- Jose
- Jul 24, 2007
-
Adrian Chiles is a broadcaster and writer
|
It was in C-SPAM (or I guess it was Election Erection at the time). Probably the Republican thread, which I think is still the current Trump thread, so they might still be somewhere in there.
since someone just quoted it here's the post
http://forums.somethingawful.com/showthread.php?threadid=3680205&pagenumber=1486&perpage=40#post450957932
|
#
¿
Oct 27, 2016 11:49
|
|
- Adbot
-
ADBOT LOVES YOU
|
|
|
#
¿
Apr 28, 2024 08:00
|
|
- Jose
- Jul 24, 2007
-
Adrian Chiles is a broadcaster and writer
|
Since my work involves charting software and i often link stuff posted itt on skype to people and i've been asked to write a blog for the company website about bad charts. Obviously a lot is obvious with that popular website infographic I posted being a great example of many ways to gently caress up a chart but was wondering what are some less obvious ways of loving them up so they're bad/meaningless/misrepresent the data
|
|
#
¿
Mar 29, 2017 08:43
|
|
- Jose
- Jul 24, 2007
-
Adrian Chiles is a broadcaster and writer
|
Some less obvious mistakes (with academic papers showing that they do screw people up, when I can remember the right ones):
1) loving with the y-axis.
http://www.cs.tufts.edu/comp/250VIS/papers/chi2015-deception.pdf
It should start from 0, and should be linear, unless there is a very good reason for it to be so, for the reasons vyelkin said in the previous post (inflates effect size, or makes effect size really hard to suss out)
2) loving with the x-axis.
If you've got time as the x-axis, it should also be linear, unless you've got a very good reason. Otherwise people are looking at the slope and saying "oh, things are slowing down and getting better," say, even if that's just an artifact of your showing data every year when you were previously showing it every decade.
3) loving with white space
http://statweb.stanford.edu/~cgates/PERSI/papers/scatter82.pdf
If you want to show that two variables are correlated, if you throw a lot of white space on either end of the charts (zooming out, in a way), it pushes the points together and makes them look more correlated than they actually are, and vice versa.
4) Failing to account for confounding variables and spurious correlations.
https://imgs.xkcd.com/comics/heatmap.png
Lots of pairs of variables may be correlated, but way fewer variables have a causal relationship. Showing charts with correlations makes people think they are causally linked, even if they should know better.
http://www.tylervigen.com/spurious-correlations
5) Intentional unnecessary complexity
A good example is the chart John Boehner's office released that was meant to show Obamacare.
http://voices.washingtonpost.com/ezra-klein/2009/07/when_health-care_reform_stops.html
Intentionally bad layout, way too many colors, and wordy labels all contribute to the idea "this is too complex for anybody to understand, so it's bad."
c.f. the redesigned chart "Do not gently caress with graphic designers" https://www.flickr.com/photos/robertpalmer/3743826461
6) Failing to follow the conventions of a particular chart genre
Often times, graphic designers will make "pseudo-charts" - stuff that looks like a chart, because chart = sciency = convincing.
http://www.ask-force.org/web/Discourse/Tal-Blinded-with-Science-Trivial-2014.pdf
However, this will mean that they sometimes do not follow the convention of the chart. I've posted a bunch of examples in this thread, but here's another:
http://68.media.tumblr.com/62113c89bfde641aee65a611d33b7712/tumblr_on1cl2IxwY1sgh0voo1_1280.jpg
It looks like a bar chart or meter chart, since it's a bunch of rectangles with numbers on them, somewhat aligned to a common axis. But it's not, it's just a bunch of rectangles with numbers on them.
7)
Beat me to it!
https://web.cs.wpi.edu/~ltharrison/files/skau2015evaluation.pdf
In general, yeah, make sure you're not conflating two visual properties of the items in a chart. Charts with circles in them tend to be the worst offenders: are you encoding value as the radius, or the area? If it's radius, then know that a circle that has twice the radius is going to have more than twice the visual area.
Some more that I thought of:
8) 3D charts are crap, don't make 3D charts. Even if you have 3D data, you still probably don't want to make 3D charts. People are not great at estimating the sizes of 3D objects and accounting for perspective and so on, and that's before you get to the problems of occlusion and perspective warping and whatnot. It's just almost never worth the cost.
9) Don't use the "rainbow" color map, where low values are bluish, and high values are reddish, and the other values are the ROYGBIV colors in the middle.
https://classes.soe.ucsc.edu/cmps261/Fall15/papers/colormapHarmful.pdf
We are way better at discriminating reds than blues, so it will make all your low values look closer together than your high values. We also separate colors into sort of discrete bins ("all the greens," and "all the reds") for instance, so you get these weird "bullseye" patterns even if your data is just increasing linearly.
well there is my blog mostly written lol. thanks
|
|
#
¿
Mar 29, 2017 16:07
|
|
- Jose
- Jul 24, 2007
-
Adrian Chiles is a broadcaster and writer
|
I'm writing a bad chart blog and was wondering if anyone has a quick example of a bad X axis that has time but its non-linear? Currently going through the thread from the start looking for one that doesn't also have poo poo loads else bad about it
|
|
#
¿
May 10, 2017 11:22
|
|
- Jose
- Jul 24, 2007
-
Adrian Chiles is a broadcaster and writer
|
I just scrolled through 50 pages of the thread and i'm defeated I'll use that one. Thanks
|
|
#
¿
May 10, 2017 12:06
|
|
- Adbot
-
ADBOT LOVES YOU
|
|
|
#
¿
Apr 28, 2024 08:00
|
|
#
¿
Oct 27, 2016 11:49
