CHI-WEB Archives

ACM SIGCHI WWW Human Factors (Open Discussion)

CHI-WEB@LISTSERV.ACM.ORG

Options: Use Forum View

Use Monospaced Font
Show Text Part by Default
Show All Mail Headers

Message: [<< First] [< Prev] [Next >] [Last >>]
Topic: [<< First] [< Prev] [Next >] [Last >>]
Author: [<< First] [< Prev] [Next >] [Last >>]

Print Reply
Subject:
From:
"cesar martin.iworkwithyou" <[log in to unmask]>
Reply To:
cesar martin.iworkwithyou
Date:
Thu, 2 May 2002 06:36:49 -0700
Content-Type:
text/plain
Parts/Attachments:
text/plain (171 lines)
This is the summary:

0. Read Tufte books.
I did it, and I love them, but the problem is there is no more information
about charts beyond Tufte and that's a problem.

1. Define objectives and think what you want to solve with the charts. This
will give you an idea about what you need to ask when you show the pies.

2. Some people think can be too expensive to test charts and a designer should
be able to solve the problem on their own.

3. For me, the best answer came form Ethan McKinney (unemployed, over-aged boy
genius). Please read more below.

In this link you can see samples with variations for "legends" in a pie chart.
http://www.chamberlainbleu.com/charts.gif

-----------------------------------------------

"Ethan McKinney" <[log in to unmask]>

Determine your objective: what information are you trying to convey with
the pie chart? What information should the user be able to extract?

Construct a test:

1)  Write questions about the information that the users _should_ be
able to answer after viewing the pie chart.

2)  Construct hypothetical pie charts as samples.

3)  Design two types of test. First, the user gets to examine the pie
chart, then, after you take the pie chart away, you ask the user a
question or series of questions about the data (obviously, these cannot
be too detailed--no exact numbers). Second, present the (different) user
with the pie chart. Let them refer to the pie chart as you ask them
questions and they answer.

Of course, you might not do the second type of test at all if you are
trying to convey information the the users need to remember over time
(several hours to several months).

For the second type of test you can also time the how long it takes the
users to find information on each type of chart. To rank the charts, you
might want to use the ordinal ranking from each test subject (first,
second, third, etc.) rather than absolute times in order to wash out
some of the variation between users. After all, some people just find
information faster than others. Of course, this method implies that
you're violating the prohibition in the next paragraph.

It's important not to have the same users test multiple pie charts,
especially those presenting the same data in different layouts and color
schemes. Learning effects will skew your results into uselessness if you
teach the users how to read the charts by exposing them to repeated
variations.

Do _not_ show them all the variations and ask them to choose. Nor do you
ask for opinions or impressions of each. That turns the exercise into
interior decoration. You may get something very pretty, but it may not
be particularly usable. So again, what's your objective.

Of course, you should start by buying and reading all three of Edward
Tufte's books, but especially the first: Visual Display of Quantitative
Information. Terrific stuff and fun as anything to read.

Ethan McKinney
unemployed, over-aged boy genius


-----------------------------------------------
"Andy Dent" <[log in to unmask]>

What are you testing for?

What are the charts supposed to convey?
- quick impression of trends
- detailed comparison of figures where some differ only slightly?
- want user to focus on a single emphasized slice?
- want to publish technically accurate but misleading intent?


If going for quick impressions, I think all-at-once would be best.

If going for details, have a set of questions ready to ask the user after
seeing
the chart (whilst seeing?). Put something bland on the screen between charts
(visual equivalent of eating dry crackers between tasting wines).

-----------------------------------------------

"Amanda Prail" <[log in to unmask]>

Honestly the best way to really test this is to test if it works...i.e
figure out the tasks that the user needs to do with the pie chart
in order to make decisions and time them on how quickly they can get answers
to the questions.  This would be better done as a between subjects design
but you could do it within subjects using order balancing and lots of
different task sets which also need careful balancing and controlling.

Showing the charts to the users may be a reasonable first pass test to
eliminate some of your options but user opinions and user performance are
very different.  Of course, it depends what the primary goals of your pie
chart and web site are as to whether user preference or user performance
should dominate.

Chances are though that this research has already been done in some form or
another and I would start looking to see what already exists.

Pie charts have been around a very long time - another first step might be
to have your prospective users describe their expectations and see what kind
of designs they would generate.

I'm not saying you will find the answers here - but Tufte's books are
excellent references on the background of graphical design for conveying
information and food for thought.

Personally, this sounds way too expensive a tradeoff between what a good
designer should be able to come up with based on what is already known and
what difference testing might make - given that I am sure the testing budget
will not really allow for a really good test.

My recommendation, if you must test - choose the design you think is best
and then test with some users to see if there is anything actually wrong
with it.  Don't try to optimise this - just satisfice and save some efforts
for other things.

Cheers

Amanda

-----------------------------------------------

"Mike Sullivan" <[log in to unmask]>

What are you trying to test for? Aesthetic appeal? Rapid comprehension? Deep
understanding? These would all imply different test methods.

-----------------------------------------------

"Tessa Johnson" <[log in to unmask]>

I would show users various versions of the charts and ask them questions
about interperting them, instead of asking which ones they like better.  If
you ask what the like they will probably tell you based on colors and
"prettiness" when what you are really looking for is which ones can they
understand and interpret the best.  The point of the charts (I'm assuming)
is to convey data, not just decorate a page so you want to make sure your
charts are working in that respect.

Hope that helps.
Tessa


=====
Usabilidad | http://home.earthlink.net/~cesarmartin/think/index.html

** Work | http://home.earthlink.net/~cesarmartin/

__________________________________________________
Do You Yahoo!?
Yahoo! Health - your guide to health and wellness
http://health.yahoo.com

    --------------------------------------------------------------
       Tip of the Day: Don't use automated replies to postings
      CHI-WEB: www.sigchi.org/web  Email: [log in to unmask]
         Vacations, holidays and other subscription changes:
             http://www.sigchi.org/web/faq.html#vacations
    --------------------------------------------------------------

ATOM RSS1 RSS2