05 June 2017

Chess Stats Year-Over-Year

Last week I used my Monday post to finish with Korchnoi's Career 1946-1977 (Crosstables; 'it's time to move to another project'), and now I'm going to take a few weeks to look at statistics on my main site m-w.com (linked on the right). In the past few months I've noticed a big drop in the number of daily visitors and I would like to know why.

Since chess has a seasonal bias -- peak interest in the winter, trough in the summer -- the best way to analyze a trend is with a year-over-year comparison. Normally I would compare stats reports from the same month in two consecutive years, but this is infeasible right now. The stats reporting package on my server host has changed twice in the last two years, as documented in two posts:-

That second post explains the main use of stats.

I look at the overall stats once a day to make sure that everything is working correctly and to identify any anomalies as quickly as possible. After watching the stats for a while, I get used to certain recurring patterns in the numbers and know intuitively when something is off.

Since something is definitely off, I need to analyze the server log files from this year and last year. While I'm at it, I'll go back one more year to identify any bias introduced by the transfer to a new host server last year. I extracted my monthly archive log files for May 2015 and May 2016, then downloaded the daily files for May 2017 and concatenated them into a single month. Here are the first numbers I derived:-

2015-05: 374144 recs, 27504 HTML, 13.6 recs/page
2016-05: 434545 recs, 31724 HTML, 13.7
2017-05: 268142 recs, 23124 HTML, 11.6

Those numbers show an increase from to 2015 to 2016 on the old server, then a decrease from 2016 to 2017 on the transfer to the new server. The last number, 'recs/page', indicates that something might have changed in the server mechanism that delivers pages. I'll investigate those numbers in another post.

No comments: