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  
bogatirev

Registered:
Posts: 1
Reply with quote  #1 

Hello everyone,

Would really appreciate your help with the following visualization challenge.

I have several categorical measures for churn analysis, and I am trying to find a way to visualize all of them in one visualization.

The visualization type I am looking for is something similar to Parallel Coordinates.
But since the data is categorical, it cannot be used. I did read about Parallel Sets and this looks like a good option, but I was wondering if anyone might have additional suggestions for visualizing categorical measures for churn analysis, I will be happy to hear them as well.

Thank you for the help in advance!
Natalia

jlbriggs

Registered:
Posts: 190
Reply with quote  #2 
This will be a lot more fruitful if you can provide

1) some example data

2) anything related to how this data is currently being displayed or reported

3) information on how this visualization will be utilized
danz

Registered:
Posts: 181
Reply with quote  #3 
Natalia,

You just remind me of a parallel like view I use for quick investigation of both categorical and quantitative data as an alternative of trivial table view. I wasn't aware of parallel set view, but obviously my solution uses a combination of parallel view and parallel set. I call this pattern discovery view, and I use this to quickly identify possible relations between different type of data. Obviously this is not helpful when the categorical data columns have a high cardinality. But for subsets of data can be useful enough. Below a snapshot of my solution where Customer Country - USA is selected.

parallel view.png

sfew

Moderator
Registered:
Posts: 802
Reply with quote  #4 
Danz,

What you're showing is a parallel coordinates plot. Out of respect for Alfred Inselberg, who invented this form of display, let's not call it by a different name.

__________________
Stephen Few
danz

Registered:
Posts: 181
Reply with quote  #5 
Stephen,

Was no intention to claim any invention, I did not see before a parallel coordinates plot used with categorical axis. Parallel set (different than parallel coordinates plot) was new for me (I just looked up for it) and I realized I made this view combining both ideas in one view. Which I posted.
sfew

Moderator
Registered:
Posts: 802
Reply with quote  #6 
Danz,

Parallel coordinates plots can be used to combine both quantitative and categorical variables, as you did in your example.

__________________
Stephen Few
danz

Registered:
Posts: 181
Reply with quote  #7 
Stephen, 

What I've seen till now were parallel coordinates plots using categorical like axis with equally space displayed labels. As R is doing in this link

My design is different, allocating the space for each label in ratio with frequency of occurrences of that category. I might be wrong, but I did not see this approach with any implementation of parallel coordinates plot views, but somehow in the parallel set Natalia just mentioned.
sfew

Moderator
Registered:
Posts: 802
Reply with quote  #8 
Danz,

I didn't notice the fact that your categorical variables reserve different amounts of space to items based on frequency. I don't believe I've seen it done this way. To do what you've done requires that you associate the categorical variable with another variable to determine the frequency (e.g., the number of orders per country). This choice has an effect on the display and the insights that we derive from it. Perhaps it is for this reason that most tools don't automatically render the plot in this manner.

__________________
Stephen Few
danz

Registered:
Posts: 181
Reply with quote  #9 

Current topic was started by Natalia to find a solution to display relations between categorical data. Parallel coordinates plot were meant to analyze patterns and relations between more than 3 quantitative variables (for two variables a scatter plot is a better choice and for three a bubble chart might be better). Current tools implementation forced the use of categorical axis in natural order (alphabetic) to simulate a parallel coordinates plot for categorical data.

Parallel coordinates plots are obtained by drawing for each row in a dataset a set of lines connecting normalized coordinates of quantitative data. My design (few years old already) uses the same idea of using a set of lines but connectivity and space representation of categorical and ordinal data is different. By allocating for each categorical (and ordinal) information a vertical space in ratio with the frequency, I have no coordinates associated to it. The lines connecting categorical data have their ends located in the center of one side of the correspondent rectangle and not on a relevant quantitative position. Not having coordinates associated for categorical information, parallel view was considered by me a more appropriate name for such of representation. I eventually end up calling pattern discovery view because of the purpose and position of this panel in the application (just under the table raw data). This allows a user to use existing table layout (columns order, visibility and widths) for a quick overview of data and possible pattern discovery without any additional setup.

To avoid a cluttered design I did not overlap different drawing entities by splitting the available horizontal space for each information in two parts: one for displaying the representation of data (rectangles for ordinal and categorical data, normalized axis for quantitative data) and the other one for displaying the relations between adjacent columns. Eventually I decided to use colors only to highlight the selected table rows.

Dan

Previous Topic | Next Topic
Print
Reply

Quick Navigation: