A case for contextual data filters?

This afternoon, I had the pleasure of sitting in a session on information visualization by Ben Shneiderman of the University of Maryland Human-Computer Interaction Lab. In his presentation, Shneiderman shared one of his mantras when it comes to visualizing information:

Overview, zoom & filter, details on demand.

Essentially, Shneiderman was saying that when visualizing information, we should provide a large overview with the entire dataset visualized, and then let the user zoom and filter to explore the data.

On surface, this sounds like a smart approach, and I'm sure there's a lot of research to prove that this is actually the most intuitive way to visualize information, but as we put the final touches on one of our new data visualization tools here at the Bank, I'm curious if there aren't some cases where it's best to present a subset of information first instead of exposing the whole dataset.

An example: let's say I've got a data visualization application on my site — hypothetically, a tool that allows people to explore various environmental indicators across geography and time — that is pervasive across various parts of the site. On the homepage of that site, which deals with environmental issues on the whole, presenting an overview of all the data makes sense.

But what about when a user clicks in to the "fossil fuel" section of the site? Does it not make sense that the tool automatically shows fossil fuel-related indicators instead of the whole data set? Does a contextual filter — of course, with the ability for the user to cancel the filter and go back to the larger overview — not make sense in this case?

I'm just thinking out loud and could be off base. Let me know if you've got some additional insight on this.

(Photo of Pie Chart by net_efekt.)