BookmarkSubscribeRSS Feed
timo
Calcite | Level 5

Structuring Member Turnover Data for Non-Profit

We are conducting an analysis to predict member turnover or churn – which members will leave –  over a one year period.

Members may be already on board at the beginning of theyear, or they may join during the year. At the end of the year, T2, their Status is either Stay or Leave.  Length_of_Membership (measured in months)will be a strong predictor, as we know that new members have a higher leave rate. Status is teh Target variable.

The challenge for us is to structure the data in a way that does not overstate the impact of Length_of_Membership. Over a year, Stayers will have a Length_of_Membership through the end of the period (year).  Leavers will have a Length_of_Membership through their LeaveMonth. 

Join Month

 

T2_Status

 

Leave_month

 

Length_of_Membership

 

May

 

Stay

 

 
9

May

 

Leave

 

July

 

3

 

Jan

 

Stay

 

 
12

Jan

 

Leave

 

July

 

6

 

Feb

 

Stay

 

 
11

Feb

 

Leave

 

April

 

3

 

The problem is when we look the proportion of Leavers at the end of the period by Length_of_Membership we may be overstating the proportion who leave at a given Length_of_Membership  level – because, for example, Stayers with Length_of_Membership=3months will be just who join in October (3 months prior to T2), while Leavers will be all those who Left with Length_of_Membership =3 months at any time throughout the12-month period. 

Length_of_Membership is an important predictor, but our current approach to structuring the data overstates its impact.  We could cut the time interval down, but that reduces the incidence of Leavers and presents a new set of challenges (as our data set is in the range of n=10000, with a 10% Leave proportion, and we are already oversampling).  Using cohorts seems to make this overly complicated…

Any suggestions/perspectives much appreciated.

Timo

1 REPLY 1
Doc_Duke
Rhodochrosite | Level 12

This is not my field, but this google search had lots of interesting articles

modelling churn site:sas.com

Doc Muhlbaier

Duke

sas-innovate-2024.png

Join us for SAS Innovate April 16-19 at the Aria in Las Vegas. Bring the team and save big with our group pricing for a limited time only.

Pre-conference courses and tutorials are filling up fast and are always a sellout. Register today to reserve your seat.

 

Register now!

How to choose a machine learning algorithm

Use this tutorial as a handy guide to weigh the pros and cons of these commonly used machine learning algorithms.

Find more tutorials on the SAS Users YouTube channel.

Discussion stats
  • 1 reply
  • 693 views
  • 0 likes
  • 2 in conversation