Flurry Blog

Flurry Product Updates

Current Articles | RSS Feed RSS Feed

Session Length Metric Changes

  
  
  

We spend a lot of time here at Flurry ensuring that the metrics we provide to you are as useful and meaninful as possible.

Since the Flurry Analytics service launched, we've offered a number of metrics around Session Length (the length of time a user uses the application) and one of them was Average Session Length. Viewing the Average Session Length over time can give you a good sense as to whether users are spending more or less time in the application as you add new features or make other changes. For all of Flurry Analytics we define the Average to be the arithmetic mean.

Unfortunately, we surveyed the distribution of Session Lengths and it turns out that for the vast majority of applications it is not a normal distribution. In statistics, a normal distribution is the typical bell-shaped distribution you expect to measure with a mean and a standard deviation. Distributions that are not normal don't necessarily follow this bell-shaped curve and hence the mean and standard deviation are not as meaningful. In fact, the mean will often be much larger than what you would otherwise consider a typical measurement.

To remedy this we removed the Average Session Length and replaced it with the Median Session Length. The Median Session Length is a session length value where half of the sessions were longer and half were shorter. The median is a much more flexible statistic and can more clearly represent the typical value of a non-normal distribution. You'll also notice that in most cases the Median Session Length is substantially lower than the Average Session Length.

We will continue to survey all of the analytics we offer for cases where we can improve and simplify. In the meantime if you have any questions on these or other metrics please let us know.

Comments

Does anyone know what a good median session length is for the average app? I'm sure it will differ for different categories i.e. games versus utilities, however I just need something to use as a benchmark.
Posted @ Monday, October 11, 2010 11:59 PM by Jess
Whilst I can understand why the median session length is a more representative value, I don't think you should have removed a metric that we, and others have been using for over a year. It's fine to add a metric, but removing one is a real problem for continuity of metrics reporting. We used it to calculate other values too, and now we can't report any of those, and have nothing to compare the median with. Please continue to report the average as well.
Posted @ Tuesday, October 12, 2010 12:43 PM by CArol
I agree. 
 
 
 
Posted @ Tuesday, October 12, 2010 6:32 PM by Jess
We definitely understand your concerns. The same average session length metric is still available via the Flurry AppMetrics API which you can read more about on our wiki: http://wiki.flurry.com/index.php?title=AppMetrics 
 
Our product development process is designed to always avoid removing any metric we offer for exactly the reasons you mention. Unfortunately in this case we could not find a way to display both the Mean and the Median on the same page without adding to the confusion. The enormous discrepancy immediately raises the question of which is accurate. Since we know the median to be more representative for these distributions we chose to go with that for display.  
 
Please do take a look at the API and if it's not sufficient for your needs let us know and we'll work on additional ways to get at those metrics. For example, would it help to add the mean into the CSV export and/or the detailed table below?
Posted @ Wednesday, October 13, 2010 7:23 PM by Sean Byrnes
Unfortunately, I don't have access to the API.  
 
 
 
Are there any statistics on median session length anywhere? I.e. is there an overall average median session length for iphone apps in general?  
 
 
 
Our app sits in the Utilities category, is there an overall average median session length for this category that we can use as a benchmark? 
 
 
 
If these stats aren't available can you please add the mean into the CSV export and detailed table?
Posted @ Wednesday, October 20, 2010 7:30 PM by Jess
I am still in the process of instrumenting my app and want to understand a little more about sessions. Is a session only defined by using the FlurryAgent.onStartSession and FlurryAgent.onEndSession calls or is there some other way of 'ending' a session.
Posted @ Thursday, October 28, 2010 4:29 PM by John
Would you please start handling background processes correctly as it relates to session length. 
 
If the user launches my app to listen to an audio stream, and then moves my app into the background but continues listening, my session length should reflect the time they spent listening in the background as well; not just the time in the foreground.
Posted @ Sunday, December 05, 2010 10:19 PM by Mike Dodson
I second Mike Dodson's request. Background processing should be measured as app usage. In light of earlier discussions regarding not changing metrics, perhaps adding two new metrics would be useful - "Background session Time" and "Total session Time" ( = the current session time + new Background session time) 
 
Also, consider a "Pause/Unpause Timer" command that we can put in the background entry/exit methods... i.e., if the app is not playing audio or is idle, Pause the timer when entering background mode.
Posted @ Thursday, December 23, 2010 11:18 AM by Gene
Comments have been closed for this article.