I recently looked at Pages/Visit and asked what would make a good engagement goal. The answer is not at all obvious since the data seemed to show a long tail distribution (aka power law distribution), where it’s really hard to boil down the data into one neat average. There are just too many outliers and the various averages, such as mode, mean and median are all different.
A characteristic of a power law is that if you plot the data on a logarithmic scale you get a linear relationship. Did you know that you can plot data on a logarithmic scale directly in Google Analytics using motion charts?
If you want to follow along:
- Grab the custom report
- Filter out 0 depths (these are flukes)
- Click Visualize
- Change the scale on both axes to Log.
Note that it’s probably not necessary to animate this data over time, but it does look pretty, doesn’t it? Just like Long Tail keywords, it does confirm though that Pages/Visit follows a power law.
Which leads me to my recommendation for using Pages/Visit. If you have a power law distribution, averages are essentially meaningless. The only approach in my opinion is to try segmenting the data.
Do visitors with few pageviews behave differently than those with many pageviews? Segment by low, medium and large page depths and see if there are differences in terms of source, behaviors or outcomes. Of course we already know that we should be segmenting our data, but when you see a power law you now know that you must segment.
Not saying that you will find anything meaningful even after segmenting, but it’s worth a try. I think this also what Mike Sullivan is saying where he looked at Page Depth against Time on Site for a new number of different types of sites.