In the course of the recent “great database fiasco”:https://www.crookedtimber.org/archives/002042.html, I took a look at the history of traffic to this site. The AWStats program gave me a the number of unique visitors for every day from our launch last July through to June 16th this year. I was interested in which posts had made the biggest splashes. Now, if I just looked at the posts that got the greatest number of visitors, there would be a bias towards posts from later in the year, because we get far more visitors these days than six or ten months ago. How can we get a fair estimate?
It’s possible to statistically “decompose”:http://www.jos.nu/Articles/abstract.asp?article=613 a time series into three components. First, there’s the _seasonal_ component: in this case, it’s the regular ups and downs caused by what day of the week it is. Generally, traffic will dip every weekend, regardless of how many visitors we’re getting on average. The average number of visitors from week to week net of the seasonal ups and downs is the second, _trend_ component. This has grown consistently over the year. And finally there’s the _remainder_ or “irregular” component, which is whatever spikes and dips are left over once seasonal fluctuations and the underlying trend are accounted for.
The nice picture above shows CT’s unique daily visitor series decomposed in this way, with the raw data at the top and the three components underneath. (You can also get this figure as a “higher quality PDF file”:http://www.kieranhealy.org/files/misc/ct-decomposition.pdf [only 34k].) As you can see, the trend is one of healthy growth. These days we typically get about seven to nine thousand unique visitors a day. But what about those spikes in the lowest panel? Which posts brought in the crowds? *Read on* for the Top 10 list. The punchline is that, even though we’re known for being a bunch of “pointy-headed academics”:http://www.matthewyglesias.com/archives/week_2004_01_25.html#002424, the out-of-the-ballpark hitters on our roster are not the ones with Ph.Ds.