[bane]

Something to keep in mind with Tufte's ideas is that, while he has many wonderful ways of thinking about visualizing data, often it's hard to generalize his concepts into a simple set of general purpose visualizations, suitable for use on a variety of problems. Instead, you find yourself seeking to find a way to display data in as "Tufte" a way as possible, which more of than not means a discrete visualization for every kind of data you have.

This becomes a problem in data visualization software, where the number of data visualizations is far less than the number of possible datasets of interest. So many datasets will tend to map to very few visualizations: bar chart, scatter chart, both with a time series, relatively uncomplex graphs of low connectivity density, etc.

You can see Tufte has this problem himself, with his singular "invention" the sparkline really just being a form of a tiny time series line graph. Which is cool, but it's not particularly earth shattering.

My takeaway has always been to be mindful of Tufte's analysis, but don't be slavish to it unless you can produce a unique visualization for each kind of data.

[andrewcooke]

thanks. that helps explain why many people i know who are vocal fans of his work produce such poor interfaces - there is nothing simple for them to learn from in a constructive way.

[forkandwait]

For a more stats oriented take on data analysis (and more pragmatically useful, I think), check out Bill Cleveland:

http://www.stat.purdue.edu/~wsc/

Note, yet ANOTHER brilliant guy to come out of Bell Labs.