Missing Data/Imputation Discussion > How to create the plausible imputed data sets for MI?
Great question, Eason! Depending on which software you are using, different types of estimation are used to create the "plausible dataset". Some software packages use maximum likelihood (ML) and others use Expectation-Maximization (EM). There are other options as well, but those are a couple of the most common. My favorite software to use for MI is R, which is an open-source software and it's free. I'll post the link to download R below. However, R is not very user-friendly and has a steep learning curve. SPSS also has options to conduct multiple imputation, but as you probably know, SPSS is very expenses, though definitely more easy to use for people who are not as familiar with computer code or syntax.
Within R, there are also several packages that will conduct multiple imputation, but my favorite one to use is "mi" (and I like to use Zelig to run my analysis, because it provides pooled estimates for most types analysis).
I hope this is helpful!
Link for R: http://cran.r-project.org/
Link for "mi": http://cran.r-project.org/web/packages/mi
Link for lavaan: http://cran.r-project.org/web/packages/lavaan/index.html
Jeremy Taylor
All, I am new for MI and question was found - As state in many literatures, the general concept of MI is to create several plausible imputed data sets for the missing value, possibly m is equal 5 or not more than 5. However, how can we define that plausible imputed data sets? Any methods that we can refer or it just a random values? Please kindly advise and deeply thanks for your explanation in this regard.
Eason