Browse beats search/combine

I'm going to come right out and say it: I think browsing is fundamentally better than search/combine (where a user combines terms from drop-down boxes to perform a search) as a way of locating information in a faceted classification system.

As Peter Morville points out in The Speed of Information Architecture, faceted systems offer a huge number of possible combinations with a relatively small set of vocabulary terms (note 1). However, this means that a faceted information space also has a potentially huge number of "empty addresses"--facet combinations for which no resource exists.

An example: Epinions used a search/combine system up until quite recently. They had 382 products in their DVD player category on April 1. They also had four (visible) facets with 59 vocabulary terms, combining for a total of 1504 possible addresses. This means that just 25% of the DVD player information space was populated. Assuming all facet terms are of equal value (note 2), someone using Epinion's search/combine system had a 75% chance of dialing up an empty address. I think Jakob would call that a search catastrophe.

Browsable systems, like the restaurants section of CitySearch, Epicurious, and Wine.com are superior because they expose their facets, but conceal the empty addresses. This completely eliminates one possible source of user error. (However, Wine.com's "Power Search" feature suffers from the same problem as Epinions.)

Until recently, Epinions used a search/combine system, which exposes all the empty addresses in the faceted information space. On the other hand, this is a good example of contingency design. (full screen capture)

Citysearch and others use browsing navigation. This lets them conceal unpopulated addresses.(full screen capture)

So I guess I might propose a couple of guidelines about using search/combine:

1. If you've got lots of content/resources/products--say 75% of all addresses are populated--test it on users and see what they think.
2. If you do use search/combine, help your users recover from a failed search by delivering at least some results (like in the Epinions example).
3. Search should be search. Don't use search/combine as a replacement for search.

1. In a simple faceted system, the number of possible addresses (F) comes from multiplying the number of vocabulary terms (x) for each facet (f1...fn), or: F = x_f1 * x_f2 * ... * x_fn.

2. In reality this isn't a great assumption, since people tend to look for some brands and features more than others. Still, since you can conceal the empty addresses through browsing, why bother using search/combine unless your information space is extremely well populated?