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

  • Multiple imputation, descriptives and outliers

    Hello, I have a few questions regarding MI and various things.I have 3 questions


    Due to multiple imputation I cannot use tab, which makes 2 by 2 tables impossible. I was wondering, does this work though?

    forvalues j=0/5{
    .   tab *_class if implicate==`j', cell nofreq
    .   }
               |                 gift_class
     age_class |       <1K       <10K      <100K      100K+ |     Total
         16-25 |      0.32       0.97       2.10       1.94 |      5.32 
         26-35 |      0.32       1.94       5.81       1.77 |      9.84 
         36-45 |      0.81       1.94      11.94       7.58 |     22.26 
         46-55 |      2.26       3.23      17.90      10.32 |     33.71 
         56-65 |      1.94       1.29      14.84      10.81 |     28.87 
         Total |      5.65       9.35      52.58      32.42 |    100.00
    And I get 6 outputs, for each implicate and the original observation. Is this an acceptable use of tab with MI?


    I have a single outlier in my key variable that I wanted to look at. I tried to use lvr2plot, but it ididn't recognize the last regression, which was

    mi estimate, esampvaryok: reg job_hours gift_received
    So I am wondering, can I use lvr2plot with MI? If not, what could I use that is helpful and allowed?

    Would cook's distnace be usable with MI?

     predict cook, cooksd, if e(sample)
    . * predict e if e(sample), resid
    . * list gift_total e cook if cook>4/74
    Would this work?


    As far as this forum and online tools have told me, most use mi sum as descriptive statistics. What if I have a single variable that has sub variables within itself? For example labour status where the variable goes from 1-9? What could I use? Would it be possible to use this command?

    mi estimate, esampvaryok: total labour_status, over(labour_status)
    Where the variable is laid over itself to give me the total of the sub populations?

  • Oscar Weinzettl
    Going over, the first question seems to be off the table... Think the output is still wrong even if it is only for each implicate. AT least it doesn't match with my graphical data.

    But does anyone know of a proven statistical way to check for outlier data when using multiple imputation?

    Leave a comment:

  • Oscar Weinzettl

    Leave a comment: