The title of my session is “Social Media Analytics: Monitor, Measure and Manage Your Reputation on the Wild Wild Web of Social Media.”  I will be sitting on a panel with the following people:

Kate Kaye, Clickz (Moderator)
Peter Corbett, iStrategyLabs
Stan Magniant, Linkfluence
Joe Mansour, David All Group

I am extremely excited about our potential discussion since we are all hands-on practitioners with different perspectives and experiences related to Social Media and politics.

My session runs from 2:00pm – 3:00pm on Tuesday, April 21st.  The conference runs from Monday, April 20 through the 21st in the Reagan Building – hopefully I will see you there!

Last week, Devaney was before Congress touting the demand for this information.  His testimony was picked up by several news outlets, including these stories at CNN, Federal Computer Week, and the L.A. Times.

In his testimony, he described traffic levels at Recovery.gov to be about “4,000 hits per second” (emphasis mine).  This means that in a day, the site would get about 350 million hits per day (4,000 * 60 seconds/minute * 60 minutes/hour * 24 hours per day).  To most people, this would imply a traffic level equivalent to every internet user in the United States visiting the site more than one time, every single day.  However, this statistic is misleading.

A “hit”, commony confused with a “page view” or a “visit”, is a single HTTP request from a web server to a browser for a specific resource.  HTML pages, images, flash objects, CSS sheets, external scripts, and other other files requested by a browser to render a page each generate a single hit upon their request.  The net effect is that single pages can generate several hits for the display of a single page to an individual user.

In the case of Recovery.gov, the home page loads 30 files to fully render.  If the 4,000 hits statistic is accurate, then there are really about 11.5 million page views per day.  Assuming that the average visitor views 5 pages on the site (which is a big assumption), the true number of visits to the site is more like 2.3 million per day, or about 27 new visits per second.

27 is a lot less impressive than 4,000, but it may be a much better picture of the true nature of traffic to the site.

I agree with Avinash in that Bounce Rate is one of the most important metrics for any type of site. A high bounce rate negatively affects Time Spent on Site in ways that may not be immediately apparent; have a look at my article on Life Tables and Time Spent on Site to see how. However, we need to take a closer look at our data collection methodology before we assign too much importance to this metric.

Bounce Rate Defined

The classic definition of a Bounce is a visitor that views a single page and then immediately leaves without viewing any other pages. A site’s Bounce Rate is the ratio of single page visits by the total number of visits. Anil Batra defines it (in his post “Bounce Rate Demystified“) this way as well, but with an added definition adding people that visit a single page, but exceed some predefined time limit on that page.

Problems with Bounce Rate

There are potential problems with Bounce Rate in that it is not consistently defined/recorded by analytics packages, it is not applicable to certain type of sites, and there is potential for the metric to be over-emphasized, particularly by those who are not aware of the shortcomings of their own analytics methods.

Bounce Rate in an E-commerce Site

As an example, if I were optimizing an e-commerce site, bounce rate would be very important to me on a number of levels. Presumably, I ensure that both my site’s homepage and sub-pages (perhaps individual products) would be indexed by search engines and other product search aggregators. I would also market my site through keywords purchases and banner advertising. I would want to closely examine individual page bounce rates to see which sites are most inviting to people to remain on the site and presumably stay on a path to a purchase. I would also closely look at the bounce rate of visits generated by each of my keywords; a high bounce rate indicates that I may want to look at spending my money on keywords that are more in-line with my site’s offerings.

Bounce Rate in a Content Site

If I were optimizing a site with a “walled garden”, a subscription-based content site, I would be much less concerned about a high bounce rate as it is defined now. Bounces to login pages would be expected, these could be attributed to visitors without access. Bounces from individual article pages could be attributed to subscription users that have individual articles bookmarked or users that email articles to another subscriber.

If I were optimizing an open content site, such as this blog, I would not have the tools at my disposal to make any conclusions based on Bounce Rate. Visitors to individual article pages come from search engines, RSS readers, social networking sites, and linkage from other blogs. A bounce from any of these sources may not be a bad thing – I want my visitors to read a single article if that is what they are seeking. The problem with our classic definition of a bounce is that there is no difference between a visitor that spends a minute to read an article and a visitor that clicks into the blog and leaves within five seconds.

