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 2 of 2      Prev   1   2
sfew

Moderator
Registered:
Posts: 827
Reply with quote  #16 

Andy,

I want to begin my response to your comments by saying that I did not invent wrapped graphs to solve problems with treemaps. The problem that I’m trying to solve is the limitation in the number of values that can be simultaneously viewed and compared in bar graphs and dot plots.

While it is true that wrapped bar and dots make it easy to see where a particular value falls in rank, this is not the primary benefit of a wrapped graph, which is to take advantage of these graphs ability to support precise and simple magnitude comparisons among a larger set of values. We can make 2-D position (e.g., dots) or length (e.g., bars) comparisons much more easily and accurately than we can make area (e.g., rectangles in a treemap or circles in a packed bubble chart) comparison. I’m trying to extend the limits of this better means of comparison so we don’t need to shift to less effective means as soon.

Regarding the annoying visual effect (moire pattern) that you’ve mentioned, it does not exist when the images are properly rendered. When viewed in a PDF document such as my article, the images are rendered in a way that introduces these visual effects. The originals do not suffer from this problem. Regarding lollipops, dot plots with lead lines that connect the labels to the dots work fine as long as the quantitative scale begins at zero, but not otherwise. For this reason, I prefer to use light grid lines that extend all the way across the plot area of the graph, which does the job in a way that does not require a zero-based scale.

Regarding negative values, it doesn’t matter how many of the values are negative versus positive. Wrapped graphs work fine in all cases. I want to point out to others that when you said that your attempt to implement a wrapped graph with negative values in Tableau did not work, you were saying that you couldn’t get the tool to work, not that negative values can’t be displayed in wrapped graphs.

Your opinion that “Whatever its advantages or disadvantages, this visualisation will be slow to be adopted, if at all,” is pure conjecture. You might be correct that bullet graphs were adopted more readily than wrapped graphs will be, but only time will tell. Whether a new form of visualization becomes adopted depends, unfortunately, relatively little on its usefulness, but rather on the willingness of software vendors to implement a good idea when they see it. If a few smart software vendors introduce wrapped graphs, I have no doubt that other vendors will follow and the people who rely on these tools will benefit.


__________________
Stephen Few
jimw

Registered:
Posts: 14
Reply with quote  #17 
I played with wrapped bar graphs a bit over the weekend and am finding them quite useful. In the below graph, I’m showing 480 products by sales. How else could you get this many products on one screen? Previously, I've grouped items into bins or categories, but this prevents shading individual rows by another measure, such as profit ratio, order quantity or product category. 

I did this example in Tableau. It’s more of a brute force approach than Andy’s example---but now that I know what I’m doing, I expect this could be done in ~30minutes. Maybe too much for the average, personal exploratory viz, but definitely about typical for something that I’d present or provide to someone else. I’ll post the detailed steps used to create this graph elsewhere or in a separate tab in the view. 

But I thought folks would like to play around with a live version of wrapped bar graphs. 

One thing I’ll improve in the next round is the labels. The product names range in length from 4 to ~30 characters. I’d like to right-align these against the bars, but when you do this in Tableau and then shorten the column width, the short names get cut off  (i.e., Tableau right-aligns, but still chooses the left-most, say, 20 characters).  

Jim

sfew

Moderator
Registered:
Posts: 827
Reply with quote  #18 
Jim,

You're getting close to the design that I've introduced. The trick is to make this functionality a click away at all times by embedding it in the product. I vote that wrapped graphs be added to Tableau as a replacement for packed bubbles. (Sorry, I couldn't resist.)

__________________
Stephen Few
Jeff

Registered:
Posts: 53
Reply with quote  #19 
Hi, Stephen. This is an intriguing idea. I like that it provides a way to show details within a bigger picture. There are a lot of interactive visualizations that allow a user to explore larger data sets, but many do so by hiding most of the data at any given moment; yours doesn't do this and it's stronger for it.

I would be interested to hear what you think of the attached image as an approach to tackling the same problem. It's a quick Excel sketch--not as refined as the illustrations in your article--but should be enough to get the idea across. 

The basic idea is that the left column is a magnified detail of the right column, the boundary of which is described by the gray box. The user would drag the box up and down to change the view in the left column. The main advantage to this approach is that it may give a better sense of the overall shape of the data set and the subset's place within it.

There are other possible refinements, too--as with the wrapped columns, the software could break the data set into several equal chunks and provide the user a quick way of navigating among them. Likewise, the view could be modified as the user applies filters to the data set.

There are probably downsides to this approach, but as I just made this I'm pretty enamored of it at the moment, so I'll let the rest of you weigh in on those.

Jeff


sfew

Moderator
Registered:
Posts: 827
Reply with quote  #20 
Jeff,

Bar graphs such as the one you've illustrated on the right above provide a nice overview of a large data set's shape, and as such they can be quite useful. When a zoom-lens is added with the ability to see the zoomed detail (focus) simultaneously with the overview (context), as you've suggested, we can explore subsets of data with greater clarity without losing a sense of how those subsets fit into the context of the larger whole. Information visualization researchers began experimenting with these focus+context displays many years ago, but commercial software tools rarely make it easy to view data in this way. What you've illustrated can be used in conjunction with wrapped graphs. For example, imagine what you've illustrated on the right above positioned on the left instead, and then several columns of wrapped bars to the right of that to show the same set of values with greater clarity.

__________________
Stephen Few
Jeff

Registered:
Posts: 53
Reply with quote  #21 
Thanks, Stephen. I can imagine that combined view being pretty useful for exploratory analysis.

Jeff

danz

Registered:
Posts: 190
Reply with quote  #22 
Example of an S&P500 graph showing the volume of transactions for one day (length of the bars) and the variation from the previous day (color) with an improved readability.

1. Splitting the values in two groups. Above are the outliers (> Q3+1.5*(Q3-Q1)) and below are the rest.
2. Different bar widths were used for the groups 
3. Labels are displayed for every bar for the outliers and for first column of the second group
4. Rank is displayed for existing labels 
5. Axes are displayed only for first column. 


sfew

Moderator
Registered:
Posts: 827
Reply with quote  #23 
Danz,

Why have you eliminated the scales for all but the first column?

__________________
Stephen Few
danz

Registered:
Posts: 190
Reply with quote  #24 
Stephen,

For the posted picture it looks for me that one axis per group for above graph can be used as reference. For small values the axis labels would hardly fit, I removed them, they brought unecessary complexity. For second graphic axis could apear on selected column only.  
sfew

Moderator
Registered:
Posts: 827
Reply with quote  #25 
Danz,

The one scale on the left will not suffice as a scale for the other columns. Each column should have its own scale so viewers can easily decode the values based on their horizontal position in relation to the scale. Also, as I illustrated in my examples, when the scale is small, such as in the rightmost column of bars, it is helpful to add minor tick marks or grid lines to support more precise comparisons of short bars.

__________________
Stephen Few
danz

Registered:
Posts: 190
Reply with quote  #26 
It works better with scales and tick marks. Here it is.

One thing I forgot to mention in my previous post, predefined symbols or abbreviations are very helpful for such of graph. Automatic end-ellipsis technique does not give always the best results.

Looking again at the picture it seems that "$" prefix could have been skipped. A simple information that scales are in million dollars would have been enough. Colors have to be explained as well.



jlbriggs

Registered:
Posts: 194
Reply with quote  #27 
"Looking again at the picture it seems that "$" prefix could have been skipped. A simple information that scales are in million dollars would have been enough."

I will most often provide the $ or % (or whatever other signs) on the first or last axis label only - seeing the symbol once makes it clear the scale is in dollars, without having to take up space with a line of text explaining it.
Tim2

Registered:
Posts: 57
Reply with quote  #28 
Hi Stephen,

Do you think if there is a large jump between adjacent values and that those adjacent values are coincidentally split between columns then something important might be missed? Your (presumably fictional) data seems to assume something fairly nice and smooth. Did you consider the pros and cons of repeating the bottom few bars of one column at the top of the next?
sfew

Moderator
Registered:
Posts: 827
Reply with quote  #29 
Tim2,

You bring up a good point. It is probably true that a significant decrease in value from one bar to the next might not be noticed when the first bar appears at the bottom of one column and the second bar appears at the top of the next column. I don't think, however, that this is something that would happen often enough to justify the redundant inclusion of the last few bars or even the last bar only at the end of one column and again at the top of the next column as well. Doing this would potentially lead to confusion (the use might not notice that  some bars appear more than once) and the loss of useful space. Intelligence could be incorporated into the algorithm that determines the break points, but even that might not be worth the effort to correct such a rare problem.

__________________
Stephen Few
BillM

Registered:
Posts: 1
Reply with quote  #30 
Hi Stephen,

Long time reader, first time poster.  Wanted to start by saying thanks for all of the educational dialogue around visualization practices. Has been very helpful in shaping my perspective around visualization tools.  The nice part about being on the vendor side of things is that you get to set your own agenda when it comes to features.  I thought this functionality was unique and a value add, so I've given it a first pass to try and embed the wrap feature into our product.  

Through a right click option, a user can enable the wrap feature to take a standard bar chart and wrap it into many on screen charts. Doing the same action a second time will disable the wrap feature.

The displayed data labels shift dynamically based on which chart series your mouse is active in.  It will show you the labels for the highlighted chart.  

It also supports applying global filters to the data set so that a user can adjust the included data points and the wrap chart will adjust accordingly.  

You can take a look at the actual chart here (plugin installation required):  
http://bit.ly/14uelDN

Happy for anyone's constructive feedback.  

Thanks again,
Bill

Attached Images
Name: WrapChart1.png, Views: 286, Size: 306.22 KB

Name: WrapChartFilter.png, Views: 283, Size: 270.35 KB


Previous Topic | Next Topic
Print
Reply

Quick Navigation:

Easily create a Forum Website with Website Toolbox.