Junk Charts

The real climategate

Climategate is all the rage at the moment. What interests me about this episode is not the integrity of certain scientists, or science in general, nor the culture of academia, and certainly not the evidence of climate change. For me, the real climategate is the woeful state of statistical education.  Let me explain.

Here is the infamous email: (via Nathan Silver, with my highlights)

From: Phil Jones
To: ray bradley ,mann@[snipped], mhughes@
[snipped]
Subject: Diagram for WMO Statement
Date: Tue, 16 Nov 1999 13:31:15 +0000
Cc: k.briffa@[snipped],t.osborn@[snipped]
Dear Ray, Mike and Malcolm,

Once Tim’s got a diagram here we’ll send that either later
today or first thing tomorrow. I’ve just completed Mike’s Nature
trick of adding in the real temps to each series for the last 20
years (ie from 1981 onwards) amd [sic] from1961 for Keith’s to
hide the decline. Mike’s series got the annual land and marine
values while the other two got April-Sept for NH land N of 20N.
The latter two are real for 1999, while the estimate for 1999
for NH combined is +0.44C wrt 61-90. The Global estimate for
1999 with data through Oct is +0.35C cf. 0.57 for 1998.

Thanks for the comments, Ray.

Cheers, Phil

What concerns me is Phil Jones' describing what he did as a "trick" to "hide the decline".  He apparently thought that he was doing something shameful. But when is it shameful to extend the plot of a time series so as to display the long-term trend, and not be misled for short-term fluctuations? This is providing statistical context to the data being examined. Lots of people are condemning this as a willful act to mislead the public but if they have some statistical literacy, they will understand that finding the appropriate time scale to look at the data is one of the most important tasks of analyzing time series data. It's a problem when even prominent scientists do not comprehend why they should be doing this.

I have always wondered why in climatology as well as in economics, we rarely see decomposed time-series plots (at least not in the public's eye). 

On the right, I found on-line a plot of a decomposition of beer sales that separates out seasonality, trend and other parts of a time series. The original data is shown up top. In practice, newspapers and blogs give us such plots all the time when they should show us the third plot down (the trend with the seasonal factor removed), unless the story is about seasonality.

Note to self: should include basic time-series decomposition in the intro stats syllabus; much too important a topic to leave to a second course.

 

Worthy of the Times

Andrew Gelman has put up a great post, discussing how he collaborated with the New York Times editors to transform his chart to a publishable form.  Some time ago, there was a discussion here about the publish-worthiness of my charts (actually, the lack thereof), and I explained that to take my charts to that level would require quite a bit of extra work.  Gelman speaks here from first-hand experience.

Here are the two versions side-by-side.  Readers on the Gelman blog are debating which version is "better" with Gelman himself voting for the revised NYT version:

Let's compile a list of the changes:

1. Removed 2000 data and relied on only 2004 data
2. Reduced number of groups of senators from 7 to 5 (with a special calling out of Senator Lieberman)
3. Re-ordered senators to facilitate classifying into Democrats, Independent, Republicans
4. Added annotation explaining the grouping of senators; in particular, qualified why the 49 Democrats and 32 Republicans were set aside
   
5. Removed scale labels on the top of the chart, retaining only the bottom labels
6. Vastly increased the area devoted to text labels, which now covered half of the chart
7. Added white vertical gridlines
8. Instead of coloring the cross marks, colored the background of the chart; introduced the yellow color for Lieberman (which Gelman originally colored blue)
9. Removed reference to the national average but explained clearly in the legend that the percentages are relative to the national average.
   

