I had such a busy week that I didn’t get a chance to reply to several interesting comments about the Paywall blog and the Customer Support blog. I also realized that I had somehow passed on my stated intention of writing a post about using econometrics to build predictive site models.
With respect to my post on analyzing Customer Support pages by analyzing the most common end-point pages for every search term, David asked whether it would be better to work from top searches to pages or from top pages to searches.
As he points out, either approach is practical but I definitely have a preference for the search to page method and I’ll explain why.
It might seem that the two methods would give identical results – but it doesn’t work quite that way.
The basic idea behind the analysis is that you can often identify the best (from the visitors point of view) answer to a given query by finding the page that was most likely to be an end-point when it was visited by users with that query. With a mart, you can create a single pass of the data that identifies this page for every single search term (though you won’t have statistically significant data on many low-volume search terms).
If you start by looking at top pages and figure out the searches they are most likely to be an end-point for, you aren’t getting quite the same thing. Doing it this way might give you these results:
Page Best Search Result
Page A Problem X
Page B Problem Y
Page C Problem Z
Page D Problem X
It’s perfectly possible for more than one page to have the same search term as its most likely end point. In the data above, both Page A and Page D perform better for Problem X than for Problem Y or Z. There’s nothing the least bit surprising about this situation.
The converse is not true.
For any given search term, there can only be one page (except in a numeric tie) which is the most common end-point. In other words, if I looked at Problem X search terms, either Page A or Page D is going to be the top end point.
Since the goal of the analysis is to optimize navigational access from search to answers, I think it's clearly better to go from Search to Page and not the other way around.