A Solution for Google Analytics Users

One (messy) workaround for Google Analytics users is to simply call the urchinTracker function after a certain delay (perhaps 15 seconds). This will count another pageview for the visit, but no longer count that visit as a bounce.

The Future of Bounce

Unless we change our methodology for measuring bounces, bounce rate will cease to be a “sexy” (or even relevant) metric for most sites due to increasing popularity of social networking, RSS, and alternate browsers such as mobile browsers, televisions, and other types of syndicated content. For example, I regularly read “newsy” blogs such as TechCrunch and Engadget. However, my path to TechCrunch is my Google Homepage >> Techcrunch and my path to Engadget is Google Homepage >> Digg >> Engadget. My personal overall bounce rate to all three of these sites is probably 95%. Deep-linking and social news will continue to drive Bounce Rates up (and the overall relevance of Bounce Rate) unless we change the way we collect and interpret bounce data.

Blogs Cited in this Post

I highly recommend Avinash’s blog, Occam’s Razor and Anil’s blog, Web Analysis, Behavioral Targeting and Advertising. Check them out.

Some Background on Life Tables

The Life Table is a tool typically used to analyze and predict patterns of mortality among populations. The most familiar measure that gets generated by a Life Table is Life Expectancy. Life Expectancy is always great fodder for news headlines; at the time of this article’s publication, the Centers for Disease Control and Prevention just announced that Average Life Expectancy for people in the U.S. is nearly 78 years. This statistic was calculated using a Life Table.

A Life Table works by taking an imaginary population cohort (typically 100,000) and “aging” them through their lives by using currently estimated Age-Specific Death Rates. Why do this to calculate life expectancy rather than looking at vital statistics (for example, averaging the age of death of everyone that dies in a given year)? Because our population is influenced by generational patterns of growth and is not uniform. There are a lot more baby boomers and children of baby boomers than people born during the depression or during the “baby bust” (like me) that followed the baby boom. These age-specific population differences would skew Life Expectancy if calculated using the average age of death low now (since the boomers are entering their retirement years) and would skew it high in twenty years, when the likelihood of their mortality most increases.

Life Tables and Time Spent on Site

Using the Life Table technique to examine Time Spent on Site is fairly straightforward. Take a look at this table and/or follow the link to the Google Spreadsheet below it.

View/Download this Spreadsheet

Working left to right, fill in the first two columns with your period bounds. In my (completely made up) example, I use minute-by-minute data to “age” the site visitors unitl there are none. Pay special attention to the first row, which are reserved for bounces. Next, fill in the period-specific site exit rate in the fourth column for each period. Again, the first row is reserved for your bounce rate.

From here, simple calculations take you home. First, calculate the visitors left on site (starting with 100,000 in the first row and subtracting visitors that left during the previous period) and visitors exiting site (calculated by multiplying visitors remaining by the current period-specific exit rate) for each period, working your way down the table until there are zero visitors left. Next, calculate the total time on site for that period by multiplying the length of the given period by the visitors remaining on site, then subtracting half of the product of the length of the given period and the number of visitors exiting the site. Accounting for people exiting the site in this manner assumes that people in each period leave the site at a uniform rate and leads to a calculation error. The scale of this error is in proportion to the size of the periods being measured.

The final two columns are where the great value in this method come to light. The first of the two, total time on site, is calculated by starting at the bottom and creating a cumulative total for all minutes spent on site in the current period and all subsequent periods. The final column, average time spent on site for everyone in this period and beyond, is calculated by dividing the total time on site by the number of visitors remaining for the current period.

Interpreting the Results

Looking at the final column, representing the total average time spent on site for this period and beyond, will give great insight into your time spent on site metrics. As you can see from my example (and has a similar pattern in real life tables with infant mortality rates), bounces have a big effect on time spent on site. In my example, the average time spent on site is about .50 minutes (which will exactly match the metric reported from my analytics software). However, factoring out bounces, you can see that the average time spent on site for a non-bounce user is 1.24 minutes. Visitors that come to my site and stay for 3 minutes will, on average, stay an additional .60 minutes.

With the growth of web video, AJAX-based site structures, and other new technologies, Time Spent on Site is growing in importance. Perhaps this method of interpreting visitor behavior can help some people understand this metric a bit better. I plan to write soon about a related topic, bounces, and what we can do to improve the measurement of them.