Of these, 2,3,4 are definite improvements. 1 is acceptable since 2000 and 2004 data are quite highly correlated (note, however, Iowa, Virginia and Ohio.) 7 looks brilliantly here although we typically don't recommend adding gridlines and such. 5 is also acceptable because of 7 -- with the gridlines, one doesn't need to have two sets of labels. 8 can be debated: I'd like to see a version with large colored squares instead of cross marks in lieu of the background; the current look is not bad. 9 can also be mulled over: would it be so bad as to insert in parentheses 73% after the words "national average" in the legend?  I suspect some feel that this may cause confusion because readers now need to understand relative versus absolute values. 6 is the most debatable because the data-ink ratio has been vastly reduced. 

There are several good features of the original chart that they left unaltered, which should also be duly noted:

• Allowed the positive side of the scale to extend to 10% even though the largest data only reached about 6%. While this creates empty space, it helps readers judge the magnitude of negative numbers relative to positive ones.
• Subdued horizontal gridlines
• Retained the title of the chart (I actually prefer that they use the title to summarize the finding as opposed to the current one which takes a neutral stance.)
• Retained the horizontal scale labels at 5% apart

One final comment on something omitted from both charts: I'd prefer that they plot all the dots for the 49 + 32 senators, on the top and bottom lines respectively, instead of just plotting the group averages.  It would be really interesting to see what the spread is around the 0% point.

Disentangle

In an article called "Off the Charts", Floyd Norris wanted to let readers know that unemployment does not hit citizens equally -- it affects some age groups and men/women to differing degrees. 

As befits the article's title, he included several charts, from which I extracted the one shown on the right.  At first glance, this seems like a normal chart.

But when one pays attention, one notices that the chart is rather complicated.  This chart is like a piece of modern music, in which the composer allows two voices to jar and talk past one another.

Think of it as a data table vying for attention with a bar chart.  The data table is a cross-tabulation of the change in employment by age and gender.  In this view, the men sit on the left, and the women on the right.

Lurking around is a bar chart, for which the point of zero change sits in the middle.  Positive growth extends to the right, while negative growth points to the left.  The gender labels at the top are irrelevant  for this bar chart: the narrow black bars indicate women, the fat colored ones, men.  The data labels are also irrelevant: see, for example, the 45-54 age group, the label for females, at -2.3, should really be placed on the left side of the middle divide!

Here is how these two charts look, disentangled: (I have converted the bar chart to a dot chart.)

 

  

Reference: "Off the Charts: Job losses mount, enduring and deep", New York Times, Nov 14 2009.

Reading

Here are some of my favorite links from other places:

A spatial journey illustrating a very long scale, created by the Genetic Science Learning Center (here)

Long scales are very difficult to deal with in charts; I have never been satisfied with log scales since it addresses the designer's challenge of trying to fit everything onto one page, bu does not deal with the reader's need to compare the elements accurately

Not sure how this helps but perhaps some of you will figure it out

Tommi left a comment about this conceptual chart by xkcd, which has been making the rounds.  Fits into our Light Entertainment category.

Says there is no optimal chart type.  A type that works very well for one data set may get hopelessly cluttered for another, similar data set.

From fellow bloggers (especially Jorge), a whole series of views of the U.S. unemployment figures by state over time.  Alternatives that are much more interesting to look at than the typically line chart. Jorge even found something in Excel that looks good.

Getting the Nobel wrong

Why do we pay so much attention to seemingly inconsequential details of a chart's scale, colors, labels, etc.?

Here's why. (Reader Omegatron pointed us to the FAIL blog that captured this beauty.)

Notice the messed-up horizontal scale, in particular, the failure to start the axis at 0.  The result: the tiniest difference presented as a wide gulf.

The graph, published by the Washington Post, has since been fixed.  See here.  Nevertheless, the comments left by readers lent witness to the confusion.  I copied the first bunch that mentioned the graphical display - there were plenty of for-and-against-the-prize comments, many assuming that the poll result was as lop-sided as the chart seemed to indicate.

By starting at a base of 45% (as of this reading), your graphic grossly misrepresents the results of your poll. The "no" bar is four times as big as the "yes" bar, giving the visual impression that the vote must have gone 80-20 against Obama's Nobel. On the contrary, as of now the vote is 53-47 against. Whoever produced this graphic should re-read Edward Tufte's "The Visual Display of Quantitative Information."

