the difference, encoded as a vertical line.

Adrienne's chart is successful because she filters our attention to a single country - the U.S. It's much too hard to drink data from nine countries in one gulp.

This then sets her up for the second graphic. Now, she presents the other eight countries. Because of the work she did in the first graphic, the reader understands what those red and green arrows mean, without having to know the underlying index values.

Two small suggestions: a) order the countries from greatest to smallest change; b) leave off the decimals. These are minor flaws in a brilliant piece of work.

 

 

Come si dice donut in italiano

One of my Italian readers sent me the following "horror chart". (Last I checked, it's not Halloween.)

I mean, people are selling these rainbow sunglasses.

The dataset behind the chart is the market share of steel production by country in 1992 and in 2014. The presumed story is how steel production has shifted from country to country over those 22 years.

Before anything else, readers must decipher the colors. This takes their eyes off the data and on to the color legend placed on the right column. The order of the color legend is different from that found in the nearest object, the 2014 donut. The following shows how our eyes roll while making sense of the donut chart.

It's easier to read the 1992 donut because of the order but now, our eyes must leapfrog the 2014 donut.

This is another example of a visualization that fails the self-sufficiency test. The entire dataset is actually printed around the two circles. If we delete the data labels, it becomes clear that readers are consuming the data labels, not the visual elements of the chart.

The chart is aimed at an Italian audience so they may have a patriotic interest in the data for Italia. What they find is disappointing. Italy apparently completely dropped out of steel production. It produced 3% of the world's steel in 1992 but zero in 2014.

Now I don't know if that is true because while reproducing the chart, I noticed that in the 2014 donut, there is a dark orange color that is not found in the legend. Is that Italy or a mysterious new entrant to steel production?

One alternative is a dot plot. This design accommodates arrows between the dots indicating growth versus decline.