I'm looking at identifying trends within my data, particularly wondering if usage of my app on new sign ups is increasing over a week to week basis. As we are constantly improving our product, I'd love to be able to identify a correlation between app usage & new feature releases.

The data I have is split into weekly cohorts. Each week I have the number of new sign ups, and how many of those accounts are still active (e.g. users are logging into the system).

Let's say I have the following data

Week 1 - 10 new trials, 3 accounts active

Week 2 - 15 new trials, 5 accounts active

week 3 - 4 new trials, 3 accounts active

week 4 - 20 new trials, 12 accounts active

week 5 - 17 new trials, 9 accounts active

In my current approach of analysis, week 3 looks amazing because 75% of accounts are still active ... In reality though, the number of new trials is extremely small compared to other weeks. As a result, I don't feel like I'm accurately comparing apples to apples in a week to week comparison.

Is there a way I can normalize the weekly data so that I am performing accurate analysis - or is percentage based really the best way to look at this data?

I am pretty new to this, so any help is much appreciated.


  • $\begingroup$ Welcome to DataScience.SE! Consider modeling your activity with a Beta random variable. $\endgroup$
    – Emre
    Jul 20, 2016 at 18:18
  • $\begingroup$ @Emre Thanks - can you point to any tutorials on how to use beta random variables for a dummy. I'm struggling to understand how i'd apply this to my problem. $\endgroup$
    – Gshock
    Jul 20, 2016 at 20:23
  • $\begingroup$ I hope this helps. You haven't mentioned how you are modeling the features though; an binary variable to indicate whether a feature was introduced that week? $\endgroup$
    – Emre
    Jul 21, 2016 at 4:41
  • $\begingroup$ I literally have no idea what I am doing, hence the ask for tutorials. I have no idea where to start. $\endgroup$
    – Gshock
    Jul 21, 2016 at 12:26
  • $\begingroup$ Maybe you need to step back and get a book on statistics? Larry Wasserman's All of Statistics isn't bad. $\endgroup$
    – Emre
    Jul 21, 2016 at 15:37

2 Answers 2


With this low number of sign ups I doubt you can profit from advanced statistics at this stage. Just chart your numbers in excel along the timeline and judge visually.

  • $\begingroup$ What if signups were in the 100's? The above data was dummy data. $\endgroup$
    – Gshock
    Jul 21, 2016 at 14:34
  • $\begingroup$ If you can have the sign ups broken down per day you could try using CausalImpact from Google or Probablistic Programming github.com/CamDavidsonPilon/… to see what the effect is. $\endgroup$
    – Diego
    Jul 21, 2016 at 23:07

Try Normalizing the trails and active accounts data with a mean or median. The resulting statistics will show week 3 as negative values which can be interpreted as a comparison to all the data. See the Excel graphs below.

mean normal mean median combo

As opposed to the data that has not been normalized. enter image description here


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.