Announcement

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

  • Multiple Imputation query

    I need some help on imputing missing observations for a variable. I am working with panel data and have 324 observations (18 years and 18 countries). My dependent variable in the regression is missing 49 observations, specifically for 4 countries there are around 12 or so years of data missing.

    I was thinking of getting Stata 14 to impute those. I use the following commands:

    mi set mlong

    mi register imputed var1

    mi impute regress var1 var2 var3, add(1)

    Stata then tells me that 49 incomplete observations were imputed. However, when I check my data, I find that Stata added 49 observations for all variables on top of the 324, and not replaced the missing values. Can I get some intuition on this? Is it possible to get Stata to just replace the missing variables with imputed variables rather than generating a whole new set?

  • #2
    It sounds like Stata did it right. The original data is left as is. Then Stata adds records for each case where values were imputed (apparently 49 records in your case). You only asked for one imputation, which is awfully skimpy. Usually people do at least 5 and more often 20 or more.

    See help mi_styles for more.

    Here are some good sources:

    https://www.ssc.wisc.edu/sscc/pubs/stata_mi_intro.htm

    https://stats.idre.ucla.edu/stata/se...stata_pt1_new/

    -------------------------------------------
    Richard Williams, Notre Dame Dept of Sociology
    Stata Version: 17.0 MP (2 processor)

    EMAIL: [email protected]
    WWW: https://www3.nd.edu/~rwilliam

    Comment


    • #3
      Thank you Richard for the links and the explanation re number of imputations. From what I understand, with the code that I have above, I have gone through 1st and 2nd steps of imputation, that is imputation and analysis(?). Would you know how I could get to the pooling stage? Is it feasible to modify the original dataset using the imputed values with Stata?

      Comment


      • #4
        Since you've only done one imputation I would not say you've completed the imputation and analysis stages -- at least you have not done them well. (In fact you have not done the analysis stage at all.)

        Stata is going to take care of the pooling for you with the mi estimate command.

        The sources I cited before would probably answer all your Qs if you go through them. My own discussion is at

        https://www3.nd.edu/~rwilliam/stats3/MD02.pdf
        -------------------------------------------
        Richard Williams, Notre Dame Dept of Sociology
        Stata Version: 17.0 MP (2 processor)

        EMAIL: [email protected]
        WWW: https://www3.nd.edu/~rwilliam

        Comment


        • #5
          Thanks Richard! I delved a little into the nature of imputation and I found your paper very useful. Also for anyone else coming across this post and trying to apply mi and without understanding the basics, I think this web page is very useful: https://www.iriseekhout.com .

          Comment

          Working...
          X