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   Page 1 of 2      1   2   Next
bpierce

Moderator
Registered:
Posts: 97
Reply with quote  #1 
In his January/February/March 2011 article, titled "Quantitative Displays for Combining Time-Series and Part-to-Whole Relationships," Stephen examines the types of graphs that are typically used to display combined time-series and part-to-whole relationships. He then describes the problems that these typical solutions suffer from and proposes alternative designs that avoid these weaknesses.

What are your thoughts on the article? Do you agree that Stephen's alternatives are more effective solutions? Are there ways that they could be improved further? Are there other designs that would work even better?

-Bryan

grebe

Registered:
Posts: 1
Reply with quote  #2 
This article was an affirmation of the difficulties of displaying data and offers some solid rational for challenging some of the default ways of displaying information.  Thanks for sharing this!

I'd highlight the point of chart interaction for those of us in the software side of information design.  The experience I've had with users of our software has shown that a few aspects can go a long way to developing an effective, usable display:

  • Ensure the chart/graphic display loads with the primary use case.
    It may not be beneficial to have every data set turned "on" by default.  Be sure there is logic and forethought in default behavior for each display.
  • Allow the user to control all aspect of the display, including which information is displayed, what the scale should be, what the measurements of the axes may be (i.e. % vs. actual)  
  • Make sure it's obvious to the user that the display can be customized.
  • Remember all of the users settings.
The above points are a bit outside of the scope of the original article, but I appreciated the article's context in examining how various chart-types communicate data vs. what piece of data is most important to communicate (which sometimes can be a dynamic goal based on the user or use-case). 

I'd be interested in hearing the learnings of others who have dealt with charts in software.

erixon

Registered:
Posts: 1
Reply with quote  #3 
Is there a downside to displaying the whole figures on the right axis and the parts figures on the left axis? That way the parts lines won't flatten out and the relational trending is easier to see than if all data sets are on the same axis or on separate graphs. 

I often do this when comparing market share (percentage) versus total industry potential (units or dollars).
elolson

Registered:
Posts: 2
Reply with quote  #4 

I like the idea to have the total displayed in a separate graph above the parts. However, I am color blind and found it difficult to quickly see the trends of the regions. Could you have a bar for each region for each time period in the lower charts?

sfew

Moderator
Registered:
Posts: 802
Reply with quote  #5 
Erixon,

Dual-scaled axes create a number of potential problems. For a full description, please read my article "Dual-Scaled Axes in Graphs: Are They Ever the Right Solution?".

__________________
Stephen Few
Thorri

Registered:
Posts: 21
Reply with quote  #6 
Thank you for a this article. It does a great job of describing the nuances of choosing and designing charts.

Is there a reason the y-axis on the Total line charts go down to zero (pages 5, 6, 8 and 9)? The Total lines on pages 8 and 9 really suffer for this.
bpierce

Moderator
Registered:
Posts: 97
Reply with quote  #7 
Hi Elolson,

If Stephen had used a bar graph to show the data for the regions instead of a line graph, it would have been very difficult to see the trends and patterns, as you can see in the attached example.

In your work, rather than avoiding line graphs with multiple lines, I recommend that you find a set of colors that are easy for you to distinguish. Colorbrewer.org is a great resource on colors and it includes the RGB values for some color sets that are "colorblind safe." Are you able to easily distinguish them?

-Bryan

Attached Images
Name: Article_Example_with_Bar_Graph.jpg, Views: 676, Size: 57.78 KB


elolson

Registered:
Posts: 2
Reply with quote  #8 
Bryan,

I am able see the colors at colorbrewer. I agree that it's still no easy to see the difference between regions with the colored bars. I'd probably need to spend a little more time on that, but overall the bars seemed okay to me.
sfew

Moderator
Registered:
Posts: 802
Reply with quote  #9 
Thorri,

It isn't necessary to start the Y-axis scale at zero with a line chart. On occasions when you don't, however, depending on your audience you might want to point out the fact that variation along the line has been exaggerated by narrowing the scale to make it easier to see the pattern of change.

