SPSS Statistics

SPSS Statistics

Your hub for statistical analysis, data management, and data documentation. Connect, learn, and share with your peers! 

 View Only

  • 1.  Non-parametric tests on weighted data

    Posted Tue April 25, 2023 05:05 PM

    Good Evening,

    I am conducting non-parametric tests (kruskal-wallis and mann-whitney) on weighted data. The weighting variable is to two decimal places. However when the output is generated I am getting the following error messages, and the 'n' number is greater than the 'n' number of the weighted dataset:

    Image previewCan anyone please advise on this error message and how it should be dealt with in my analysis?

    Thank you.



    ------------------------------
    Gillian Campbell
    ------------------------------


  • 2.  RE: Non-parametric tests on weighted data

    Posted Wed April 26, 2023 07:20 PM

    It depends on which procedure you are running, but for nonparametric tests, the weights are generally rounded to the nearest integer.  I can't see any error message in your post, though.



    ------------------------------
    Jon Peck
    ------------------------------

    Original Message


  • 3.  RE: Non-parametric tests on weighted data

    Posted Thu April 27, 2023 03:21 AM

    Hi John,

    Hi Jon,

    The error message states:
    'Frequency weight data values must be positive integers'
    'Positive non-integer frequency weight data values are encountered. These values are rounded to the nearest integers for analysis.'

    The purpose of applying the weightings variable in my analysis is to make the sample representative of the total population. Therefore can you please advise what impact the rounding of the weighting values to integers will have on my analysis and any approaches that I should take? 

    Thank you. 



    ------------------------------
    Gillian Campbell
    ------------------------------

    Original Message


  • 4.  RE: Non-parametric tests on weighted data

    Posted Thu April 27, 2023 10:01 AM
    That sounds like just a warning.

    Without knowing the distribution of the weights, it's impossible to say what the impact of rounding is.  In the extreme, it's possible that the rounding could make all the weights equal to 1, 

    One way to see the impact would be to run the equivalent parametric tests with and without weighting to see how that impacts the test results.  Another thing to try would be to multiply the weights by 100, so rounding would be insignificant and then run the tests.  But you would need to adjust the significance for the false sample size.

    --



    Original Message