Posted by: threefab | October 9, 2009 8:33 AM

Someone made comment that the graph is misleading, but it really isn't if you know how to read a graph.

This type of graph highlights the difference, not the complete number of votes and is appropriate when viewing percentages.

Posted by: gconrads | October 9, 2009 8:39 AM

why is graph not drawn to scale? The vote is 51 - 49 no and the graph looks like an overwhelming number of "voters' said no

Posted by: spitts1 | October 9, 2009 8:41 AM

The graph is purposely designed to make it appear that there are a huge number of no votes and demonstrates obvious bias.

Posted by: fingersfly | October 9, 2009 8:41 AM

I'm looking at the graphic here and cannot figure out what you guys are trying to show?
The tiny slice of blue is supposed to be 50% and the huge slice of red is supposed to be 50% and the scale at the bottom is......
WHAT?

Posted by: Tomcat3 | October 9, 2009 8:53 AM

It is my understanding that the President did not "win", but was "awarded" the prize...much like a bonus given to you from your employer...deserving or not. In addition, I could not help but notice that after casting my vote, the graphic scale indicated 49% yes and 51% no; however, the graph lines seemed much more disproportionate for only a 2% spread with emphasis on “no”…

Posted by: jonbwnfd | October 9, 2009 9:40 AM

Why does the graph make the vote look so lopsided?

Posted by: subwayguy | October 9, 2009 9:48 AM

The graph is not lopsided .. the numbers are. Read the posts and you will see the posts are against the idea not for.

Posted by: T-Tom | October 9, 2009 9:53 AM 

Who designed the bar graph? The difference between the "yes" votes and the "no" votes is two percentage points, yet the "no" bar is three times as long as the "yes" bar.

In fact, "yes" is 49% and "no" is 51%: the graph is supposed to illustrate reality, not. . . well, just what DOES it illustrate?

Posted by: cmcintyr | October 9, 2009 10:02 AM

Inconsequential detail?  Big fail?  You decide.

A tale of four charts

Speaking of rules for making charts, I think the most important is "if at first you don't succeed, try, try and try again."  It's absolutely essential to produce multiple looks before settling on the one that helps tell the story.

While researching U.S. consumer credit this week, I came across these four views of presumably the same data set.

The chart on the top left (via Business Insider) shows a downward sloping line, with a steep decline on the right edge of the chart.  The authors clearly wanted to show us that consumer credit in the U.S. was collapsing.  The bottom was falling out.  To aid in this effort, they chose to:

• start the plot at the peak of the time series (2000);
• set the minimum value of the vertical scale to coincide with the lowest value attained thus far (2009) -- we have reached rock bottom, ladies and gentlemen; and
• pay no particular attention to the 0% line, which is placed in the lower half of the chart, rather than the middle, thus obscuring the fact that credit grew at faster rates during the height of the boom than it has been declining during the recession.

(Thanks to Excel's default setting, we are treated to a centipede effect on the horizontal axis.  Also thick lines and line shadows.)

Readers who previously complained about my willingness to draw a line through things that shouldn't be connected will be none too amused by the use of a line chart to plot year-on-year changes.  Thus, the zig-zagging of the line represented the change of change, which has been dubbed the "second derivative" during a period when optimists used technical wizardry to show us "green shoots".  To read this chart properly, one should focus on the actual annual decline (the dots) rather than the line.

If one takes a longer-term view, as in the chart on the top right (via Rolfe Winkler), the recent drop in consumer credit can be put in perspective.  Since the 1960s, consumer credit has been positively exploding with few years of decline.  Note, again, that the falling line between 2000 and 2008 represented a slowing rate of growth, not a decline.  The question to ask is: after so many years of almost continuous growth, is the current correction such a big cause for alarm?

