ASA Connect

 View Only

  • 1.  Determining zero-inflation

    Posted 09-06-2017 08:22

    Dear Colleagues,

     

    Is there a specific percentage (of zeroes among values of a variable) that warrants classification of a variable as zero-inflated?

    Can anyone give a reference that defines what is meant by zero-inflation in a (predictor or outcome) variable?

     

    Regards

     

    Novie

     

    ************************************************

    Novie Younger-Coleman PhD

    Statistician/Senior Lecturer

    Caribbean Institute for Health Research

    (formerly Tropical Medicine Research Institute)

    University of the West Indies

    Kingston 7

    Jamaica

    Tel: 876 977 0746 or 927 2471 ext 227

    Fax: 876 927 2984

    e-mail: novie.younger@uwimona.edu.jm

    ********************************************************************************************

    The Jamaica Statistics Symposium and Preconference Workshop Series (JASSYM) 2017

    October 24-27, 2017  in Kingston, Jamaica

    THEME: Statistics for Success: Ethics, Data Security and Quality

    For more information send e-mail to jssbiennial.rc@gmail.com

    (You can't afford to miss It!!)

    *********************************************************************************************

     



  • 2.  RE: Determining zero-inflation

    Posted 09-07-2017 05:08
    Hi Novie,

    Zero-inflation is typically determined by the shape of the distribution not a percentage of O's per se. I believe you can test for zero-inflation using a constant only model and observing residuals and fit statistics. I have only used zero-inflated Poisson and Negative Binomial models. However, zero-inflation can also produce log-normally distributed data. I found this article some time ago, it makes specific reference to zero-inflated Poisson vs simple Poisson. I think it will help.

    A Score Test for Zero Inflation in a Poisson Distribution

    Jan van den Broek
    Biometrics
    Vol. 51, No. 2 (Jun., 1995), pp. 738-743
    DOI: 10.2307/2532959


    Blessings,
    Ayesha Johnson, PhD



    Original Message


Related Content