Announcement

Collapse
No announcement yet.
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • Simanova command output

    Hi,

    A first attempt to ask a question here in the forum, which I have otherwise used a lot to find solutions to coding issues.

    My question is about the simanova command (https://stats.idre.ucla.edu/stata/ad...anova-hlp-htm/) that is used as an alternative to standrad ANOVA when there are different sized groups and variables with unequal variances.

    I am having some difficulties with the output, as I get two rows with p-values from each analysis. Here is an extract from my code (this is, as you can see from the `x', part of a loop-code, where I analyse several variables in a number of different population groups):

    ***
    simanova `x' row_individual, seed(74593) reps(5000) nomp(0.05) level(95) fstar wtest

    Results of Standard ANOVA

    ----------------------------------------------------------------------
    Dependent Variable is cost_drug and Independent Variable is row_individual
    F( 7, 808.00) = 2.836, p= 0.0063
    ----------------------------------------------------------------------

    5000 simulated ANOVA F tests
    --------------------------------
    Nominal Simulated Simulated P value
    P Value P Value [95% Conf. Interval]
    -----------------------------------------
    0.0063 0.0090 0.0066 - 0.0120
    0.0500 0.0614 0.0549 - 0.0684

    ***

    The first row (yellow) appears to be somehow derived from the standard ANOVA-results, while the second row (green) is my selected “nomp-value” (which should be the nominal p-value, if I understand the help page correctly). Why do I get both and how can I interpret the different rows? Which is the correct one to report/use if I just want to know the p-value for this comparison of [drug] costs between groups? I have not found any information on this either here in the Statalist or in searching e.g., google. I hope someone here in the forum can guide me?

    Kind regards,
    Hanna Gyllensten, postdoc at the University of Gothenburg in Sweden

  • #2
    Hanna:
    welcome to the list.
    The correct p-value to report is research report/paper is the actual p-value.
    See -rnethelp "http://stats.idre.ucla.edu/stat/stata/ado/analysis/simanova.hlp"- for further details.
    As an unrequested advice, I would have used -regress- with -bootstrap- standard errors.
    Kind regards,
    Carlo
    (Stata 19.0)

    Comment


    • #3
      Thanks, I hope I interpret this correctly now that I should use the row:
      Originally posted by Hanna Gyllensten View Post
      0.0063 0.0090 0.0066 - 0.0120
      So that I should report 0.0090.

      Maybe I can even remove the nomp-part of the code?

      Does regress work well in these types of analyses also with a categorical x-variable? I have three groups of patients that I compare, instead of two (where I would use t-test to compare costs), thats why I used ANOVA, and during a review-phase for a manuscript I was asked why I used the unequal variances-option for the two-group comparisons but not for three group-comparisons. That's when I came up with this solution, and it was approved by that reviewer at least. But I would be very happy to find a better solution!

      Comment


      • #4
        May be useful to say also that my patient categories are not in any way ordered, they are just different groups based on some characteristic, such as where in a country someone is living or if they live in a large or small city et cetera.

        Comment


        • #5
          Hanna:
          yes, -regress- works fine with categorical predictors, too (see -fvvarlist- foer an efficient notation).
          -glm- is an option too.
          Kind regards,
          Carlo
          (Stata 19.0)

          Comment


          • #6
            Thanks for your help!

            Comment

            Working...
            X