Discussion


Note: Images may be inserted into your messages by uploading a file attachment (see "Manage Attachments"). Even though it doesn't appear when previewed, the image will appear at the end of your message once it is posted.
Register Latest Topics
 
 
 


Reply
  Author   Comment  
destraynor

Registered:
Posts: 5
Reply with quote  #1 
Hey guys, 

I'm writing to ask you, has anyone seen any great ways to visual a cohort study. 

A cohort study is when you study variables across different groups, where each group has one thing in common. 

A typical example would be customer retention. You'd group customers by the month they sign up, and then measure the retention rate in following months. This data might show things like the following

* Our retention rate is increasing (e.g 90% of last months visitors have returned this month)
* Our customers have a lifespan (e.g. Only 5% of the customers during any given month are still with us 6 months later). 

Typical visualisations for this include line charts http://www.avc.com/a_vc/2009/10/the-cohort-analysis.html and matrices. http://skitch.com/bokardo/dfhs6/metrics-driven-design-notes.key

I'm designing one of these, and I've gone through my literature (Tufte, Few, Suda, etc) and I've yet to see anything worthwhile or different. 

I'm turning to you guys in the hope you'll have ideas. 

Regards, 
Des
sfew

Moderator
Registered:
Posts: 814
Reply with quote  #2 
Des,

Your question isn't easy to answer, because cohort studies can't all be displayed in the same way. The best way to display the data depends on the story that you're trying to tell with the data. Given the fact that all cohort studies produce time-series data, line graphs are often appropriate, but only when you're trying to tell a story that involves change over time. Cohort studies can produce important stories about the final result (i.e., not how things change through time), etc. Even when you want to display change through time, the nature of the data or the things that you want to emphasize might require something other than a line graph.

What I'm saying is that, it's hard to answer your question in general, but it might be easy to suggest an effective way to display a specific set of cohort study results for a particular purpose for a particular audience. If you describe a specific case, I'll gladly suggest a way to display it.

__________________
Stephen Few
destraynor

Registered:
Posts: 5
Reply with quote  #3 
Thanks for your reply Stephen, I understand your point (and the error of my question). 

Could you suggest a good way to visualise something like this example: http://img.skitch.com/20100530-nrgtisuj4f7s77336ade1yaiwn.jpg

The question the visual needs to answer is "Is our retention getting better over time", i.e. "Are we losing less of our new customers this month than we were 6 months ago"

Thus far, I've looked at adding a sparkline showing the rate of change in each column. It's a start, but it still feels like we're leaving a lot of eyeballing work to the user. 

I've looked at changing type of  data, so that we instead show only the percentage of previous month kept. (You can see that data for the above example here: https://skitch.com/destraynor/fbexa/microsoft-excel )

I've looked at showing the percentage dropped for month to month 

Any pointers would be really appreciated. 
destraynor

Registered:
Posts: 5
Reply with quote  #4 
Just as a follow up, what I think there are multiple questions here, regarding retention. There is the drop after one month, 3 months, 6 months, etc. 

I've considered splitting the data, so that each type of loyalty can be looked at separately. E.g. if our focus is one month retention, then let's analyse that by itself. 
Here is a very quick mockup (using different data to the above) https://skitch.com/destraynor/fbeig/untitled

I'd love to hear anyones thoughts. 
idev

Registered:
Posts: 1
Reply with quote  #5 
Hi destraynor

When I looked at your data example I immediately thought of a Heat Map - I'm not sure if this is the best way but I think it would show a high level trend that you could then drill down to the data.

See attached image of the sample data (well hopefully I've attached it correctly:)

Attached Images
Name: cohort_example.png, Views: 1374, Size: 8.64 KB


bpierce

Moderator
Registered:
Posts: 98
Reply with quote  #6 
Hi Des,

If what you really want to highlight is how your one-month, three-month, and six-month retention rates have changed over time, I think your most recent mock up, which is similar to a cycle plot, is a good way to show that. The one thing I would suggest is that, rather than letting people choose their interval with a drop down box, allow them to choose multiple intervals with checkboxes and stack the graphs one over the other, with their dates aligned. This way people can also look for similar patterns between different intervals.

-Bryan
destraynor

Registered:
Posts: 5
Reply with quote  #7 
Hey guys, 
Sorry for replying so late - for some reason I stopped receiving email notifications. 

Anyways, big thanks to both idev & bpierce for their input too. 
You can see a snippet of where it ended up 

here: https://skitch.com/destraynor/f8eqw/screen-shot-2011-06-29-at-16.04.51

Looking into Cycle plots now. 

Thanks again, 
Des
destraynor

Registered:
Posts: 5
Reply with quote  #8 
Hey guys, 
Just to complete the circle here, I wrote a blog post about my learnings
Regards,
Des
Previous Topic | Next Topic
Print
Reply

Quick Navigation: