Web analytics tools commonly have a Time Spent metric. Understanding how long people have spent reading a page is a really valuable thing for some businesses. For publishers, the quality of engagement with content is vital, given they’re effectively selling the attention of their readers.
What many people don’t realise is that the way this metric is calculated is critically flawed on the most popular web analytics tools, Adobe Analytics and Google Analytics. These tools calculate Time Spent by counting the time between pageview events.
In this example we see a user who:
So the actual Time Spent is 2 minutes 45 seconds. However, Google Analytics and Adobe Analytics will allocate zero seconds to the final pageview, because these tools only work from the difference in timestamps between pageview events and there is no pageview event following the final page. So Google Analytics and Adobe Analytics will record 1 minute 15 seconds for this session.
The pathological example of this problem is for a single page session:
In this example, the user enters the site, views the content for 30 seconds and then leaves. Traditional web analytics tools will record zero seconds against this session as there is only the single pageview event.
Many publishers now receive a huge amount of traffic in the form of single-page visits, primarily coming from aggregators and social networks. This means despite the fact your content may be receiving significant attention, your analytics will be showing very low Time Spent and a high bounce rate.
1 2 snowplow('enableActivityTracking',5,5); snowplow('trackPageView');
So once the page loads, a ping is sent every five seconds recording that the page is still open. That gives an accuracy of at least five seconds to calculating the actual time a user spent.
For tools that use the traditional mechanism of measuring Time Spent, there are some workarounds to get better numbers. Though none are ideal. The biggest problem is that there is no reliable mechanism to ensure a pixel is sent out when a user leaves your site.
Chartbeat uses a similar approach for data collection, although they attempt to measure actual engagement by monitoring user activity in the window as well. It would be interesting to apply this approach to the Snowplow page ping. Shouldn’t be too hard to update the tracker to support this.
Page pings create a large number of event rows inside your Snowplow database, consistent with the way Snowplow does things. That opens up a bunch of different ways for you to calculate the Time Spent metric. In the next blog, Mike will go through different approaches for modelling the page ping data to analyse Time Spent.
Until then you can check out Snowplow’s Measuring content page performance article.
Subscribe to this blog via RSS.