I am in the data preparation stages of conducting a multiple regression analysis of US health survey data. The first task I have decided to do is impute missing values from the dataset of 8 variables (1 dependent 7 independent).

I have exactly 6000 entries for each variable in the dataset and am trying to impute for the missing values in variable x (continuous numerical variable) and y (categorical factor variable).

x = 828 missing values
y = 239 missing values

I can't seem to understand what imputation method I should use. I understand that MI is better for when there is greater than 10% missingness but only one of the two variables above is missing by greater than 10% of its total entries...

Is it that I am supposed to use a mixture of both? (MI for x and HD for y)

Any help would be much appreciated.

The five remaining variables consist of:

One Dependent variable

One Continuous numerical Independent Variable

Three Categorical Factor Independent Variables

  • 2
    $\begingroup$ This may help stats.stackexchange.com/questions/48668/… $\endgroup$
    – user83346
    Dec 27, 2017 at 4:49
  • 1
    $\begingroup$ Some info about hot-deck stats.stackexchange.com/q/307339/3277 $\endgroup$
    – ttnphns
    Dec 27, 2017 at 7:58
  • 1
    $\begingroup$ The answer depends on what you want to know. If you just want to do exploratory analysis, you can get away with almost anything. But if you want to do inference of any kind (p-values, confidence intervals) then multiple imputation is superior in general. $\endgroup$ Dec 27, 2017 at 13:39
  • $\begingroup$ Yes, I'd like to make inferences about the explanatory power of the predictor variables in the regression model. Essentially, I'm struggling to realise why I would pick one over the other... $\endgroup$ Dec 27, 2017 at 14:50


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.