Layered donuts have excess fats and oils
An enjoyable romp through the movies

Butcher: which part of the leg do you want? Me: All of it, in five pieces please

This ABC News chart seemed to have taken over the top of my Twitter feed so I better comment on it.

abcnews_trumptransition

Someone at ABC News tried really hard to dress up the numbers. The viz is obviously rigged - Obama at 79% should be double the length of Trump's 40% but not even close!

In the Numbersense book (Chapter 1), I played the role of the Devious Admissions Officer who wants to game the college rankings. Let me play the role of the young-gun dataviz analyst, who has submitted the following chart to the highers-up:

Redo_abcnews_trumpfav_accurate1

I just found out the boss blew the fuse after seeing my chart. The co-workers wore dirty looks, saying without saying "you broke it, you fix it!"

How do I clean up this mess?

Let me try the eye-shift trick.

Redo_abcnews_trumpfav_hollowfave1

The solid colors draw attention to themselves, and longer bars usually indicate higher or better so the quick reader may think that Obama is the worst and Trump is the best at ... well, "Favorability on taking office," as the added title suggests.

Next, let's apply the foot-chop technique. This fits nicely on a stacked bar chart

Redo_abcnews_trumpfav_onecut1

I wantonly drop 20% of dissenters from every President's data. Such grade inflation actually makes everyone look better, a win-win-win-win-win-win-win proposition. While the unfavorables for Trump no longer look so menacing, I am still far from happy as, with so much red concentrated at the bottom of the chart, eyes are  focused on the unsightly "yuge" red bar, and it is showing Trump with 50% disapproval.

I desperately need the white section of the last bar to trump its red section. It requires the foot-ankle-knee-thigh treatment - the whole leg.

Redo_abcnews_trumpfav_onebigcut1

Now, a design issue rears its head. With such an aggressive cut, there would be no red left in any of the other bars.

I could apply two cuts, a less aggressive cut at the top and a more aggressive cut at the bottom.

Redo_abcnews_trumpfav_twocuts1

The Presidents neatly break up into two groups, the top three Democrats, and the bottom four Republicans. It's always convenient to have an excuse for treating some data differently from others.

Then, I notice that the difference between Clinton and GW Bush is immaterial (68% versus 65%), making it awkward to apply different cuts to the two neighbors. No problem, I make three cuts.

Redo_abcnews_trumpfav_threecuts1

The chart is getting better and better! Two, three, why not make it five cuts? I am intent on making the last red section as tiny as possible but I can't chop more off the right side of GHW Bush or Reagan without giving away my secret sauce.

Redo_abcnews_trumpfav_fivecuts1

The final step is to stretch each bar to the right length. Mission accomplished.

Redo_abcnews_trumpfav_fivecuts_rescaled1

This chart will surely win me some admiration. Just one lingering issue: Trump's red section is still the longest of the group. It's time for the logo trick. You see, the right ends of the last two bars can be naturally shortened.

Redo_abcnews_trumpfav_fivecuts_logo1

The logo did it.

***

Faking charts can take as much effort as making accurate ones.

The ABC News chart encompasses five different scales. For every President, some percentage of dissenters were removed from the chart. The amount of distortion ranges from 15% to 47% of respondents.

Redo_abcnews_trumpfav_distortion1

 

 

 

 

Comments

jlbriggs

"Faking charts can take as much effort as making accurate ones."

Based on this walk-through, I would say it takes decidedly more effort to fake it!

Erik Terner

Why would they want to fake it?
Do you really believe ABC is shilling for Trump?

Ken

Rather than faking it, they are probably driven by a need to make the chart look right. Making the lengths correct would either require small text or a wide chart with lots of white space. After all, including the numbers fairly much gives the game away.

Alex

I don't know. In my experience, Hanlon's razor of "never attribute to malice that which is adequately explained by stupidity" [1] is a very good guide for real life.

I think what really happened here is that the designer was desperately trying to find a way to display the names overlaying the bars in a way that they area readable even on old NTSC TV sets. So what he did is, he took your original bars and then just clipped and clinched them horizontally such that they only take up half of the screen to give enough space for the names. And then, to make it look good, he used the white bars as background for the names on the left hand side. This is also why he chose the confusing color scheme: With red bars (instead of red "emptiness") the names would probably become illegible on old TV sets.

I made you an illustration:
https://framapic.org/j0oLz9JEruSV/sUhGcxl0xNcJ.png


[1]: https://en.wikipedia.org/wiki/Hanlon%27s_razor

Kaiser

Alex: Interesting idea that they may have chopped off both ends. Note that I did not attribute any malice on the ABC designer - I have no claim on knowing why they did what they did, and deliberately chose to present this as a general comment on how to fake charts. So you can say I used Hanlon's razor.
More generally, I don't know how one can square Hanlon's razor with plausible deniability which is widely practiced. The same thing with manipulation of college rankings that I discussed in the book. There are lots of ways to nudge the statistics replete with normalizing explanations for those acts. One quick example: many schools switched to "Common App" saying that this levels the playing field - a clear effect of this is to increase drastically the number of applications received, which directly improves selectivity rate.

Antonio

I'm on Alex side. I found the five axis cuts very intricated. So, even if I could give some credibility to malice, I think that Alex recreation is the best because it is far better simpler than the "five axis cuts": Occam's razor holds besides Hanlon's razor.
On the other hand (on the side of the sister blog), I wonder what the chart means. In particular I have a question, since I do not know the victory percentage of each presidential election.
Does it mean that democratic presidents have won their elections with greater margin than republican ones, or that republican electorate is more democratic than democratic one (excuse me the word pun), since it acknowledge the other party victory more frequently?
Moreover, are we sure that the polls behind the chart are trustworthy? Indeed, we are speaking of a president who is making what he has promised, and previous polls about him have been proved not so affordable. As a foreigner observer, I'm shocked by the paroxysmal and furious press campaign against Trump.

The comments to this entry are closed.