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 4      Prev   1   2   3   4   Next
wd

Registered:
Posts: 167
Reply with quote  #16 

Agree with nixnut and his logic.  Two separate plots definitely preferred, even if only to prevent people from thinking about correlation.  After all, correlation should be viewed with a scatter plot - one of the 7 common quantitative relationships  ...


__________________
Bill Droogendyk
grasshopper

Registered:
Posts: 245
Reply with quote  #17 
Hmm ... show me a better way!  :)

The data is inline in the code, if you want to give it a try...
http://robslink.com/SAS/democd25/pax.sas


grasshopper

Registered:
Posts: 245
Reply with quote  #18 
More fuel for discussion:

I decided to go back and see what my favorite graphics book (which I used
way back in grad school) says about charts with 2 y-axes.  The book is
"Statistical Graphics: Design Principles and Practices" by Calvin F. Schmid,
year 1983.  He doesn't say these charts _shouldn't_ be used, and he
gives the following rules-of-thumb for their use...

From page 62:
"Column charts and other types of rectilinear coordinate time charts
with two or more multiple-amount scales can be difficult to interpret,
as well as being confusing.  Because of these potential pitfalls, a few
caveats are in order:
1) Charts of this kind should be constructed with meticulous care.
2) They should be used with discretion.
3) Multiple-amount scales normally should be limited to two scales.
4) The zero or other base line should never be omitted.
5) The curves and/or columns should be gauged from a common
    base line, and the divisions of both scales should be spaced in
    the same manner."

(he then gives an example of a good chart, and a bad chart, and
explains what's good and bad about them)

metman

Registered:
Posts: 5
Reply with quote  #19 
As a scientist, I have no problem using dual-axes graphs. There are a couple of reasons to overlay - to show that there is or is not a correlation, and to avoid forcing the reader continually going back and forth. I'll be the first to admit that a non-scientist casual reader may have a harder time.

As long as the colors are distinct, and the representation (bars vs dots), there is little chance of confusion. To make the chart even easier would be to have the temperature scale in blue, not just the label, and as a final touch have the font color of the bottom caption blue for "Maximum daily Air Temperature" and red for "Daily Rain Totals".

Additionally, while the chart may belong in the "water and habitat" section of the report, that should not be the chart title. Again it could be bi-colored to lock the association of color and variable.

Oh, and a note about dot plot for correlation - while it would indeed show the lack of correlation, it removes the chronology, which in some applications could provide important context.
Derek_C

Registered:
Posts: 69
Reply with quote  #20 
There are a couple of reasons to overlay - to show that there is or is not a correlation, and to avoid forcing the reader continually going back and forth.

The point is that overlaying does not actually show a correlation, or rather, that it all too often falsely appears to show one when there is none there.
grasshopper

Registered:
Posts: 245
Reply with quote  #21 

Derek - so are you saying that it is also not ok to overlay charts with a single/same axis?

Derek_C

Registered:
Posts: 69
Reply with quote  #22 
I believe there's an element of constraint in that case that makes comparison valid (as I think Tufte says, one of the most important questions you can ask about a quantity is "compared to what?").

The problem with uncoupled parallel axes in my opinion is that the constraint has gone away, to be replaced by the choices made by the graph designer, who is at liberty to cherry-pick from alternate views to create a view that fits the thesis. To be fair, I'm not always clear why simply putting the curves on separate graphs is supposed to make that better :-)

I'm not dead set against using them, just using them to show correlation. Correlation is best shown using orthogonal axes instead of parallel ones.
grasshopper

Registered:
Posts: 245
Reply with quote  #23 
Here are a couple more examples which seem at least somewhat useful.
I found the first one on the Internet, and I created the 2nd one myself
(using fabricated data):








sfew

Moderator
Registered:
Posts: 803
Reply with quote  #24 
I believe this information could be presented much more clearly and with less confusion using two graphs positioned one above the other--one for total items shipped and total items shipped on time, encoded as lines rather than bars, and one for the percentage of items shipped on time.
__________________
Stephen Few
grasshopper

Registered:
Posts: 245
Reply with quote  #25 
Stephen - As you can tell, I still haven't bought-in yet, but maybe you can
make a believer of me, if you can "show me" the better graph :)

Here is the data, if you care to take a shot at it ...

15feb2005 553 1.0000
15mar2005 507 1.0000
15apr2005 586 1.0000
15may2005 488 1.0000
15jun2005 404 1.0000
15jul2005 306 1.0000
15aug2005 323 0.9905
15sep2005 531 0.9959
15oct2005 677 0.9600
15nov2005 695 0.9624
15dec2005 867 0.9229
15jan2006 557 0.9888
sfew

Moderator
Registered:
Posts: 803
Reply with quote  #26 
Grasshopper,

The information that you provided isn't the same as the information that appears in your graph.

__________________
Stephen Few
Derek_C

Registered:
Posts: 69
Reply with quote  #27 
grasshopper refers to two graphs, but I only see one. Could the data be related to the unseen second graph? Here's a version based on the graph shown.

Attached Images
Name: shipped2.png, Views: 905, Size: 2.80 KB


grasshopper

Registered:
Posts: 245
Reply with quote  #28 
Derek - thanks for pointing out that only 1 graph was showing up!
I had inadvertantly pasted my local/intranet path to the 2nd graph,
so it was showing up for me, but not for anyone else!

I've edited the URL, and it should now be working correctly!

sfew

Moderator
Registered:
Posts: 803
Reply with quote  #29 
Grasshopper,

Here's my dual-graph solution to your second example (the one for which you provided data).

Attached Images
Name: Two_Graphs_to_Replace_Dual-Scaled_Graph.jpg, Views: 904, Size: 129.51 KB



__________________
Stephen Few

Derek_C

Registered:
Posts: 69
Reply with quote  #30 
As the on time is consistently near 100%, I'd graph the "not on time" and use a scale that only covers 10% or so.

And just to show I'm not dead set against them, here's the dual scale version. It's harmless enough, I suppose. My biggest problem with it is that although it doesn't do much harm, neither does it deliver much extra value.

Attached Images
Name: delivery.png, Views: 880, Size: 1.84 KB

Name: delivery2.png, Views: 883, Size: 1.56 KB


Previous Topic | Next Topic
Print
Reply

Quick Navigation: