Pooling results in Multiple Imputations

[Magdalena Rendic]

Hi everyone,

Im an inexperienced Stata user, and I was wondering if you could help me with something: I’m trying to replicate a simulation of the closing of the gender gap (in labor income per hour and working hours) using the MI method. It uses pmm, with a model with variables like age, years of education, among others, and the imputed variable is the labor income per hour (or the working hours) but with missing values for women. It makes 20 imputations. I understand that the next stages in MI are the completed-data analysis, performing the desired analysis separately on each imputation, and then the pooling step, combining the obtained results into a single multiple-imputation result. What is not clear for me is how I do the pooling step. Since what I want to do is a simulation of women incomes (based on men incomes), and then recalculate the household incomes to see the effect in poverty and inequality, I would like to have a single multiple-imputation result, but I guess it’s not as simple as calculating the mean of the 20 imputations.

Could anyone guide me a little? Thank you very much.

[Marcos Almeida]

Hello Magdalena,

Welcome to the Stata Forum.

I'm not sure whether this is the information you want from the Stata Manual:

mi estimate using is convenient when refitting models using mi estimate would be tedious or time consuming. In such cases, you can perform estimation once and save the uncombined, individual results by specifying mi estimate’s saving(miestfile) option. After that, you can repeatedly use mi estimate using miestfile to estimate linear and nonlinear transformations of coefficients or to obtain MI estimates using a subset of saved imputations.

mi estimate using performs the pooling step of the MI procedure;

Hopefully that helps.