Avg. Time per Visit – Standard definition

The Web Analytics Association (Jason Burby and co.) recently releases 26 Standard Definitions to Promote Consistency among us Vendors (PDF) – this is GREAT news and should make it much more straightforward to communicate web metrics among not only traditional report consumers, but also more experienced analysts! However we are not there yet …and please see this post as an inspiration to continue the splendid work. One of the standard definitions are “Visit Duration” – specifically described as follows:

WAA Standard definition: Visit Duration

Definition/Calculation
The length of time in a session. Calculation is typically the timestamp of the last activity in the session minus the timestamp of the first activity of the session.

Comments
When there is only one piece of activity in a session (a single-page visit or singleevent visit), no visit duration is typically reported.

This is a fine definition and one that I certainly agree on. BUT then comes questions on the subsequent metric that we could call “Visit Duration Average” – first question is what to name the metric and second question is on how to calculate the metric. Before concluding anything. Let us take a look at Google Analytics and IndexTools (because that is funny enough what I use here on my blog – still waiting for those Omniture blokes to call me ;-)

- Avg. Time per Visit 4m 46s (IndexTools)
- 00:07:09 Avg. Time on Site (Google Analytics)
Two different naming conventions and two different calculations (assuming we both collect and accurately calculate the metric). *send me your definition from Omniture, Visual Sciences or other decent Web Analytics vendor and we might be able to help out our good friend Jason. I searched through my WebTrends Marketing Lab account and could not find anything. Anybody?

So first (and I am completely biased here) – I do not really think the Google Analytics “Avg. Time on Site” naming convention is clear enough confirming that we talk about a Visit (session).

But more importantly – WHY do the numbers differ?

Google Analytics discounts Bounces (or Single Page View Visits as the new WAA standard definitions calls it) and IndexTools includes ALL visits into the calculation. So to be clear, if we have:

1000 visits
300 bounces
2300 minutes spent on site

Remember that page tagging solutions like IndexTools and Google Analytics do not count time spent on site for bounces (Single Page View Visits). Thus in the above example the 2300 minutes are spent on 1000-300 = 700 visits. Even though we know for a fact that those 300 bounces (Single Page View Visits) in fact did spend time on the site.

  1. IndexTools = 2300 minutes / 1000 visits = A Visit Duration Average on 2.30 minute. (or Avg. Time per Visit 2m 30s as we would write it)
  2. Google Analytics = 2300 minutes / 700 visits = A Visit Duration Average on 3.29 minute. (or 00:03:29 Avg. Time on Site as my friends at Google write it)
So NOW the questions goes, who is right and who is wrong? :-)

In point of fact, I think both calculations are justified and I see a reason to include both of them. It might be a bit clumsy to start pre-segmenting visits (visits with MORE than 1 page view) as our friends from Google Analytics have done (sorry Avinash and Brian). Looking at a calculation where we discount bounces, we are actually just saying that we want to look at the, to some extent, more engaged users (those who decided to dig further into the site). I therefore prefer the original calculation (as Google Analytics used to have it as well) where we look at ALL visits. Then one can always segment to separate more or less engaged users from the others, whether that is page views or not.

As an example. Let us look at May 2007 (I can of course recreate these results with on the fly segments):

1 page views or more segment = 04.11 (the complete data set)
2 page views or more segment = 10.54
3 page views or more segment = 17.38
4 page views or more segment = 22.30

As you can see the more page views people look at the more time they spend on the site. And all of a sudden starting on 2 page views or more seems awkward.
Conclusion (this is for you guys at the WAA):
If you add another definition to your standard set about visit duration, I suggest you call it: “Visit Duration Average” and the calculation is the total length of time spent in all visits (sessions) divided by the total number of visits (session). Hereafter one can segment on page views or more intelligent engagement metrics.

Further fine comments on the subject:

  • Clint

    Interesting. So if Index Tools includes all visits (regardless of bounce status) in the calculation of Avg. Time per Visit and does not include time spent by bouncers, aren’t you essentially diluting the avg time metric? In other words, you’re making the assumption that bounces also spend 0 seconds on the site. I don’t know if that’s true…I suspect not.
    The advantage of the Google method (and others I am sure) is that it is working with known (as Ian Houston might say) values whereas the Index Tools calculation makes an assumption, which may or may not be true.

    The problem with the way that the GA calculation works is (I think, someone tell me if I am wrong) that what it is actually calculating is the minimum known avg time. So we know that, on average, visitors spent a minimum of x minutes on the site.

  • Claire Stokoe

    I was aware that the figures had altered due to the drop of bounce rates from the calculation, but seeing the actual calculation alongside the Index Tools sum, I can now appreciate the need for a re-think on Google’s part.

    I do agree that the new avg time metric, makes a lot more sense, based more on that which is tangible rather than that which is probable. It will be whole lot easier explaining this to my clients, who am I’m sure would rather see a figure based on actual time on site as apposed to a (as Clint pointed out) a diluted figure mixing hypothetical average times into the equation.

    Thanks Dennis, fantastically informative post

  • Dennis R. Mortensen

    Hi Clint,

    Thank you very much for your comment. Much appreciated.

    I think my “problem” lies within

    A) the combined definition and calculation (this including my IndexTools) and
    B) pre-segmenting (as in removing 1 page view visits)

    >>Interesting. So if Index Tools includes all visits (regardless of bounce status) in the calculation of Avg. Time per Visit and does not include time spent by bouncers, aren’t you essentially diluting the avg time metric?

    YES! And I think we should; we being the WA industry and analysts in general. Simply because of the fair assumption that the more pages people visit the more engaged (as in time spent on site) they are. Have a look at my quick segment from 1 page or more to 4 pages or more per visit – the time spent increases dramatically. And I think it is fair to included 1 page view visits at 0 seconds to dilute the metric when calling it “Avg. Time per Visit”.

    Furthermore, it leaves me with the full dataset and if I want to exclude bounces I can segment them away later. I cannot ADD them in e.g. Google Analytics.

    >>The problem with the way that the GA calculation works is (I think, someone tell me if I am wrong) that what it is actually calculating is the minimum known avg time

    I think it is difficult to put “minimum” and “average” in the same sentence :-) however; I really like your definition of “KNOWN” – that makes a lot of sense to me. So with your input we are now talking about:

    - Visit Duration Average (not segmented)
    - Known Visit Duration Average (segmenting to exclude bounces)

    Great reply Clint.. Thanks.

    Dennis

  • Dennis R. Mortensen

    Hi Claire,

    Thanks for the comment and heads up!

    Cheers

  • angie

    Hi Dennis, Angie here from the Standards Committee. Thank you for such a detailed post: outlining issues like this so specifically will help us to make better definitions in the future, and you definitely have made points worth considering.

    I encourage you (and your readers) to send feedback like this directly to “standards at webanalyticsassociation.org” to make sure it gets visibility. We’re trying to keep up with what’s out there in the blogosphere but I know we won’t find everything.

    Thanks again,
    Angie

  • Dennis R. Mortensen

    Hi Angie,

    >>I encourage you (and your readers) to send feedback like this directly to “standards at webanalyticsassociation.org”

    Will do! :-)

    By the way, are you attending Emetrics Washington? – as I have a set of questions/inputs from a Vendor point of view..

    Cheers
    Dennis

  • Pingback: Google Analytics Changes How They Calculate Time On Site (Again)