Skip to main content
Solved

Unable to understand N-day retention.

  • September 28, 2022
  • 3 replies
  • 374 views
J

The retention table given in this article:- 

https://help.amplitude.com/hc/en-us/articles/360050153551-Interpret-your-retention-analysis  looks like this..

 

 

 

This article says that day 1 retention = (day 1 users)/ total users =2,480/ 10,830 = 22.9%

day 2 retention = (day 2 users)/ total users = 2,339 / 10,830 = 21.6%

and so on… so everything is clear.

 

but in my case 

 

week 1 retention should have been = week 1 users/ total users =3708/ 10998= 33.71%

week 2 retention should have been = week 2 users/ total users = 2811/10998 = 25.5%

 

but in my retention table (see button of screenshot for table) the retention for week 1 is 38.8% and for week 2 this is 35.1% .. WHY THIS IS DIFFERENT.. Why the Denominator us changing in this case?

 

 

PLEASE HELP

 

 

 

Best answer by Yosimy.Cortes

Hi @jyoti , happy to help!

For weekly retention, we use a daily basis. This means all timestamps/times for events are floored to the start of the day.

For each user, they are Week X retained if and only if they perform an event…

  • Week X - between [7 * X, 7 * X + 7) days from any starting event (floored to the start of the day)

For roll-up (the overall number), it’s important to note that we try to keep only users that have had enough time to retain to avoid skew. This means a user is only included in the Week X point if week X (as defined above) is COMPLETED before the current time. For example, if it is currently Oct 4, and a user performed the starting event on Oct 3, we would not count this user in the Week 0 retention (even if they have already been retained) because Week 0 for this user [Oct 2, Oct 8) is not complete.
 
The reason that the percentages don't add up is that we only include users who have had enough time to retain. If you take a look at this week's percentages in your chart, you'll notice that they have an asterisk next to them. This means that those weeks have incomplete data. For example, some users may have only triggered the starting event today and so they will not be included in the denominator for week 0 as they have not finished their week 0.

If you want to see how the computation for retention is determined with visual examples of the breakdown table, you can check out this article:


Hope this clears things up!

View original
Amplitude Admin
J
2 people like this
  • Amplitude Admin
    Amplitude Admin
  • J
    jyoti
Did this topic help you find an answer to your question?

3 replies

Saish Redkar
Expert
Forum|alt.badge.img+10
  • Saish Redkar
  • Expert
  • 1382 replies
  • September 28, 2022

Hi @jyoti 

Since your time frame says ‘last 12 weeks’, it’s most likely that you will have some incomplete data points and uneven user counts within each week.
This FAQ doc and the posts below will help you understand this issue in a bit more detailed manner and different perspectives-

 

Hope this helps.

Saish | Connect with me on topmate.io/saish_redkar for supercharging your Amplitude skills
Amplitude Admin
1 person likes this
  • Amplitude Admin
    Amplitude Admin
J
  • jyoti
  • Author
  • New Member
  • 1 reply
  • September 29, 2022
Saish Redkar wrote:

Hi @jyoti 

Since your time frame says ‘last 12 weeks’, it’s most likely that you will have some incomplete data points and uneven user counts within each week.
This FAQ doc and the posts below will help you understand this issue in a bit more detailed manner and different perspectives-

 

Hope this helps.

Hi Saish

 

thanks for the answer.. but I think the total changes in every retention curve..what I mean if it is 30% retention in 7th day so this 30% is not of Day 0 users rather its 30% of SOME OTHER NUMBER..

 

you can see that in any of your retention curve also..

 

Regards

Amplitude Admin
1 person likes this
  • Amplitude Admin
    Amplitude Admin
Y
Team Member
Forum|alt.badge.img+5
  • Yosimy.Cortes
  • Team Member
  • 41 replies
  • Answer
  • October 4, 2022

Hi @jyoti , happy to help!

For weekly retention, we use a daily basis. This means all timestamps/times for events are floored to the start of the day.

For each user, they are Week X retained if and only if they perform an event…

  • Week X - between [7 * X, 7 * X + 7) days from any starting event (floored to the start of the day)

For roll-up (the overall number), it’s important to note that we try to keep only users that have had enough time to retain to avoid skew. This means a user is only included in the Week X point if week X (as defined above) is COMPLETED before the current time. For example, if it is currently Oct 4, and a user performed the starting event on Oct 3, we would not count this user in the Week 0 retention (even if they have already been retained) because Week 0 for this user [Oct 2, Oct 8) is not complete.
 
The reason that the percentages don't add up is that we only include users who have had enough time to retain. If you take a look at this week's percentages in your chart, you'll notice that they have an asterisk next to them. This means that those weeks have incomplete data. For example, some users may have only triggered the starting event today and so they will not be included in the denominator for week 0 as they have not finished their week 0.

If you want to see how the computation for retention is determined with visual examples of the breakdown table, you can check out this article:


Hope this clears things up!

Jeremie Gluckman
1 person likes this
  • Jeremie Gluckman
    Jeremie Gluckman

Reply


Helpful Members this Week

Terms & ConditionsCookie settings

Cookie policy

We use cookies to enhance and personalize your experience. If you accept you agree to our full cookie policy. Learn more about our cookies.

 
Cookie settings