Sitemeter is broken – no Open Parachute stats.

I see that Martyn Bradbury finally noticed that I dropped the public accessible statistics about The Standard in April at Open Parachute.

He characteristically titled it with his usual humility as “Blog Rankings: TDB second largest blog in NZ”. I thought that maybe we should start a movement to label him as “New Zealand’s Trump”. But I think that one of the big swinging dickheads locked in a current defamation trial already has his head stapled to his scrotum using the much larger Whaleoil sleaze and clickbait blog.

The reason that The Standard dropped on Ken’s list was quite simple and one that Bomber should have spotted himself earlier if he was less interested in bombastic self-promotion and more interested in how his site operates. Over the previous 4 months, the readings from Statcounter for page views had gotten increasingly inaccurate. Since Sitemeter had already displayed unexplained wild variances in recent years and most of the other public site stats packages have significiant other flaws on highish volume sites, we stopped running any.

As well as Statcounter, this site runs Google Analytics, WordPress Stats, and most accurately a slow analysis using the Apache logs. I tend to use Google Analytics as the usual, reference as it always closest to my occasional checks running the exhaustive Apache log analysis.

You can see the problem off the graph and the table below. Page views on Statcounter is got way below Analytics in February and never recovered. Over the last few years, the maximum variation between Statcounter and Google Analytics page views has been within +/- 15,000 for the range of page views per month of 300,000 to 600,000. Normally it is +/- 10,000. Even when we hit election peaks of more than 800,000 it is typically less than +- 20,000. Roughly 5% at the worst case.

Those are acceptable variances for something that is as tricky as measuring page views using questionable javascript, client side browsers, and various network filters and firewalls. When it consistently kicks up over 15% over a number of months then you know that something is going wrong.

I noticed it in December and made time to look at the coding for problems. I couldn’t see any. I thought that there was a possible issue that would come right. January was less worse but on a smaller holiday volume. I started mentioning the issue to authors in the internal discussion posts.

In February there was a funny spike when it looked like someone at Statcounter dumped a pile of numbers into the page views in one hit – about an extra 30 thousand for us.

The Standard daily page views from Statcounter for February 2017

We weren’t alone. The same thing happened at the Daily Blog (an extra 22k) and a number of the other NZ sites that use Statcounter.

The Daily Blog daily page views from Statcounter for February 2017

March brought a lot of issues with slow page loads specifically related to statcounter. See the conversations here and here and here. I was somewhat busy that week at work (being a blogger and doing 60+ hour working weeks doesn’t always mesh), so I didn’t start tracking it until the weekend. But I got no useful responses from Statcounter.

Anyway, April was still bad. So when I had a break at work, I cut the connection to Statcounter and fixed the links in the mobile and desktop sites.

On a complete side issue, the other problem with the blog rankings is that it orders the ‘rankings’ on the wrong statistic. What Ken and Sitemeter calls ‘Visits’ is completely different to Sitemeter’s ‘Unique Visits’ and Analytics ‘Sessions’. Despite them trying to measure the same thing (how long a person is actually reading pages in a visit or session) and having very similar published parameters, they vary a lot. This shows up in the page views per visit or session.

Roughly speaking for our site in the same month,  the differences are stark. On our site and apparently on Kiwiblog, Sitemeter shows a consistent ratio of between 1.3 to 1.7 page views per ‘visit’ over a number of months. Statcounter shows 2.0 to 2.2 page views per unique visit over the same months. But Google Analytics shows 2.9 to 3.6 page views per session over the same months.

Bearing in mind that Google Analytics can show you pathways of each anonymous individual through the site visit including time stamps, then I’d say that google is correct. When I look at both Statcounter and Sitemeter logs, I have absolutely no idea where they cut off a ‘visit’. I suspect that is because they don’t post-process their logs.

But these are monumental differences, The effect of ordering on uncorrected ‘Visits’ from different stats packages is to make the ranking pretty useless. They don’t measure the same thing. Page views are a whole lot more similar between packages. However they can be affected by page design.

I don’t have the Sitemeter for the same period, but here are the visits and sessions for sitemeter and google analytics. Notice the big bump in February when Statcounter dropped that day’s extra visits in.