Discussion


Note: Images may be inserted into your messages by uploading a file attachment (see "Manage Attachments"). Even though it doesn't appear when previewed, the image will appear at the end of your message once it is posted.
Register Latest Topics
 
 
 


Reply
  Author   Comment  
sfew

Moderator
Registered:
Posts: 838
Reply with quote  #1 
This discussion thread complements an article in my blog titled "When Are 100% Stacked Bar Graphs Useful?" Please post your own examples of 100% stacked bar graphs that you find useful that are different from those that I described as useful in the article.
__________________
Stephen Few
acotgreave

Registered:
Posts: 18
Reply with quote  #2 
This is a great topic introduced with an interesting post. 

I'm with Cole: I don't think bars are as bad as you say in all circumstances (even when used for time series). I do find the "part-to-whole" is easier to interpret with bars than lines. I expect we're going to disagree. To link to one of your blog posts last like, "Exploring part-to-whole relationships" would be a great research topic.

Anyway - I've been playing with population pyramids recently, and I think a % stacked bar can work very well if the question is "What is the gender ratio through different age groups in a country?"

Gender ratio.png    


If the question is about RATIO then I think this succeeds. 

The above is missing context about the actual numbers. That can be easily added with extra, contextual bars below the % stacked bar:


Population stacked bar.png 


But I did want to test whether a line could be better. So I tried showing % of males as a line instead of a stacked bar:

gender ratio line.png    

I do agree the line has a more pure clarity to it, but I think it requires more cognitive effort to comprehend the line. That should be taken into consideration. If you're building a chart for a lay person who's only going to give the chart a glance, I think the bar will succeed. For those with more familiarity and domain expertise, the line is best.


sfew

Moderator
Registered:
Posts: 838
Reply with quote  #3 
Andy,

I'm looking for examples of useful 100% stacked bars that are different from those that I've already identified as useful. In the article I identified and explained the usefulness of 100% stacked bar graphs that consist of two segments only, as you've illustrated above.

__________________
Stephen Few
danz

Registered:
Posts: 190
Reply with quote  #4 
Stacked Bars, 100% Stacked Bars, Stacked Area, 100% Stacked Area are among the usual chart types designers love to use to catch the attention. In my opinion, they are rarely necessary, so Stephen request of showing possible useful different example is kind of challenging. 

Below an example I think worth to be mentioned. It is a 100% stacked bar chart for a list of 60 categories of products showing sales margins of underlying products, where the margin (price-cost)/cost*100 is divided in five intervals, not necessarily connected to Likert scale. I use a sequential palette for margin intervals (legend at the bottom) and a top-bottom order based on the average margin per category (lower in top). This design is a variation of already mentioned "sum of multiple parts among multiple bars", but in this case any of the partial sums from first to right or from last to left make sense (margin <=10% first color, <=20% first two colors, <=30% first three colors, ...).

Why would I design that?
1. Main  reason: the bars being stacked in the order of margin intervals, all partial sums from first to the right and from last to the left make sense.
2. A clustered bar chart or five side by side charts showing comparison between each margin interval would not fit properly in the same available space.
3. The comparison of individual margin segments across categories is less relevant. Is no special message to send in comparing, let's say, 10-20% margin interval across the categories.
4. Combined with an average margin order (in top the lowest), I can scan similar margin distribution patterns for adjacent items (see KVM vs KVB Cables).
5. For interactive charts, the selection of distinct segments is helpful for further investigation (tooltip or selection). 
6. Somehow the diversity (richness and evenness) can be detected. Terminals - two major margin intervals, Lenovo one margin interval, Books one margin interval...

 
Margin Stacked Bar Horiz.png 


Even if the above chart does not reveal a new major reason for designing stacked charts, it shows a slightly possible different scenario by using a binned quantitative measure for segmentation. Personally, I don't immediately see a better alternative for the above design.

Dan

acotgreave

Registered:
Posts: 18
Reply with quote  #5 
Steve - 
Regarding my first post: Fair point, you're looking for more categories. It just happened that I was looking at population data at the time of your post, and thought it worthwhile to share.

Regarding my comment on the blog, I hypothesise that Cole's intention was to give the reader an understanding of the growth of missed targets over time. If that's her intention, the stacks work fine: I see the growth of missed targets. The other bars are secondary to that intention. If her  intention had been to allow accurate comparison of every data point, then your lines are better. 

However, if Cole wanted to focus on just the growth of missed targets, then she could just draw those and not the others. I do this a lot when I want to focus on % growth of one particular category:

Dashboard 1.png 

acotgreave

Registered:
Posts: 18
Reply with quote  #6 
I had a look in my work for stacked bars. here's an example from my election project:
percent of total by org.png 

I acknowledge the problems with this chart: it's very very hard to see which organisation tweeted most about UKIP, or the Lib Dems, or any of the other central segments. But, I put it to you, my intention was NOT to allow comparison of every segment. My intention (as shown by the title of the view) was to highlight which organisations were tweeting most about the Conservative party. That was my prime goal. The Conservatives are the left-most segment and sorted in descending order. Any other information viewers can glean from this is secondary to my prime purpose and therefore was underplayed. 

I acknowledge the stacked bar isn't perfect, but I don't know how else I could have designed the chart so that it answered my prime intention (% of tweets about Conservatives) and allowed the viewer to see secondary information. How would you have redesigned it? If you wish, the data is here.

jlbriggs

Registered:
Posts: 200
Reply with quote  #7 
Quote:
Originally Posted by acotgreave

However, if Cole wanted to focus on just the growth of missed targets, then she could just draw those and not the others. I do this a lot when I want to focus on % growth of one particular category:


While a line of just the one data series would obviously work to convey that message, I think the whole point here is that the purpose of the stacked chart is not to focus solely on the "missed" data.

The purpose is to highlight the "missed" data, in the context of the full data set.

The stacked chart does that in a way that no other solution so far does.


Harley_Ellenberger

Registered:
Posts: 3
Reply with quote  #8 
Quote:
Originally Posted by acotgreave
I acknowledge the stacked bar isn't perfect, but I don't know how else I could have designed the chart so that it answered my prime intention (% of tweets about Conservatives) and allowed the viewer to see secondary information. How would you have redesigned it? If you wish, the data is here.


Andy,

If the intent is to draw attention to the % of tweets about conservatives (and if it's important to show the part to whole relationship), then I think your 100% stacked bar is the best solution. It's a traditional chart type that most people are familiar with, so there's some value in that.

Here's a slightly different (maybe non-traditional) approach that still draws attention to your main theme (% of tweets about conservatives) while also allowing a more direct comparison on tweets about other parties.  That each circle originates from a zero base allows the reader to more directly compare those values.  The other party data is slightly subdued based on the circle size.  I didn't include a color legend, but you get the idea.  Of course with this approach you completely lose the part-to-whole context however.

Curious to hear your thoughts on this approach.


2016-01-13 08_47_05-Tableau - Book2.png 

acotgreave

Registered:
Posts: 18
Reply with quote  #9 
Quote:
Originally Posted by jlbriggs


While a line of just the one data series would obviously work to convey that message, I think the whole point here is that the purpose of the stacked chart is not to focus solely on the "missed" data.

The purpose is to highlight the "missed" data, in the context of the full data set.



Here's the problem: who are we to judge the purpose of someone else's chart? If a chart is successful in one purpose but not another, who are to make a judgement based on the unsuccessful purpose? What if it was designed only for the one that succeeded. 

My single line chart's purpose was to show the change in missed targets over time. For that purpose, it only needs one line. I agree, that means it doesn't succeed in many other purposes, but that doesn't make it wrong. If we judge every chart based on intentions not covered, then we'll end up criticising everything.
acotgreave

Registered:
Posts: 18
Reply with quote  #10 
Quote:
Originally Posted by Harley_Ellenberger

Curious to hear your thoughts on this approach.


2016-01-13 08_47_05-Tableau - Book2.png 


This is great. It certainly solves a couple of problems with the original: overwhelming colours and labels. You've cleverly used size and better palettes to maintain the emphasis on Conservatives while showing the detail of the others.

Your chart highlights the crossover between Labour and Conservative (Red and Blue). 

I think it still suffers by being unusual in that it breaks the part-to-whole nature, which is the similar issue I have with Steve's redrawing of Cole's using lines. That results in it taking more time to grasp what's being shown. But once you grasp that, you can easily compare any mark against another. Which is great, if that's the desired intention! 
acotgreave

Registered:
Posts: 18
Reply with quote  #11 
I worked a bit more on my Tweet stacked bar. The colours and labels were too overwhelming in the original. So here's a version with more subtler colouring.

My primary intention is to show the relative amount of tweets about the Conservatives:

  % of total.png 

sfew

Moderator
Registered:
Posts: 838
Reply with quote  #12 
jlbriggs,

Your statement that the 100% stacked bar graph highlights the missed goals in the context of the other categories (met and exceeded) "in a way that no other solution so far does" goes without saying. The question, however, is "Does the 100% stacked bar do this in a way that works more effectively than the line graph?" I believe the answer is "No, it does not." I can make the line that represents missed goals stand out above the other lines as much as necessary. In doing so, I can present the data in a way that provides many benefits over the 100% stacked bar graph.

__________________
Stephen Few
Previous Topic | Next Topic
Print
Reply

Quick Navigation:

Easily create a Forum Website with Website Toolbox.