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 
Hello,

I am new to data viz and I am trying to prepare a graph to show the decrease in unit customer acquisition costs despite an increase in spend. Is the graph below the best way to show it or can i do better? I would also like to show how ROI has also increased but not sure how I can fit it on this graph. Any thoughts and advice are much appreciated. 

[image] 


jlbriggs

Registered:
Posts: 194
Reply with quote  #2 
First and foremost, you have no y axis labeling, so we have no way of knowing what is being plotted.

Are these two series on the same scale, measuring the same units?  If not, that's a big problem.

Either way, label the y axis so we know what's what.

Secondly, don't worry about how to fit ROI onto the same chart - worry about making however many charts are necessary to properly display the data.

It's hard to get into any more detail without knowing what the purpose, audience, and underlying data are for this.
aruns

Registered:
Posts: 22
Reply with quote  #3 
Hi,
We're essentially comparing the % changes in two variables. These variables could be on the same or on different axes, that is immaterial to this representation. I've modified it a bit to try and make it clear, please look at it and tell me if that can change your opinion by a bit. Thanks. Tell me what you think


Screen Shot 2015-12-04 at 12.48.42 am.png 



acraft

Registered:
Posts: 51
Reply with quote  #4 
Whether these variables are on the same axes or different axes is actually very material to this representation: In your example, I don't know if the intersection at Month 2 implies something significant or not.  Having a labeled y-axis would clear up a LOT here.

Also, the curvature of the lines in the second version doesn't make the chart any clearer, it actually makes it kind of less clear.
danz

Registered:
Posts: 186
Reply with quote  #5 
Aruns,

Remove the arrow, is no need for that, direction of change is obvious.
Avoid using smooth techniques for lines, they bring no added value.
Remove any background color or ornaments, at least during design phase (in my opinion, remove them for good).
Make the y axis visible and mention the unit of measure.
Your variables suppose to start both at 100% (original reference), one goes up to 200% (var +100%), the other one goes to -100% (var -200%), so they do not intersect as they were represented in your chart.

My last comment is related the way % of change is used. For instance, if certain variable increases from 75 to 150, the % of change is +100%. But if the same variable decreases next month from 150 to 75, the % of change is -50%. The comparison between 100% and -50% usually leads to confusion. I find useful to use % of change referenced to original value, rather than on consecutive values.

Dan
aruns

Registered:
Posts: 22
Reply with quote  #6 
Hi Dan,

I'm a little confused with respect to the comment above 
  1. Let's take both variables on 2 different axes. Now their units of measurement maybe same or different
  2. Both start from a particular point in their own domain. In the next 5 months, the % change observed in both is as follows:       +100% in Spend, -200% in CPT
  3. Now if I've scaled the axes in a way that enables the primary trend to cross over the secondary trend because one is upward and the other is downward, then what is the problem with that?
Anyway I'm providing you the data I'm trying to plot without showing the absolute numbers (reason for hiding axis and ticks). Appreciate it if you can send me your version of how this could be consumed. In the meanwhile, I'll make some changes to the visualization sent above
MonthSpend CPT
Apr₹ 27,21,292₹ 2,522.05
May₹ 37,44,460₹ 2,036.14
Jun₹ 51,11,190₹ 1,953.07
Jul₹ 59,02,438₹ 1,321.05
Aug₹ 59,02,335₹ 1,165.09





aruns

Registered:
Posts: 22
Reply with quote  #7 
Have added the labels
Screen Shot 2015-12-04 at 2.24.07 pm.png 

danz

Registered:
Posts: 186
Reply with quote  #8 
Aruns,

Jlbriggs already mentioned that using one graph with two different y axis for different measures is wrong. I thought you are trying to show the percent of change for two different measures, which is fine for one graph (they share the same meaning, unit of measure and scale). But it seems you just fit in available vertical space two variations they do not share the same scale. This is wrong, please read Stephen article about this matter.

One more remark related to percent of change comment you made. What you call a "decrease of 200%" is in fact a decrease of 50%. You need to be consistent with the reference value which in both cases are April (1st month) values. I have no idea what AdWyze is, I just assumed it is a technology or procedure. If this is the case, a more appropriate conclusion would be "Since April, when AdWyze was implemented, the Expenses have been doubled, while Cost Per Transaction has decreased by more than a half".  

Below you may see the graph designed according with the suggestions made in my previous post. Values displayed below are optional, I made them available for a better clarification.

CPTvsExpenses.png  



PS. You latest post, with updated design I saw only after I posted this one. I insist to remove arrows and gradient. Added vertical labels do not help the original design at all.

jlbriggs

Registered:
Posts: 194
Reply with quote  #9 

I skipped ahead, and have not yet read the last few replies.
I will go back and read them, and possible comment further, but before that I need to respond to this:


"These variables could be on the same or on different axes, that is immaterial to this representation."

That is as wrong as wrong can possibly be. In fact, that was a pretty big part of my point.
Surely you've seen similar charts in the news lately, or posted on various datavis sites. 

If these two series are not sharing a common scale, then the chart is at best meaningless, and at worst an outright lie.

Aside from that, the axis needs to be labeled so that the viewer can understand the magnitude of the changes plotted.

