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
aruns

Registered:
Posts: 22
Reply with quote  #1 
Hi guys,

It's wonderful to find you contributing quality thought to my projects. Thanks for it all. However, I am back with another visualization to view variation of CTR (a continuous metric ranging from 0% to 4%) and Spend with each day of the week as well as each hour of each day. Please go through the description and image that follow and share feedback/ suggest your own visualizations.

Please find below the drawing I made (Please bear with it). The idea is - 
  1. to have x and y axes as time scales (hour of day and day of week respectively) so that a slot is fixed for each hour of each day,
  2. to have a circle for each slot that represents performance through 2 elements - 
  • Size - Radius of circle proportional to spend
  • Color - Color key ranges from red to green while going from least to highest CTR value
  • Note that there would be no circle for spendless hours
  • All circles will be colored varying from red to green

Attached Images
Name: 2015-12-29_13-46-00.png, Views: 131, Size: 256.94 KB


sfew

Moderator
Registered:
Posts: 812
Reply with quote  #2 
Aruns,

Tell us a bit more about how this is used. First of all, what does CTR mean? Is Spend measured in dollars? What are the conditions, patterns, or trends that people should be able to easily see when viewing this? Should pattern of change over time be easy to see? Do these measures tend to follow particular patterns across the hours of a day? What are the primary comparisons that people make when viewing this? What do people do in response to this? Why did you choose to use circle sizes and colors to represent the values?

__________________
Stephen Few
aruns

Registered:
Posts: 22
Reply with quote  #3 
In the digital advertising domain,
  1. CTR is Click-Through-Rate or Clicks/Impressions. Impressions are number of ads served and clicks is self-explanatory
  2. Spend can be measured in any currency (specific to business)
  3. Purpose of the chart - hours and weekdays having high CTR and high spend should stand out, similarly low CTR and high spend, and indicate clearly as to which hours are good or bad for CTR
  4. Users want to primarily find out what hours/days are good/bad for CTR
  5. Chose colors to make CTR stand out, circle size to indicate the amount of spend in that hour
acraft

Registered:
Posts: 51
Reply with quote  #4 
I wouldn't use colored circles, since it's much harder to examine trends using color scales or circle radii.  Circle radii are extra difficult, since the tendency when comparing circles is to compare their areas, and we all know that our brains are bad at comparing areas.

Your Y-axis as day-of-week isn't really a scale - it's really just rows in a matrix, each of which could contain a separate line chart (with all the charts aligned on the same X-axis, hour of day).  Arranging it that way frees up your Y-axis (that is, the Y-axis in each chart) to be something else, like spend, or CTR, or maybe CTR / spend.
sfew

Moderator
Registered:
Posts: 812
Reply with quote  #5 
Aruns,

I'm still interested in knowing what people will do in response to seeing these conditions that you've mentioned (e.g., low CTR and high spend) and if they need to see patterns of change across the hours of the day. As acraft pointed out, circle sizes and colors display quantitative information in an imprecise way and not in a way that clearly reveal patterns of change. Also, regarding colors, keep in mind that 10% of men are unable to distinguish green from red.

__________________
Stephen Few
aruns

Registered:
Posts: 22
Reply with quote  #6 
Thank you Stephen and acraft,

CTR might not be the best metric to measure, another use case which is likely to be more popular is visualizing spends/revenue e.g. i might be spending money in certain time periods but not get any returns during those times. If users can identify hours of wasteful spends they will manage their spends better by pausing advertising campaigns during those time periods.


sfew

Moderator
Registered:
Posts: 812
Reply with quote  #7 
Aruns,

The primary concern that acraft and I are expressing has to do with the fact that you are encoding time-series values in a way that is difficult to perceive compared to other means. Our brains cannot easily decode and compare sizes or colors. They do a good job, however, of decoding and comparing variations in vertical positions along lines. Separate line graphs for each day of the week arranged one above the other such that the hours of the day are aligned for all of the graphs, would provide a richer picture of the data, especially of patterns of change are of interest.

The fact that you switched from CTR to a different variable when I asked you to tell me what people would do in response to the data suggests that you are looking for a reason to use your design rather than looking for the best way to display particular data for a particular purpose. Always begin with a particular set of conditions that you're trying to display for a particular purpose and then design a means of solving that problem. Don't begin with a particular design and then look for a reason to use it.

__________________
Stephen Few
aruns

Registered:
Posts: 22
Reply with quote  #8 
Stephen,

The metric itself is not important because the user has flexibility to choose his metric of interest. What I do want to enable is comparison of 2 metrics across hours and days as mentioned above. The most fundamental ones I'd want to display by default are CTR and Spend.

The line chart solution would require primary as well as secondary axes for the 2 metrics. This would result in 7 charts with 2 trends on each, wouldn't that be difficult to consume? 
jlbriggs

Registered:
Posts: 194
Reply with quote  #9 
@aruns - do you have some example data you could post to work up examples with?

jlbriggs

Registered:
Posts: 194
Reply with quote  #10 
I don't think it has to be as complex as it sounds to plot this in the manner suggested.

One way to go about it would be to use opposing areas - one for CRT, one for Spend, like this:

crt_vs_spend_daily.png 

The benefit over using circles where color is one metric, and size is another, is that it removes the confusion and the difficulty in determining what a particular size and shade represents.

The benefit to using the area instead of just a line is that it still functions in the almost heatmap fashion, by noting increased activity with increased blocks of color.

Patterns by time of day or day of week should jump out pretty clearly (though with this randomly generated normal data, nothing jumps out...)

I would avoid any red/green coloring, and would keep the colors fairly low key - there's going to be a lot of color in the display, and you don't want it to be overwhelming.

sfew

Moderator
Registered:
Posts: 812
Reply with quote  #11 
Aruns,

In part, I'm trying to point out that the metric always matters. There is no generic solution that fits all metrics. For example, the solution that jbriggs illustrated above works when one metric has positive values (CRT) and the other has negative values (spend), but it would be quite confusing otherwise. You always begin with a specific set of data and a particular set of questions that need to be asked of the data. Generic solutions are of limited use.

__________________
Stephen Few
jlbriggs

Registered:
Posts: 194
Reply with quote  #12 
Stephen - that's a very good point, which I should have mentioned in my post.

I would only use this set up in situations like this, where spend can be logically associated with a negative value, and because the two metrics can be seen as 'opposing', and so can be plotted as opposing as well.

Simply aligning any two random metrics in this manner would not make sense.
acraft

Registered:
Posts: 51
Reply with quote  #13 
@aruns:
Quote:
The metric itself is not important because the user has flexibility to choose his metric of interest.

While it's technically feasible to swap out one metric for another and while such a feature may turn out to be useful with a carefully-selected list of available metrics, it's not a very useful design approach to just generalize your use cases.  That's how most BI vendors seem to approach the problem.

I think you still need to consider Stephen's concerns about how your visualization is to be used, what people will do in response to certain conditions, etc.  This research is how you determine the best way to visualize your data.

Quote:
This would result in 7 charts with 2 trends on each, wouldn't that be difficult to consume?

I'm not sure why it would be any harder to consume than a grid with 168 circles in it (if you are in fact showing all 24 hours in each day of the week).  It's not as easy as it seems to look at a screen full of circles and quickly notice which ones are red vs. green, or big vs. small.  And it's nearly impossible to examine trends this way.  You are essentially limiting your users' ability to analyze the metrics to only examining one day's hour at a time.

On the other hand, it's actually quite easy to look at a timeline and quickly identify highs and lows, even if you have 7 of them, in addition to seeing when metrics start to rise or fall.  It's the same data, but a lot more information is conveyed this way, and clearly.  You also have a lot more flexibility this way, depending on your users' needs to see averages, control limits, comparisons to prior time periods, etc.
danz

Registered:
Posts: 186
Reply with quote  #14 

Quote:
Purpose of the chart - hours and weekdays having high CTR and high spend should stand out, similarly low CTR and high spend, and indicate clearly as to which hours are good or bad for CTR


Even if information is time related, it does not look for me that your audience primary goal is to study the evolution in time of the two variables, but to indicate highly correlated or non correlated values.

1. Instead of a crosstable use a scatterplot of 24x7 points. Use colors to identify days of the week. Beside an accurate encoding of both variables, any negative or positive correlation will stand out, be sure. 

2. If above looks too cluttered, use 7 scatterplots of 24 points. You may have several options to encode colors. Use a divergent series of colors to encode CTR/SPEND ratio (slightly redundant). Do not use a gradient palette, but a fixed amount of colors (4, 5 or 6). Add a histogram under each scatterplot counting the occurrences of each point in its performance interval.

You can indefinitely go for more angles of your data. Time variation - separate or same chart if it makes sense, cycle plots (how did behave your variable every day of the week at the same hour, 24 line charts), aggregated values and so many...

In order to develop your visualization skills, there were some readings jlbriggs and I suggested you in another post. If you would have checked those suggestions, your subject would have been less general, more accurate described and, consequently, our comments more effective.


aruns

Registered:
Posts: 22
Reply with quote  #15 

@jlbriggs:

Thanks for the effort. Agreed.

@acraft:

  1. All metrics I'm considering here fall under 2 categories w.r.t. positive correlation with spend-
    1. Good - e.g. high spend, high CTR
    2. Bad - e.g. high spend, high Cost per Transaction
  2. If I can figure visualizations for these 2 cases, I think I can manage to include metrics that satisfy these
  3. Agreed about the trends v/s circles

@stephen:

  1. Questions that need to be answered:
    1. Which hours are consistently bad/good performing over the week?
    2. Which days are better/worse than other days?
  2. Quote:
    For example, the solution that jbriggs illustrated above works when one metric has positive values (CRT) and the other has negative values (spend), but it would be quite confusing otherwise.
  3. I don't quite understand what you mean by the above quote as all metrics are positive, just that the meaning of their positive correlation with spend changes. Could you please explain? Thank you all.

 

Previous Topic | Next Topic
Print
Reply

Quick Navigation: