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 3      1   2   3   Next
bpierce

Moderator
Registered:
Posts: 99
Reply with quote  #1 
For the April/May/June 2014 Visual Business Intelligence Newsletter article, Stephen examines three different ways of Displaying Change Between Two Points in Time. He describes and illustrates each approach, and evaluates their specific strengths and weaknesses. He also shows how combinations of the three approaches can often serve people's needs better than any one graph could.

What are your thoughts about the article? We invite you to post your comments here.

-Bryan
Berry

Registered:
Posts: 6
Reply with quote  #2 

In the logarithmic slopegraph, you should really add ticks at each 10'000 - otherwise the viewer might easily miss that the scale is logarithmic.

And why do you have to "zoom in" at the smaller values and "zoom out" at the larger values (which is what a logarithmic display does) to see rates of change? I see rates of change on a linear scale as well.
To be able to exactly compare them in a slopegraph, you would have to let all values start at 100% of the starting value anyways - which is not done by a logarithmic scale either.
Regards,
Berry
wd

Registered:
Posts: 167
Reply with quote  #3 
I think the key lesson here is that one graph (e.g. those on page 1 (often displayed as full page)) is not necessarily sufficient to display the entire story that's hidden in the data.  It's far better to show 2 or 3 or 4 (or more) smaller charts (on one page) from different perspectives and plainly bring out all that's meaningful - and that's preceded by good analysis.  I like what Kathy Rowell wrote here: http://ksrowell.com/blog-visualizing-data/2014/01/10/the-one-trait-all-successful-data-analysts-must-possess/.

Bill

__________________
Bill Droogendyk
bpierce

Moderator
Registered:
Posts: 99
Reply with quote  #4 
Hi Berry,
 
Regarding the tick marks, you're correct that although minor tick marks (unlabeled tick marks that appear between the labeled tick marks) usually aren't useful in graphs with linear scales, they can serve to help clue people into the fact that they're looking at a logarithmic scale. However, I assume Stephen didn't think this was necessary in the article, because the graph with the log scale appeared in a section that specifically talked about log scales, and was introduced as such, so it wasn't a surprise.
 
Regarding your second point, although there are other ways of comparing rates of change besides using a log scale, you can't accurately compare rates of change using a standard line graph with a linear scale. You can if you use a log scale. With a log scale, any two lines with the same slope have the same rate of change.
 
In the following example, the data has been displayed on the left with a linear scale and on the right with a logarithmic scale.

Comparing Rates of Change.jpg  
In the left graph, it appears that Hardware is increasing faster than Software. It's certainly changing more in magnitude, but using the linear scale we can't accurately compare the rates of change. Using the logarithmic scale on the right, however, we can see that Hardware and Software have exactly the same rates of change, evidenced by the fact that we could overlay one line on top of the other and they would be identical. In a graph that uses a log scale, any two line segments with the same slopes exhibit the same rates of change. This is true regardless of what base the log scale uses. 

-Bryan

Berry

Registered:
Posts: 6
Reply with quote  #5 
Wow - why did nobody teach me this? This is great.
I mean, I always try to avoid logarithmic axes as they're sometimes hard to interpret for layman.
But for rates of change it's the only senseful way indeed.
And it actually works for all bases - I guess I have to get back to the base (pun intended^^) and refresh my knowledge of logarithms...

Thanks for humbling me and teaching me some pretty cool stuff!

If somebody wants a simpler dataset with several bases, here it is:

logchanges.png 

ulrichw

Registered:
Posts: 4
Reply with quote  #6 
In my humble opinion, even the missing features "Easy to compare values at the same point in time" and "Easy to spot changes in rank" which were marked with "N/A" in the Deviation Bar Graph can be resolved by using the following chart type which was designed according to the IBCS-A standards - Ulrich

FEW_IBCS_2014-05-08_v01-depois_01_8.png

sfew

Moderator
Registered:
Posts: 823
Reply with quote  #7 
Ulrich,

This chart from IBCS-A can be improved in several ways.
  1. The horizontal bar graph on the left is not a particular good way to show change through time. Lines or range bars do a better job of this. Also, notice the confusing visual effect caused by placing bars on top of one another rather than side-by-side. Comparing a full bar to a sliver of another bar is perceptually difficult.
  2. In the third graph, there is no reason to use thin lines with squares at the end (i.e., what Rolf Hichert calls a pin) rather than bars. Bars are easier to see and compare.
  3. Placing numbers at the ends of the bars creates unnecessary clutter. It usually works best to include quantitative scales instead. When individual numbers are required for precision, it works best to place them in their own columns next to the item labels for easy look-up and comparison. Notice that placing the numbers at the ends of the bars made it impossible to show numbers for both sets of bars in the left graph.
Although Rolf Hichert's standards (a.k.a., IBCS-A) include many valuable suggestions, they also exhibit many examples of ineffective design. This stems from the fact that they were not based on a clear understanding of visual perception and cognition, but solely on the Hichert's experiences and judgments, which are of high quality but not complete. I appreciate Hichert's fine work, but cannot endorse it as a set of international standards. Assuming that international standards in the design of quantitative graphs are needed, which I'm not convinced is the case, they should be developed as a collaboration between international experts in the field rather than the work of one individual.

__________________
Stephen Few
GiovanniMilan

Registered:
Posts: 4
Reply with quote  #8 
Some time ago I devised this alternative to slope graphs but I it was too unusual, so I used it only privately:
xy-Graph.jpg

Pros: it shows both pre- and post-ranks, it allows to compare the variations of close points (e.g. Caffe Latte vs Mint, Regular Espresso vs Green Tea).
Cons: it may become too cluttered, variations of distant points are difficult to compare (variations are not ranked).

ulrichw

Registered:
Posts: 4
Reply with quote  #9 
Quote:
Originally Posted by sfew
Ulrich,

This chart from IBCS-A can be improved in several ways.
  1. The horizontal bar graph on the left is not a particular good way to show change through time. Lines or range bars do a better job of this. Also, notice the confusing visual effect caused by placing bars on top of one another rather than side-by-side. Comparing a full bar to a sliver of another bar is perceptually difficult.
  2. In the third graph, there is no reason to use thin lines with squares at the end (i.e., what Rolf Hichert calls a pin) rather than bars. Bars are easier to see and compare.
  3. Placing numbers at the ends of the bars creates unnecessary clutter. It usually works best to include quantitative scales instead. When individual numbers are required for precision, it works best to place them in their own columns next to the item labels for easy look-up and comparison. Notice that placing the numbers at the ends of the bars made it impossible to show numbers for both sets of bars in the left graph.
Although Rolf Hichert's standards (a.k.a., IBCS-A) include many valuable suggestions, they also exhibit many examples of ineffective design. This stems from the fact that they were not based on a clear understanding of visual perception and cognition, but solely on the Hichert's experiences and judgments, which are of high quality but not complete. I appreciate Hichert's fine work, but cannot endorse it as a set of international standards. Assuming that international standards in the design of quantitative graphs are needed, which I'm not convinced is the case, they should be developed as a collaboration between international experts in the field rather than the work of one individual.



Stephen,

thank you very much for your comments. Let's keep the discussion alive:

Ad 1: I agree with you. Line graphs are better for showing change through time. In my example although I referred to the suggested combination of the two deviation bar graphs. Adding a bar graph with absolute previous year and actual year data increases the understanding of the deviation values. Placing the PY bar on top makes sense as the position below is reserved for budget or forecast data sets. For sure, the secondary bar graph could be broadened to 50% or 100% of the primary bar.

Ad 2: One of the basic ideas of IBCS is that every element in a graph or table has its meaning: Therefore a broader bar is for absolute values, a narrow one is for absolute deviations and the pin with the square is for relative deviations. Off course, the selection of the different forms can still be discussed. But we should keep in mind, that “Similar content should be visualized in a similar manner and vice versa; what looks the same should also mean the same.”

Ad 3: Placing numbers indeed is for precision purposes only. And placing them outside the bar makes sense when we deal with small numbers which wouldn’t fit in the bar. As in the mentioned bar graph we work with absolute values and absolute deviation values there is not necessarily a need for placing the PY value, as it can be calculated using the value of the second graph. If we broadened the PY bar,  as you suggested, we also could have placed the PY values there.

First of all I think we should thank Rolf Hichert for his fundamental work. At least he prepared a written basis for a broader discussion of international standards for the business communication area, which are still lacking. When we have a look at the inumerous powerpoint presentations held at investor days, shareholder meetings, general assemblies etc. and we see the abundance of highly creative graphs and tables, but without any visible use of notation, we can feel that there is a need for some kind of standardization, at least for quantitative business communication.    

sfew

Moderator
Registered:
Posts: 823
Reply with quote  #10 
Ulrich,
  1. Adding a bar graph with absolute previous year and current year data doesn't actually increase understanding of the deviation values compared to a slope graph. I'm not sure what you're suggesting by the comment "...the secondary bar graph could be broadened to 50% or 100% of the primary bar."
  2. Requiring that every type of data be visualized in a different manner (e.g., actual values with a broader bar, absolute deviations with a narrow bar, relative deviations with a pin) over-complicates the IBCS standards and dramatically limits their application. One of the problems with these standards is the fact that they are narrowly focused on financial data. Assigning specific conventions to common items of financial data may be possible, but once you open the door to other types of data, it is not possible or useful to established fixed conventions for every type of data for there are far too many types. I agree that similar content should be visualized in a similar manner. The problem is that deviations along a nominal scale of all types are similar, but ICBS standards insist on treating them differently (e.g., bars vs. pins for absolute vs. relative deviations). I understand the intention of these standards, but the differences in graph conventions that they advocate go too far.
  3. I think you've misunderstood my suggestion regarding the placement of numbers in graphs. Placing them anywhere in the plot area of the graph creates undue clutter and makes them harder to compare than they would be if they're placed outside of the plot area entirely, such as a row of numbers directly below the x-axis labels.
I am grateful to Rolf for his fundamental work. I don't believe, however, that it can serve as the basis for international charting standards and I'm not convinced that these standards are needed. Unlike notation standards for specific areas of engineering, quantitative graphs are used for a vast range of data domains and purposes, and as such, they aren't amenable to rigid notation standards. What's needed is more widespread training in graph design based on a firm understanding of human perception and cognition. Unfortunately, Rolf's standards are not rooted in this understanding, which is why they sometimes advocate ineffective design practices such as positioning bars on top of one another.