__________________
Stephen Few
Poffes

Registered:
Posts: 3
Reply with quote  #10 
Thank you for the interesting article. I have just taken over a project and plan to switch to the recommended displays.

I did think that the set of charts on the final page could be improved upon though. As it stands, the relationships of of the different types of tea to each other or the different types of coffee to each other are easily discovered, but I cannot see any way to compare coffee to tea, or either one to the total. If the top graph had lines for the coffee and tea product types I think it would be more complete. Alternatively, you could graph the percentages of the total rather than the product types, but I think that would not be as clear.    
sfleming

Registered:
Posts: 2
Reply with quote  #11 
Thanks for the wonderful review and suggestions for a time series and part-to-whole combination. In the data I work with the part-to-whole relationship is ordinal (test score is on target, off target, or far off target). My co-workers are adamant that these data have to be stacked in order to be understood. What I have suggested is labeling the lines by multiple categories when they are stacked. For example, the first line tracks far off target; a stacked line tracks far off target or off target. What do you think of this approach? What else could work?
sfew

Moderator
Registered:
Posts: 802
Reply with quote  #12 
Sfleming,

If you post an example, I'll be able to respond more usefully.

__________________
Stephen Few
sfleming

Registered:
Posts: 2
Reply with quote  #13 
Let's say student test scores are assigned a performance level on either on track, off track, or far off track. Image001 shows the percentage of students in each of these categories over 5 years. Because there is a natural order to these categories (off track is better than far off track), my clients prefer to see image002, a stacked version with categories off track and not far off track (on track and off track combined). Do you think this is a clear way to represent data of this type?

Attached Images
Name: time_series_with_ordinal_categories_8720_image001.png, Views: 594, Size: 17.06 KB

Name: time_series_with_ordinal_categories_8720_image002.png, Views: 598, Size: 15.28 KB


sfew

Moderator
Registered:
Posts: 802
Reply with quote  #14 
Sfleming,

I'm confused. In the second version of the graph, you've combined "Off Track" and "Far Off Track" into "Not Far Off Track." How can "Far Off Track" become part of "Not Far Off Track", it's opposite?

I understand the usefulness of displaying "On Track", "Off Track", and "Far Off Track" in order, but doing this using stacked lines or areas makes it difficult to compare them or see how all but the bottommost of the three is changing through time. Rather than using stacked lines or areas, perhaps you could reinforce the ordinal nature of the three lines visually by using lines of three ordinal intensities, either using three shades of color from light to dark or three or three stroke weights from thin to thick.

__________________
Stephen Few
vizwiz

Registered:
Posts: 3
Reply with quote  #15 
Stephen,

I've been having an on-going debate about whether a line chart should start at zero and I'm a bit torn/confused.

In Show Me the Numbers, page 169 zero-based scale you say: "When you set the bottom of your quantitative scale to a value greater than zero, differences in values will be exaggerated visually in the graph."  You then say that you should "generally avoid" the non-zero axis, but when you do you should call it out to the reader.   
 
But then in this post (http://www.perceptualedge.com/example14.php), you said "If you look at the vertical axes on all of these bar graphs, you’ll notice that none of them start at zero. This would be fine if lines or dots were used to encode the values."  This article doesn't make any mention of "generally avoiding" or making the reader aware.  
 
Attached is an example of where I get confused.  I added the black line so that you can see how the slope gets exaggerated when you do not start at zero.

I certainly agree that not including zero gives you more detail, that can't be argued.  However, I know when I look at a chart, I see the lines first, then I look at the axis, which is why I tend to prefer the zero-based axis.  
 
I don't think either is "wrong", but if I went with the non-zero axis, I would call it out to the user to ensure they don't misinterpret the steepness of the slope.  

I guess my question is, is it generally acceptable to use a non-zero axis for line charts?

Attached Images
Name: image.png, Views: 263, Size: 33.85 KB


Previous Topic | Next Topic
Print
Reply

Quick Navigation: