INFORMS Open Forum

  • 1.  Kaplan Meier Curve to answer business questions

    Posted 06-22-2023 15:43

    Does anyone have experience or pointers when using Kaplan Meier curve to answer business questions? Seems silly to use a survival analysis tool from epidemiology to answer a business research question but I think it might be useful in my case. I am looking at cancelled vs. completed (event), completed_record_date (time), and independent variable (1 or 0 as treatment or compare). I like Kaplan Meier because I can get a probability of survival table. I want to know what the probability or odds of a record being cancelled when using x independent variable. Thoughts?



    ------------------------------
    Alberto Aparicio
    Data Analyst
    Charitable Adult Rides & Services, Inc.
    CA
    ------------------------------


  • 2.  RE: Kaplan Meier Curve to answer business questions

    Posted 06-23-2023 09:21

    This reminds me of the day when I was a statistics undergraduate; to me, I don't see a reason why you shouldn't proceed with your current setup.

    I guess you will be interested in reading this paper which applied Cox model to the problem of customer churn (attrition).

    Do personalized economic incentives work in promoting shared mobility? Examining customer churn using a time-varying Cox model

    https://doi.org/10.1016/j.trc.2021.103224

    Let me just make a direct quote for our easy reference.

    Survival analysis was originally developed to estimate the lifetime of patients and is now widely used in various fields to estimate different durations (Kleinbaum and Klein, 2010). This approach can be applied to model the time of customer churn, where birth is defined as the time that a customer joins the program, and death is defined as the time that a customer terminates the program or becomes inactive.



    ------------------------------
    Vincent Tsz Fai Chow
    Department of Logistics and Maritime Studies
    The Hong Kong Polytechnic University, Hong Kong
    Email: tsz-fai.chow@polyu.edu.hk
    https://tfvchow.github.io
    ------------------------------

    Original Message


  • 3.  RE: Kaplan Meier Curve to answer business questions

    Posted 06-23-2023 12:07

    I have used survival analysis in R. See this nice writeup.

    https://www.emilyzabor.com/tutorials/survival_analysis_in_r_tutorial.html

    As the article points out, survival analysis is about 'censoring', how to handle cases that begin but do not finish a situation, eg someone who puts an item in his online shopping cart but then does not buy it. The 'death' event never happens. Or someone who puts the item in his cart and then removes it before buying rather than just never doing anything with it.  Lots of examples.  

    I encourage you to look into Kaplan-Meier curves. They are just basic statistics, nothing obscure or medical.  It happens they are of great use in Ed's medical applications, his specialty.



    ------------------------------
    Bruce Hartman
    Professor
    University of St. Francis
    Tucson, AZ United States
    bruce@ahartman.net
    website:http://drbrucehartman.net/brucewebsite/
    ------------------------------

    Original Message


  • 4.  RE: Kaplan Meier Curve to answer business questions

    Posted 06-23-2023 22:25

    The Kaplan-Meier product-limit estimator (KMPLE) is commonly

    used in medicine, epidemiology, and reliability, but it can

    be used in any application in which there is time-to-event

    data with random right censoring.  

    Here is one example.  Let's say you want to determine the

    probability distribution of the lifetime of an airline.

    Some airlines have failed (like Pan Am, which was founded

    in 1927 and ceased operations in 1991, constituting an observed

    lifetime of 64 years).  But other airlines are still

    operating (like Delta Airlines, which was founded in 1925

    and is still operating, constituting a right-censored time of

    98 years).  The KMPLE can combine these two types of

    observations to form a point estimate of the survivor function,

    which is one minus the cumulative distribution function.

    To extend the application a bit further, you might want

    to know whether discount airlines have shorter lifetimes

    than the larger carriers.  In this case you could plot

    the KMPLE of the lifetimes of the larger airlines and the

    KMPLE of the lifetimes of the discount airlines on the same

    set of axes, then use the log rank test to see if there is

    a statistically significant difference between the two

    survivor function estimates.

    So to answer your question more directly, yes, the KMPLE

    is appropriate for use in business applications.



    ------------------------------
    Lawrence Leemis
    Professor of Mathematics
    College of William and Mary
    Williamsburg VA
    ------------------------------

    Original Message


  • 5.  RE: Kaplan Meier Curve to answer business questions

    Posted 06-24-2023 10:22

    Good question on Kaplan and survival curves.  This is  a good application.  Kaplan is part of the general area of life data analysis - its original applications were in industry similar to your question.   The best book I am aware of is Applied Life Data Analysis by Wayne Nelson - part of GE's best in class statistics team lead by Gerry Hahn.



    ------------------------------
    Ken Fordyce
    director analytics without borders
    Arkieva
    Wilmington DE
    ------------------------------

    Original Message


  • 6.  RE: Kaplan Meier Curve to answer business questions

    Posted 07-25-2023 01:46

    Amazing responses! Really. I took a lot in and ran with it. Used completed records as censored. Cancellation as event.

    Look what I came across in the county library. 



    ------------------------------
    Alberto Aparicio
    Data Analyst
    Charitable Adult Rides & Services, Inc.
    San Diego
    ------------------------------

    Original Message


Related Content