__________________
Stephen Few
danz

Registered:
Posts: 190
Reply with quote  #11 
Ulrich,

If I have to compare Stephen graphs from the article with IBCS template submitted by you, I can only say IBCS sample fails in too many aspects to consider (for now at least) IBCS as a standard reference. Even if I will repeat some of Stephen remarks, I'll make below my full list of comments.

1. Partial bars overlapping is not a practice I would consider in any scenario.
2. Labels are absolutely not necessary, scales being more appropriate for graphs and improved readability.
3. A table like approach with alternate row backgrounds or thin grid lines would improve the readability of IBCS example. 
4. I see no benefit in IBCS example from using different graph types for different measures. I do consider myself dot graphs as alternative to bar graphs in some cases, but to consider as standards pins or bars for different types of measures looks arbitrary to me. 
5. Colors (if they are needed at all) are too vivid.
6. Vertical axes are too thick.
7. Titles are too close to the graphs area.
8. It is not an aesthetically pleasing layout, making the overall impression not favorable.


I have to say that I wasn't aware of "International Business Communication (IBCS) Association" till two days ago. Being intrigued by your IBCS template submission in this thread, I took my time to navigate through ibcs-a.org site to read more about this organisation and its purpose.
I agree that many of the success rules I found there have sense and I do appreciate the effort of collecting and organizing them in one place. At this early stage of the organisation, "standards" is, in my opinion, an ambitious word, "recommendations" would have been more appropriate. Certifying consultants, applications, software or software providers by IBCS or any other similar organisation requires much more than what I see for now on ibcs-a.org site.

Dan 


sfew

Moderator
Registered:
Posts: 823
Reply with quote  #12 
I think it's appropriate to remind ourselves that this particular discussion topic was set up to discuss my recent article, "Displaying Change Between Two Points in Time." We've gone off on a tangent. I didn't know about the International Business Communication Standards (IBCS) that Hichert & Partner are proposing until last week. Juergen Faisst, the partner at Hichert & Partner, attended my Visual Business Intelligence Workshop in the Netherlands last week, and he introduced me to their efforts during breakfast together one morning. Although I appreciate Rolf Hichert's work, I don't think it is appropriate to base international charting standards on it. I believe that the venture, however well intentioned, is ill-conceived, in part because the world is not asking for international charting standards of the type that IBCS proposes. These standards are too rigid, too focused on finance, and too closely aligned to the vision of one man. I understand Rolf's interest in giving his work more lasting permanence as he approaches retirement, but I believe that his legacy is already well established and will only be undermined by the IBCS program.
__________________
Stephen Few
ulrichw

Registered:
Posts: 4
Reply with quote  #13 
Dan,

regarding your suggestions on the bar graphs I designed an example according to your points. In comparison to my former contribution I think we lost quite a lot of precision and information....
- Ulrich


FEW_IBCS_2014-05-22_v02-depois.png 

sfew

Moderator
Registered:
Posts: 823
Reply with quote  #14 
Ulrich,

Dan did not suggest that you place the bars in the first graph on top of one another. They should occupy separate space, in this case by placing them one directly above the other. Also, your quantitative scales should be complete rather than ending below the highest value and the intervals should be equal in size. If you include grid lines, as you have, they should not be placed on top of the bars and they should be very light, just visible enough to assist the eye. The scales on the first and last graphs should have more intervals, such as intervals of 10 rather than 30 in the first, and intervals of 25 rather than 50 in the last. The axis line at 0 should be consistent in appearance in all three graphs, rather than black in one and gray in the others. I suggest a gray of about 25%. If precise values are needed, they can be placed to the left of each plot area, right-aligned in a column, rather than at the ends of the bars. For example, for the graph on the left, numbers could appear to the right of the labels, to the left of the zero-based axis line.

__________________
Stephen Few
danz

Registered:
Posts: 190
Reply with quote  #15 
Ulrich,

While we focus on current year performance I would keep current year revenue, absolute change and percent of change. If values are important, a table like below would provide enough information. Even if it introduces some redundancy using scale and labels, is not that heavy. Even if I usually prefer a reduced notation for scales (20k instead of 20,000), I preferred in this case to be consistent with the format used for labels.

chg-time.png 

Dan

Previous Topic | Next Topic
Print
Reply

Quick Navigation:

Easily create a Forum Website with Website Toolbox.