This situation of exploding growth followed by a slight correction is better visualized in the third chart (lower left, via chartingtheeconomy.com)  The author also answered a question asked by Rolfe, which is to compare the total consumer credit to the population, as he plotted the per-capita consumer credit.  An eyeball estimate tells us that consumer credit jumped by more than 800 percent since the 1970s, and the current retrenchment is a blip if we take a long view.  The explosion in consumer credit has no doubt enhanced wellbeing in the U.S. for decades; even though credit might well have been over-extended in the recent past, it is far from something evil.

The final chart (bottom right, via The Big Picture) should carry a health warning.  It really should not be shown by itself, or without comment.  Both charts on the right, in fact, came from the same presentation, by David Rosenberg.  This chart, not surprisingly, is the most dramatic -- the chart designer is going for the jugular.  The decline on the right side is much more exaggerated than in any of the other three.  The trick here is:

• to plot the dollar value of credit change, rather than percentage value, and thus ensuring that the further back the time, the more insignificant the change;
• to connect dots which bring out the steep decline when in fact, the steep drop reflected the second derivative; and
  
• to put the 0% line below the middle of the chart, which causes the bottom "half" of the chart to be smaller than the top "half", which plays with our perception so that we may not realize that there were multiple years of growth above the absolute level of decline recently experienced.
  

If this chart were to be believed, our focus should not be on the cliff-diving at the far right -- instead, the chart has the hallmark of a system getting completely out of control, and oscillating to oblivion.  One hopes that is not the message intended by its creator.

For anyone creating charts, it would be a great idea to have attempted all of these versions, and more.

A world full of bubbles

As promised, we stick to bubbles.  Like the street artist blowing soap bubbles at passers-by, this map -- published in the Guardian (UK) -- is a gift of bubbles.

And our reader Frederic M. is not amused.  "A tremendous failure", he said.

In terms of conveying the data, a simple bar chart would do a better job in exposing the biggest polluters, as well as the relative magnitude between the biggies and the small fish.

The chart reveals more problems if one clicks on, say, Europe, and sees the following:

For starters, compare the bubbles labeled 858, 468, 586, 418 with those labeled 23, 20, 18.  And look at the little ones in the periphery labeled 133, 174, 128.  Baffling, isn't it?

What they did was to print the ranks for every country, except the top four in Europe for which the ranks are placed next to the country name (in small font), and the actual amounts are placed in the middle of the bubbles.  The ranks, of course, are pretty useless, and they obliterate the scale of the differences between countries.

Besides, the bigger the polluter, the smaller the rank but the larger the bubble.  This built-in disconnect can also be disorienting.

Every bubble chart typically contains lots of data labels, and the reason is that the bubble form lacks self-sufficiency.  Without the data labels, the reader has trouble comparing the areas.

Reference: "The Carbon Atlas", Guardian, Dec 9 2008.

Bubble after bubble

I finally checked the Junk Charts mailbox again, and I found an uprising against bubble charts and pie charts.   It appears that despite their shortcomings amply demonstrated here and elsewhere, editors everywhere continue to believe that the public has a lovefest with these creatures.

I will start off the parade with this one from the Wall Street Journal, purportedly showing that the Bank of England has continued to inject cash into the economy, and at ever increasing rates.  The headline said Bank of England to expand bond-buy plan.


This chart has a variety of problems, in addition to the use of overlapping bubbles.  As has been documented, it is almost impossible to gauge the relative sizes of circular areas, especially when they are overlapping.

If we remove all but one of the data labels, the chart is non-functional.  This is what we mean by not self-sufficient: the interpretation of this chart requires, indeed demands, that all the underlying data be printed on the same chart.  The only way readers can understand what is going on is by reading the data itself!

The horizontal axis (indicating time) is also non sensical.  The separation from month to month is variable.  Besides, and this is the key flaw of the chart, the projected number is a three-month total cumulative growth being treated like a monthly figure.

 

Since the Bank is projected to inject 175 50 billion extra pounds in the next three months, that would work out to be roughly 60 16 billion per month.  That would turn the story upside down: one would conclude that the Bank is gradually slowing the rate of injection.  The following bar chart points this out with little fuss:

 

When bars are used, there is no need to print every single data point.  The relative lengths of the bars can be estimated easily.  The months are equally spaced.

One final point: the exchange rate cited is not very helpful.  What would have been more useful for readers would be the scale of the cash injection with respect to each nation's GDP.

Reference: "Bank of England Expands Bond-Buy Plan", Wall Street Journal, Aug 7 2009.

PS. Per Andrew's comment, here is a line chart, where the growth/decline in the injection is encoded in the slope of the line segments:

Degrees of likeness 1

The NYT team just put up a fantastic visualization of the American Time Use Survey data, which purports to measure how the average American spends the time of a day.  (Apparently, thousands of people recalled what they did every minute of an average day.)  The amount of data collected is massive, and this graphic allows readers to explore the data in intuitive ways.

The chart shows for each minute of the day (horizontal axis) the proportion of people doing specific activities.  Not surprisingly, we spend more time sleeping than any other type of activity.  The axis and data labeling as well as gridlines are very restrained.

Normally, I am not a big fan of these proportional area charts because the only relevant dimension to look at is the vertical distance from one curve to the next but the focus on areas put equal weight on the horizontal and vertical distances.   The horizontal distance is meaningless, and thus the area is meaningless.

These designers found a solution to the problem, and good for them!  Because of the mouse-over effect, I could not save the actual appearance -- here, I show what it looks like.

By mousing over different parts of the graph (say, moving vertically), we can compare the actual proportions.  Terrific!

The key interest of this graphic is the following legend.

While the above graphic shows the use of time by all Americans in aggregate, this panel allows us to zoom in on specific groups of Americans.  How alike are Americans?

Reference: "How Different Groups Spend Their Day", New York Times, July 31 2009.

 

Space and time

When it comes to space or time in graphics, old habits die hard.   When we have spatial data, the default is to put it on a map.  When we have a time series, the default is to plot time along the horizontal axis.  Sometimes, these defaults work; other times, breaking up the map or straight-time-line works better.

Thanks to a reader, I noticed that Google put up a "Flu Trends" website to help us track the flu season.  They use two main charts to plot the data, as shown below.




On the right side is the time series, showing the severity of flu cases from month to month.  There are many great things about this chart and one serious flaw.  I love the fact that they did not plot time on the horizontal axis; they realize the seasonality and they create overlapping lines.  They make good use of foreground and background; it's easy for us to compare year to year differences.

The serious flaw: no vertical scale.  This was a problem with Google Trends from day one (see my post here).  They still haven't fixed it.  Because of this, we don't know if the peak shown was 5 cases or 5000 cases.  While for Google key word searches, one can excuse them for trying to protect commercial secrets.  I would imagine that this public health data is, well, public.  Since the apparent purpose of this chart is to allow citizens to declare a flu epidemic (say, when they see the current trend depart from the historical norm), not having the scale is a huge problem.

I also disagree with shifting the months around for the Northern Hemisphere so that the peaks of the graphs are aligned towards the middle.  It is better for the peaks to appear on the left and let the order of the months conform to our expectation.  (The "peak" would be split on the sides and the chart would look like a valley, which presumably is why they did it this way.)

The charts on the left side plot the spatial data, not surprisingly on maps.  Sadly, the standard exhibited on the time-series charts is nowhere found on these maps.

First, the legend is seriously deficient.

Second, the gradation of the colors is not fine enough, or put differently, the aggregation to the state/province level is much too coarse for any interesting pattern to be seen.

This poorly aggregated map becomes a farce when applied to the U.S.  There is not much left to be said, is there?