Hacker News new | past | comments | ask | show | jobs | submit login
Visualizations that make no sense (wtfviz.net)
137 points by denzil_correa on Aug 30, 2013 | hide | past | web | favorite | 38 comments

Really amusing, but there is a deeper interesting point here: people are susceptible to the "Rhetoric of Data". Apart from simple mistakes, it seems clear that people are getting utility that is not related to the information presented out of these graphics, and I hypothesize that utility has everything to do with the rhetorical utility of fancy looking graphics, bright colors, and lots of numbers. "Oh," the audience thinks, "I don't understand this but it looks complicated, so the speaker must be very smart and the point they are making very profound."

This is why skepticism is only second to honesty as the most important quality in an organization. Some organizations punish skeptics as being disruptive, but the simple fact is that many times people don't know what the fuck they are talking about, and skeptics are an organization's immune-system. If someone is presenting on something complicated and there are not probing questions, then something is terribly wrong.

With a little bit of context, the 3D elevation plot actually makes a lot of sense: http://www.nasa.gov/topics/earth/features/haiti20101014img.h...

It turns out it's like a continuous contour plot (showing deformation from a Haitian earthquake). Thanks to Google image search, such poorly documented images can be traced back to their source - and context! - quite quickly. This one was on the first page of results.

I feel like this site is missing a lot of content from the news media: http://mediamatters.org/research/2012/10/01/a-history-of-dis... (this is specific to FOX but the same can be applied to most other major news sources, too.)

I really want to know the context of some of these, like the social media 'chart': http://wtfviz.net/image/59427832631

I agree. Of course, including the context would make a number of the visualizations make sense, which would defeat the purpose of the site. Of course, visualizations that make sense with no context are the sort I guess the creator of this site likes. Someday, we'll all get our news in nice graphics, without all that bothersome reading and thinking that goes along with context.

I physically twitch reading these. It's worst than cringing. It really physically hurts having so much data that I can interpret. Some of this would be interesting info graphs but then I can't glean anything from any of it and that bothers me to no end.

Ah, the sort of graphs you use when you don't want to be understood.

I remember, back when I was doing some research having to explain to people what a standard deviation was and what confidence meant.

There are ways to talk to people who don't know statistics - for instance if you talk about percentages people glaze, however, if you talk about 25 people in every hundred, that's something they can tend to engage with better.... If you use a bar chart, people tend to be used to looking at those. Drop 'correlates with' and use 'relates to' - that sort of thing.

It seems silly, it's all the same stuff. But some people have addition, division and multiplication as more or less the extent of their mathematical skills - because that's all they've been called upon to use since school. There's nothing wrong with that most of the time, I mean it's entirely understandable, it just means that if you're talking to them half the job is putting it in forms that they can understand.

I sympathize with the NASA chart. It was clearly created using IDL, which is powerful but also suffers from allowing lazy scientists to use automatic plot parameters.

Even worse is when said lazy scientists plot time series data as a set of images where each image is autoscaled.

A basic statistics course should be required for a college degree.

It often is. Whether or not students learn anything is the issue.

Perhaps you could say that basic statistical competency should be required for a college degree.

Actually, the 'Becoming a Data Scientist' one is awesome. The caption derisively says 'Disconnected subway map? Sequential, linear relationships?', but that is exactly what it is. I searched for it and read up on it, and the author thought he'd represent learning paths as subway routes. Data science is now a lot of tools, and I got into reading this map. I know probably 20% of the tools there at the moment. Actually, I found and bookmarked that viz, because it's going to be handy. Sure, it's not perfect, but I like it!

Could this also be called "data twerking" ?

I actually might pay to watch Miley Cirus do this.

Love the pretzel to illustrate "soft"

And the pretzel to illustrate "flexible"

What's worse than visualizations that make no sense is visualizations that are good at misrepresenting data.



That is a bad example of a bad example. It is clearly labeled and shows what it wants to show: The previously hottest years and the large jump of that record in 2012. How would you have done it better? What is it misinterpreting?

The worst thing is that the years are not in order, though one would assume that they were. It makes it look like the years are getting hotter and hotter as time moves from left to right.

Which is what you did. There's no "large jump" in 2012. 2012 was one degree warmer than 1934, which in turn was a quarter degree warmer than 1999.

Besides, using the most extreme six points in the last century to paint a trend is awfully stupid anyway. With all of the data that was omitted, the temperature could have averaged 53.5 degrees from 1921 to 2006, then dropped to 45 degrees from 2007 to 2011 - indicating that we were suddenly falling into a new ice age punctuated by sparse years of extreme heat - and this graph(esque monstrosity) would look exactly the same.

This is a perfect example of information presentation that makes the people exposed to it stupider than they were before they saw it, because now they think that they know something.

edit: to be clearer - all this graphic is saying is that 2012 was the hottest year in the last 100 years.

Here's a mind-experiment; find a record value of anything that occurs in sequence. Take the previous 99 values of that sequence, and eliminate all but the top five values. Arrange those values from lowest to highest from left to right, ending with the record value.

  Number of Participants in the Sleepy
  Hollow Hog-Calling Competition During
  Record Years of Participation in the
  Sleepy Hollow Hog-Calling Competition
  56                           **
  55                           **
  54                      **   **
  53            **   **   **   **
  52       **   **   **   **   **
  51  **   **   **   **   **   **
     1991 1915 1914 2006 1931 2012

Have we learned anything?

We learn that there has never been such an interest in the hog-calling competition before and that in 2012 it was significantly higher than ever before.

The axis should start at 0 degrees, not 53. The 3 dimensionality distorts it as the orange bars slightly angle to the left and makes 2012 look much bigger in relation to the other bars. If you actually plot all the hottest years on a 2d bar chart starting from 0, it gives a much clearer picture of climate change while still revealing that 2012 was indeed the hottest year on record..

> The axis should start at 0 degrees, not 53.

If its showing relative magnitude of change, 53 degrees is better than 0. Whether how much farther this is from other "hot" years than they are from each other is a key point is at least debatable.

For any purpose for which a zero is needed (which doesn't include showing magnitude of change), for temperature 0 degrees F (which presumably this is -- a real problem with this is that it just specifies degrees, which is ambiguous) is just as arbitrary and inappropriate as 53 degrees F; if you are doing a visualization for which it is important to show a meaningful zero, you'd actually want absolute zero (0K). But for most things related to climate change, there's no reason for "zero" as the baseline.

There's certainly purposes for which a meaningful baseline other than 0K or 53F would be appropriate -- what that baseline would be depends on the purpose.

I vehemently disagree that 0 should be the baseline. That would make absolutely no sense for this plot. I would use something like the average or median of the relevant temperatures as baseline. Using 0 as baseline gives you huge columns and does not highlight the big step at all. I would rather have relative value changes to the lowest temperature I chose to display, so in this case: 0 0.12 0.33 0.50 0.51 1.49

3D columns are bad of course, but that is neglible.

Also (this may be my personal opinion), but I think people are used to seeing the x-axis encode time. That visualization makes it look like its slowly getting hotter year after year if you don't pay attention to the labels.

visualization rule #1: read the labels

Excellent "enterprise" presentation skills! (and this morning's best posted story, IMO)

er, I meant "presentment", rather than "presentation".

Some illustration forms are bad.

However, if they were demonstrated in an interactive media, e.g. a touch screen tablet, I bet the situation would be quite different, given that some components were designed to be highlighted through user interactions.

What's wrong with 'The layered pie slices.'?

Apple, at 34%, is almost as big as Google, at 89%.

The percentges add up to 130%, too.

Assuming a simple typo, where the 89 should be 59, the pie slices still look wrong to me. Looking at the relative sizes of the 4% and the 3% slices, that might be a rounding issue (could of instance be 3.7% and 3.3%), but the Apple and Android ones look too big to me.

Yeah, a typo can explain the total percentage issue.

But I don't see how wrong it would be to use the pie graph in such a way per se. IMHO, the foremost feature of pie graphs, the angle of sector, is kept.

It turns out you can have your pie and eat 30% of it, too.

Sorry, but I could not get it.

If Apple is at 34%, with Google/Android 89%, Windows Phone 4%, and Blackberry 3%, it would be a total of 130%, but 130% of what?

Oh! I think the point is: if you're comparing pieces to the whole, you should make a normal pie chart where the pieces are all on the same level. Stacking them up that way isn't really a pie chart at all, it's just a round bar chart.

Except that you most likely shouldn't use a pie chart to begin with. Unless you know you shouldn't, then it might be ok.

(apart from the numbers adding up to 130%...)

One issue is that the area of the figure should correspond to the number it represents. For example, the area for 3% should be 75% of the size of the area for 4%, but it clearly isn't in this illustration. If they don't match up, what is the point of having a visual at all? To show that 3% is smaller than 4%?

Even if some of the others match up (I can't tell for sure), their shape is so different that it really doesn't give us any better understanding of how 34% compares to 89%. And again, what is the point of a visualization that doesn't aid in understanding?

interesting. the anti-grammar of graphics.

This is hilarious

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact