AnalyticsTowards a True Bounce Rate

Towards a True Bounce Rate

A recent study by Covario's Marcos Richardson has found that the current way analytics software vendors are measuring bounce rate is potentially misleading. Richardson has lobbied the industry to accept his proposal for a new calculation.

The definition of ‘Bounce rate‘ is the total number of single page visits divided by the total number of visits but analytics packages may calculate it slightly differently. However what has been uncovered by Covario and is currently in discussion with industry leaders is the addition and inclusion of a variable page load speed to more strongly differentiate bounce rate from, what are known in the industry as, ‘short clicks’.

A recent study by Covario‘s Marcos Richardson has found that actually the current way analytics software vendors are measuring bounce rate potentially has an inaccuracy at the heart of it. Richardson has lobbied key players in the industry to concede there is potentially an issue and accept the proposal for his new calculation.

By all accounts, his project has been successful and Google were the first web analytics provider to offer a new “adjusted bounce rate” metric in July 2012. Richardson spoke to Search Engine Watch (SEW) and gave us the full story on his project.

Gateway Metrics

In the nerdiest circles of web analytics (of which I count myself), there is a concept of gateway metrics – namely metrics that, if altered, have a knock on effect on tier two metrics. Instinctively, if you are experienced in analytics, you will know what they are: page impressions, referral sources and visitors.

However one of the most underused and most recently updated gateway metric is bounce rate. Bounce rate is the metric designed to measure non-interactions, the zero engagements, the users who came to your site and did nothing at all or the pages on your site that delivered little value.

That said, bounce rate is not able to measure non-interaction in definitive terms because the current calculation defines bounce rate more in terms of the absence of any other factor, as opposed to the presence of any specific properties.

Perhaps, by being able to describe almost nothing, this is why bounce rate remains the most misunderstood gateway web analytics metric.

Furthermore, bounce rate can actually be wrong. Whilst the conventional wisdom is that a user who “bounces” did nothing on your site, it’s quite possible that certain types of site architecture can generate a conversion, without registering another page load.

In such cases a user would register as a single page visit, when they have actually taken an action which forwards your business goals. For example, just reading this story on SEW could be counted as a conversion because users reading our stories is the most basic and fundamental success metric of a publishing business.

So, the implicit assumption that ‘doing nothing’ is bad, is complicated by the fact that doing nothing could just as easily be taken as an indicator of customer satisfaction.

It is not hard to imagine a scenario where “no further action is required” by the user as they found exactly the information they were looking for. In Google’s own post on adjusted bounce rate, Alexey Petrov, from the Google Analytics Insights Team does a good job of explaining when a bounce isn’t always a bounce:

While working perfect for most websites, there are categories of sites where this metric is not enough.

Imagine you’re promoting a blog post that describes all the benefits of your company. The visitor might read the whole post and remember your company and products really well – they might even go to search for your product on one of the search engines straight away. However, since the visitor only looked at 1 page (exactly where the blog post is) they will be recorded as bounced visitor.

Another example if you have a description of the product right on the landing page, and your phone number on the same page. The visitor might study the description and call straight away – again, they will be recorded as a bounced visitor, as only 1 page was viewed. There are many more examples, and even traditional websites may benefit from the method described below as opposed to the standard bounce rate.

Therefore, with all of the above in mind, equally (and confusingly), bounce rate does seem to measure some kind of engagement.

It compares the number of single page visits to your site, direct or from any source, to visits that led to a user clicking another page on the site. Therefore, at worst it registers a page load of your site, which at best reflects some kind of engagement with your brand from other sources.

But What if the Page Didn’t Load??

Did you catch what smidgen of data a bounce rate could be measuring? Yes, a page load. In the absence of all other factors, a 90% bounce rate still means that 90% of your users loaded a single page of your website. Huzzah!

But what if the page didn’t load? Then, arguably, it doesn’t really count as a true single page visit.

And if single page visits are counted even when a page does not load, then the current calculation does not signify the true bounce rate. At best, the current bounce rate would in effect only be comparing a server ‘request’ for your webpages versus users who visited another page on your site.

If this is the case then it would seem that, unwittingly, we are not comparing apples with apples as we first thought.

That is the question Richardson put to the web analytics community. The possibility that the page may not have loaded is the potential inaccuracy at the heart of the current bounce rate calculation.

Richardson details a new calculation and custom report filter for Google Analytics in his white paper that will enable marketers to look historically through their data and get a sense of what the true bounce rate was. Similarly, that the page did actually load in one of the single page visits is, in essence, the solution Google has provided in their adjusted bounce rate Google Analytics script to reflect a true bounce rate going forward.

Short Click Vs Long Click

In a nutshell, the current calculation of bounce rate hasn’t differentiated between what the web analytics industry calls the “short click versus long click” problem. Short-clicks are when a user immediately hits the back browser when they land on a page, potentially before it has fully loaded.

Although it may be common sense that those robots/crawlers that can trigger a page visit should definitely be eliminated from analytics data, when humans display similar behavior the question of what ‘really counts’ arises again.

With human browsers, it is arguable whether a user who didn’t even manage to load a page should be counted as a bounce. Equally, in terms of the long click, sometimes it takes a user more than 10 seconds to get a sense of the contents of a page before they click that back button. Both of these behaviors are weighted euqally in the current calculation.

Similarly, some visits will take users to another site element, such as watching an embedded video or clicking through to a Facebook page via the widget, and these can register as a bounce. Also many types of ad codes can register as a bounce. Similarly, websites which place analytics code at the top of the website could be registering a visit before the page has fully loaded.

The inherent difficulties in calculating bounce rate mean that it remains an underused metric.

However, as it is a gateway metric, it affects other downstream metrics which we tend to naturally have more confidence in. Return visitors, page views per user /and multi channel funnels are all affected. In the latter case, how goal attribution works from first to final click can affect the accuracy of your multi channel funnel. Time on site metrics are also affected.

The Benefits of Adjusting Your Bounce Rate Calculation

If you are spending money on paid advertising, whether that be paid search clicks, social media clicks or even measuring the clicks from your display ads then your tier two metrics will change and you may see improvements in engagement metrics from those sources by applying the adjusted bounce rate calculation. Average time on site and page views per user should increase.

Equally, you could find that you are getting more short-clicks from a source than you anticipated. And this can be from any source – paid search, SEO, local listings and display ads.

Before you hit the panic button, bear in mind that the responsibility as to the quality of the clicks from paid search is essentially shared amongst both webmasters and advertisers. Furthermore this concept of shared accountability is built into most paid search platforms via the “quality score” mechanic found in AdWords and Bing Ads – that calculation is data driven indicator as to how both parties are ensuring click quality and doing their upmost to serve the interests of the consumer.

Earlier this year, Jonathan Alferness, Director or Product Management at Google explained to me why the concept of quality score is so fundamental to Google’s relationship with it’s advertisers:

“While it may have seemed like a black box for ad pricing Alferness told SEW it is an important game mechanic within AdWords to enable Google to have a permanent data-centric dialog with advertisers on how to improve their campaigns and create a better experience for potential customers. Quality Score helps to maintain the integrity of every single keyword level auction and mediate the interests of all the parties – namely, Advertisers, Google and Users. If Google can help temper advertiser ambitions to focus on a better user experience (on not just Google but their own website too), then this win for the user should ultimately translate into more profitable campaigns.”

In seeking a second opinion, I asked Mike Grehan, Publisher of SEW and ClickZ and currently President of SEMPO, what the adjusted bounce rate findings could mean for the industry.

Grehan pointed out that not only does the responsibility of generating a page load lie with the host and the webmaster (rather than the deliverer of traffic), but also that the myriad of devices and access points inherently compromise page load speeds. He said, “Safari on an iPad in an airplane Vs Firefox on a T1 or faster connection at an ISP Vs IE inside Microsoft headquarters in Seattle, Vs Chrome inside Google. Do we all have a device and utility with the same responsiveness and measurement metric? I don’t think so.”

In fact, Google’s own page speed service, which aims to close the loop on such delays in order to drive a higher quality relationship between the consumer and the advertiser, is based on their own protocol called SPDY.

SPDY seeks to reduce the web page load latency of HTTP by reducing the number of connections required to deliver content. By contrast, HTTP creates so many variables which affect and delay page load speeds, that even with the adjusted bounce rate calculation, we can still only reach an approximation of your website’s true performance for the consumer.

