R help - Feb 2011 - Re. When is *interactive* data visualization useful to use?

Hello Tal,

You asked *When is it helpful to use interactive plots? Either for data
exploration (for ourselves) and data presentation (for a "client")?*

My answer: It's helpful for checking data quality, for exploration with and
without "clients", for checking results, and for data presenting.

Notes:
(1) It's difficult to explain interactive data visualization in print,
demonstrations are so much more effective.
(2) Interactive data visualization is fun, both for the analyst, and more
important, for the dataset owners.  You not only get better interaction with the
data, you get better interaction with the scientists you cooperate with.  They
are prepared to contribute, because they can understand what is going on.  That
is not always the case with statistical models.
(3) The key is not "animation" but "direct manipulation". 
The aim is to be able to directly interact with all statistical objects in a
graphic: querying, linking, reordering, reformatting, zooming, whatever.
(4) You write of point-based graphics, what about area-based graphics like
histograms, barcharts and mosaicplots?  For categorical data the ability to
select groups and look at spineplots of other variables to compare proportions
is very effective. (And don't forget linking to maps for spatial data.)
(5) You mention outliers.  How do you decide what is an outlier?  Interactive
parallel coordinate plots are extremely useful, either for identifying outliers
or for checking ones found with an analytic approach.
(6) Interactive data visualization is not in competition with other approaches,
it complements them.  Results found with models should be checked graphically
and results found graphically should be checked analytically.  Your comment
about data dredging is important, though why people think this only happens with
graphics and not with modelling approaches always puzzles me!
(7) There are often interesting features of a dataset (not just errors and
outlier groups) that can be found graphically that would be difficult or
impossible to find analytically.

Have a look at Interactive Graphics for Data Analysis: Principles and Examples
by Martin Theus and Simon Urbanek (Chapman & Hall).  There are some
excellent explanations and case studies there.

I could go on (and on), but what you really need is a good demo.

Best regards

Antony

PS Have you reported the bugs in GGobi and Mondrian you have found to the
software authors?

Antony Unwin
Professor of Computer-Oriented Statistics and Data Analysis,
Mathematics Institute,
University of Augsburg, 
86135 Augsburg, Germany

Hello Antony,
Thank you very much for your detailed answer!
All of your points are valid and interesting to reflect upon.

Regarding your note number (6), it's a very good point - I didn't think
of
it.
It might be argued that because interactive data visualization can be faster
then analytical programming, it might invite more "visual hypothesis
testing" without "counting your hypothesis" were as when doing it
with
printed output - it's much clearer what the number of your hypothesis tests
were.

Regarding your note number (7) - I'd be happy for (non spatial) examples for
interesting patterns found by interactive methods.

Regarding the bug reports.
So far I've filed one for ggobi:
http://code.google.com/p/ggobi-documentation/issues/detail?id=37
Although since the last time the software was built was 2008, I'm not sure
if anyone is even going to respond to the ticket...

Cheers,
Tal



----------------Contact
Details:-------------------------------------------------------
Contact me: Tal.Galili@gmail.com |  972-52-7275845
Read me: www.talgalili.com (Hebrew) | www.biostatistics.co.il (Hebrew) |
www.r-statistics.com (English)
----------------------------------------------------------------------------------------------




On Fri, Feb 11, 2011 at 3:55 PM, Antony Unwin
<unwin@math.uni-augsburg.de>wrote:
> Hello Tal,
>
> You asked *When is it helpful to use interactive plots? Either for data
> exploration (for ourselves) and data presentation (for a
"client")?*
>
> My answer: It's helpful for checking data quality, for exploration with
and
> without "clients", for checking results, and for data presenting.
>
> Notes:
> (1) It's difficult to explain interactive data visualization in print,
> demonstrations are so much more effective.
> (2) Interactive data visualization is fun, both for the analyst, and more
> important, for the dataset owners.  You not only get better interaction
with
> the data, you get better interaction with the scientists you cooperate
with.
>  They are prepared to contribute, because they can understand what is going
> on.  That is not always the case with statistical models.
> (3) The key is not "animation" but "direct
manipulation".  The aim is to be
> able to directly interact with all statistical objects in a graphic:
> querying, linking, reordering, reformatting, zooming, whatever.
> (4) You write of point-based graphics, what about area-based graphics like
> histograms, barcharts and mosaicplots?  For categorical data the ability to
> select groups and look at spineplots of other variables to compare
> proportions is very effective. (And don't forget linking to maps for
spatial
> data.)
> (5) You mention outliers.  How do you decide what is an outlier?
>  Interactive parallel coordinate plots are extremely useful, either for
> identifying outliers or for checking ones found with an analytic approach.
> (6) Interactive data visualization is not in competition with other
> approaches, it complements them.  Results found with models should be
> checked graphically and results found graphically should be checked
> analytically.  Your comment about data dredging is important, though why
> people think this only happens with graphics and not with modelling
> approaches always puzzles me!
> (7) There are often interesting features of a dataset (not just errors and
> outlier groups) that can be found graphically that would be difficult or
> impossible to find analytically.
>
> Have a look at Interactive Graphics for Data Analysis: Principles and
> Examples by Martin Theus and Simon Urbanek (Chapman & Hall).  There are
some
> excellent explanations and case studies there.
>
> I could go on (and on), but what you really need is a good demo.
>
> Best regards
>
> Antony
>
> PS Have you reported the bugs in GGobi and Mondrian you have found to the
> software authors?
>
> Antony Unwin
> Professor of Computer-Oriented Statistics and Data Analysis,
> Mathematics Institute,
> University of Augsburg,
> 86135 Augsburg, Germany
>
>
>
>
	[[alternative HTML version deleted]]