SEO

Tick Tock: The Limitations of Average Time on Page and Average Time on Site in Google Analytics [EXPERIMENT]

I’ve been told that I don’t have a very good poker face, and I’m ok with that.  I find it actually helps my clients, since they can tell almost immediately when I don’t agree with something or when a point needs more clarification.  For example, there are times I see marketers heavily focus on time-based metrics like Time on Page and Time on Site (as if those metrics were 100% accurate).  Unfortunately, this isn’t the case and the numbers reported can be far from accurate.

The confusion about how time metrics are calculated usually leads to some interesting conversations (as clients see various times listed in their reporting and want to start comparing pages and visitor segments.)  That’s hard to do, since there are serious limitations with the way time metrics are calculated in web analytics (for now anyway).  Even worse, people can subsequently make decisions based on the data.  This is why I tend to focus on key actions visitors can take versus other measurements of success.  I track these actions via conversion goals and events versus focusing heavily on time-based metrics.  I’m not saying that time-based metrics are useless, but they need to be taken with a grain of salt (at least until we find a more accurate way to track them).

For example, let’s say a marketing manager analyzed a new piece of content and was blown away with an Avg. Time on Page of 5:26.  This is where my (lack of a) poker face leads to a deeper conversation about how the actual Time on Page could be much different.  My comment is typically something like, “I think it’s great that visitors who decided to visit another page on your site spent 5:26 on the page.”  I often get a strange look at this point.  Then I start to explain that Time on Page is actually calculated by page jumps.  So, if someone reaches a certain page and exits from that page (or bounces), then Time on Page is 0:00.  No, that’s not a typo.  Time on Page is 0:00 whether they spent 10 seconds on the page or 10 minutes.  I’ll explain more about how time metrics are calculated below, but I wanted to give a quick example now so we’re on the same page.

time metrics stepping stones Tick Tock: The Limitations of Average Time on Page and Average Time on Site in Google Analytics [EXPERIMENT]

Note, although I’m focusing on Time on Page and Time on Site in this post, there are ways to calculate time metrics like Elapsed Time for certain processes that occur on a page.  For example, check out my tutorial for using TimeTracker to calculate elapsed time (or Time to Complete).  It’s an interesting way to extend event tracking to record how long it takes to finish a process like completing an elaborate form.

Access To Data Can Be Powerful But Interpreting Data Incorrectly Can Be Dangerously Powerful

I never want to rain on anyone’s parade, but it can be dangerous to walk around your office telling everyone something that might not be entirely accurate (and can be way off).  When I start helping clients with analytics, I typically make a point to grab a conference room with a white board and start explaining how Time on Page and Time on Site are calculated.  After I explain how the metrics are calculated, there are times that I hear a level of skepticism, since there isn’t hard evidence that the metrics are really calculated this way.  And I’m totally cool with that. I’d be a hypocrite to say that being skeptical is wrong, since I test everything.

In online marketing, I don’t necessarily take anyone’s word without testing it myself.  Which brings me to this post…  Based on what I explained above, I decided to run an experiment to test Time on Page and Time on Site in Google Analytics.  My goal was to document the hard numbers so you can see how the metrics are reported.  I think it’s important to understand the limitations of the metrics, so you can provide context in your analysis.  If you provide context, then your clients can better understand the analysis you provide.

The G-Squared Lab

When setting up this experiment, my goal was to isolate directories and pages and then document the exact time that I spent while visiting those pages (and sets of pages).   If each visit was isolated, then I could drill into the reporting to see exactly how Google Analytics tracked time for each page and each visit.  I’ll detail the paths I took in the experiment below.

Quick Introduction: The Formulas for Time on Page and Time on Site:

Before we hop into the experiment, let’s start with a quick introduction to how Time on Page and Time on Site are calculated.  Both metrics are based on page jumps (or moving from one page to the next).  By calculating the difference between visiting one page and the next page during the visit (via the page timestamp), Google Analytics can determine how long you stayed on the preceding page.  Calculating Time on Page via page jumps is where a major limitation is exposed.

That’s because if you visit a page and then exit the site, the Time on Page is 0:00.  As mentioned earlier, that’s whether you spent 10 seconds or 10 minutes on the page.  On a similar note, if you visit just one page on the site and leave (which is a bounce), the Time on Page is also 0:00.  Time on Site is calculated by adding the time that a visitor spent on the site by combining the Time on Page for each page that was part of the visit.  Average Time on Site simply divides total time by number of visits and does not exclude bounces.  You can see the formulas below.

Time on Page A = Page B Time – Page A Time (based on the timestamp)

Average Time on Page = Total Time on Page A/(Pageviews – Exits)

Total Time on Site (For a Specific Visit) = Page A + Page B  (or the sum of Time on Page for each page in the calculation)

Average Time on Site = Total Time on Site/Visits (this metric includes bounces)

Example:

As a simple example, let’s say that someone visited Page A for 2:00, then visited Page B for 3 minutes, and then exited the site.  The Time on Page for Page A would be 2:00.  However, the Time on Page for Page B would be 0:00.  Google Analytics can identify the Time on Page for Page A only because there was a jump to Page B (and it calculates the difference in time between the two pages).  Since the visitor exited the page from Page B, there is no way for Google Analytics to know how much time they spent on the page.  The Time on Site would also be 2:00, since the time on Page B cannot be determined.  Based on this simple example, can you tell why Time on Page and Time on Site need to be taken with a grain of salt?  What if that visitor spent 7:00 on Page B?  You would think by just looking at the reporting that visitors were more engaged with Page A, when in reality, they were spending more time on Page B!

The Experiment: 5 Scenarios Plus a Bonus

For the experiment, I set up several directories with four pages in each directory.  All the pages within each directory are linked together via simple text links.  For each visit, I identified how many pages I would visit and how long I would spend on each page (give or take a few seconds).  Everything was documented so I could cross reference my notes to see if the reporting matched up.  I will list each scenario below and show you the reporting in Google Analytics.  Then you can see for yourself how the metrics are calculated.

Scenario 1: Bounce After 20 Seconds

For the first visit, I wanted to simply bounce, but after a short visit.  I visited the first page, stayed for just 20 seconds, and then exited the site.  I wanted to show you how the Time on Page for a bounce is reported as 0:00.  After checking the reporting, it surely was.

time experiment1 Tick Tock: The Limitations of Average Time on Page and Average Time on Site in Google Analytics [EXPERIMENT]

Scenario 2: Bounce After 5 Minutes

I’ve heard some confusion about what determines a bounce in Google Analytics.  For example, some people think that if you spend enough time on a page, it’s not considered a bounce.  So, I spent five minutes during the next visit, but stayed on the landing page.  Once again, the time on page was 0:00 and it was considered a bounce.  I just wanted to clarify that point.

time experiment1a Tick Tock: The Limitations of Average Time on Page and Average Time on Site in Google Analytics [EXPERIMENT]

Scenario 3: Land on Page A, Spend 1:17, Click Through to Page B, Spend 1:30 and Then Exit

For the third visit, I wanted to visit the first page in the directory (Page A), wait 1:17 and then click through to Page B.  I would spend another 1:30 on Page B and then exit the site.  Based on what I explained earlier, the Time on Page for Page A should be 1:17 and the Time on Page for Page B should be 0:00.  In addition, the Time on Site for this visit should only be 1:17, even though I actually spent close to 3:00.  The reporting confirmed this.

time experiment2 Tick Tock: The Limitations of Average Time on Page and Average Time on Site in Google Analytics [EXPERIMENT]

time experiment tos 2 Tick Tock: The Limitations of Average Time on Page and Average Time on Site in Google Analytics [EXPERIMENT]

Scenario 4: Land on Page A, Spend 1:23, Click Through to Page B, Spend 0:58, Click Through to Page C, Then Exit

For the fourth visit, I wanted to visit Page A for 1:23, Page B for 0:58, then click through to Page C.  I would spend 1:00 on Page C and then exit.  The reporting should show a Time on Page of 1:23 for Page A, 0:58 for Page B, and then 0:00 for Page C (even though I spent over a minute on the final page).  Also, Time on Site should be 2:21, even though I spent 3:21 on the site.  The reporting confirms this.

time experiment3 Tick Tock: The Limitations of Average Time on Page and Average Time on Site in Google Analytics [EXPERIMENT]

time experiment tos 3 Tick Tock: The Limitations of Average Time on Page and Average Time on Site in Google Analytics [EXPERIMENT]

Scenario 5: Combining Visits to Show Average Time on Site


I mentioned above that Average Time on Site (ATOS) includes bounces, so I wanted to show you how this looks when you combine multiple visits.  So, I visited the same set of pages three times.  The durations for each visit were 0:00 (a bounce), 1:11, and 2:14.  Based on the calculation for Average Time on Site, the bounce should be included and the ATOS should be 1:08.  The reporting confirmed this.

time experiment atos Tick Tock: The Limitations of Average Time on Page and Average Time on Site in Google Analytics [EXPERIMENT]

Bonus: The Virtual Pageview

I mentioned earlier that if you exit from a page (including a bounce), then Time on Page is 0:00.  But that’s not always the case.  Welcome to web analytics.  icon smile Tick Tock: The Limitations of Average Time on Page and Average Time on Site in Google Analytics [EXPERIMENT]  There’s something called a virtual pageview that enables you to trigger a pageview, but in reality, a page wasn’t really loaded.  It’s a versatile feature in Google Analytics and can help you track clicks off your site, conversions that don’t require a page to load, and other clicks you want to track that don’t necessarily load pages.  You can also use a virtual pageview to track conversion goals (since you can use that virtual pageview as the destination URL for the conversion goal).  Again, you can read my blog post about conversion goals and events to learn more about this functionality.

The Virtual Pageview Reporting:

In this scenario, I visited one page and bounced.  However, I triggered a virtual pageview after staying for 1:28 (before I left the page).  So, does triggering a virtual pageview impact Time on Page?  It absolutely does (see the screenshot below) and it’s important to understand this when analyzing your reporting.  The bounce rate was 0% and Time on Page was 1:28.  If the virtual pageview was not triggered, the Time on Page would have been 0:00 and the bounce rate would have been 100%. Keep this in mind when analyzing your reporting.

time experiment5 Tick Tock: The Limitations of Average Time on Page and Average Time on Site in Google Analytics [EXPERIMENT]

Moving Forward With Time on Page and Time on Site

There you have it.  We’ve taken a detailed look at Time on Page and Time on Site in Google Analytics, while exposing some of the limitations with time-based metrics.  I hope the results from my experiment help you better understand how the metrics are actually calculated.  To be clear, I’m not saying to forget about Time on Page and Time on Site, but you just need to take the metrics with a grain of salt.  As you’ve seen in this post, the actual numbers may be off (and way off for certain situations).  This discrepancy makes it challenging to determine if the time reported was good or bad, which can definitely inhibit making solid decisions on the data.

Some key takeaways regarding Time on Page and Time on Site:

  1. Don’t obsess over Time on Page or Time on Site.  Unfortunately, the metrics are flawed and can skew your analysis.  Keep the limitations in mind while analyzing site performance.
  2. Time on Page does not include exits (or bounces), and can inaccurately report actual time on the page.  It can be much lower or much higher than reported…
  3. Time on Site does include bounces, but still cannot determine the actual length of time spent on exit pages.  Therefore, this number can also be way off.
  4. Virtual pageviews enable Google Analytics to calculate Time on Page, even for exits (from the time the virtual pageview is triggered).
  5. Try and focus on key actions that visitors can take on the site (in the form of conversion goals and events.)  Then you can use time metrics to help support your findings (if it makes sense for the site in question).

Now it’s your turn to help us accurately report the Time on Page for my post.  Please visit another page on Search Engine Journal so I know how long you actually spent reading my post! icon smile Tick Tock: The Limitations of Average Time on Page and Average Time on Site in Google Analytics [EXPERIMENT]

 Tick Tock: The Limitations of Average Time on Page and Average Time on Site in Google Analytics [EXPERIMENT]
Featured SEO Writer for SEJ Glenn Gabe is a digital marketing consultant at G-Squared Interactive and focuses heavily on SEO, SEM, Social Advertising, Social Media Marketing, and Web Analytics. Glenn has over 18 years of experience and has held leadership positions both in-house and at a global interactive agency. During his career, Glenn has helped clients across a wide range of industries including consumer packaged goods (CPG), ecommerce, startups, pharmaceutical, healthcare, military, education, non-profits, online auctions, real-estate, and publishing. You can follow Glenn on Google+ here.

Comments are closed.

24 thoughts on “Tick Tock: The Limitations of Average Time on Page and Average Time on Site in Google Analytics [EXPERIMENT]

  1. @Glenn i love the post, but it also highlights another interesting aspect of a complex metric “exit rate”. It is really an annoying metric as it is also tied to how well GA is reporting bounce rates and visitors are using your site.

    Scenario 3:
    1 visitor could be tracked exiting the site after 2 pages

    Scenario 4
    1 visitor could be tracked exiting the site after 3 pages

    Exit rates seem to be useful as part of a signup process or a page that is likely to get a user interaction such as a click.

    I know i've talked about using it on projects, but its not certain if it works as it should but using time based EventTracking loading which wouldn't inflate your page views…

    1. Thanks David. I’m glad you liked my post. I found the metrics were really confusing people. I hope the results of my experiment help clear up that confusion.

      Regarding exit rate, the numbers were calculated correctly above in my experiment screenshots. It shows 100% exit rate from the pages where I exited the site. I actually wrote a post about exit rate and bounce rate, to help clarify the difference between the two metrics. Maybe this will help: http://www.hmtweb.com/blog/2007/08/bounce-rate-and-exit-rate-what-is.html

      1. Ah yes but a did have a followup test that might be interesting, it shouldn't technically affect the tracking because it works on a cookie, but how does “open in new tab” or “open in new window” impact on results?

        I ask because it is fairly common for sites to have the TARGET=”_blank” code automatically applied to links. So would these split the visitor, count as an exit or would it even possibly record it as a bounce?

  2. Thanks for an excellent post Glen. The “time on site” and “time on page” information provided by Google Analytics has confused me in the past. It is very true that decisions based on incorrect information can be damaging. Your post clarifies the issue and will greatly assist me with my analytical interpretations going forward.

    1. Thanks Heinrich. I'm glad my post could clear up some of the confusion with Time on Page and Time on Site. I think the numbers are hard to prove until you isolate visits like I did in my experiment.

      Regarding making decisions based on incorrect data, context is critically important in analytics. That's why it's essential to have a solid understanding of how these metrics are calculated.

  3. interesting post Glen, but i want to be able to take something concrete away from this.

    would you say it could almost be a best practice to always enable virtual pageviews for websites that have no real event or conversion to track, in order to get the most accurate results when it comes to reporting acutal time on page and time on stie, correct?

    1. Great question. Seeing how virtual pageviews enable you to track actual time on page does provide a compelling reason to do this. I think it depends on the page at hand. If you are extremely interested in understanding time on page for a specific piece of content, then I would recommend adding a vpv. That said, if someone simply closes the browser window or enters a new URL, then you're out of luck. This is why I try and focus on key actions versus time-based metrics. I hope that helps.

  4. Thanks Gabe — Very detailed description of time on page / site. The scenarios and screen captures really validate your points. Excellent post!

    1. Thanks Mike. I'm glad the scenarios and screenshots helped. I think isolating visits like I did in the experiment definitely helps make this topic clearer. There are still too many people relying on TOP and TOS without a full understanding of how the metrics are calculated!

  5. Thank you Glenn :-) I really appreciate the the way you explained this 'time on page' etc. issue… I “got” it. Your explanations and examples clearly outline how this works — I'm glad I ran across your article.
    Andy :-)

    1. Hey, thanks Andy. I appreciate it. I'm glad my post could help. Understanding how analytics packages calculate time-based metrics definitely helps provide context when analyzing site performance. Without context, it's easy to think that the content with the highest Time on Page is the most engaging. And that's unfortunately not the case!

  6. That was excellent. It is always a pleasure to read something new – it doesn't happen every day in this filed. Also, I really appreciate the fact I can read the whole article on Google Reader and not just a simplre preview. That's a bit OT but still, many people really enjoy that feature.

    1. Excellent, thanks Erica. I'm glad my post was helpful. I definitely plan to write more posts about web analytics in the future, so stay tuned.

    1. Thanks Dennis. I wanted to make sure that everyone had a good understanding of the metrics and the calculations before showing the results from my experiment. I'm glad you made it to the end. :)

  7. Excellent, thanks Erica. I'm glad my post was helpful. I definitely plan to write more posts about web analytics in the future, so stay tuned.

  8. Thanks Dennis. I wanted to make sure that everyone had a good understanding of the metrics and the calculations before showing the results from my experiment. I'm glad you made it to the end. :)

  9. Ah yes but a did have a followup test that might be interesting, it shouldn't technically affect the tracking because it works on a cookie, but how does “open in new tab” or “open in new window” impact on results?

    I ask because it is fairly common for sites to have the TARGET=”_blank” code automatically applied to links. So would these split the visitor, count as an exit or would it even possibly record it as a bounce?

  10. Great article, this made it very clear why the times are 0:00 when tracking our embeded flash players on other sites – there is no Page B to compare the stamp to.

  11. This is a great post. I'm in the midst of putting together a web analytics reporting dash board as I was suspicious when my analytics report was detailing 100% bounce rates but not time on that page. I thought this cant be right. This post has cleared it up form me. Nice one Glenn :-)

  12. Excellent post, really enlightening. It explains all those 0:00 on my blog (if google analytics still calculates that way that is?) it was explained in an easily understandable manner and I only installed google analytics yesterday!