Calculating time on a site is a thorny issue in web analytics. Let's hope Google's documentation and support can raise standards.
In my last article, I talked about Google Analytics and the wide range of features it contained. With such an extensive product, the idiosyncrasies of individual features will take time to emerge.
Web analytics standards are in a primitive state and leave a great deal of latitude to the designer of a system in terms of how they process some events. This directly impacts the numbers produced by the system, and Google Analytics is no exception. It is important to remember that all web analysis reports, from any system, are approximations of what really happened. To create an image in your mind of what your visitors are doing, you have to understand how the web analytics system you use interprets things.
Recently, I encountered two issues when using Google Analytics that prompted me to question its support staff. I needed to understand better how Google Analytics processed information in order to gain a clear picture of what was happening on my client's sites. If you use Google Analytics, you may find this information useful.
Bouncing the average time on site
The amount of time people typically spend reading your site is a very important item of information. This number sits at the center of a range of assessments of visitor behavior. Understanding how long people spend on your site and comparing it with how long you would like them to spend is a critical and recurring part of web analysis.
Calculating time on a site is a thorny issue in web analytics. To work out how long someone spent on a site, we add up all the times they spent on each individual page. In modern web analytics, we do this very badly. We could simply record when someone started looking at a page, when they stopped looking at the page and work out how long there was in between. That would tell us how long they spent on the page. Do this for all their pages, and we have the total duration of their visit. We could do this, but we don't. What we do is simply record the time they first started with a page, then record when they went to the next page. We calculate the time on the first page based on when they went to the second. In other words, we only record the time spent on one page when someone goes to the next. This means we never count the time spent on the last page, which means we under-record the duration of every visit by one page. It also means we record a duration of zero for people who bounce (only look at one page).
To a web analytics system, if you only view one page, you spend no time doing it. Whether you spent 20 seconds or 20 minutes, the time recorded is zero. This makes it impossible to analyze the behavior of bounced visitors.
Why do we do this? If we record when a page is opened, why don't we record when a page is closed? It is perfectly possible to do this with modern page-based tracking systems, including Google Analytics. We know we can, but we act like we can't. In fact, the international standard definition for a visit won't allow us to measure the closing of the page. This is because the standard was written when all web analysis was done via log analysis. Server logs can only record the opening of a page, they never know when one is closed. Even though the world has moved on, and no one uses server logs any more, and the standard is still stuck in that era. All the web analytic systems stick to the standard even though their technology has always been capable of going further.
So how do we calculate the Average Duration? Obviously, we add up all the durations and divide them by the number of visits. The question every designer of a web analytics system has to ask is: do we include bouncing visits (which have a duration of zero)? This is a very important question. If a high percentage of the visits bounced, the Average Duration will be dragged down substantially. This is a problem if you confuse "average' with "typical," as most people do. If a site has an Average Visit Duration of five minutes, most people will take that to mean that people who did not bounce spent around five minutes on the site. However, if the bounce rate was 50 percent, people spent 10 minutes on the site. It is thus critical to know whether calculations for Average Visit Duration included bouncing visitors.
There is no standard for calculating the Average Visit Duration. While there are standards for the individual visit, a system can legitimately include or exclude zero-duration bouncing visits because the standard is silent.
Google Analytics responds
I questioned Google Analytics on this issue and received the following answer from the Google AdWords Team:
"Previously, the Average Time on Site had been calculated as the total time on site for all visits divided by the total number of visits. Both the total time on site and total number of visits included bounces.
As of July 20, 2007, we began reporting the Average Time on Site as the total time on site for all visits (excluding bounces) divided by the total number of visits (excluding bounces). This change also affected data from earlier dates, not just newer data.
Because bounces have been removed from this calculation, you may notice a significant increase in the average amount of time spent on your site. Please be assured that we have in no way changed your data, only the way that the Average Time on Site is being calculated."
So, if you have reports filed away from before July 20 that contain Average Time on Site, you'll need to retrieve that information again. If you look now you will find the duration for that time period is now much higher than it was.
Next Page >>