ASA Connect

 View Only

  • 1.  Number of Ties: Non-parametric Statistics

    Posted 09-13-2021 12:08
    Dear ASA Connect Community,

    Many non-parametric statistical procedures for location parameters such as two samples Mann-Whitney test often referred the terms "no ties", "just a few ties", "many ties", etc. Are there any specific numbers to differentiate among "no ties", "just a few ties", "many ties"? Your ideas and expertise in this matter are greatly appreciated.



    ------------------------------
    Sincerely,

    Achut Adhikari
    Department of Statistics
    Miami University, Oxford OH
    ------------------------------


  • 2.  RE: Number of Ties: Non-parametric Statistics

    Posted 09-14-2021 06:56
    Specific numbers? Probably not. But in the Mann-Whitney case there are two distinct scenarios. In one scenario the data values are "conceptually continuous" with a different value for every observation ("no ties") or almost every observation ("just a few ties"). The other scenario is ordinal categorical data with K response categories, often represented by a 2-by-K crosstabulation. In this scenario there are "many ties." Mann-Whitney applies to both scenarios.

    ------------------------------
    S. Wright
    ------------------------------

    Original Message


  • 3.  RE: Number of Ties: Non-parametric Statistics

    Posted 09-14-2021 09:30
    Dear Achut,

    I answered this question partially for the one-sample case in a simulation study. Please see the published article linked below.

    https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0200837

    Good luck!
    Monnie McGee

    ------------------------------
    Monnie McGee
    Associate Professor
    Southern Methodist University
    Dallas,TX United States
    ------------------------------

    Original Message


  • 4.  RE: Number of Ties: Non-parametric Statistics

    Posted 09-14-2021 12:19
    Here's a quick approach.  Since ties are often the result of rounding data (or can be thought of as such), a quick approach is to "un-round" the data by adding random numbers uniformly distributed over the ranges that would have been the source for the rounded values.  These ranges may have different widths for different values, for example when we have left-censored data.  Do the computations you like with the unrounded data.  If you like, repeat a few (or many) times, and average the results.

    ------------------------------
    Charles Davis
    EnviroStat
    ------------------------------

    Original Message


Related Content