The first was to say that a stepped polar line could have mitigated the weakness of polar line graphs, by making it clear that the line actually was falling then rising (instead of giving the appearance of just taking a short cut across the circle. But I then said that this only worked because Jon’s data points were hourly, and the real numbers were a continuum: the higher the resolution Jon chose (hits per minute, hits per second), the richer the cartesian line graph would look, but the polar graph would get weaker, even in its stepped form.

My second comment was that the cycle plot had a problem with aspect ratio: as it is, the daily trends are too spiky to be discerned, but if the graph were expanded horizontally until the daily trend was rich and clear, then the hourly trend would be too flat (that’s a problem with simple time series too– there is often no one aspect ratio that works on every scale).

I put it better the first time, but c’est la vie electronique.

Sorry about the colors. They look distinct on my main monitor so I didn’t spend a lot of time with them, but on my secondary monitor the colors are not so distinct. Probably your screen is set up more like my secondary screen.

I attempted the monthly totals to average out the day-of-the-week fluctuations, but as mentioned, my database was incomplete. Usually I track a 7-day moving average to minimize the daily effect. I know that’s not really rigorous, but for Q&D analysis, it suits my needs.

The line charts are much easier for me to understand and I like your correlation to the workday. Makes sense. The other conclusion I would draw, since the linear patter is similar in both charts, is that most visitors are hitting the same (one or two) pages. And that would make sense for a blog-oriented site; most folks are likely just hitting the most recent updates.

One minor nit…at first I was drawn into the charts wondering why Thursday traffic would be so much lower than other weekdays. Then I realized that Sunday and Thursday colors are similar and I was confused; even more so because I was lazy and the Thursday entry in the legend is awfully close to the endpoint for Sunday’s data. Anyway, colors a little more distinct would be helpful.

Jon- Your data is cyclical with weekly cycles (Sat. and Sun. so much lower than weekdays.) Cleveland introduced cycle plots for time series data that had the cyclical component removed. For more information I’d look at Visualizing Data or one of these papers:

Cleveland, William S. and Irma J. Terpenning. 1982. “Graphical Methods for Seasonal Adjustment.” Journal of the American Statistical Association, 77:52-62.

Cleveland, R.B., W. S. Cleveland, J.E McRae, and I. Terpenning. 1990. “STL: A Seasonal-Trend Decomposition Procedure Based on Loess.” Journal of Official Statistics, 6:3-73.

Here are cycle plots for a week last month. The day-by-day variation is hard to read clearly.

I also plotted monthly totals by hour of day, but these were distorted because I started in the middle of August, and the service has missed some weekly reports (one in September, two in December, and one in March). It’s a free service and you get what you pay for.