u500k.erinye.com - the AOL data analyzed

Analyzing Click-Through Behaviour

Thelma's Factor

With 454 total queries and 239 unique queries, for each unique query recorded from user 4417749, there were roughly 1.9 total queries. Curiously, it turns out that this factor of 1.9 yields a very good estimate when determining the number of users with more than a given number of unique queries in the neighbourhood of user 4417749 (in terms of searches performed in three months), i.e. the number of users with more than a given number of total queries is very close to the number of users with more than that same number divided by 1.9 of unique queries.

factor 1.9

User 4417749's factor of 1.9 means that, on average, to search for any term, she spent 1.9 clicks. One of those clicks is the initial entry of the query, the remaining 0.9 are clicks on resulting links (including "next page"). Sadly, AOL users act much more erratic than statistics, so 1.9 is not a good estimate for how many clicks users spend searching for any term. The following figure shows that 2.3 would be a better estimate of this average.

factor 1.9

Note how AOL users that spend a lot of time online (as expressed by many recorded searches) act much more erratic than users with less online time. I guess this is not because the internet compromises your mind but because shared accounts would show up in that region, though. Also, users that spend very little time online seem to have less patience with AOL search.

Clickthroughs by Rank

It is conventional wisdom that the first result from a search gets the most clicks and that results on the second page are only rarely chosen. The following figure shows that this is true. Beware of the logarithmic scale - without it, results on the second and subsequent pages wouldn't even be visible. The first result gets more than an order of magnitude more clicks than the tenth.

Furthermore, the figure shows that last results on a page get more hits than those immediately before, which could be explained by laziness (looking at the last result first). However, the amount of clicks spent on last results is still much less than what is spent on the first few. The spikes for first results repeat, that is, making first place on the second page is still measurably better than second place. With the first and second pages being an exception, being listed first on a page seems to be better than being last on the preceding page. The first three "first result on a page" spikes are pointed out by the green arrows.

Another interesting conclusion is that users who will look at more than the first few results seem to prefer the last page of search results linked from each page over the others. This is pointed out by the blue arrows.

And finally, with the data covering ranks from 1 to 500, note how the last result is selected much more often than those preceding it. As pointed out by the magenta arrow, making last position in AOL search will get you more visitors than any other position below rank 250.

clicks by rank

For those curious enough, I've prepared a detail view covering the first 100 results. It is very evident that the second page gets less hits than the first by more than one order of magnitude.

clicks by rank, detail

Top Search Results

Thes are the top 20 web sites AOL users go to from searching. Note how two other search engines appear within the top three destinations.