As Grehan explained with characteristic candor, “There can be no such thing as an ‘industry standard bounce rate’ in an industry built on a protocol designed to suit a native environment. It’s about as accurate as American Indians doing Analytics on smoke signals and looking for a standard… It all depends on which way the wind is blowing.”

As the true concept of a quality click has to be a shared responsibility, although the adjusted calculation can tell the story slightly more eloquently; it can still only tell half the story.

To Bounce or Not to Bounce?

Bounce rate is an intrinsically rhetorical metric. Whatever number you tell people says more about how much you care about it than what it really could mean.

Your bounce rate can be as much about how you define your macro and micro conversions (put simply, those large and small actions your customer takes which further the goals of your website or business) and to some degree you can very easily force a bounce rate improvements simply by adding more custom ‘event’ tracking in your web analytics interface.

And, to be clear, you are not really forcing or faking anything. By adding or using ‘event tracking’ to illustrate a micro conversion you are arguably adding a layer of business intelligence that will ultimately help you understand your user better.

Nonetheless, when you see the dramatic improvement in your site wide average bounce rate, you will certainly feel like you faked an improvement.

To illustrate this with a personal example, at SEW, we actually define a micro conversion as a user reading a story – which is pretty much the most fundamental and implicit action you can take on the site. Yet, as a site that is more or less completely made up of stories, one could argue that it doesn’t make sense to turn effectively turn everything into a conversion. However, the intelligence yielded from turning every page load into a conversion means that SEW can actually gauge and breakdown broad interests by source.

SEW category interest by social source

At the very least, even without “humanly possible” page load data factored in, the chart above gives a useful reading of how different social sources drive skew towards different interest categories. We’ll keep you updated as to whether there is a significant change as a result of the new bounce rate calculation.


Whilst the connection seems obvious, there is increasing evidence that bounce rate is one of the strongest indicators as to the likelihood of a conversion taking place on your site.

However, it is still too crude a measure which is why it remains misunderstood. In it’s current form, at best it tells you someone tried to engage with your site, but there are so many variables at play that it says almost nothing useful until you take other factors into account and segment your data. Avinash Kaushik better describes this idea as non-flirts vs potential lovers.

This latest re-calculation of the bounce rate metric does add a much needed ‘intention’ benchmark to turn it into more of a signal for both search engines and webmasters.

All parties involved in the research and discussion should be commended for their investigation into this area and uncovering this blind spot – especially Covario who spearheaded the research and Google who have been the quickest to adopt the calculation.

It is also worth noting that the debate on how best to easier bounce rate still continues in web analytics circles. In discovering the limits of what bounce rate can truly tell us, we may also have just stumbled upon the limits of the HTTP protocol. We might need to invest in more real time analytics solutions – perhaps XMPP or RTP hold the solution? Honestly, I don’t know but we definitely have to explore the possibility idea of more strongly defining metrics in terms of absence of other factors and presence of specific properties.

In my opinion, whilst we may be closer, in order to get to a true bounce rate web analytics systems may have to adapt another level and into a two way data stream. If webmasters really want a ‘high definition’ picture of the customer user journey, we are going to need to send data back to the service providers we work with, much like the social Data Hub partners in Google Analytics. Whilst companies are building on-site black boxes to glean data from or disrupt search engine black boxes, arguably the technology we need goes beyond this.

What seems clear to me is that for bounce rate to be a more useful metric – and thus a stronger predictor of conversion – it has to be a two way provision. To get a true picture of bounce rates the entire industry needs to move towards something we can truly call real-time data sharing.


The 2023 B2B Superpowers Index
whitepaper | Analytics

The 2023 B2B Superpowers Index

Data Analytics in Marketing
whitepaper | Analytics

Data Analytics in Marketing

The Third-Party Data Deprecation Playbook
whitepaper | Digital Marketing

The Third-Party Data Deprecation Playbook

Utilizing Email To Stop Fraud-eCommerce Client Fraud Case Study
whitepaper | Digital Marketing

Utilizing Email To Stop Fraud-eCommerce Client Fraud Case Study