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 July/August/September 2013 article, titled Wrapping Graphs to Extend Their Limits, Stephen shows how bar graphs or dot plots can be wrapped across several columns in a way that optimizes screen space and allows many more values to be displayed at once than would be possible with standard bar graphs and dot plots.
 
What are your thoughts about wrapped bar graphs and wrapped dot plots? We invite you to post your comments here.
 
-Bryan
neenertronics

Registered:
Posts: 1
Reply with quote  #2 
That is interesting – I think the bar graphs work better than the dot plots; it’s tougher to find the point of reference with the dots. I was hoping he’d give a real-life example of when one would need an irreducible 250+ data points on one graph. 
sfew

Moderator
Registered:
Posts: 805
Reply with quote  #3 
Neenertronics,

You've reminded me that I failed to include light grid lines in my examples of dot plots, which I have now done. The grid lines solve the problem of linking the labels to the associated dots.

It isn't difficult to think of examples that require 250 values or more in a single graph. A graph that compares all of the world's countries comes immediately to mind. Or, how about a comparison of sales by product. Many companies have hundreds and even thousands of products. Or, how about a graph that a salesperson might use to compare her customers.

__________________
Stephen Few
yersinia

Registered:
Posts: 2
Reply with quote  #4 
Wrapping graphs is a great idea!  I would like to try using this type of visualization when profiling large data sets.  Can this be done using Tableau somehow?
sfew

Moderator
Registered:
Posts: 805
Reply with quote  #5 
yersinia,

Two Tableau experts have tried to reproduce wrapped graphs in Tableau, but it cannot be done. Some aspects of the design can be implemented with some effort in Tableau, but several essential features cannot. Even if wrapped graphs could be supported in Tableau with some programming effort, they would be useful in that form. They would only be useful if they were implemented as a standard feature that could be easily turned on or off. Having to do a bunch of work to produce a wrapped graph wouldn't make sense.

__________________
Stephen Few
yersinia

Registered:
Posts: 2
Reply with quote  #6 
Thanks, Stephen.  Until there is a button or functionality to wrap graphs in our favorite tools, how do you suggest I wrap some graphs?  How did you mock up the exmaples in your article? 
sfew

Moderator
Registered:
Posts: 805
Reply with quote  #7 
Yersinia,

I suggest that you encourage your vendor to add functionality for wrapped graphs if you find them useful.

__________________
Stephen Few
danz

Registered:
Posts: 183
Reply with quote  #8 
Wrapped graphs are very useful to display more values then regular graphs. They can be indeed a good replacement of treemaps especially when negative values are involved or better comparison precision is required.

Should be possible to switch between no labels, one at a time column labels and all columns labels.
For saving space, abbreviations or truncated labels can be used in a delimited space. For a better readability labels area should be equal for all columns.

In case of even more elements (1000+), an horizontal scroll bar can be provided to access the rest of the columns. If this will be the case,  I would suggest an adjustable common scale per view (visible columns) to increase the visibility of very small or very large values. The common scale does not mean common axis ranges for visible columns, these have to stay calculated per columns.  
acotgreave

Registered:
Posts: 18
Reply with quote  #9 
I agree that the attempts to reproduce Wrapped Bars in Tableau are cumbersome, difficult and impractical. Personally my attempt was not to produce a working model that can be transferred easily, but as an exercise in flexing Tableau to see how far it could go. Indeed, Wrapped Bars would only function if they could be turned on/off with the click of a mouse.
wvdv

Registered:
Posts: 2
Reply with quote  #10 

I would suggest to fix the # columns to 4 such that each column corresponds to a quartile - no need for the 25/50/75 lines ?

Still -fundamentally : is an axis with, say, 400 elements still discrete/categorical ? How about binning ? Or a simple line chart with markers ? After all, if you have to hover over a bar to read its label, a simple marker would achieve the same purpose ?

How about this alternative : a stacked bar chart where data points 1..4 would be stacked in bar 1, 2..7 in bar 8..11 in bar 3, and so on : you would need 'only' 100 bars, all data points are in sequence and visible (perhaps colored by # mod 4)

danz

Registered:
Posts: 183
Reply with quote  #11 
wvdv,

I assume that quartile approach is ideal, indeed. But I would not limit the technique to 4 columns only, in some cases few or more columns can be a better choice.

A line chart I cannot use for comparison, even if a hint would provide the desired information in place.