Assuming that those two lines *are* in fact on the same scale, both representing % change, we have no way of understanding what % change has actually occurred, because there are no labels.

The entire change plotted could represent a range of only a small fraction of a %, or could be multiple hundreds of a % - obviously the viewer will need to know this in order to make any sense of the chart at all.




jlbriggs

Registered:
Posts: 194
Reply with quote  #10 
Quote:
Originally Posted by aruns
Hi Dan,

  1. Let's take both variables on 2 different axes. Now their units of measurement maybe same or different
  2. Both start from a particular point in their own domain. In the next 5 months, the % change observed in both is as follows:       +100% in Spend, -200% in CPT
  3. Now if I've scaled the axes in a way that enables the primary trend to cross over the secondary trend because one is upward and the other is downward, then what is the problem with that?



What's wrong with it is that it is a lie.

Plain and simple.

You have created an interaction between the two series, where the two cross, that implies a meaningful milestone occurred at that point.

But in fact it is completely meaningless.

If you want to chart the % change of two different things, then you need to:

1) measure it in terms of the actual % change, and plot it accordingly.

2) label your y axis to reflect how much of a percent change is being plotted.

In addition, as Danz has pointed out, all of the odd background effects are nothing but a distraction that make your chart harder to read.

I have no problem with the arrow heads, personally. They're fairly small and function almost the same as a normal point marker at the end of the series.

But the bottom line is, show the data in as clear a way as possible, without misrepresenting it. What you have done here is to wildly misrepresent that actual data.

In addition to the article Danz posted, I'll leave these links here as further reading:

* http://www.thefunctionalart.com/2015/10/if-you-see-bullshit-say-bullshit.html

* http://www.thefunctionalart.com/2015/10/double-axes-double-mischief.html

* http://junkcharts.typepad.com/junk_charts/2015/11/egregious-chart-brings-back-bad-memories.html
aruns

Registered:
Posts: 22
Reply with quote  #11 
I see what you mean. Thanks for the education. Spending time with your chart, I am convinced that it is more accurate than mine as also you've found an interesting way to communicate changes in 2 different variables (of 2 different units) by using a common starting point for both (100%). Thanks for the links. Agreed about the effects also. 

Learning: When trying to show % changes for different metrics over time, I can think of them being in a 'null' state (the common starting point) so as to say where each metric is at a certain default value and this default state is the point of origin for the trendline, and from then on the % changes in the values are an intelligent way to show by how much the future state varied from the null.


aruns

Registered:
Posts: 22
Reply with quote  #12 
I was trying another viz. to convey the same insight, as you'll see below.
I've tried including 3 dimensions - Spend, Transactions and ROAS. Spend and ROAS make up the axes, while I'm using radius of the circle to represent number of transactions.

Please comment on how I could make this more consumable/ share thoughts on how miserably I've failed. Thanks a ton!

Screen Shot 2015-12-06 at 6.58.02 pm.png  

danz

Registered:
Posts: 186
Reply with quote  #13 
Aruns,

I strongly suggest you to read some literature (you can start with Stephen books) before to continue to design data visualizations. Without having basic knowledge, it would be difficult for you to follow our advices.

You were right about your latest graphic, it totally failed. If you would have had the basics, you would have understood that:

Size of the bubbles are harder to compare than bars length.
Triple size of a bubble does not mean triple diameter. Triple diameter means 9 times size.
Is no need to compare all different measures in one view.
Graphs require scales.
Colors have to be used with care. Background colors for tables have the same sense as for graphs: none.
Alignment of different views/elements on screen matters.
Center alignment of labels or values in a table are rarely useful.
Font types, sizes, color, styles should have a limited diversity

...


aruns

Registered:
Posts: 22
Reply with quote  #14 
You're so right man! Damnn!
jlbriggs

Registered:
Posts: 194
Reply with quote  #15 

I agree with danz on the technical aspects of his comments.

However, I think the concept of what you've done in your last graphic is perfectly valid.

I would caution against the instinct to try creating on killer graphic that will wow everyone into understanding, which is what it seems you're going for here. But there's no reason that the three variables involved can't be displayed in this type of chart.

While it's true that bubbles are not very good in comparison to bars, when you need to encode a third variable on a chart where x and y are spoken for, it can be perfectly justifiable (assuming you scale them properly, by multiplying the radius rather than the diameter).

The question comes down to does it work in this case, or are there better ways?

But we can get back to that point later.

What this chart is truly suffering from is a need for better graphic design.

As danz mentioned, colors, type faces, spacing, alignment, are all integral parts of a design, and make no mistake - data visualization is very much a design field.

As danz also mentioned, reading Stephen's work is a good place to get started in understanding a lot of the concepts needed.
You can do that right here on his site, starting with the examples:

http://www.perceptualedge.com/examples.php

And looking through the newsletters section here on the board.

I would also direct you to Storytelling With Data:

http://www.storytellingwithdata.com/gallery/

For some good, clear, basic tips and examples.
Take note of how effective and clear Cole's charts are, and what they all share: clean, simple design, with limited use of color, great use of typography. Simple design choices with a fairly easy learning curve that make a huge impact on the finished product.





 

 

Previous Topic | Next Topic
Print
Reply

Quick Navigation: