Measure what is relevant: A critical look at bounce rates | Hacker News Reader

Measure what is relevant: A critical look at bounce rates

briancray.com

12 pointsbriancray8 comments

Bounce rates are also ridiculously sensitive to traffic source. You can do things which are screamingly net-beneficial for the business and impact bounce rates in a negative or counter intuitive manner.

If your company blog gets picked up on Reddit or CNN, your bounce rate just went through the roof.

If your ongoing SEO triples the number of visitors you get from Google every month and double revenues, you probably now rank for at least a few phrases which you are not a good fit for, causing bounces

If you make changes to your site's design which clearly suggest to people that this is not the Facebook login they are looking for, your bounce rate will increase as your blood pressure decreases.

All of the above suggests ditching bounce rates and concentrating on metrics which actually matter to the business. Conversions and revenues are two excellent choices -- most of the scenarios where one says "Our revenues are up by 100%... DANG IT ALL" are fairly contrived. The only reason bounce rate persists is because it is very easy to measure without getting buy-in from anyone else in the organization -- sort of like hits.

I think we've successfully embarrassed anyone who would quote hits as a meaningful metric out of the industry. Bounce rate should probably be next. (Full disclosure: at least one guy who knows what he is doing about metrics -- Avinash Kaushik -- disagrees with me violently on this one.)

I respectfully disagree on revenue - it's also largely a vanity metric. I know it doesn't feel that way (especially when there is little of it), but think about the seasonality problems and all the things that may compete for users attention like the Olympics. Revenue is a good indicator over several months, but it's largely useless for day-to-day decision making.

Far more useful metrics would be engagement among the people who have proven to be the core audience, likelyhood to recommend the product to others, time-to-first-satisfactory-outcome, feature use frequency. I am contemplating a diabolical plan of turning features off and measuring the relative frustration rates.

Well said! Thanks for the comments!

A time-based bounce rate measurement would largely minimize all the downsides listed in the article. (That plus some sort of event tracking for outside links.) It's unfortunate that's not the standard, default measurement for bounce rates.

when we made tmedweb (http://tmedweb.tulane.edu), one of our initial goals was a high bounce rate and minimum time on site. We were simply tying a calendar held elsewhere to course material held elsewhere. The basic design was our interest. You know you're doing it right when your biggest detractors set you as their homepage. :-)

By "time-based" do you mean how long they spent on the page?

Exactly. Some analytics programs go by whether they viewed a single page, some go by visits that viewed page(s) and only stuck around for less than 3 to 5 seconds. Most, if not all, analytics programs measure "time on site", so it's theoretically possible to segment out the groups that only viewed a single page and spent only a few seconds on the site, it's just inconvenient to set up if that's not the default.

In Kaushik's book, Web Analytics an Hour a Day, he defines bounce rate as "The percent of traffic that stayed on your website for fewer than 10 seconds." So that's probably why he's so keen on bounce rate, using his definition you avoid some of the issues listed in that article.

Sure - I've read that - but even if you record >10 second visits, you're still not finding out if users did what you wanted them to do based on bounce rate, unless your metric is time on site.

I don't think the 10 second rule minimizes all other points in the article.

Reply on news.ycombinator.com