A collection of 100 stacked bars (every consecutive 4 values as far as I understood) has many downsides. It does not have the same alignment for consecutive values making the comparison very difficult. A stacked bar has an obvious cumulative effect we might not need to achieve (I do not see any reason to "cumulate" 4 or any consecutive values from a distribution). The required space is even larger then in the case of wrapped bars (adding first 4 ranked values 1, 2, 3 4 would produce a much longer bar then adding the values ranked 1, 101, 201, 301...). 

sfew

Moderator
Registered:
Posts: 805
Reply with quote  #12 
wvdv,

Restricting the number of columns to four would not produce optimal use of screen space or sizing of bars.

A line chart is not appropriate with discrete values that do not represent either a time-series or a frequency distribution.

Regarding your idea about layered bars (not stacked), Danz (Daniel Zvinca) created a design some time ago that he will be introducing on this site in the next month or so that works along this line.

__________________
Stephen Few
grasshopper

Registered:
Posts: 245
Reply with quote  #13 
Interesting idea, and certainly has some possibilities for exploring data that has lots of values.

One possible problem to watch out for - a user who is "untrained" in reading this kind of graph might look at it and assume that the bars that line up horizontally are all related somehow (as is often the case with a grid of charts).

Will each user need to be educated in how to read the chart, or is there perhaps something "visual" that can be done to help intuitively prevent the user from interpreting the layout incorrectly?
sfew

Moderator
Registered:
Posts: 805
Reply with quote  #14 
Grasshopper,

Yes, you make a good point. As with any new form of display, training will be needed to prevent people from misunderstanding. I've toyed with ways to circumvent this visually, but haven't found a means that doesn't introduce problems that override the benefit.

__________________
Stephen Few
acotgreave

Registered:
Posts: 18
Reply with quote  #15 
Great stuff, Stephen. Here's my thoughts. 

1. The fear of the new

Stephen has created wrapped bars to fix problems in treemaps. I confess I initially looked at wrapped bars and disliked them. But was this the fear of the new? They clearly fix a problem that exist in treemaps, but I initially thought, “Well, treemaps are good enough, why do we need something else to fix a smallish problem with them?”

But what if Ben Schneiderman had invented Wrapped Bars, not treemaps, back in the 1990s? We’d be comfortable with those, by now, I suspect. And then imagine Stephen Few comes along and fixes a problem with Wrapped Bars by giving us the treemap. Would my reaction still be that the new thing was inferior? I suspect yes. This is something one should always be mindful of when assessing new visualisation types.

With that in mind. what is the sweet spot that wrapped bars fix? In essence it’s finding the exact relative magnitude of dimension members at lower ranks. For example, what is the rank of the red mark in the treemap below?

It’s very hard to answer that. Now check out a Wrapped Bar version with 60 rows in each column.

It’s much easier, isn’t it? I am interested to know how big an issue this is for users of visual analysis tools. I don’t think I have ever felt a need for a middle ground between the treemap and bar chart. Switching between those two has sufficed for me so far.

2. Moire pattern

The bars on the first few pages of his article create a moire pattern that disturbs my eyes as I scan the graphic. I would recommend using lollipopsfor wrapped bars: they reduce the amount of ink while preserving the link from the dot back to the axis origin. Incidentally, lollipops should only be used when the scale begins at zero. If you are showing another value, then the line should extend right across the plot area.

 3. Negative values?

Stephen addresses negative values and draws an example where a small number of values are negative. This works well enough and is an improvement on treemaps. It would be an interesting exercise to see how well they work if, say, 50% of values are negative. I tried it in Tableau but my implementation did not work with negative values.

5. Is it for presentation or exploration?

This is the key question for me. Whatever its advantages or disadvantages, this visualisation will be slow to be adopted, if at all it is. Slower than bullet charts, at any rate, I suspect. I don’t expect I will see it in publications or printed materials any time soon. Indeed, Stephen does not intend this to be a printed chart: it is built for interactvity.

However, would things be different if it was part of something like Tableau’s Show Me and could be switched to and from quickly? In this case, I think there is more potential success. There are times during a visual analytic cycle that you do want to understand the lower ranked members of a dimension and I believe the wrapped bar might have more value there.

What do you think? I am always very wary of my first reactions to new visualisation types. People are naturally resistant to change. Are my concerns part of a natural fear of the new, or are they correct?

Previous Topic | Next Topic
Print
Reply

Quick Navigation: