1. Forum
  2. Usenet
  3. COMP.SOFT-SYS.STAT.SPSS

  • multinomial regression warnings

    From Heba Dabbour@21:1/5 to All on Sun Jan 12 10:21:37 2020
    I have a question about multinomial regression using spss. I have a dependent variable of 3 categories and other independent variable which are also categories and on scale (age). I used automatic coding using spss then I run the the multinomial
    regression and get this warning" Unexpected singularities in the Hessian matrix are encountered. This indicates that either some predictor variables should be excluded or some categories should be merged"
    and also this warning" There are 372 (62.9%) cells (i.e., dependent variable levels by subpopulations) with zero frequencies."

    I don't have empty cells.
    Can anyone help me please as how to proceed.

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)
  • From Rich Ulrich@21:1/5 to [email protected] on Sun Jan 12 18:23:42 2020
    On Sun, 12 Jan 2020 10:21:37 -0800 (PST), Heba Dabbour
    <[email protected]> wrote:

    I have a question about multinomial regression using spss. I have a dependent variable of 3 categories and other independent variable which are also categories and on scale (age). I used automatic coding using spss then I run the the multinomial
    regression and get this warning" Unexpected singularities in the Hessian matrix are encountered. This indicates that either some predictor variables should be excluded or some categories should be merged"
    and also this warning" There are 372 (62.9%) cells (i.e., dependent variable levels by subpopulations) with zero frequencies."

    I don't have empty cells.
    Can anyone help me please as how to proceed.

    If you think that you have no empty cells, /and/
    SPSS thinks that you have 372 empty cells, /then/
    you and the procedure have a /major/ disagreement
    about the design and the data.

    That is the error message you need to deal with.
    How many cells do you think you have, in all?

    590+ is a big, big number of predictors; I have
    doubts about the efficacy of such a design,
    generally speaking, even if your sample is the
    ten or twenty thousand or more it needs.

    If you used "automatic coding" that caused one
    continuous variable to be treated as categories,
    that could generate this surprise.

    - If the above doesn't let you solve your problem,
    please post your SPSS code with your further
    question.

    --
    Rich Ulrich

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)
  • From Rich Ulrich@21:1/5 to [email protected] on Mon Jan 13 17:49:07 2020
    On Sun, 12 Jan 2020 18:23:42 -0500, Rich Ulrich
    <[email protected]> wrote:



    If you used "automatic coding" that caused one
    continuous variable to be treated as categories,
    that could generate this surprise.

    Or, more precisely - If somebody's "automatic coding"
    sees a categorical variable with /whatever/ number of
    values between 1 and 90, it /might/ decide use the largest
    value and - thus - there will be 90 categories.

    I think it is "AUTORECODE" that will produce categories
    of 1-k in a new variable, with the original values (originally
    either numeric or string) as Value Labels.

    --
    Rich Ulrich

    --- SoupGate-Win32 v1.05
    * Origin: fsxNet Usenet Gateway (21:1/5)