Announcement

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

  • winsorize

    Dearall
    Thank you for everything
    Please, can someone tell me what is the correct command for the panel data to winsorize all variables in the model at the 1st and 99th percentiles? And if I have a dep.var dummy (0,1) please how can I separate the sum statistic into two group id with one and id without Thank you so much in advance.
    Best regards


  • #2
    Winsorizing either makes no difference for indicator variables or at worst maps 0 to 1 or 1 to 0, which surely is not a good idea..

    For other reasons Winsorizing every variable in sight is so extreme a strategy that I decline to advise how to do it.

    Outliers are best identified in multivariate context!
    Last edited by Nick Cox; 24 Sep 2020, 02:48.

    Comment


    • #3
      Dear Nick
      Thank you for your replay .
      I mean I would to winsorize all variables in the model at the 1st and 99th percentiles (not the dummy ) , I mean my expalntory varibels which not dummy (panel data) . Thank you

      Comment


      • #4
        OK, but my main comment stands as an honest view.

        Comment


        • #5
          Dear Nick ,
          please, I dont understand your comment very well, sorry

          Comment


          • #6
            Fair enough. But Winsorizing has often been discussed here. For one take please see my comments at https://www.statalist.org/forums/for...sorizing-level

            I don't recall any discussion in the history of Statalist that defended or justified Winsorizing in as much detail as I've discussed its limitations and better alternatives. As always, that comment is open to refutation.

            Comment


            • #7
              While agreeing completely with Nick's concerns, some disciplines do demand winsorized variables.

              Whenever you hit this kind of problem, the first place to go is to the command line of Stata where you enter findit winsorize . This will bring up a variety of related winsorize related items – a Stata procedure if Stata itself deals with the problem, Stata Journal or ado files if a user written procedure deals with the issue, etc. If this doesn't provide an answer, it is often worthwhile googling the problem - so you might enter a Google search looking for winsorize Stata .

              If you take the trouble to enter findit winsorize from the command line, you will find a procedure called windsor which does the winsorizing for you.

              Comment

              Working...
              X